78621 Trade in Value Added Developing New Measures of Cross-Border Trade a THE WORLD BANK edited by Aaditya Mattoo, Zhi Wang and Shang-Jin Wei TRADE IN VALUE ADDED Developing New Measures of Cross-Border Trade Trade in Value Added: Developing New Measures of Cross-Border Trade Copyright © 2013 by The International Bank for Reconstruction and Development/The World Bank 1818 H Street, NW, Washington, DC 20433, USA ISBN: 978-1-907142-58-1 All rights reserved The findings, interpretations, and conclusions expressed herein are those of the author(s) and do not necessarily reflect the views of the Executive Directors of the International Bank for Reconstruction and Development/The World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this publication is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The International Bank for Reconstruction and Development/The World Bank encourages dissemination of its work and will normally grant permission to reproduce portions of the work promptly. For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance Center Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; telephone: 978-750-8400; fax: 978-750-4470; Internet: www.copyright.com. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, The World Bank, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2422; e-mail: pubrights@worldbank.org. Copyedited and typeset by T&T Productions Ltd, London Published in association with the London Publishing Partnership www.londonpublishingpartnership.co.uk Centre for Economic Policy Research The Centre for Economic Policy Research is a network of over 800 Research Fellows and Affiliates, based primarily in European universities. The Centre coordinates the research activities of its Fellows and Affiliates and communi- cates the results to the public and private sectors. CEPR is an entrepreneur, developing research initiatives with the producers, consumers and sponsors of research. Established in 1983, CEPR is a European economics research orga- nization with uniquely wide-ranging scope and activities. The Centre is pluralist and non-partisan, bringing economic research to bear on the analysis of medium- and long-run policy questions. CEPR research may include views on policy, but the Executive Committee of the Centre does not give prior review to its publications, and the Centre takes no institutional pol- icy positions. The opinions expressed in this report are those of the authors and not those of the Centre for Economic Policy Research. CEPR is a registered charity (No. 287287) and a company limited by guarantee and registered in England (No. 1727026). Chair of the Board Guillermo de la Dehesa President Richard Portes Chief Executive Officer Stephen Yeo Research Director Lucrezia Reichlin Policy Director Richard Baldwin The World Bank The World Bank Group is a major source of financial and technical assis- tance to developing countries around the world, providing low-interest loans, interest-free credits and grants for investments and projects in areas such as education, health, public administration, infrastructure, trade, financial and private sector development, agriculture, and environmental and natural resource management. Established in 1944 and headquartered in Washing- ton, DC, the Group has over 100 offices worldwide. The World Bank’s mission is to fight poverty with passion and professionalism for lasting results and to help people help themselves and their environment by providing resources, sharing knowledge, building capacity and forging partnerships in the public and private sectors. Trade in Value Added Developing New Measures of Cross-Border Trade edited by AADITYA MATTOO, ZHI WANG AND SHANG-JIN WEI Contents List of Figures x List of Tables xiv Acknowledgements xxiii 1. Measuring Trade in Value Added when Production is Fragmented across Countries: An Overview 1 Aaditya Mattoo, Zhi Wang and Shang-Jin Wei 2. Towards the Measurement of Trade in Value-Added Terms: Policy Rationale and Methodological Challenges 17 Sébastien Miroudot and Norihiko Yamano 3. The Importance of Measuring Trade in Value Added 41 A: Imperatives from International Trade Theory Gene M. Grossman B: Why Measuring Value-Added Trade Matters for Developing Countries Judith M. Dean C: Implications for Macroeconomic Policy Mika Saito and Ranil Salgado 4. Accounting for Intermediates: Production Sharing and Trade in Value Added 69 Robert C. Johnson and Guillermo Noguera 5. Estimating Domestic Content in Exports when Processing Trade Is Pervasive 105 Robert Koopman, Zhi Wang and Shang-Jin Wei 6. Foreign and Domestic Content in Mexico’s Manufacturing Exports 135 Justino De La Cruz, Robert B. Koopman, Zhi Wang and Shang-Jin Wei viii Trade in Value Added 7. Gravity Chains: Estimating Bilateral Trade Flows when Trade in Components and Parts Is Important 161 Richard E. Baldwin and Daria Taglioni 8. Using Trade Microdata to Improve Trade in Value-Added Measures: Proof of Concept Using Turkish Data 187 Nadim Ahmad, Sónia Araújo, Alessia Lo Turco and Daniela Maggioni 9. Developing International Input–Output Databases: IDE-JETRO and OECD Experiences 221 Satoshi Inomata, Norihiko Yamano and Bo Meng 10. A Three-Stage Reconciliation Method to Construct a Time Series International Input–Output Database 253 Nadim Ahmad, Zhi Wang and Norihiko Yamano 11. Direct Measurement of Global Value Chains: Collecting Product- and Firm-Level Statistics on Value Added and Business Function Outsourcing and Offshoring 289 Timothy J. Sturgeon, Peter Bøegh Nielsen, Greg Linden, Gary Gereffi and Clair Brown 12. Integrating Value-Added Trade Statistics into the System of National Accounts 321 A: Perspectives from the World Trade Organization Andreas Maurer B: Perspectives from the United Nations Ronald Jansen C: Perspectives from the Organisation of Economic Cooperation and Development Nadim Ahmad D: Perspectives from the US Bureau of Economic Analysis Robert E. Yuskavage List of Figures 2.1 The difference between US exports of intermediate inputs Fix ‘Continued’ to China and US imports of assembled iPhones. 24 figure and table captions before 2.2 Export share by industry and category: China, 1995 and 2009. 32 press stage. 2.3 Export share by industry and category: USA, 1995 and 2009. 32 3.1 Foreign content share (%) of Chinese exports, 2002. 50 3.2 Export similarity and vertical specialisation, 1997 and 2002. 53 3.3 World exports relative to production (percent of GDP). 61 3.4 Foreign contents in gross exports. 63 3.5 Foreign contents in gross exports: high-tech sectors. 64 3.6 Source of change in exports of advanced countries (1995–2005). 64 3.7 Simulated impact of exports by sector. 66 3.8 Contribution to adjustment in trade balance. 66 4.1 Composite sector shares of gross exports and value-added exports, by country (2004): (a) manufactures; (b) services. 89 4.2 Between–within decomposition of aggregate VAX ratios, by country (2004). 90 4.3 Value added to gross trade ratios for the USA and Germany, by partner (2004). 91 4.4 Bilateral trade and value-added balances for the USA, by partner (2004). 92 5.1 Input–output table with separate production account for processing trade. 111 6.1 US–Mexico goods trade. 136 6.2 US and Mexico manufacturing production, 2000–10. 137 7.1 GDP coefficients for Factory Asia countries, 1967–2008. 173 7.2 Coefficient for the size variables measured as 1−σ ln((Yot /Ωot )(Edt /Pdt )). 178 8.1 Q–Q plot of intermediate import ratio against export share. 207 8.2 Distribution of value added per unit of output by firm size. 215 8.3 Distribution of value added per unit of output by firm ownership. 216 x Trade in Value Added 8.4 Q–Q plot of intermediate import ratio against export share by firm ownership. 217 9.1 Similarity to the Japanese IO table. 227 9.2 Sample format of questionnaire. 233 9.3 Layout of the AIO table. 234 9.4 Linking of national IO tables (two-country case). 235 9.5 Adjustment procedure. 236 9.6 Distribution of CT error. 236 9.7 Format of an OECD inter-country input–output model. 241 9.8 Estimation procedures for harmonised format IO. 243 9.9 Export share by industry and category (world, 2009). 249 9.10 Export share by industry and category. 250 9.10 Continued. 251 10.1 Comparing data sources for goods and services: world imports plus exports (various sources as a percentage of National Accounts data). 273 10.2 Reporter reliability and mean absolute percentage adjustment of total exports, 1995–2009. 281 10.3 Reporter reliability and mean absolute percentage adjustment of world goods by product, 1995–2009. 281 11.1 Geography of value added in a Hewlett-Packard notebook computer. 294 11.2 Basic data needed for product-level GVC studies. 296 11.3 R&D and engineering functions sourced internationally by enterprises in selected European countries, 2001–6. 304 11.4 Employment trends by type of function sourced internationally, Denmark, 2000–7. 307 11.5 Data collection grid for outsourcing and offshoring by business function. 310 11.6 Location of business functions as a percentage of costs of goods or services sold (all cases, n = 306). 312 11.7 Location of outsourced/offshored business functions as a percentage of costs of goods or services sold: F1K cases, n = 86. 312 11.8 Location of outsourced/offshored business functions as a percentage of costs of goods or services sold: private sector non-F1K cases, n = 104. 313 11.9 Percentage of international costs by type of location (operating costs in relation to the USA) and business function, 2010, organisations engaged in international sourcing (n = 58). 314 List of Tables 2.1 Countries that provide intermediate inputs into the iPhone 4. 19 2.2 US trade balance in iPhones. 24 2.3 Country coverage of OECD Input–Output 2009 edition (as of May 2011). 29 2.3 Continued. 30 2.4 OECD IO industry classification. 31 2.5 Current BEC and SNA classes of goods. 34 3.1 Export shares, processing trade and pollution intensity by Chinese industrial sector, 2006. 55 3.2 US trade balance (percent of GDP). 60 3.3 China’s external balance, 2008 (percent of GDP). 60 3.4 Share of foreign value added in gross exports. 62 3.5 Hub’s VA contained in gross exports. 63 3.6 Simulated long-run impacts of relative price shocks on external balances: base year = 2008 (percent of national GDP, unless otherwise noted). 65 4.1 VAX ratios by country and sector. 100 4.1 Continued. 101 4.1 Continued. 102 4.2 Aggregate and manufacturing VAX decompositions. 103 4.3 Bilateral VAX ratio: bilateral HIY versus production sharing adjustment. 103 4.4 Decomposing trade: absorption, reflection and redirection. 104 5.1 Major trade share parameters used in estimation, 1997–2008. 118 5.2 Shares of domestic and foreign value added in total exports (%). 120 5.3 Domestic and foreign value added: processing versus normal exports (as percentage of total exports). 121 5.4 Shares of domestic value added in exports by firm ownership (%), 2002 and 2007. 122 5.5 Domestic value added share in manufacturing exports by sector, 2002. 124 5.5 Continued. 125 5.5 Continued. 126 xii Trade in Value Added 5.6 Domestic value-added share in manufacturing exports by sector, 2007. 127 5.6 Continued. 128 5.6 Continued. 129 5.7 Total domestic value-added share in Chinese gross merchandise exports to its major trading partners (%), 2002 and 2007. 130 6.1 Mexico’s processing manufacturing exports, 1996–2006. 142 6.2 Mexico’s total imports for processing exports, by leading markets, 2000–6. 142 6.3 Mexico’s total processing exports, by leading markets, 2000–6. 143 6.4 Domestic and foreign value added in Mexico’s manufacturing exports: three-digit NAICS versus four-digit NAICS (in percent of total manufacturing exports). 144 6.5 Domestic value-added share in Mexico’s manufacturing exports by three-digit NAICS, 2003 (sorted by total foreign value added (weighted sum 2) in descending order). 147 6.6 Domestic value-added share in Mexico’s manufacturing exports by four-digit NAICS, 2003 (sorted by total foreign value added (weighted sum 2) in descending order). 148 6.6 Continued. 149 6.6 Continued. 150 6.6 Continued. 151 6.7 Domestic and foreign content in Mexico’s gross exports, 2003, computed directly from the Mexico IO table with a separate maquiladora economy account. 154 6.8 Domestic and foreign content in Mexico’s gross exports, 2003, estimated from aggregated Mexico IO table by our mathematical programming model. 155 7.1 Bilateral flows of total, intermediate and final goods, 187 nations, 2000–7. 170 7.2 Classification for intermediate and final goods. 171 7.3 Bilateral flows of total goods among Factory Asia nations (1967–2008). 172 7.4 Estimates for EU15, and USA, Canada, Australia and New Zealand, 1967–2008. 174 7.5 Interactions with share of intermediates in total imports, full sample. 176 7.6 All countries, 2000–7, by share of intermediate imports. 177 7.7 New mass proxies with share of intermediate, all nations, 2000–7. 180 List of Tables xiii 7.8 New mass proxies with intermediate deciles, all nations, 2000–7. 181 8.1 Use of imported intermediates and output breakdown by firm type in China. 197 8.2 Merchandise trade by large economic sectors (as a percentage of total trade in 2009 or latest available year). 200 8.3 Comparison of results, 2002. 202 8.3 Continued. 203 8.4 Country coverage of OECD Input–Output 2009 edition (as of May 2011). 205 8.4 Continued. 206 8.5 Correlation table between selected indicators. 207 8.6 OECD IO industry classification. NACE Classification – Rev. 1.1. 208 8.6 Continued. 209 8.7 Distribution of export shares (%). 210 8.8 Distribution of intermediate import ratios (%). 211 8.8 Continued. 212 8.9 Summary of results. 214 8.10 Differences in intermediate import ratios between exporters and non-exporters. 218 9.1 Different features and characteristics in national IO tables across the AIO target economies. 225 9.2 Similarity in the presentation format. 227 9.3 Similarity in the industrial classification number. 228 9.4 Responsiveness to the 1993 SNA. 228 9.5 Data sources for OECD inter-country IO model. 242 9.6 Country coverage of OECD Input–Output 2009 edition (as of March 2012). 245 9.6 Continued. 246 9.7 OECD IO industry classification. 247 9.7 Continued. 248 10.1 Countries/regions included in World Input–Output Database. 267 10.2 Product Classification of World Input–Output Database. 268 10.2 Continued. 269 10.3 Industrial classification of World Input–Output Database. 270 10.4 Comparisons of world goods and service trade (various sources as a percentage of National Accounts data). 271 10.5 Comparing merchandise trade data for selected countries (various sources as a percentage of National Accounts data). 272 xiv Trade in Value Added 10.6 Comparing services trade data for selected countries (various sources as a percentage of National Accounts data). 274 10.7 World trade in total (share of imports over exports by source). 276 10.8 Reporter reliability indexes, initial inconsistency and mean absolute percentage adjustment of total exports and imports, 1995–2009. 280 10.9 Mean absolute percentage adjustment of national statistics. 283 10.10 Mean absolute percentage between WIOD industry-by-industry WIOTs and adjusted ICIO tables, 2005. 284 11.1 The location of value added and capture for a ‘Tea Party Barbie’ doll, 1996. 293 11.2 Business functions sourced internationally by manufacturing enterprises in selected European countries, 2001–6: share of enterprises carrying out international sourcing (%). 305 11.3 Business functions sourced internationally by services enterprises in selected countries, 2001–6: share of enterprises carrying out international sourcing (%) 305 11.4 Organisation and offshoring: four possibilities. 309 11.5 Average share of employment (in percent) by business function and organisation type, December 2011 (US-owned firms’ US operations). 311 List of Contributors Nadim Ahmad is the Head of the Trade and Business Statistics Division of the OECD’s Statistics Directorate, where he leads work on trade statistics, entrepreneurship and productivity. He also leads the OECD’s horizontal activ- ity on trade in value added. He joined the OECD in 2000 to develop the OECD’s input–output database, and since then has worked in a number of positions, notably in the national accounts, where he played an important role, as the OECD’s representative on the Inter-Secretariat Working Group on the National Accounts, in the development of the new 2008 System of National Accounts. Prior to joining the OECD, Nadim worked at the UK’s Office for National Statis- tics and the UK’s Finance Ministry. Sónia Araújo is an economist at the OECD Economics Department, working on monitoring the implementation of structural policy reforms in OECD and BRIICS countries, and assessing their impact on employment, productivity and income growth. Since joining the OECD as a Young Professional, Sonia has contributed to several studies related to international trade and invest- ment. In addition to looking at the role of investment guarantees and polit- ical risk insurance provisions in Foreign Direct Investment, she has worked on capital-flow regulation in BRIICS countries and the construction of quan- titative indicators of globalisation, while helping with the development of the OECD framework for analysing cross-border trade issues. Through her projects, Sonia has also examined the determinants of infrastructure invest- ment and efficiency, including public–private partnership practices and the regulatory and competition environment, and evaluated the impact of large infrastructures on economy-wide investment and economic growth. Richard E. Baldwin has been Professor of International Economics at the Grad- uate Institute, Geneva since 1991, Policy Director of CEPR since 2006, and Editor-in-Chief of Vox since he founded it in June 2007. He was Co-managing Editor of the journal Economic Policy from 2000 to 2005, and Programme Director of CEPR’s International Trade programme from 1991 to 2001. Before that he was a Senior Staff Economist for the President’s Council of Economic Advisors in the Bush Administration (1990–1), on leave from Columbia Uni- versity Business School, where he was Associate Professor. He did his PhD in economics at Massachusetts Institute of Technology (MIT) with Paul Krugman. He was visiting professor at MIT in 2002/3 and has taught at universities in Italy, Germany and Norway. He has also worked as consultant for numerous xvi Trade in Value Added governments, the European Commission, OECD, World Bank, EFTA and USAID. The author of numerous books and articles, his research interests include international trade, globalisation, regionalism and European integration. He is a CEPR Research Fellow. Peter Bøegh-Nielsen is Head of Division for Structural Business Statistics, Statistics Denmark. He chairs a working group within the European Statistical System (ESSnet) on developing globalisation indicators and measurement of global value chains and international sourcing. He is also a member of the Bureau of the OECD Working Party on Globalisation of Industry. Clair Brown is Professor of Economics and Director of the Center for Work, Technology, and Society at the University of California, Berkeley (UCB). The industries she has studied in the field include semiconductors, consumer elec- tronics, ICT and high-tech start-ups. Clair, together with Timothy Sturgeon of Massachusetts Institute of Technology (MIT), developed a General Social Sur- vey module to study workers’ perceptions of the impact of trade on domestic jobs compared with the possibility their jobs might be offshored based on job characteristics. Funded by the NSF, Clair and Tim are now working on a rep- resentative survey of US firms that documents their domestic employment by business function and their location of activities domestically and glob- ally in order to study the relationship between jobs, organisation structure and innovation. Funded by USAID, Clair heads a UCB group that is develop- ing Ready-Made Impact Assessment, an open source online tool for low-cost effective project evaluation. Her books include American Standards of Living, 1919–1988 (Blackwell, 1994), and (with coauthors) Work and Pay in the United States and Japan (Oxford University Press, 1997), Economic Turbulence (Uni- versity of Chicago Press, 2006) and, with Greg Linden, Chips and Change: How Crisis Reshapes The Semiconductor Industry (MIT Press 2009, 2011). Judith Dean is Professor of International Economics in the International Busi- ness School at Brandeis University. Her teaching and research focus on inter- national trade and economic development, and in particular the interrelation- ships between trade, the environment and poverty. In a series of empirical studies using Chinese data, she has been exploring the possibility that trade growth, foreign investment and production fragmentation may have beneficial effects on the environment. In other work, she studies trade and fragmenta- tion, the effects of non-tariff barriers and the implications of trade preferences for economic development. Prior to joining Brandeis, Judy held positions as Senior International Economist in the Office of Economics, US International Trade Commission, Associate Professor of Economics at SAIS, Johns Hopkins University, and Assistant Professor of Economics at Bowdoin College. She has been a consultant to the World Bank and the OECD, and a Visiting Scholar at the Indian Statistical Institute, New Delhi, India. Her recent work has included collaboration with the USITC, Tsinghua University and the India Development Foundation. List of Tables xvii Justino De La Cruz is an International Economist in the Office of Economics at the US International Trade Commission (USITC), where he conducts research and writes and leads reports for the US Trade Representative and Congress. Prior to joining the International Trade Commission, Justino taught interna- tional trade and finance, econometrics and statistics, and business forecast- ing at the University of Texas. He has been a visiting scholar at the Banco de Mexico (Mexico’s Central Bank), and an economist at the Inter-American Development Bank and at Wharton Econometrics Forecasting Associates. Gary Gereffi is Professor of Sociology and Director of the Center on Glob- alization, Governance, and Competitiveness at Duke University. He received his BA from the University of Notre Dame and his PhD from Yale Univer- sity. Gary has published numerous books and articles, including The New Offshoring of Jobs and Global Development (International Institute of Labor Studies, 2006), Global Value Chains in a Postcrisis World: A Development Per- spective (The World Bank, 2010) and ‘Shifting End Markets and Upgrading Prospects in Global Value Chains’ (a special issue of the International Jour- nal of Technological Learning, Innovation and Development, 2011). He has recently completed a three-year project on economic and social upgrading in global value chains, financed by the UK’s Department for International Devel- opment, and is working on global value chains in emerging economies and new methodologies for measuring value chain upgrading. Gene M. Grossman is the Jacob Viner Professor of International Economics at Princeton University, Chair of Princeton’s economics department and the Director of Princeton’s International Economics Section. He received his BA in economics from Yale University and his PhD from the Massachusetts Insti- tute of Technology. Gene is a Fellow of the Econometric Society and of the American Academy of Arts and Sciences, a Life Member of the Council on Foreign Relations and holder of an honorary doctorate from the University of St. Gallen. He has written extensively about many aspects of international trade. Satoshi Inomata joined the Institute of Developing Economies–Japan External Trade Organization (IDE-JETRO) in 1992, after receiving his MSc in develop- ment economics from the University of Oxford. His major research interests include international input–output analyses and global value chains analyses. He was involved in the construction of the Asian International Input–Output Tables (AIOTs) for 1990, 1995 and 2000. In 2007, he initiated and organised a new project, alongside the 2005 AIOT, for constructing the 2005 BRICs Inter- national Input–Output Table that covers BRICs economies plus Japan, the USA and the EU. He is a co-editor of the recent joint publication by the WTO and IDE-JETRO, Trade Patterns and Global Value Chains in East Asia. He is a mem- ber of the council board of the International Input–Output Association and the editorial boards of Economic Systems Research, and the United Nations Handbook of Input–Output Table Compilation and Analysis. xviii Trade in Value Added Ronald Jansen is Chief of the Trade Statistics Branch of the United Nations Statistics Division in New York, responsible for statistics of international trade in goods and services and tourism statistics. He has over 20 years of experi- ence in the field of international trade statistics. His main interests include the measurement of international trade and economic globalisation, linking trade statistics to business registers, foreign affiliate statistics, tourism statistics, trade in services statistics and the revision of the classification of broad eco- nomic categories. He also served as Chief of the Capacity Development Section and has conducted training workshop and expert meetings on issues related to international trade statistics in many countries around the world. Ronald gained a degree in statistics and psychology at the University of Groningen in 1984. Thereafter, he taught statistics and did research at the University of Nijmegen, obtaining a PhD in mathematical modelling of human information processing in 1990. Robert Johnson is an Assistant Professor in the Department of Economics at Dartmouth College and Faculty Research Fellow at the National Bureau of Economic Research. His research interests are in international economics, at the intersection of international trade and macroeconomics. He has written on trade and product quality with heterogeneous firms and the implications of cross-border production sharing for measurement of trade linkages and transmission of shocks. He received his PhD in economics from the University of California, Berkeley, his MSc in global market economics from the London School of Economics and his BA in economics from Northwestern University. Robert B. Koopman is the Director of Operations at the US International Trade Commission, overseeing the Commission’s trade policy research and pro- viding negotiation assistance to the US Trade Representative and Congress. He also oversees anti-dumping and countervailing duty investigations, unfair imports adjudication, and maintenance of the Harmonized Tariff System. Bob previously served as the Chief Economist and Director of the USITC Office of Economics. In this position he provided economic analysis and support for the Commission’s role as adviser to Congress and the President on international trade matters. Greg Linden is a Research Associate at the Institute for Business Innovation, a research unit at the Haas School of Business, University of California, Berkeley (UCB). He has authored numerous articles about the globalisation of industry, and co-authored Chips and Change: How Crisis Reshapes the Semiconductor Industry (MIT Press, 2009). He holds a master’s degree in public policy and a PhD in economics, both from UCB, and has worked as a consultant on projects in Asia to develop industrial policy for high-tech industries. Alessia Lo Turco is Assistant Professor at the Faculty of Economics of the Polytechnic University of Marche, Italy. She joined the Faculty in 2005, lec- turing on international economics. Her research interests lie in the fields of List of Tables xix empirical international trade, productivity and the labour market. Alessia has been working on the impact of international integration on the Turkish man- ufacturing sector at the firm and firm–product level. She holds a PhD in eco- nomics from the Polytechnic University of Marche, and an MSc in economics from the University of Warwick. She has been a visiting fellow at University College Dublin and at the United Nations Economic Commission for Latin America and the Caribbean (ECLAC, Chile). She has also worked as a consul- tant for the ‘Centro de Análisis y Difusión de la Economía Paraguya’ (CADEP, Paraguay) and the ‘Red Mercosur’. Daniela Maggioni is post-doctoral fellow at the Faculty of Economics of the Polytechnic University of Marche, where she has also been lecturing interna- tional economics. Daniela earned her PhD in January 2010 at the same uni- versity. Her research interests lie in the fields of international trade, produc- tivity, innovation and the labour market. She has been studying the trade and productivity nexus at the firm level for the Turkish and Italian manufactur- ing sectors. Daniela has also been research fellow at the Centro Studi Luca d’Agliano, where she has worked on the European Firms in the Global Econ- omy (EFIGE) project. She was awarded the Polytechnic University of Marche’s Scholarship for PhD students and has won the second prize Sakip Sabanci international award for her paper on the Turkish economy under the global crisis. Aaditya Mattoo is Research Manager, Trade and Integration, at the World Bank. He specialises in trade policy analysis and the operation of the World Trade Organization (WTO), and provides policy advice to governments. Prior to joining the Bank in 1999, Aaditya was Economic Counsellor at the WTO. Between 1988 and 1991, he taught economics at the University of Sussex and at Churchill College, Cambridge University. He holds a PhD in economics from the University of Cambridge, and an MPhil in economics from the University of Oxford. He has published widely in academic and other journals on trade, trade in services, development and the WTO, and his work has been cited extensively, including in the Economist, Financial Times, New York Times and Time magazine. Andreas Maurer is Chief of the International Trade Statistics Section at the World Trade Organization (WTO). He studied economics, specialising in pub- lic finance and statistics/econometrics, at the University of Hohenheim, Ger- many, and holds a doctorate in economics. In 1990, he joined the United Nations Economic Commission for Europe, where he worked in the Secre- tariat of the Conference of European Statisticians. He was editor-in-charge of the Statistical Journal of the UN ECE. In 1994 he moved to GATT/WTO. His current interests include measuring trade in services flows, especially Mode 4, and trade in value added. Bo Meng is a Research Fellow at the Institute of Developing Economies–Japan External Trade Organization (IDE-JETRO). He received his PhD in information xx Trade in Value Added science from Tohoku University, Japan in 2005. In 2009 and 2013 he was a visiting scholar at the OECD and USITC, carrying out joint research with IDE-JETRO on global value chains. He has been an expert-group member of the OECD Trade in Value-Added Related Scientific Committee from 2012. His research interests include international (inter-regional) input–output analy- ses, spatial CGE models, low-carbon economics and issues related to global value chains. He is one of the co-authors of the 2011 WTO and IDE-JETRO joint publication, ‘Trade Patterns and Global Value Chains in East Asia: From Trade in Goods to Trade in Tasks’. Sébastien Miroudot is Senior Trade Policy Analyst in the Trade in Services Division of the OECD Trade and Agriculture Directorate. He holds a PhD from Sciences Po Paris in international economics. Before joining the OECD, he was research assistant at Groupe d’Economie Mondiale and he taught in the mas- ter’s degree programme at Sciences Po. His research interests include trade in services, trade and investment and trade flows within global value chains. He is currently working at OECD on the measurement of trade in value-added terms and the construction of a Services Trade Restrictiveness Index. He has published several journals articles and contributed several chapters to vari- ous books dealing with trade policy issues. Guillermo Noguera is a Postdoctoral Research Scholar at the Graduate School of Business at Columbia University. His research is in international trade with a focus on cross-border production fragmentation, the effects of trade costs on value added trade, and the gravity model. His work has appeared in the American Economic Review and the Journal of International Economics and has been mentioned in The Economist and VoxEU among others. He received his PhD in economics from the University of California, Berkeley, in 2011 and will join the Department of Economics at the University of Warwick as an Assistant Professor in 2013. Mika Saito is a senior economist at the International Monetary Fund (IMF). She joined the Fund in 2003 and taught at the IMF training centres for a number of years before taking up bilateral and multilateral surveillance work on develop- ing, emerging and advanced economies. Most recently, she has been leading projects assessing the role of trade finance since the 2008–9 global financial crisis, and measuring trade spillovers and real effective exchange rates in the world with global supply chains. Before joining the Fund, Mika was an assis- tant professor at the Mendoza College of Business at the University of Notre Dame. She holds a PhD in economics from Cornell University. Ranil Salgado is a division chief in the International Monetary Fund (IMF), responsible for trade issues. He has been at the IMF for over sixteen years, also serving in the Asia and Pacific Department (including Australia, India, Myan- mar and Singapore), the Research Department (multilateral surveillance and the World Economic Outlook) and the Western Hemisphere Department (US, List of Tables xxi Canada and Dominica). Prior to joining the IMF, he worked in a strategy man- agement consulting firm and as teaching and research assistants at the Univer- sity of Pennsylvania, Johns Hopkins University and the Federal Reserve Board of Governors. He attended Harvard University, and has graduate and doctoral degrees from Cambridge University and the University of Pennsylvania. Timothy J. Sturgeon is a Senior Research Affiliate at the Industrial Perfor- mance Center (IPC) at the Massachusetts Institute of Technology MIT, co- organiser of the Global Value Chains Initiative. His research focuses on the process of global integration, with an emphasis on offshoring and outsourc- ing practices in the electronics, automotive and services industries. Tim has made significant contributions to global value chain (GVC) theory, and is work- ing to improve the metrics and methods available for globalisation research. His work explores the implications (for employment, industrial development, technological learning, and policy) of deepening GVC linkages between indus- trialised and developing economies. Tim works actively with the policymakers in international development agencies, industrialised countries and develop- ing countries to disseminate his insights and collaborate on actionable, effec- tive policy responses to global integration. Tim is co-editor (with Momoko Kawakami) of Local Learning in Global Value Chains: Experiences from East Asia (Palgrave Macmillan), and has published his research in Studies in Com- parative International Development, Industrial and Corporate Change, Review of International Political Economy, Journal of East Asian Studies and Journal of Economic Geography. Daria Taglioni is a Senior Economist at the World Bank. Previously, she worked at the European Central Bank (2006–11) and at the Organisation for Economic Cooperation and Development (2002–6). Her recent research focuses on international competitiveness, the economics of value chain pro- duction and the link between financial markets and trade. Daria is an Ital- ian national and holds a PhD in International Economics from the Graduate Institute, Geneva, a master’s degree in economics from the College of Europe, Bruges, and an undergraduate degree from the University ‘La Sapienza’, Rome. She speaks Italian, English, French and German. Zhi Wang is a lead international economist at Research Division, Office of Economics, US International Trade Commission (USITC). He obtained his PhD in applied economics at University of Minnesota with a minor in computer and information sciences in 1994. After graduating, he worked as a consultant for the World Bank, an economist at Purdue University, the Economic Research service of US Department of Agriculture, and the Bureau of Economic Analysis of US Department of Commerce, and as a senior research scientist at School of Computational Sciences of Gorge Mason University, before joining USITC in 2005. Before he came to the USA he was a research fellow at Chinese Academy of Agricultural Sciences and served on the board of directors of the Chinese Economists Society during 1992–93. xxii Trade in Value Added Shang-Jin Wei is the Director of the Jerome A. Chazen Institute of Interna- tional Business, Professor of Finance and Economics and N. T. Wang Professor of Chinese Business and Economy at Columbia University’s Graduate School of Business, and a Research Associate at the National Bureau of Economic Research (US), and Research Fellow at the Centre for Economic Policy Research (Europe). Norihiko Yamano is responsible for the current edition of the OECD’s Input– Output and related economic models. He has previously worked at the Central Research Institute of Electric Power Industry, Japan, as a research economist, where he developed various types of economic models using originally devel- oped regional public capital stock, labour force statistics and inter-regional IO tables. He has also served as an external consultant on many projects involv- ing the use of economic, energy and demographic forecasting models. Robert E. Yuskavage was the Associate Director for International Economics at the Bureau of Economic Analysis (BEA) before retiring in February 2012. He is currently a consultant on international economic statistics. As associate director, Robert oversaw the balance of payments and international invest- ment position accounts, statistics on the operations of multinational com- panies and BEA’s extensive international data-collection programmes. Previ- ously he served as chief of BEA’s Balance of Payments Division, where he led efforts to modernise and enhance the international accounts. Earlier, Robert served in BEA’s Industry Accounts Directorate as a researcher and manager. As chief of the GDP by Industry programme, he directed efforts that signif- icantly enhanced BEA’s industry accounts and led to major new initiatives. Prior to joining BEA in the early 1990s, he conducted tax policy research at the US Treasury Department. Acknowledgements This volume includes papers that were first presented and discussed at a workshop on ‘The Fragmentation of Global Production and Trade in Value- Added: Developing New Measures of Cross Border Trade’, held at the World Bank in Washington, DC, on 9–10 June, 2011. This workshop would not have been possible without the brilliant organisation and leadership of Francis Ng. The editors thank the authors of individual chapters, the discussants and participants in the workshop for insightful comments and suggestions that helped improve the papers. The editors also thank Justin Lin and Martin Raval- lion for their guidance and support. Anna Regina Bonfield provided outstand- ing administrative support, with valuable inputs from Michelle Chester. Aaditya Mattoo, Zhi Wang and Shang-Jin Wei, Editors 1 Measuring Trade in Value Added when Production is Fragmented across Countries: An Overview AADITYA MATTOO, ZHI WANG AND SHANG-JIN WEI 1 What is the ‘country of origin’ of an iPad? If you order a new model of iPad today from Apple’s online store, you will notice that your device will be shipped out of China by a company called Foxconn, so China is officially the country that exports iPads. Of course, the product is designed in California and uses lots of components from Japan, Korea and other countries. When a product is produced by a global production chain in which a number of countries participate in different stages of production and supply different parts and components that make the final product what it is, the concept of ‘country of origin’ is not very useful. The world is increasingly like this: more and more products traded internationally use parts and components from multiple countries before being assembled into the form of final prod- ucts. Indeed, between one-half and three-quarters of overall world trade in goods and services consists of trade in intermediate goods (see Chapter 2 by Sébastien Miroudot and Norihiko Yamano). In this volume, we take stock of what we know about this phenomenon based on a body of active and expanding research. Most of the work reviewed here was presented at a conference at the World Bank on this topic in 2011, and subsequently revised and expanded. First, we explore the implications for economic policies of the increasing divergence between countries that export a given product and the countries that provide the value added that goes into that product. Second, we examine several approaches to estimating domes- tic and foreign value added in a country’s exports (and imports). Third, we present ideas about modifying the existing national and international stat- istical infrastructure that can lead to better measurement of trade in value added. As this is still an evolving field, we hope that this volume will provide a framework for comparing existing approaches and also inspire new ones. 1 The views in the papers are those of the authors, and do not necessarily reflect the official position of the institutions with which the authors are associated. 2 Trade in Value Added 1 IMPLICATIONS FOR ECONOMICS AND ECONOMIC POLICIES It is important to go beyond official trade statistics and to gauge the true exports in value-added terms for a number of reasons. First, and most fun- damentally, it would help us to understand better the changing nature and consequences of international trade. As argued by Gene M. Grossman in Chap- ter 3, thanks to the dramatic improvements in communication technology, international specialisation is no longer at the level of the good or industry, but at ever finer levels, such as specific tasks. Therefore, new but still largely untested theory is posing traditional questions, such as where production takes place and what effect trade has on factor prices and the distribution of income, not at the level of the industry but at much finer levels of economic activity. Apart from showing the continued importance of factors such as dif- ferences in relative factor prices, new determinants of the location of activity are being identified, such as the ease with which a task can be performed at a distance (relating to its prevalence and amenability to codification) and whether tasks are complementary (which affects whether they need to be co- located or can be separated). And new predictions about gains from trade are being generated: offshoring can generate productivity gains that are shared by all domestic factors of production so that the traditional conflict of inter- est between factors with regard to trade may be alleviated. The accuracy of these predictions can only be assessed if we go beyond trade data based on gross flows. Economic activity is best measured by value added not gross out- put, as national accountants have known for a long time. As Grossman and Rossi-Hansberg (2008) concluded, the globalization of production processes mandates a new approach to trade data-collection, one that records international transactions, much like domes- tic transactions have been recorded for many years. Second, and perhaps even more urgent than the positive questions are the normative questions raised by the fragmentation of production across bor- ders. Miroudot and Yamano (Chapter 2) and others have identified a list of policy-relevant issues that crave answers. • More accurate understanding of bilateral trade imbalances. Standard measures of bilateral trade balances are based on gross trade. Consider the Chinese and Japanese trade surpluses against the USA in recent years. These are overstated because Chinese production for exports to the USA uses lots of imported inputs from Japan, Korea, USA and elsewhere. In comparison, US production for its exports to China uses mostly US-made inputs. It is estimated that the true Chinese trade sur- plus in value-added terms is about 25–40% lower than the surplus mea- sured by gross trade (Maurer and Degain 2010). In contrast, the Japanese surplus against the USA is understated, since Japan also exports value Measuring Trade in Value Added: An Overview 3 added indirectly by exporting intermediate goods to China and other countries that are used in these countries’ exports to the USA. • Incidence of trade policies. Because high-income countries are more likely to be at the upstream end of global production chains, their imports from low-income countries are more likely to contain their own value added (through their exports of machinery and other interme- diate inputs to these low-income countries). For example, this is the case for US imports from China and Mexico, both of which are heavy users of the processing exports mode that uses lots of imported inputs from the USA (and other high-income countries) in their exports to the USA. Given this structure, an increase in barriers by the USA on imports from China or Mexico is also likely to hurt US-based firms. The extent to which this is true varies by sector, as the US content in US imports from China and Mexico also varies by sector. Similar statements can be made for trade policies by most other high-income countries. For example, a study of the Swedish National Board of Trade on the Euro- pean shoe industry highlights that shoes ‘manufactured in Asia’ incor- porate between 50% and 80% of European Union value added. In 2006, anti-dumping duties were introduced by the European Commission on shoes imported from China and Vietnam. An analysis in value-added terms would have pointed out that EU value added was in fact subject to anti-dumping duties (National Board of Trade 2007; see also Chapter 2). • Employment content of trade. Policymakers often wish to find out the effect of a given trade policy, say an across-the-board tariff by the USA on imports from China, on employment in China and the USA. This analysis would be misleading if one did not realise either that the gross trade statistics do not accurately reflect true value added from the exporting country or that the importing country’s value added could be embedded in the exporting country’s gross exports. (Of course, the analysis could also be misleading if one did not realise that the general equilibrium effect on employment in the longer run is different from the partial equilibrium effect in the short run.) The EU shoe industry example can be interpreted in terms of jobs. Traditional thinking in gross terms would regard imports of shoes manufactured in China and Vietnam by EU shoe producers as EU jobs lost and transferred to these countries. But in value-added terms, one would have to account for the EU value added; and while some workers may have indeed lost their jobs in the EU at the assembly stage, there could be a higher number of jobs in the research, development, design and marketing activities that exist because of trade (and the fact that this fragmented production process keeps costs low and EU companies competitive). • Trade and competitiveness. Indicators of competitiveness such as Bal- assa’s (1965) revealed comparative advantage (RCA) has proven to be 4 Trade in Value Added useful in many research and policy applications. The problem of mul- tiple counting of certain value-added components in the official trade statistics suggests that the traditional computation of RCA could be noisy and misleading. For example, Koopman et al (forthcoming; hence- forth KWW2) show that using gross exports data suggests India had a strong revealed comparative advantage in finished metal products (ranked fourth among the 26 countries in the sample). However, when looking at domestic value added in that sector’s exports, its ranking in RCA drops precipitously to fifteenth place. This change in rank causes the sector to switch from being a comparative advantage sector to a comparative disadvantage sector for India. • Trade and the environment. Concerns over greenhouse gas emissions and their potential role in climate change have triggered research on how trade openness affects CO2 emissions. The fragmentation of pro- duction requires a value-added view of trade in order to understand where imported goods are produced and thus where CO2 is generated. For example, initial work by Dean and Lovely (2010) argues that the emergence of global supply chains may have had surprisingly bene- ficial effects on China’s environment. This is because China’s exports have been shifting over time towards highly fragmented sectors (such as office and computing machinery and communications equipment) that are less polluting, and away from traditional exports that are less fragmented and also more polluting. Third, and related to the policy dimension discussed above, is the inter- national surveillance and policy analysis dimension, illustrated in the macro- economic context by Mika Saito and Ranil Salgado of the International Mone- tary Fund (Chapter 3). They argue that using accurate value-added trade data would improve exchange rate assessments because real effective exchange rates based on value-added trade weights would more accurately measure competitiveness than those based on gross trade weights. And these assess- ments would in turn improve our ability to estimate the impact of changes in relative prices, including on global rebalancing. They suggest, for example, that changes in relative prices would result in asymmetric rebalancing effects between downstream and upstream countries (eg China and the USA, respec- tively) in terms of value chains. This is because the larger share of foreign value added in the exports of the downstream country mitigates the impact of exchange rate changes. Fourth, and perhaps the greatest benefit of measuring value-added trade, is in understanding and responding to the development challenge, as argued by Judith Dean (Chapter 3). She notes that in principle the more production can be split globally and tasks dispersed based on comparative advantage, the more low-income countries can participate in these production chains. More disaggregated value added trade data could help us understand the extent to Measuring Trade in Value Added: An Overview 5 which developing countries are already participating in global supply chains, and the extent to which global chains are promoting indirect exports from developing countries. The latter could be happening if such chains help small and medium-sized enterprises to overcome financial and other constraints. We could also understand what role developing countries play in such supply chains and why. For example, research by Antras (2005) and Feenstra and Han- son (2005) suggests that improved property rights and better quality control may help developing countries ‘move up’ the supply chain. Beyond the issues raised by Dean are certain normative questions. We need to understand better whether and why it matters from a development per- spective where you are in the supply chain (or where you add value). This is the counterpart in fragmented production space of the traditional question of whether it matters what you export. Are certain tasks (or types of value adding activities) associated with greater scope for learning-by-doing or knowledge spillovers? In fact, what it means to ‘move up’ the value chain and whether it is desirable in normative terms is itself an open question that craves analysis with better data. Finally, even if location in the value chain matters, how far can it be influenced by policy? The issue of whether developing countries can mould their production structures into more dynamic forms through policy intervention is again not a new question. But there is an added richness to this question, and even greater empirical challenge in providing credible answers, when we think in terms of tasks rather than entire products. 2 APPROACHES TO DISTILLING VALUE ADDED IN TRADE FROM STANDARD TRADE STATISTICS A first hint of the relevance of intermediate trade is evident from the behaviour of the gravity model. The gravity model of trade volume is per- haps one of the most successful empirical specifications in international eco- nomics both in terms of goodness of fit (typically in excess of 60% even in a cross-sectional context) and in terms of compatibility with leading eco- nomic theories (as it can be justified by the theory of differentiated trade, the standard theories of comparative advantage and new theories based on heterogeneous firms). It has been used to analyse a myriad of trade policies, such as the effectiveness of the World Trade Organization in promoting trade (see for example, the debate between Rose (2004) and Subramanian and Wei (2007)). However, as Richard E. Baldwin and Daria Taglioni (Chapter 7) point out, while the gravity model works well for nations and time periods where most exports are consumer goods, it works poorly when trade in components and parts is important. More precisely, the standard practice of using the GDP of the exporting and importing countries as the ‘mass’ variables in the grav- ity equation is inappropriate for bilateral flows where trade in intermediate goods is a nontrivial part of overall trade. As long as producer demand devi- 6 Trade in Value Added ates from consumer demand, one would need to use something other than, or in addition to, GDP to proxy for the total demand and supply. Baldwin and Taglioni document the following patterns. First, the estimated coefficients on the log GDPs are lower for nations where parts and components trade is important. Second, the coefficients on log GDPs fall over time, as parts and components trade grows in importance. Third, in those cases where the GDPs of exporters and importers lose explanatory power, one can find a role for demand by third countries. Two parallel lines of work in the literature attempt to estimate value added in trade. The first attempts to measure the degree of vertical specialisation and imported foreign content in a country’s exports using a single country input–output table, and started with the pioneering work of Hummels et al (2001) (henceforth HIY). The second approach traces value added at various stages of production across countries using an inter-country input–output (ICIO) table. It tries to measure one country’s exports of value added to a second country, both by excluding foreign value added embedded in parts and components used in the production of the exports, and by adding indirect exports of value added embedded in the first country’s exports of components and parts to a third country, which in turn uses them to produce products that are exported to the second country. This approach is described in detail by Robert C. John- son and Guillermo Noguera. 2 They provide a formal definition for value-added exports: which is value added produced in a country but absorbed in another country. In contrast to HIY’s measure of foreign content in exports, they pro- pose a measure of the ratio of value added to gross exports, or the VAX ratio, to measure the intensity of inter-country production sharing. They find that exports of manufactures have lower VAX ratios, and imbalances, at the bilat- eral level, measured in value-added terms can differ substantially from gross trade imbalances. As an example, they show that the US–China trade imbal- ance in 2004 is 30–40% smaller when measured in value-added terms. In the HIY framework for estimating vertical specialisation or foreign con- tent, it is assumed that the input–output coefficients in the production for exports and those in the production for domestic market are the same. This of course does not have to be true in general, but it is especially likely to be violated in processing trade where imported machinery and other imported inputs are extensively used to produce for the export market. Many coun- tries offer outright tariff exemption for imported inputs used in processing exports, or at the very least streamlined customs duty drawbacks. In some countries, processing exports can be a significant part of overall exports. For example, in China, processing exports account for about half of over- all exports. Wholly foreign owned firms and Sino-foreign joint ventures are heavy users of the processing export scheme. In Mexico, processing exports 2 See Johnson and Noguera (2012), reproduced here as Chapter 4. Measuring Trade in Value Added: An Overview 7 account for an even greater share, largely due to the prevalence of exports by Maquiladora firms. One way to address this issue is by tracking separately the two sets of input– output coefficients in production for the export market versus the domestic market. Of course, most countries do not officially publish separate input– output coefficients. Robert Koopman, Zhi Wang and Shang-Jin Wei 3 devel- oped a framework that allows one to estimate these two sets of coefficients by combining information on processing trade share at the sector level from a country’s customs data and the country’s existing official input–output table. Their methodology has two parts. First, they derive the equations necessary to do the computation. The key part of the derivation is to split the standard input–output (IO) table into two parts: one that focuses on domestic pro- duction and trade, and one that focuses on processing exports. This yields four coefficient matrices: one for domestic production used for processing exports, one for domestic production used for processing activity, one for imported inputs used for processing exports, one for imported inputs used for non-processing activities. Second, they develop an approach to estimate these coefficient matrices. The estimation essentially attempts to keep the coefficients as close as possible to those implied by official trade and national account statistics with some proportionality assumptions, while at the same time also satisfying the supply and use balance conditions and adding-up constraints. Koopman et al apply their methodology to data from China. There are four main findings. First, foreign value added in China’s overall manufacturing exports was about 50% in 2002, which is more than double what would be obtained by a straightforward application of the method from HIY. Second, the foreign value added in non-processing exports was compar- atively small, about 10% in 2002 and 16% in 2007. Third, and most interest- ingly, those sectors that are labelled as relatively sophisticated or “high-tech’, such as electronic devices and communication equipment, have particularly low domestic content (about 30% or less). Finally, the share of domestic value added of Chinese manufacture exports increased to about 60% by 2007, just five years after China joined the World Trade Organization (WTO). The latter finding is of particular interest. Over time, different forces pull the share of domestic value added in total exports in different directions. On the one hand, with the falling tariff rates and non-tariff barriers, espe- cially since China’s accession to the WTO in December 2011, all exporters in China might use more imported inputs, which would lower the share of domestic value added in China’s exports as evidenced by the 6% increase of foreign content in China’s normal exports. China may also increase the share of those sectors that use more imported inputs, which would also result in a reduction in the domestic value added in its total exports. On the other hand, 3 See Koopman et al (2012), reproduced here as Chapter 5, and referred to as KWW1 in subsequent discussions. 8 Trade in Value Added as domestic producers of intermediate goods become technologically better and stronger, exporters might also choose to source more inputs locally, sub- stituting Chinese-made inputs for previously imported inputs, which would increase the share of domestic value added in China’s exports. A concrete channel through which this occurs is that more foreign-owned intermedi- ate input producers have chosen to relocate from abroad to China in recent years, enhancing the capacity in China of producing sophisticated parts and components needed in China’s exports. In addition, as China’s domestic mar- ket grows in size relative to the world market, more producers reorient their sales towards the Chinese market, resulting in a decline in processing exporters (which sell almost exclusively in the world market) relative to ordi- nary exporters (which sell in both the Chinese and foreign markets) and a decline in the use of imported inputs in the production. While the net effect could go either way, the data uncovered by KWW1 indicate that, on balance, the second effect dominates, and the share of Chinese value added in China’s total exports increases over time. However, KWW1 only address a special case of aggregation bias caused by product and firm heterogeneity when using industry level data. Even if national statistical agencies exceptionally publish the four key input–output matrices at industry-level as Mexico does, as long as different firms and prod- ucts within an industry have different imported input use intensities, using industry-level data will still generate a bias in the measurement of domestic value added in exports. Many recent firm-level studies show that exporters differ in many dimensions from non-exporters, including in their choice of inputs, and there is large heterogeneity in the import penetration rates among firms, especially between those actively engaged in trade and those that pro- duce only for the domestic market. Exporters are more likely to use more imported inputs than domestic firms. Kee and Tang (2012) complement the analysis of KWW1 by using firm-level data on exports and imports for Chinese processing exporters over 2000–6. In particular, instead of relying on the standard input–output data and assum- ing the same input–output coefficients in the production for exports and for domestic sales, the firm-level data could allow for heterogeneous input– output coefficients at the firm level. Of course, we would still need to aggregate the information at some level, otherwise there may be too much individual- level information and insufficient group-level information. If imported inputs by a given firm are primarily used by the firm to produce for exports and contain little Chinese value added, and if the domestically sourced inputs by processing exporters contain no imported value added, then one can com- pute the share of domestic value added in exports for this group of firms by simply looking at the ratio (exports − imports)/exports. By this methodology, Kee and Tang find that the average share of domestic value added in China’s processing exports rose from 52% in 2000 to 60% in 2006. The trend over time Measuring Trade in Value Added: An Overview 9 is very similar to those reported by KWW1, though the former focus only on processing trade. Justino De La Cruz, Robert B. Koopman, Zhi Wang and Shang-Jin Wei (Chap- ter 6) apply KWW1’s methodology to Mexico’s exports. Mexico uses the pro- cessing trade scheme (under the Maquiladora and PITEX programmes) even more extensively than China. In fact, their chapter improves on KWW1, as Mexico has direct measures of the input–output coefficients for processing exports, eliminating the need to estimate them and therefore reducing one margin of error. On average, the share of domestic value added in Mexico’s manufactured exports is 34%. Those manufacturing industries whose share of domestic value added is 50% or less account for about 80% of Mexico’s total manufacturing exports. Similarly to Kee and Tang (2012), Nadim Ahmad, Sónia Araújo, Alessia Lo Turco and Daniela Maggioni (Chapter 8) use firm-level data in Turkey to estimate the share of domestic value added in Turkish exports in 2005. The use of firm-level data allows Ahmad et al to permit separate input–output coefficients for firms that sell primarily to the domestic market and those that sell primarily to the world market. They estimate that the share of for- eign content in Turkey’s exports in 2005 was about 27%, which is 6 percent- age points higher than the share estimates from the official IO table based on aggregated data at the industry level. However, this number is lower than estimates for China and Mexico, most likely because Turkey engages in fewer processing exports than the other two countries. These studies demonstrate that estimates based on IO tables and firm-level data each have their advantages and shortcomings in estimating domestic and foreign content in exports. The methods are not substitutes but comple- mentary. Because any empirical work based on real world data has to involve some degree of aggregation (even with the most detailed plant level data), such ‘aggregation bias’ cannot be completely eliminated; it can only be reduced or minimised. The challenge for empirical researchers is how to minimize the ‘aggregation bias’ based on the particular research issue at hand and the information available at the time when the research is conducted. 3 CLARIFYING THE CONNECTIONS AND DISTINCTIONS AMONG ALTERNATIVE CONCEPTS RELATED TO TRADE IN VALUE ADDED Because the research on measuring trade in value added and quantifying the degree of vertical specialisation is active and dynamic, a number of concepts have been proposed. Some of them have similar names but distinct content. It may be useful to take stock of these concepts, pointing out both connections and distinctions. We have already mentioned the first measure of vertical specialisation, pro- posed by HIY (2001), which refers to the share of the imported content in a 10 Trade in Value Added country’s exports. This measure, commonly labelled as VS, includes both the directly and indirectly imported input content in exports. A second measure, also proposed by HIY (2001) and labelled as VS1, looks at vertical specialisation from the export side, and the value of intermediate exports sent indirectly through third countries to final destinations. However, HIY did not provide a mathematical definition for VS1. A third measure is the value of a country’s exported goods that are used as imported inputs by the rest of the world to produce final goods that are shipped back home. This measure was proposed by Daudin et al (2011). Because it is a subset of VS1, they call it VS1∗ . A fourth measure is value-added exports, which is value added produced in source country s and absorbed in destination country r . Johnson and Noguera (Chapter 4) defined this measure and proposed using the ratio of value-added exports to gross exports, or the ‘VAX ratio’ as a summary measure of value- added content of trade. However, the domestic content share in a country’s exports and the VAX ratio are, in general, not equal to each other. (In other words, the value of domestic content in exports and the value of a country’s value-added exports can be different.) Koopman, Wang and Wei (forthcoming; subse- quently referred to as KWW2) propose a methodology that decomposes a country’s total exports into four buckets (or nine components in total with a few terms in each bucket). The first bucket gives a country’s value-added exports, exactly as defined by Johnson and Noguera (Chapter 4). The sec- ond bucket gives the part of a country’s domestic value added that is first exported but eventually returned home. The third bucket is the value of for- eign value added used in the production of a country’s exports. The fourth bucket consists of what they call ‘pure double counted terms’, arising from intermediate goods being traded back and forth multiple times. Some of the terms in the fourth bucket double count value added originating in the home country, whereas other terms in the fourth bucket double count value added originating in foreign countries. KWW2 define ‘domestic value added in a country’s exports’ as the sum of the first and second buckets. This concept only looks where the value added is originated, regardless where it is ultimately absorbed. In comparison, a country’s ‘value-added exports’ refers to a subset of ‘domestic value added in a country’s exports’ that is ultimately absorbed abroad. The ‘domestic content of a country’s exports’ is defined by KWW2 to be even broader than ‘domestic value added in a country’s exports’. It is the sum of the first and second buckets, and those items in the fourth bucket that reflect pure double counting of value added that originate in the home country. Symmetrically, the ‘foreign content of a country’s exports’ is the sum of the third bucket and those items in the fourth bucket that reflect pure double counting of value added that originates in foreign countries. Measuring Trade in Value Added: An Overview 11 Such definitions have two attractive properties. First, KWW2 verify that the ‘foreign content of a country’s exports’ is mathematically identical to the VS measure proposed by HIY (2001) in multi-country settings but without HIY’s restrictive assumption of no two-way trade in intermediate inputs. Second, the sum of the domestic and foreign contents of a country’s exports is equal to that country’s total gross exports. As stated above, KWW2’s approach can completely decompose a country’s gross exports into the sum of nine components (or the sum of four buckets). Once one has the decomposition, other measures of vertical specialisation such as VS, VS1 and VS1∗ , in addition to ‘value-added exports’, can also be expressed as linear combinations of some subsets of the nine components. In this sense, the KWW2’s gross exports accounting method provides a compact and precise way to characterise the relationships among the major existing measures in this literature. The KWW2 decomposition also provides information on the structure of double counting in gross trade statistics (in addition to the total amount of double counting). The structural information can be useful in delineating a country’s position in the global production chain. For example, in some sec- tors, China and the USA may have a similar number of value-added exports but a different composition of the double counted terms. For China, the dou- ble counted terms may show up primarily in the form of the use of foreign components (eg foreign product designs or machinery) in the final goods that China exports. For the USA, the double counted terms may show up primar- ily in the form of domestic value added finally returned and consumed at home (eg product designs by Apple that are used in the final Apple products produced abroad but sold in the US market). The ratio of these two types of double counted terms offers a convenient measure of a country’s position in the global value chain. 4 SUGGESTIONS FOR OFFICIAL STATISTICAL INFRASTRUCTURES In the previous section, we reviewed research that takes the existing work by the official statistical agencies as given and seeks the best way to estimate trade in value added by combining information from trade data and national input–output tables. In this section, we discuss proposals for modifying the way official data is collected and reconciled that can improve the accuracy of the estimated trade in value added. Satoshi Inomata, Norihiko Yamano and Bo Meng (Chapter 9) review the compilation approaches of an inter-country input–output table for selected major economies in Asia (Asia IIO table for short) and the inter-country IO table produced by the OECD (OECD IIO table for short). For some of the cov- ered countries, the underlying data involves a periodic survey of firms that use intermediate imports. The Asia IIO table also harmonises sector defini- tions for the IO tables of the participating countries by the means of a survey 12 Trade in Value Added of individual countries’ input–output-table-compiling agencies. The Asia IIO table then reconciles the discrepancies to produce a consistent and balanced inter-country input–output table. The OECD IIO table covers more countries (58 countries). A major feature of the OECD IIO table is that information on the flow of intermediate inputs across constituent countries is available by both sectors and end-use categories. It is well known that international trade statistics do not balance at the global level, giving rise to the humorous anecdote of Earth trading with Mars or the Moon to explain the net surplus (or deficit). At the national level this can generally be ignored, the perspective being that the inconsistencies are in some other country’s accounts. But when considering global accounts, and in particular in relation to the estimation of trade in value added, these incon- sistencies have to be eliminated. In Chapter 10, Nadim Ahmad, Zhi Wang and Norihiko Yamano demonstrate how this can be accomplished by a three-stage data reconciliation model. In the first stage, their model reconciles total goods and services exports and imports recorded in each country’s GDP by expen- diture accounts with trade statistics at the product group level recorded in each country’s supply and use tables. It results in a consistent time series of country- and product-group-level total exports and imports, which satisfy the condition that world total exports plus a shipping cost margin (including insurance and freight) equal to world total imports. The use of international shipping services is also balanced with its supply from producing industries at the global level. In the second stage, their model benchmarks each coun- try’s supply and use tables with each country’s GDP by expenditure account, using globally consistent export supply and import demand estimates from the first stage as controls. In the final stage, their model allocates bilateral trade flows to producing/using industries and final users in each country based on international bilateral trade statistics broken down by end-use cat- egories, resulting in a time series of bilateral trade statistics within a global supply–use table that is consistent with global control totals estimated in the first stage. They use mirrored trade statistics as interval constraints in the final stage with a quality-based reliability index for each bilateral trade flow by product group, to arrive at a balanced global table that is consistent with the major components in each country’s GDP. Preliminary empirical tests of the model using WIOD data and aggregate trade statistics from official national accounts, as well as bilateral trade data from OECD, produced encouraging results. Ahmad et al show that imposing global consistency and eliminating ‘exports to the Moon’ will make no signif- icant changes to reported GDPs and other major aggregate national account statistics in the final database. If estimating value added from IIO tables can be called a ‘top-down’ approach, Timothy J. Sturgeon, Peter Bøegh Nielsen, Greg Linden, Gary Ger- effi and Clair Brown (Chapter 11) suggest two ‘bottom up’ approaches: product-level global production chain studies and business function surveys. Measuring Trade in Value Added: An Overview 13 A product-level study can decompose an individual product into an exhaus- tive list of components and parts, and trace the country of production of each component/part. The advantage of this approach is that one obtains more detailed information at the component level. However, a major disadvantage of this approach is that there are only a limited number of products for which such an approach is feasible. Another disadvantage is that tallying up all phys- ical inputs would not give a complete list of all inputs for most products, since a range of intangible support functions (R&D, marketing, IT services, etc.) also contribute to the final value of the product and they have a share of domestic value that is between 0 and 1. The second bottom-up approach is to expand a typical survey of firms by asking for information on how and where each business function is sourced (in addition to where each physical compo- nent is sourced). The second approach can avoid the shortcoming of the first approach of missing value added in R&D, IT services and other support func- tions. However, organising a periodic survey of firms across countries for this purpose in a consistent manner is an expensive undertaking and goes beyond what statistical agencies do currently. Five separate contributions, comprising Chapter 12, by Andreas Maurer, Ronald Jansen, Nadim Ahmad and Robert E. Yuskavage, from five different government institutions, respectively propose additional ideas on how the System of National Accounts (SNA) could be modified to integrate data collec- tion and lead to better measures of trade in value added. There is a consen- sus among the five contributors that conventional trade measures have major limitations for assessing inter-country linkages and bilateral trade balances. Therefore, developing trade in value-added statistics that could ultimately be included as supplementary measures in the SNA should be supported. For example, both Maurer and Ahmad believe that the OECD and the WTO are now ‘in a position to coordinate efforts towards the estimation of trade flows in value-added terms based on official trade statistics and national accounts’. (Maurer; see page 323 of this volume). There is also a consensus among the five contributors that direct measure- ment of value-added trade is extremely difficult if not impossible, primarily because the information is not available in business record-keeping systems. Therefore, conventional gross trade statistics should remain as the featured measures of cross-border trade and ‘will remain a necessary input for many analytical purposes’ (Jansen, this volume; see page 326). While data on value added at the firm level are useful to have, they are too expensive to collect because the current business record-keeping system does not contain such information. (An individual firm does not need to know how much imported content is contained in a domestically sourced intermediate good; a multi- product firm also does not typically track how the value of imported inputs is distributed across its different products or business functions.) As pointed out by Yuskavage (page 333): ‘in general, US business firms do not maintain information in their accounting systems that would allow them to readily 14 Trade in Value Added identify whether their material inputs are from domestic or foreign sources. Firms typically obtain their material inputs from wholesale suppliers and dis- tributors and are not necessarily concerned about the country of origin for these materials.’ Because of such difficulties in data collection, he believes (see page 333) that ‘the most promising approach to develop comprehensive and consistent value-added trade measures that go beyond case studies of indi- vidual high-profile products involves the use of world IO tables’. Jansen (see page 329) advocates linking enterprise survey data to detailed merchandise trade statistics via business registers to improve current official data collec- tion and the international standardisation of the compilation of IO tables as ‘two parallel and mutual supportive developments’. It is clear from the chapters in this volume that measuring the value-added content of trade requires a global input–output table. Such a table would inte- grate official national accounts and bilateral trade statistics on goods and services into a consistent accounting framework. Conceptually, it is a natural extension and integration of the SNA. In statistical practice, it requires rec- onciling each individual country’s supply and use tables with official bilateral trade statistics. New official statistics of trade in value added could be esti- mated under such an accounting framework to completely distribute value- added production to their original sources and final destinations at either the countrywide or industry average level. Because supply and use tables and input–output accounts are already a central part of 1993 and 2008 SNA, which by international consensus is the best framework for data gap assess- ment and GDP estimation, 4 it provides a workable and cost-effective way for national and international statistical agencies to remedy the missing infor- mation in current official trade statistics without dramatically changing the existing data-collection practices of national customs authorities. To mainstream the production of statistics on trade in value-added statis- tics, beyond knowing the conceptual definitions of the objects we wish to measure, we have to ask how official statistical collection can be amended in a cost-effective way to generate a consistent time-series of IIO tables of acceptable quality. Existing conceptual work has established a formal and precise relationship between value-added measures of trade and official trade statistics, and allows various value added and double counted components in a country’s official gross exports statistics to be correctly identified and estimated. It opens the possibility for the System of National Accounts (SNA) to accept the concept of trade in value added and provides a feasible way for international statistical agencies to report value-added trade statistics regularly in a relatively low cost fashion. 4 SNA (1993) recommended using a supply and use table as a coordinating framework for economic statistics, both conceptually and numerically to assure consistency for data drawn from different sources, especially in reconciling GDP estimates from production, expenditure and income sides (see SNA 1993, pp. 343–371). Measuring Trade in Value Added: An Overview 15 Some additional effort is useful in this regard, including the following. (i) Helping more developing countries to generate and publish supply and use tables regularly; for example, a good initial set of countries would be those emerging economies that are in the G20. (ii) Harmonising supply and use tables across countries: a common indus- try and product classification needs to be included in national IO statis- tics. (iii) Improving classification systems to properly identify intermediate in- puts in imported services and dual-use products, such as fuels. (iv) Improving the allocation of imported inputs (of both goods and services) into sector users within each country by making official use of firm-level data from the current economic census and industry surveys as well as customs transaction level data. (v) Constructing improved estimates of bilateral trade in services. An accurate assessment of value added in trade has to go beyond a single country’s effort, as it requires information on cross-border input–output rela- tionships. Therefore, an internationally coordinated approach is needed and it could best be achieved with an inter-secretariat approach that brings together a number of international agencies that are able to tap into their existing insti- tutional networks of official statistics. Otherwise, ‘practical problems would arise if each country were responsible for compiling its own value-added trade statistics’ (Yuskavage; see page 335). Constructing an annual IIO database is time consuming and resource inten- sive. An appropriate division of labour among major international agencies is necessary to make the best use of limited resources and avoid duplication of effort. National statistical agencies are the major source of raw data. More technical assistance and capacity building initiatives in developing countries, such as that by the Asian Development Bank, 5 can improve the statistical abil- ity of developing countries to fully implement 1993 SNA recommendations. This is also consistent with UNSD’s objective to improve national account and GDP estimation across countries. On 15 March 2012, the WTO and OECD launched a joint initiative: ‘mea- suring trade in value added’. 6 The work is designed to provide a means to develop these new metrics of trade on an ongoing and long-term basis. In order to improve the quality and timeliness of the estimates, the programme also seeks improvements in the inputs from national authorities. It will cap- italise on existing networks and build new ones. The agreement between the OECD and WTO is the most visible example, supported by collaboration with 5 The Asian Development Bank organised a project with the participation of 17 devel- oping countries (RETA 6483) in the Asia Pacific region to construct supply and use tables for each participating country. 6 See http://www.oecd.org/trade/valueadded. 16 Trade in Value Added other agencies, such as IDE-JETRO and US-ITC. The programme is designed to standardise and routinise the production of statistics on trade in value added, generating the global IIO table and value-added trade estimates periodically, and making them a permanent part of the statistical landscape. The first offi- cial release of major trade in value-added indicators was in January 2013. To summarise, the trade economist community and the trade policy world have reached a near consensus that official trade statistics are deficient and the deficiency grows with the deepening global division of production chains. There has been a burgeoning interest in developing new measures of both value added trade and the structure of double counted trade flows. We may be on the verge of breaking new ground in developing feasible new measures that can illuminate trade policy discussions. We hope this volume makes a contribution to that effort. REFERENCES Antras, P. (2005). Property Rights and the International Organization of Production. American Economic Review Papers and Proceedings May, 25–32. Daudin, G., C. Rifflart and D. Schweisguth. (2011) Who Produces for Whom in the World Economy? Canadian Journal of Economics 44(4), 1409–1538. Dean, J. M., and M. E. Lovely (2010). Trade Growth, Production Fragmentation, and China’s Environment, in China’s Growing Role in World Trade (ed. R. Feenstra and S. Wei). Chicago: NBER and University of Chicago Press. Feenstra, R., and G. Hanson (2005). Ownership and Control in Outsourcing to China: Estimating the Property-Rights Theory of the Firm. Quarterly Journal of Economics 120(2), 729–761. Grossman, G., and E. Rossi-Hansberg (2008). Trading Tasks: A Simple Theory of Off- shoring. American Economic Review 98(5), 1978–1997. Hummels, D., J. Ishii and K. Yi. (2001). The Nature and Growth of Vertical Specialization in World Trade. Journal of International Economics 54, 75–96. Kee, H.-L., and H. Tang. (2012). Domestic Value Added in Chinese Exports. Working Paper, World Bank and Tufts University. Koopman, R., Z. Wang and S.-J. Wei (forthcoming). Tracing Value-Added and Double Counting in Gross Exports. American Economic Review. A longer version is available as Working Paper18579. Maurer, A., and C. Degain (2010). Globalization and Trade Flows: What You See Is Not What You Get! World Trade Organization Staff Working Paper ERSD-2010-12, Geneva. National Board of Trade (2007). Adding Value to the European Economy: How Anti- Dumping Can Damage the Supply of Globalised European Companies: Five Case Studies from the Shoe Industry. Kommerskollegium, Stockholm. Rose, A. (2004). Do We Really Know That the WTO Increases Trade? American Economic Review 94(1), 98–114. SNA (1993). System of National Accounts. United Nations Statistics Division. URL: http://unstats.un.org/unsd/nationalaccount/sna1993.asp. Subramanian, A., and S.-J. Wei (2007). The WTO Promotes Trade, Strongly but Unevenly. Journal of International Economics 72(1), 151–175. 2 Towards the Measurement of Trade in Value-Added Terms: Policy Rationale and Methodological Challenges SÉBASTIEN MIROUDOT AND NORIHIKO YAMANO 1 Since the 1990s, a fundamental change has been taking place in the structure of world production and international trade. Production has become increas- ingly fragmented across countries that trade intermediate inputs before exporting final products. This reality for businesses is not reflected well in trade statistics, which attribute the full value of a good or service to the last country that contributed to its production, at the end of the ‘value chain’. Misperceptions associated with the inability to identify the country where value added originates can lead to misguided decisions. In the context of stalled multilateral trade negotiations, slow growth and continued economic uncertainty, this is important to provide a better understanding of sources of productivity and competitiveness at the international level and to encourage structural reforms that take into account the new landscape of international trade and global production. Several papers, workshops and conferences have now addressed the issue of the measurement of trade flows in the context of the fragmentation of world production. At this stage, what seems important is to provide a frame- work in which international organisations can build on the pioneering work done so far and move forward to provide new data and indicators that answer the concerns raised on standard statistics. Against this backdrop, this chapter has two objectives. The first one is to present the policy drivers that motivate the production of new trade statistics in value-added terms. The second is to address some of the methodological issues in such work, based on OECD experience in the compilation of a global input–output model of trade and production. The chapter concludes with the main challenges ahead. 1 This chapter draws on the work of the OECD Secretariat (Nadim Ahmad, Koen de Backer and Colin Webb) and the WTO Secretariat staff (Christophe Degain, Hubert Escaith and Andreas Maurer). The views expressed are those of the authors and do not necessarily reflect those of OECD or WTO member countries. 18 Trade in Value Added 1 WHY DO WE NEED NEW TRADE STATISTICS AND INDICATORS TO ACCOUNT FOR GLOBAL PRODUCTION NETWORKS? 1.1 The Issue with Conventional Trade Statistics With the globalisation of production, there is growing awareness that conven- tional trade statistics may give a misleading perspective of the importance of trade to economic growth and income and that “what you see is not what you get” (Maurer and Degain 2010). This reflects the fact that trade flows are mea- sured in gross terms, and so the value of products that cross-borders several times for further processing will also be included several times in trade flows. The past decades have been characterised by declining trade costs as a consequence of technological progress and trade policy reforms. Inventions such as the container ship or the Internet have revolutionised trade in several ways, but an important step was also service trade liberalisation. Key sectors that are part of the global logistics chain (transport, finance, telecommunica- tions, etc ) have seen their regulatory barriers reduced. This process led to the ‘fragmentation’ of production (Jones and Kierzkowski 2001), ie the possibility for firms to split up the production process in several countries to maximise the benefits of vertical specialisation. The emergence of global production networks and rise of trade in intermediates explain why there is increasing concern with the gross valuation of trade flows in current statistics. An often cited case study that illustrates the issue well relates to the pro- duction of an Apple iPod (Linden et al 2009). The study showed that of the $144 (Chinese) factory-gate price of an iPod, less than 10% contributed to Chinese value added, with the bulk of the components (about $100) being imported from Japan, and with much of the rest coming from the USA and Korea. Box 2.1 revisits the Apple example with the iPhone 4. Box 2.1. Who Bites the Apple? The iPhone Example Revisited. Several studies have illustrated the concept of value-added trade using Apple’s emblematic devices: first the iPod (Linden et al 2009) and then the iPhone (Xing and Detert 2010) and the iPad (Linden et al 2011). All these high- tech products are assembled in the People’s Republic of China, and so make a not insignificant contribution to China’s exports. But Chinese value added represents only a small share of the value of these electronic devices, which incorporate components from Germany, Japan, Korea and other economies that manufacture intermediate inputs. Based on estimates provided by iSuppli and Chipworks, Table 2.1 illustrates this by identifying those countries that provide intermediate inputs into the iPhone 4. But this does not tell the full story. The table only shows the value of the intermediate inputs produced by the firms, but they themselves will no doubt have used intermediate imports in their production or sourced intermediate Towards the Measurement of Trade in Value-Added Terms 19 Table 2.1: Countries that provide intermediate inputs into the iPhone 4. Country Components Manufacturers Costs ($) Chinese Taipei Touch screen, Largan 20.75 camera Precision, Wintek Germany Baseband, Dialog, Infineon 16.08 power management, transceiver Korea Applications LG, Samsung 80.05 processor, display, DRAM memory USA Audio codec, Broadcom, 22.88 connectivity, Cirrus Logic, GPS, memory, Intel, Skyworks, touchscreen Texas controller Instruments, TriQuint Other Other Misc. 47.75 Total 187.51 Source: Xing and Detert (2010), iSuppli, Chipworks. goods from domestic suppliers, who in turn would have used intermediate imports. Identifying these flows is equally important, particularly in the con- text of the example above, because some of those imports may have originated in China. Moreover, while the country indicated is the country where the firms producing the components are headquartered, these inputs are often pro- duced in other countries. Infineon, for example, has several factories in China. Chinese value added may therefore not only be limited to the assembly costs. To fully decompose the value added of the iPhone, and therefore ascribe it to individual countries, one cannot rely on a list of component suppliers. Information on all of the suppliers and their suppliers, and their suppliers’ suppliers, and so on, is equally important. What is needed therefore is a data set that is able to link production processes within and across countries: in other words, a set of international input–output tables with bilateral trade links (a global input–output table). Naturally, input–output tables developed by statistical offices the world over aggregate firms into groups (sectors) of firms that produce similar products, and thus input–output tables will not be able to reveal the total domestic value added generated by the production of an iPhone in any country. However, they will be able to provide such estimates for the whole economy and indeed by the sectors. The iPhone example also highlights that, beyond trade flows, more informa- tion on royalty payments and income flows is required to answer the question 20 Trade in Value Added of who benefits from trade. One should also look at ownership: Foxconn, the company that assembles iPhones in China, is a Chinese-Taipei-owned firm and the value added by mainland China in the example is split between wages paid to Chinese workers and income for shareholders in Chinese Taipei. Three main issues can be identified with conventional trade statistics. The first issue is the implicit multiple counting of intermediate goods and services. When world trade is calculated as an aggregation of all bilateral trade flows measured in gross terms, the value of the same primary or intermediate input is implicitly counted as many times as it crosses a border for further process- ing, reflecting its embodiment in the good as it goes through the processing chain. The second issue is perhaps the most important. The gross recording of trade flows and the fact that exports increasingly embody intermediate inputs sourced from abroad makes it difficult to identify the real contribution a given export may make to an economy’s material well-being, be that in terms of income or employment creation. Moreover, conventional trade statistics are not able to demonstrate those sectors of the economy where value added originates. In developed economies a large share of the total value added gen- erated by manufactured exports originates in the service sector, disentangling the domestic value chain into its sectoral components can therefore shed new light on the sources of international competitiveness. One final issue goes beyond ‘value added’, which has been the focus of most, if not all, of the contributions made so far. Value added in a national accounts sense reflects the compensation of labour and the compensation for produced and non-produced non-financial assets and natural resources used in production. However, measuring flows of value added reflects only part of the ‘global trade’ story. The fragmentation of production processes often involves fragmentation within a multinational enterprise. In that sense, part of value added, or at least part of what is referred to as operating surplus in the national accounts, may be repatriated to the parent company. This may be a straightforward transfer from the affiliate to the parent (recorded as distributed income) or it may reflect payments for the use of intellectual property products that are not recognised as produced assets in the national accounts. Either way, the point is that even estimates of value added in trade may not provide the full picture of the importance of trade to an economy. Increasingly, there is recognition that a focus on flows of value added embodied in trade flows provides more meaningful measures of the impor- tance of trade to economic growth. The underlying concept is in and of itself not relatively contentious, and there is widespread agreement that it reflects, for a given export, the percentage or amount of domestic value added that is generated by the export, throughout the production chain. In other words, any given export can be decomposed into value-added contributions from different domestic industries and different foreign industries. Towards the Measurement of Trade in Value-Added Terms 21 Measuring trade in value added closes the gap between research and statis- tics. The recent contribution of researchers to the understanding of interna- tional economics (the so-called new ‘new’ trade theory) emphasises the leading role of firms and business strategy in shaping international trade. In today’s industrial economy, dominated by global manufacturing and international supply chains, countries do not exchange goods, but ‘trade in tasks’. 2 Measur- ing trade in value added is a significant step in reflecting in official statistics the reality of economics capitalising on advances in academic research and data. A particular challenge is to disentangle domestic and foreign value added in the context of highly fragmented production networks where ‘circular’ trade takes place: inputs are shipped abroad and then come back as more processed products. Such a circular trade is particularly important in North America (especially between Mexico and the USA) and in Eastern Asia. National accounts do not provide a measure of domestic and foreign value added in trade flows. Therefore, input–output tables from different countries have to be harmonised and linked to create a global input–output table in order to estimate the share of domestic value added both in exported and in imported goods and services. In addition, when working on bilateral balances in value- added terms, one needs to fully decompose foreign value added according to the ultimate source country. Indeed, part of the value of the imports from the last known exporting country may originate from third countries (and even, as mentioned, include reimports from the domestic economy). This work requires a full set of inter-country input–output tables, where all bilateral exchanges of intermediate goods and services are accounted for. A last remark is that, despite their shortcomings for understanding inter- national trade linked to global production networks, traditional trade statis- tics tracking the physical movement of goods (gross accounting) remain fully relevant from an analytical point of view. The concept of ‘value added’ is use- ful in order to understand where economic activity and jobs are generated. But, on the demand side, gross trade flows tell us how much consumers have spent on imported goods and services. As consumers pay the full price in a single currency, the gross trade flows also matter in addressing currency or exchange rate issues, although, even here, some care is needed, as the goods and services recorded in conventional trade statistics do not always change ownership, particularly if the products are processed within affiliates of a multinational enterprise or they are, as is increasingly the case, sent abroad for further processing without any cash transaction occurring for the under- lying goods to be processed. 2 See Lanz et al (2011) for more on ‘trade in tasks’. 22 Trade in Value Added 1.2 A Brief Overview of the Literature on Trade in Value Added Although the literature on trade in value added is quite technical, it has attracted a lot of attention from policymakers. What could first look like a concern for trade statisticians is now understood as a key issue for the policy debate. For example, Word Trade Organization (WTO) Director-General Pascal Lamy notes that 3 the statistical bias created by attributing commercial value to the last country of origin perverts the true economic dimension of the bilateral trade imbal- ances. This affects the political debate, and leads to misguided perceptions. Even though global manufacturing through international supply chains may have became a major characteristic of international economy in the past 20 years, reflections about the global nature of production date from much earlier. A first intent to formalise it is attributed to Leontief in the 1960s (Leontief and Strout 1963). However, current reflections on the value-added content of international trade stem from two streams of economic literature. The first one deals with the importance of trade in intermediate goods and services. This is not a new topic, as Sanyal and Jones noted in their seminal 1982 paper that the bulk of international trade is in intermediate products and that trade in intermediates mainly consists not of raw material or pri- mary inputs but of products that have already received some value added (Sanyal and Jones call them ‘middle products’). Today, trade in intermediates accounts for about 56% of world trade in the case of goods and 70% in the case of services (Miroudot et al 2009). The growth of trade in intermediates has been highlighted in various recent surveys, in particular in Asia (see for exam- ple, Hayakawa 2007). Looking at trade in intermediate goods and services is the first step in the measurement of trade in value added. Following the definition introduced by Hummels et al (2001), the second stream of literature focuses on ‘vertical trade’. The latter expression refers to the vertical specialisation of trade, which is the consequence of the interna- tional fragmentation of production. There is vertical trade when three condi- tions are met: (i) a good (or service) is produced in two or more sequential stages; (ii) two or more countries provide value added during the production pro- cess; and (iii) at least one country uses imported inputs in the process and some of the output is exported. When taking into account both direct and indirect imported inputs, as sug- gested by Hummels et al (2001), the vertical specialisation (VS) share of world trade is about 25%. 3 ‘ “Made in China” tells us little about global trade’, Financial Times, 24 January 2011. Towards the Measurement of Trade in Value-Added Terms 23 The literature on vertical trade aims at measuring sequential trade in ver- tical production chains by looking at the import content of exports. Trade in value added is a broader concept but shares with this literature a com- mon concern: how can we distinguish the foreign and domestic value added in gross exports. Coefficients from imports and domestic matrices in input– output tables are used to operate this distinction. One issue that has been identified is the use of the same coefficients for the production sold on the domestic market and for exports, particularly in countries with a high level of ‘processing trade’, such as China (see Koopman et al 2008). The first papers to explicitly refer to the value added of trade (with some empirical measurement) are Daudin et al (2009), Johnson and Noguera (2012), Koopman et al (2010) 4 and Foster et al (2011). The first three studies rely on the Global Trade Analysis Project (GTAP) database to calculate trade flows in value added, while Foster et al (2011) is based on preliminary results from the World Input–Output Database (WIOD). Daudin et al (2009) identify ‘who pro- duces what and for whom’ by reallocating the value added contained in final goods to each country participating in their production. In addition to the VS share of Hummels et al (2001), the authors calculate the share of exports used as inputs to further exports and the domestic content of imports (that is, domestic value added that comes back to the country through intermediates originally exported and reimported within more processed products). John- son and Noguera (2012) present similar calculations, but based on a different decomposition of value-added exports. They focus on bilateral trade flows and calculate the ratio of value added to gross exports, a measure of the inten- sity of production sharing. As an illustration, they show that the US–China bilateral imbalance in 2004 is 30–40% smaller when measured in value-added terms. As opposed to Hummels et al (2001), their framework allows two-way trade in intermediates (each country can both import and export intermedi- ates, while in the VS framework the last country exports final goods only). Koopman et al (2010) provide a full decomposition of value-added exports in a single conceptual framework that encompasses all the previous measures. Exports are first decomposed into domestic value added, returned domestic value added (domestic value added that comes back incorporated in foreign inputs produced with domestic inputs) and foreign value added. Domestic value added is then split between exports absorbed by direct importers and indirect exports sent to third countries. By taking into account the returned domestic value added and the indirect exports to third countries, two sources of indirect value-added exports are taken into account and the decomposition is complete (thus matching standard trade data in gross terms when all the decomposed values are aggregated). Foster et al (2011) prefer, however, to focus on ‘net trade’ in value added to account for two-way trade in interme- 4 See Chapters 4 and 5 in this volume. 24 Trade in Value Added Components TWN Upstream input 229 207 suppliers 161 DEU USA CHN 800 ? KOR Assembly 413 65 ROW Final good 1875 Figure 2.1: The difference between US exports of intermediate inputs to China and US imports of assembled iPhones. Table 2.2: US trade balance in iPhones. US trade balance in iPhones with: CHN TWN DEU KOR ROW World Gross −1,646 0 0 0 0 −1,646 Value added −65 −207 −161 −800 −413 −1,646 diate inputs and to maintain the consistency between a country’s net exports in value-added and gross terms. Between the pioneering work of Hummels et al (2001) and these latest stud- ies, the conceptual framework has been greatly enhanced and we now have a better understanding of what constitutes trade in value-added terms. The field is therefore not only extremely relevant, but also mature to be included in official statistics (Escaith 2008). 1.3 Policy Drivers What can we expect from developing these new statistics on international trade? There are at least six areas where measuring trade in value added brings a new perspective and is likely to impact policy choices; the principal areas are as follows. Box 2.2. The Balance of Trade in Gross and Value-Added Terms (The iPhone Example Continued) It is easy to observe, that all calculations concerning the balance of trade are founded on very uncertain facts and suppositions. Hume (1985) To understand how the measurement of trade in value added affects bilat- eral trade balances, we can use the setting of the iPhone example described Towards the Measurement of Trade in Value-Added Terms 25 in Box 2.1. Assuming that 10 million iPhones are exported from China to the USA, the iPhone trade represents a trade deficit of $1,646 million for the US economy (this is simply calculated as the difference between US exports of intermediate inputs to China ($229 million) and US imports of assembled iPhones ($1,875 million; see Figure 2.1)). In gross terms, there is only a deficit between China and the USA. In value-added terms, one has to take into account that China adds a small share of domestic value added to the iPhone, corresponding to the value of the assembly work. As highlighted in the list of costs presented in Box 2.1, most of the components of the iPhone are sourced from outside China. Let assume that Chinese assembly costs are $6.50 per iPhone (and are part of the miscellaneous costs in Box 2.1). In value-added terms, Table 2.2 shows that the trade deficit is not only with China but also with Chinese Taipei, Germany, Korea and the rest of the world. The overall trade deficit (vis-à-vis the world) stays unchanged at $1,646 million. In this example, we do not take into account the suppliers of the suppliers. It is likely that what Chinese Taipei, Germany and Korea manufacture incorpo- rates further foreign inputs. The above calculation would have to be adjusted to fully take into account the value added by each country in the supply chain. This is why we need to add on the above figure upstream input suppliers and why the calculation can only be done if we have all the information about all the producers involved. • Global imbalances. Accounting for trade in intermediate parts and com- ponents and taking into account ‘trade in tasks’ does not change the overall trade balance of a country vis-à-vis the rest of the world, but it redistributes the surpluses and deficits across partner countries (see Box 2.2). When bilateral trade balances are measured in gross terms, the deficit with final goods producers (or the surplus of exporters of final products) is exaggerated because it incorporates the value of for- eign inputs. A WTO report calculates that the US–China trade balance in 2008 would be about 40% lower if calculated in value-added terms. 5 The true imbalance is therefore also with the countries who have supplied inputs to the final producer. As pressure for rebalancing increases in the context of persistent deficits, there is a risk of protectionist responses that would target countries at the end of global value chains on the basis of an inaccurate perception of the origin of trade imbalances. • Market access and trade disputes. Measuring trade in value added sheds new light on today’s trade reality, where competition is not between nations but between firms. Competitiveness in a world of global value chains means access to competitive inputs and technology. The 5 See Maurer and Degain (2010). Koopman et al (2010) find that the domestic value added of Chinese exports is on average 60%. 26 Trade in Value Added optimum tariff structure in such a situation is flat (little or no escalation) and reliable (contractual arrangements within supply chains, especially between affiliated establishments, tend to be long term). Outsourcing and offshoring of elaborate parts and components can only take place in situations where intellectual property is respected. Moreover, in the context of the fragmentation of production and global value chains, mercantilist-styled ‘beggar your neighbour’ strategies turn out to be ‘beggar thyself’ miscalculations. As mentioned, domestic value added is found not only in exports but also in imports: some goods and ser- vices are intermediates shipped abroad, whose value comes back to the domestic economy embodied in imports of foreign products. As a con- sequence, tariffs, non-tariff barriers and trade measures (such as anti- dumping rights) are likely to impact domestic producers in addition to foreign producers. For example, a study of the Swedish National Board of Trade on the European shoe industry highlights that shoes ‘manufac- tured in Asia’ incorporate between 50% and 80% of European Union value added. In 2006, anti-dumping rights were introduced by the European Commission on shoes imported from China and Vietnam. An analysis in value-added terms would have pointed out that EU value added was in fact subject to the anti-dumping rights (Swedish National Board of Trade 2007). • The impact of macroeconomic shocks. The 2008–9 financial crisis was characterised by a synchronised trade collapse in all economies. Many authors have discussed the role of global supply chains in the trans- mission of what was initially a shock on demand in markets affected by a credit shortage. In particular, the literature has emphasised the ‘bullwhip effect’ of global value chains (see Escaith et al 2010; Lee et al 1997). When there is a sudden drop in demand, firms delay orders and run down inventories, with the consequence that the fall in demand is amplified along the supply chain and can translate into a standstill for companies located upstream. A better understanding of value-added trade flows would provide tools to help policymakers anticipate the impact of macroeconomic shocks and adopt the right policy responses. Any analysis of the impact of trade on short-term demand is likely to be biased when looking only at gross trade flows. • Trade and employment. Several studies on the impact of trade liber- alisation on labour markets try to estimate the ‘job content’ of trade. Such analysis is only relevant if one looks at the value added of trade. What the value-added figures tell us is where exactly jobs are created. Decomposing the value of imports into the contribution of each econ- omy (including the domestic one) can give an idea of who benefits from trade. The EU shoe industry example can be interpreted in terms of jobs. Traditional thinking in gross terms would regard imports of shoes manufactured in China and Vietnam by EU shoe producers as EU jobs Towards the Measurement of Trade in Value-Added Terms 27 lost and transferred to these countries. But in value-added terms, one would have to account for the EU value added and while workers may have indeed lost their job in the EU at the assembly stage, there is a higher number of jobs in the research, development, design and mar- keting activities that exist because of trade (and the fact that this frag- mented production process keeps costs low and EU companies competi- tive). When comparative advantages apply to ‘tasks’ rather than to ‘final products’, the skill composition of labour embedded in the domestic content of exports reflects the relative development level of participat- ing countries. Industrialised countries tend to specialise in high-skill tasks, which are better paid and capture a larger share of the total value added. 6 • Trade and the environment. Another area where the measurement of trade flows in value-added terms would support policymaking is the assessment of the environmental impact of trade. For example, con- cerns over greenhouse gas emissions and their potential role in climate change have triggered research on how trade openness affects carbon dioxide (CO2 ) emissions. The unbundling of production and consump- tion and the international fragmentation of production require a value added view of trade to understand where imported goods are produced (and hence where CO2 is produced as a consequence of trade). An OECD study notes that the current relocation of industrial activities has a high impact on differences in consumption-based and production-based measures of CO2 emissions (Nakano et al 2009). • Trade, growth and competitiveness. Likewise, indicators of competi- tiveness such as the ‘revealed comparative advantage’ are affected by the measurement of trade in gross terms. Going back to the iPhone example, China seems to have a comparative advantage in producing iPhones on the basis of traditional trade statistics, while its comparative advantage is in assembly work. Bearing in mind development strategies and the concerns of policymakers to identify export sectors and pro- mote industrial policies, the analysis of the export competitiveness of industries cannot ignore the fragmentation of production and the role of trade in intermediates. The above examples make a compelling case for the production of trade statistics in value-added terms. There is no doubt that such analysis is highly relevant from a policy perspective. We believe that international organisa- tions should invest resources in the development and improvement of such trade statistics, in cooperation with national statistics offices and research projects. There are several challenges in producing statistics that would fully decompose the value of exports according to the country where it was added, 6 See WTO and IDE-JETRO (2011) for an illustration with global value chains in East Asia. 28 Trade in Value Added but such an exercise could enhance our understanding of trade and all areas where trade matters, starting with growth and jobs creation. 2 HOW TO CALCULATE THE VALUE-ADDED CONTENT OF TRADE? As emphasised in the previous section, measuring the value-added content of trade requires a global input–output table. Constructing such a table is data-intensive process and presents numerous challenges. In this section, we first describe the work undertaken at the OECD to harmonise single-country input–output (IO) tables and then apply multi-regional input–output model techniques to produce an inter-country input–output database that can be used to estimate trade in value-added terms. The rest of the section dis- cusses techniques to estimate bilateral trade flows of intermediate goods and services and explores how inter-country IO tables and trade statistics can be refined to produce more robust estimates of the value-added content of trade. 2.1 The Construction of Inter-Country Input–Output Tables The following steps describe how an inter-country input–output table is being built in the OECD. The data sources at OECD are harmonised input–output tables and bilateral trade coefficients in goods and services. The model spec- ification and estimation procedures can be summarised as follows. (i) Preparation of national IO tables for reference years using the latest published data sources eg supply and use tables, national account and trade statistics. (ii) Preparation of bilateral merchandise data by end-use categories for ref- erence years. The published trade statistics are adjusted for analyti- cal purposes (namely, confidential flows, re-exports, exclusion of waste and scrap products and manual adjustment of high-value valuables). Trade coefficients of utility services are estimated based on cross-border energy transfer. Other trade coefficients of service sectors are based on OECD Trade in Services and UN Service Trade statistics. However, many missing flows are filled by econometric model estimates; (iii) Conversion of cost, insurance and freight (CIF) price-based import fig- ures to free on board (FOB) price-based imports to minimise the incon- sistency issues of mirror trade (import = export) in the international IO system. (iv) Separation of import matrices of each country by cleaned trade coeffi- cients. (v) Total adjustment (missing sectors, trade with rest of the world, etc ) and minimisation of discrepancy columns using biproportional methods. Towards the Measurement of Trade in Value-Added Terms 29 Table 2.3: Country coverage of OECD Input–Output 2009 edition (as of May 2011). OECD Mid-1990s Early 2000s Mid-2000s Australia 1994/95 1998/99 2004/05 Austria 1995 2000 2005 Belgium 1995 2000 2005 Canada 1995 2000 2005 Chile 1996 — 2003 Czech Republic 1995 2000 2005 Denmark 1995 2000 2005 Estonia 1997 2000 2005 Finland 1995 2000 2005 France 1995 2000 2005 Germany 1995 2000 2005 Greece 1995 2000 2005 Hungary 1998 2000 2005 Iceland — — — Ireland 1998 2000 2005 Israel 1995 — 2004 Italy 1995 2000 2005 Japan 1995 2000 2005 Korea 1995 2000 2005 Luxembourg 1995 2000 2005 Mexico — — 2003 Netherlands 1995 2000 2005 New Zealand 1995/96 2002/03 — Norway 1995 2000 2005 Poland 1995 2000 2005 Portugal 1995 2000 2005 Slovak Republic 1995 2000 2005 Slovenia — 2000 2005 Spain 1995 2000 2005 Sweden 1995 2000 2005 Switzerland — 2001 — Turkey 1996 1998 2002 United Kingdom 1995 2000 2005 USA 1995 2000 2005 Harmonised Input–Output Tables for Reference Years The OECD has been updating and maintaining harmonised IO tables, split- ting intermediate flows into tables of domestic origin and imports, since the mid-1990s, usually following the rhythm of national releases of benchmark IO tables. The process of compiling OECD’s IO database greatly depends on cooperation with national statistical institutes. Ideally, national authorities provide the latest supply–use tables and benchmark symmetric input–output tables (SIOTs) at the most detailed level of economic activity possible, with a basic price valuation and, preferably, separating domestically produced and imported intermediate goods and services. The first edition of the OECD IO Database dates back to 1995 and covered 10 OECD countries with IO tables spanning the period from the early 1970s to the early 1990s. The first updated edition of this database, released in 2002, 30 Trade in Value Added Table 2.3: Continued. Non-OECD Mid-1990s Early 2000s Mid-2000s Argentina 1997 — — Brazil 1995 2000 2005 China 1995 2000 2005 Chinese Taipei 1996 2001 2006 India 1993/94 1998/99 2006/07 Indonesia 1995 2000 2005 Romania — 2000 2005 Russia 1995 2000 — South Africa 1993 2000 2002 Thailand — — 2005 Vietnam — 2000 — Malaysia∗ 2000 Singapore∗ 1995 2000 2005 A dash means that the available year data is not available. ∗ Not published (internal use only). increased the country coverage to 18 OECD countries, plus China and Brazil, and introduced harmonised tables for the mid-1990s. Tables are now available for 46 countries, 7 (33 OECD and 13 non-OECD countries) with tables for the mid-2000s (mainly 2005) now available for most of them (Table 2.3). The input–output tables show transactions between domestic industries, but as a complement to these tables are supplementary tables that break down total imports by user (industry and category of final demand). Some countries provide these import tables in conjunction with their input–output tables, but in some cases they are derived by the OECD Secretariat in produc- ing input–output tables directly from supply–use tables, which requires the use of assumptions that may have a significant impact on the results of trade in value-added analysis, particularly at the industry level. The main assumption used in creating these import matrices is the ‘propor- tionality’ assumption, which assumes that the share of imports in any prod- uct consumed directly as intermediate consumption or final demand (except exports) is the same for all users. Indeed, this is also an assumption that is widely used by national statistics offices in constructing input–output tables. Improving the way that imports are allocated to users will form a central part of future work of the OECD. This will require a better understanding of how countries estimate their import-flow matrices, and indeed an attempt to moti- vate better methods of allocation, at the national level, where possible. The industry classification used in the current version of the IO database is based on ISIC Rev. 3 (Table 2.4), meaning that it is compatible with other industry-based analytical data sets, and in particular with the OECD bilateral trade in goods by industry data set (derived from merchandise trade statis- tics via the standard Harmonized System to ISIC conversion keys). The system, 7 For more details, see http://www.oecd.org/sti/input–output. Towards the Measurement of Trade in Value-Added Terms 31 Table 2.4: OECD IO industry classification. ISIC Rev. 3 ISIC Rev. 3 code Description code Description 01,02&05 Agriculture, hunting, 40&41 Utility forestry and fishing 10–14 Mining and quarrying 45 Construction 15&16 Food products, beverages 50–52 Wholesale & retail trade; and tobacco repairs 17–19 Textiles, textile products, 55 Hotels & restaurants leather and footwear 20 Wood and products of 60–63 Transport and storage wood and cork 21&22 Pulp, paper, paper 64 Post & telecommunications products, printing and publishing 23 Coke, refined petroleum 65–67 Finance & insurance products and nuclear fuel 24 Chemicals 70 Real estate activities 25 Rubber & plastics products 71 Renting of machinery & equipment 26 Other non-metallic mineral 72 Computer & related products activities 27 Basic metals 73 Research & Development 28 Fabricated metal products, 74 Other business activities except machinery & equipment 29 Machinery & equipment, 75 Public admin. & defence; nec compulsory social security 30 Office, accounting & 80 Education computing machinery 31 Electrical machinery & 85 Health & social work apparatus, nec 32 Radio, television & 90–93 Other community, social & communication equipment personal services 33 Medical, precision & 95 Private households with optical instruments employed persons 34 Motor vehicles, trailers & semi-trailers 35 Other transport equipment 36&37 Manufacturing nec; recycling (including furniture) by necessity (ie to maximise inter-country comparability), is relatively aggre- gated. Differentiating between types of companies within a given sector is, however, essential in order to improve the quality of trade in value-added 32 Trade in Value Added Figure 2.2: Export share by industry and category: China, 1995 and 2009. Figure 2.3: Export share by industry and category: USA, 1995 and 2009. results (particularly in the context of exporting and non-exporting compa- nies), and so part of future work will be to explore ways of using microdata that could improve the quality of results (see Ahmad and Araujo 2011). Measuring Bilateral Trade in Intermediate Inputs Central to the construction of a global input–output database is the estima- tion of flows between countries. The OECD has developed a Bilateral Trade Database by Industry and End-Use Category (BTDIxE), 1988–2009, derived from OECD’s International Trade by Commodities Statistics (ITCS) database Towards the Measurement of Trade in Value-Added Terms 33 and the United Nations Statistics Division (UNSD) UN Comtrade database, where values and quantities of imports and exports are compiled according to product classifications and by partner country (Figure 2.2 for China and Figure 2.3 for USA). The OECD International Trade by Commodities Statistics (ITCS) database is updated on the basis of annual data submissions received from OECD mem- ber countries and, in some cases, from Eurostat. Due to the convergence of OECD ITCS and UNSD Comtrade 8 updating processes, data sharing and other related cooperation between the two organisations, tables can also be com- puted for non-OECD members as declaring countries, notably the countries which belong to the OECD Enhanced Engagement Programme, namely Brazil, China, India, Indonesia and South Africa. In ITCS and Comtrade, data are classified by declaring country (ie the coun- try supplying the information), by partner country (ie origin of imports and destination of exports) and by product (ie according to Harmonized System (HS)). In both data sources, trade flows are stored according to the product classification used by the declaring country at the time of data collection. In general, source data are held according to Standard International Trade Clas- sification (SITC) Rev. 2 for the time period 1978–87, the Harmonized System (1988) for 1988–95, HS Rev. 1 (1996) for 1996–2001, HS Rev. 2 (2002) for 2002–2006 and HS Rev. 3 (2007) from 2007 onwards. To generate estimates of trade in goods by industry and by end-use cate- gory, six-digit product codes from each version of HS from ITCS and Comtrade need to be assigned to a unique ISIC Rev. 3 industry and a unique end-use cat- egory according to the Broad Economic Categories (BEC) classification, and hence SNA basic classes of goods (see Table 2.5). Thus, eight sets of con- version keys have been estimated using classification correspondence tables, developed internally or available from UNSD. There are several thorny issues to be considered, including the following. • Confidential trade: there is currently a different treatment in ITCS and UNSD Comtrade. Standard conversion keys from HS do not account for confidential trade, although if defined at two-digit HS chapter level (eg the difference between reported two-digit data and sum of six-digit com- ponents) it can be allocated to ISIC and BEC codes. • Re-exports: adjustments are required for re-exports that are significant for major continental trading hubs. Sufficient data are available in order to adjust for reported trade between China and the rest of the world via Hong Kong, but not currently for other major hubs such as Belgium, the Netherlands and Singapore, and this will need to be investigated. • Identifying used/second-hand capital goods: HS codes, and thus report- ed trade in ITCS and Comtrade, cannot differentiate between new and 8 See http://unstats.un.org/unsd/comtrade/. 34 Trade in Value Added Table 2.5: Current BEC and SNA classes of goods. Classification by Broad Economic Categories SNA: Use class 1 Food and beverages 11 Primary 111 Mainly for industry Intermediate 112 Mainly for household consumption Final Consumption 12 Processed 121 Mainly for industry Intermediate 122 Mainly for household consumption Final Consumption 2 Industrial supplies not elsewhere specified 21 Primary Intermediate 22 Processed Intermediate 3 Fuels and lubricants 31 Primary Intermediate 32 Processed 321 Motor spirit Intermediate/Final Consumption 322 Other Intermediate 4 Capital goods (except transport equipment), and parts and accessories thereof 41 Capital goods (except transport equipment) Capital 42 Parts and accessories Intermediate 5 Transport equipment and parts and accessories thereof 51 Passenger motor cars Capital/Final Consumption 52 Other 521 Industrial Capital 522 Non-industrial Consumption 53 Parts and accessories Intermediate 6 Consumer goods not elsewhere specified 61 Durable Consumption 62 Semi-durable Consumption 63 Non-durable Consumption 7 Goods not elsewhere specified Not classified Source: UNSD, ESA/STAT/AC.124/8, New York, April 2007. old capital goods (such as second-hand aircraft and ships). Estimating international trade in these flows in a value-added context requires an elaboration of the input–output framework that allows these flows to be recorded in a way that aligns with total global value added produced in a given period. • Final consumption goods as intermediates: goods identified as con- sumer goods in the BEC/SNA classes may be used as intermediates in service activities, eg pharmaceuticals (medical services) and various foodstuffs (catering services), and it will be important to fine-tune the estimation here using feedback loops with input–output data. • Unidentified scrap and waste: certain types of waste and scrap do not Towards the Measurement of Trade in Value-Added Terms 35 have separate six-digit HS codes, eg PCs and other electrical equipment exported (often to developing countries) for recycling. While the development of a database of bilateral trade in intermediate inputs can provide a finer allocation of imports by exporting country to users (intermediate consumption, household final demand, and investment), this is only the first step. Improving the quality of inter-industry trade flows in the global input–output matrix requires further refinements. Two of those considered by the OECD are presented below. 2.2 How Can We Refine the Analysis? Improving the Quality of the Assumptions Used to Allocate Imports to Users The Trade by Enterprise Characteristics (TEC) exercise 9 is a joint project of the OECD and Eurostat which disaggregates trade values (imports and exports) according to the characteristics of trading firms. This is achieved by linking customs data and business statistics at the level of the firm and covers virtu- ally the entire population of a country’s business and (internationally) trad- ing population. Customs data provide volume and value and HS codes of the products traded at the six-digit level, together with the identification of the business entities involved in the international transaction. This information is then matched with company-level information available in countries’ busi- ness registers, which contain information on firm size and turnover, activity (industry) and ownership. Linking these two sources of firm-level information allows one to calculate firm-level value added and uncover the characteristics of the firms engaged in value-added creation through exports and/or imports. Thus, the TEC database provides a unique opportunity to further refine the quality of the import data used in the input–output tables and also to create sub-categories of industry groups that discriminate between export intensive, import intensive, import/export intensive firms and other firms, allowing for a more detailed understanding of international production networks. One of the challenges in using the TEC database in this way relates to the fact that many exporting and importing companies are classified according to the wholesale sector, even if the wholesaler just reflects the distribution or purchasing arm of a manufacturer. Linking these wholesalers to the man- ufacturing part of the company therefore will form an important part of the work. 9 More information on the TEC exercise can be found in the OECD Statistics Brief No. 16 (2011) and the Eurostat website: http://epp.eurostat.ec.europa.eu/statistics_explained/ index.php/International_trade_by_enterprise_%0Acharacteristics. The resulting database, which displays aggregate trade values due to confidential rules, is accessible through the OECD website: http://stats.oecd.org/Index.aspx via the submenu Globalisation/Trade by Enterprise Characteristics. 36 Trade in Value Added Constructing Improved Estimates of Bilateral Trade in Services This is perhaps one of the most challenging statistical issues faced in the construction of a global input–output table, as bilateral raw trade in services data is generally only available for most countries (in a comparable way) at the total services level. Some countries are able to provide breakdowns of trade in services using the Extended Balance of Payments (2002) breakdown (which has recently been revised, EBOPS 2008) but not typically on a bilateral basis. The EBOPS classification has a very weak correspondence with ISIC indus- tries used in input–output tables. Moreover, when a breakdown is available for EBOPS categories, a large share of trade remains unallocated (on average for OECD countries, disaggregated data total up to 70% of total trade flows). To construct an Estimated Bilateral Trade in Services by Industry database, the OECD is using both econometric estimations (based on gravity modelling) and optimisation techniques to decompose all bilateral trade flows according to the ISIC classification and consistently with imports of intermediate and final services as reported in national accounts. The TEC database also offers con- siderable potential scope for allocating international trade in services between industries when constructing global input–output tables. 3 CONCLUDING REMARKS: CHALLENGES AHEAD Estimating trade in value added is clearly of high policy interest and has been the subject of considerable analysis in recent years. There are several projects which aim to produce international input–output tables that can be used to calculate the domestic and foreign content of bilateral trade flows, such as the WIOD project previously mentioned, or the OECD project. There are also existing international IO tables that have been used to analyse trade in value- added terms, in particular the Asian IO tables from IDE-JETRO and the GTAP database. It is therefore important in the future to find some convergence on the way data are collected and estimated, and define best practices for both the data collection and the measurement methods. The identification of ‘best practices’, a common procedure in official statistics, would greatly reduce the cost of replicating and extending present initiatives. 10 Some of the above-mentioned projects are limited in time, and one concern should be to institutionalise the construction of trade statistics in value-added terms. This is why further cooperation should be encouraged between international organisations, as well as with national statistics offices and other research institutions, in order to complement the work that has already been done and 10 Some regions, such as Africa and Western Asia, are still absent from a systematic coverage based on official data, despite the fact that they would probably benefit most from a better understanding between vertical trade, trade in tasks and development. Towards the Measurement of Trade in Value-Added Terms 37 converge to a set of commonly accepted computation methods and imputa- tion techniques that could form, for the time being, the ‘best practices’ for estimating trade in value added. Clearly the key challenges in the immediate future concern the quality of trade statistics and the assumptions made to allocate imports to users (indus- tries/consumers). But the challenges do not stop there. There are a number of challenges that arise from the recent revision to the System of National Accounts (2008 SNA) and Balance of Payments Manual (BPM6) which provide the underlying basis for international trade transactions and indeed those recorded in input–output tables. Chief among these concerns are changes made to the recording of ‘goods sent abroad for processing’ and ‘merchant- ing’. But other important changes have been made too, such as the recogni- tion that ‘research and development (R&D)’ expenditures should be recorded as investment, which directly changes value added. Indeed, the recognition of R&D as investment shines a spotlight on other intellectual property products and on the importance of flows of income as opposed to only value added. Moreover, given the considerable advances made in the field, it also seems timely to consider whether the approach could be extended beyond measuring purely trade in value added and consider income flows. In this context there are two important, albeit related, issues that merit consideration. The first reflects payments for the use of intellectual property and the second reflects value added or income generated by foreign-owned firms. Getting some han- dle on these flows, which are not typically included in general trade statistics but are included in balance of payments statistics, is a logical next step in the work that starts with trade in value added. Finally, an important question is how the data on trade in value added should be conveyed to policymakers. Out of the construction of a global input– output table, we will have a tool that can be used to measure trade in value- added terms. But, concretely, this tool will be a series of matrices providing coefficients disaggregating trade flows according to the country from which the value added is sourced. While a single figure can summarise exports from country A to country B, switching to value added implies that we need to look at the contribution of all countries to exports from A to B. The challenge, therefore, is to come up with synthetic indicators that can be useful in trade policymaking. REFERENCES Ahmad, N., and S. Araujo (2011). Measuring Trade in Value-Added and Income using Firm-Level data. Mimeo. Ali-Yrkkö, J., P. Rouvinen, T. Seppälä and P. Ylä-Anttila (2011). Who Captures Value in Global Supply Chains? Case Nokia N95 Smartphone, ETLA Discussion Papers 1240, February. Daudin, G., C. Rifflart and D. Schweisguth (2009). Who Produces for Whom in the World Economy?, Document de travail de l’OFCE 2009-18, July. 38 Trade in Value Added Dedrick, J., K. L. Kraemer and G. Linden (2010). Who Profits from Innovation in Global Value Chains?: A Study of the iPod and Notebook PCs’. Industrial and Corporate Change 19(1), 81–116. Escaith, H. (2008). Nouvelle économie industrielle, commerce international et valeur ajoutée: implications statistiques et comptables, Actes du 12ème colloque de l’Association de comptabilité nationale, INSEE-Méthodes, pp. 225–251. Paris: INSEE. Escaith, H., N. Lindenberg and S. Miroudot (2010). International Supply Chains and Trade Elasticity in Time of Global Crisis, WTO Staff Working Paper ERSD-2010-08. Eurostat, IMF, OECD, UNCTAD, UNSD and WTO (2010). Manual on Statistics of Interna- tional Trade in Services (MSITS 2010). Washington DC: United Nations Publications. Foster, N., R. Stehrer and G. de Vries (2011). Patterns of Net Trade in Value Added and Factors’, Paper for the 7th EU-Framework Project ‘WIOD: World Input–Output Database: Construction and Applications’. Johnson, R. C., and G. Noguera (2012). Accounting for Intermediates: Production Shar- ing and Trade in Value Added. Journal of International Economics 86(2), 224–236. Jones, R., and H. Kierzkowski (2001). A framework for fragmentation. In Frag- mentation: New Production Patterns in the World Economy (ed. S. Arndt and H. Kierzkowski), pp. 17–34. New York: Oxford University Press. Hayakawa, K. (2007). Growth of Intermediate Goods Trade in East Asia. Pacific Eco- nomic Review 12(4), 511–523. Hume, D. (1985). Of the Balance of Trade. In Essays, Moral, Political and Literary (ed. E. F. Miller). Indianapolis: Liberty Classics. Original essay published in 1742. Hummels, D., J. Ishii and K.-M. Yi (2001). The Nature and Growth of Vertical Special- ization in World Trade. Journal of International Economics 54(1), 75–96. Koopman, R., Z. Wang and S.-J. Wei (2008). How Much Chinese Exports Is Really Made in China: Assessing Foreign and Domestic Value Added in Gross Exports. NBER Working Paper 14109. Koopman, R., W. Powers, Z. Wang and S.-J. Wei (2010). Give Credit to Where Credit is Due: Tracing Value Added in Global Production Chains. NBER Working Paper 16426. Lanz, R., S. Miroudot and H. Nordås (2011). Trade in Tasks, OECD Trade Policy Working Paper 117. OECD Publishing. Lee, H. L., V. Padmanabhan and S. Whang (1997). The Bullwhip Effect in Supply Chains. Sloan Management Review 38(3), 93–102. Leontief, W., and A. Strout (1963). Multiregional Input–Output Analysis. In: Structural Interdependence and Economic Development (ed. T. Barna), pp. 119–150. New York: St-Martin’s Press. Linden, G., K. L. Kraemer and J. Dedrick (2009). Who Captures Value in a Global Inno- vation Network? The Case of Apple’s iPod’. Communications of the ACM 52(3), 140– 144. Linden, G., K. L. Kraemer and J. Dedrick (2011). Who Captures Value in the Apple iPad?’. Mimeo, March. Maurer, A., and C. Degain (2010), Globalization and Trade Flows: What You See Is Not What You Get! World Trade Organization, Staff Working Paper ERSD-2010-12. Economic Research and Statistics Division, Geneva. Meng, B., N. Yamano and C. Webb (2010). Application of Factor Decomposition Tech- niques To Vertical Specialisation Measurements, IDE Discussion Paper 276. Institute of Developing Economies. Miroudot, S., Lanz, R. and A. Ragoussis (2009). Trade in Intermediate Goods and Ser- vices. OECD Trade Policy Working Paper 93. OECD Publishing. Nakano, S., A. Okamura, N. Sakurai, M. Suzuki, T. Tojo and N. Yamano (2009). The Mea- surement of CO2 Embodiments in International Trade. Evidence from Harmonised Towards the Measurement of Trade in Value-Added Terms 39 Input–Output and Bilateral Trade Database. OECD Science, Technology and Industry Working Paper 2009/3. Parikh, A. (1979). Forecasts of Input–Output Tables Using the RAS Method. Review of Economics and Statistics 61(3), 477–481. Sanyal, K., and R. Jones (1982). The Theory of Trade in Middle Products. American Economic Review 72(1), 16–31. Swedish National Board of Trade (2007). Adding Value to the European Economy. How Anti-Dumping Can Damage the Supply of Globalised European Companies. Five Case Studies from the Shoe Industry. Kommerskollegium, Stockholm. Walmsley, T. L., and C. Lakatos (2008). Regional Input–Output Data. In Global Trade, Assistance, and Production: The GTAP 7 Data Base (ed B. G. Narayanan and T. L. Walmsley), Center for Global Trade Analysis, Purdue University. WTO and Commission des Finances du Sénat (2011). Actes de la conference ‘Mon- dialisation des chaînes productives industrielles et mesure du commerce en valeur ajoutée’. Geneva. WTO and IDE-JETRO (2011). Trade Patterns and Global Value Chains in East Asia. Report. World Trade Organization and Institute of Developing Economies–Japan External Trade Organization, Geneva. Xing, Y., and N. Detert (2010). How the iPhone Widens the United States Trade Deficit with the People’s Republic of China, ADBI Working Paper 257, December. 3 The Importance of Measuring Trade in Value Added A: Imperatives from International Trade Theory Gene M. Grossman 1 INTRODUCTION I have been asked to talk about the importance of measuring trade in value added to international trade theory. To that end, I will briefly review some recent developments in trade theory, with an emphasis (of course) on my own work. I will discuss why the new models have been developed, how they have been constructed, and whether they make a difference to our theoretical conceptualisation of trade. Then I will talk about how these models might be ‘tested’ or calibrated for use in quantitative exercises. Here, the importance of measuring trade in value-added terms becomes obvious. 2 WHY A NEW THEORY? The ‘old trade theory’ associated with the names of Ricardo, Heckscher and Ohlin addressed the determinants of comparative advantage that lead coun- tries to specialise their production in some industries rather than others. The ‘new trade theory’ of the late 1970s and the 1980s brought a focus on prod- uct differentiation and increasing returns to scale, but still the emphasis was on trade in final goods or important intermediates, such as steel and textiles. It was standard to take the industry as the starting point for the analysis (described by a production function for a ‘good’), and to treat the object to be traded (automobile, clothing, wine) as technological given. This made sense until the ‘second unbundling’, to borrow the term coined by Richard E. Bald- win (2006). Baldwin’s first unbundling began centuries ago, but accelerated in the nine- teenth century. It was facilitated by technological advances in transportation, 42 Trade in Value Added which made it possible to separate the production of a good from where it was consumed. Following the industrial revolution and the innovations that it brought, workers became ever more specialised in performing particular tasks in the production of a good. Adam Smith described the British pin factory as a well-known example. The technological mandate for specialisation gave rise to an organisational innovation, namely the industrial ‘factory’. Workers came together under one roof to perform their separate tasks and to pro- duce a finished good, be it a final good or an important intermediate. With the improvements in land and sea transport, it became less and less essential that the factory be located close to the ultimate market. The finished goods could be traded across long distances. But communications at the time were no faster than the transportation. Correspondence was delivered by ship, rail or carriage, much like the finished goods. This made it impractical, if not impossible, to separate geographically the workers involved in producing some good. Production required the coor- dination of the various tasks. Coordination, in turn, required proximity. If an adjustment had to be made in the efforts of workers performing different tasks, it was not practical to wait a week or a month for the interaction to take place. This remained true, to a lesser extent, even after the telegraph and telephone made rapid communications of certain types possible. The second unbundling is associated with advances in communication and the IT revolution, and it is still unfolding. With the development of the fax, the email and a common communications protocol, and with high capacity computing power to govern information management, instructions can be delivered [almost] instantaneously, and coordination of production tasks can take place in real time. Workers can perform their tasks in different places, discuss the problems that arise via email or teleconference, make adjustments to product design and distribute new instructions to workers throughout the globe. Now, increasingly, production can be separated from production, just as production was separated from consumption in the first unbundling. Increasingly, international specialisation is no longer at the level of the good or industry, but at ever finer levels, perhaps even the task. Trade theory for- merly asked ‘Where will a particular be good produced?’ Increasingly, it must ask instead ‘Where will a particular task be performed?’ And, increasingly, the item that is to be traded is itself endogenous; it is possible to perform a greater or smaller number of tasks in a given location, thereby determining the margin at which ‘goods’ or value added are traded. 3 WHAT NEW THEORY? A number of authors have developed new models of trade in which the point at which trade takes place is endogenous. These models emphasise a finer division of the production process than was common in earlier trade theory, The Importance of Measuring Trade in Value Added 43 which incorporated only trade in final goods, or perhaps final goods and a single intermediate. Early examples of the new type of model include Dixit and Grossman (1982) and Feenstra and Hanson (1999). Given the shortness of time, I will mention just a few more recent examples. Grossman and Rossi-Hansberg (2008) conceptualise the production pro- cess as a large number (or continuum) of tasks. Each task must be performed by some factor of production; ie there are tasks for unskilled labour, tasks for skilled labour, tasks for capital, etc. Production of a unit of some good requires the performance of all of the tasks that go into its making. Some goods may use tasks performed by skilled labour more intensively, others tasks performed by unskilled labour. So, the model is a lot like the factor- proportions models familiar from neoclassical trade theory, except that the production function for the good has been replaced by a technology specified in terms of tasks. The key assumptions are that (i) tasks can be performed remotely, so that the production of a good can be internationalised, (ii) offshoring is costly in the sense that performing a task at a distance requires a greater factor input than if the task is performed nearby, and (iii) tasks differ in their costs of remote performance. In this setting, there is ‘trade in tasks’ as well as trade in final goods; in every industry, some tasks are performed locally (near the firm’s headquarters), while others are performed at a distance. The decision of what tasks to per- form offshore depends on factor prices in the home and foreign countries and on the communications technology. Improvements in communications reduce the cost of offshoring tasks in all industries and result in a more globalised production process for every good. Whereas our first paper focuses on globalisation of production among coun- tries at different stages of development—different technologies, different fac- tor endowments and different factor prices—our second examines task trade between similar countries (see Grossman and Rossi-Hansberg 2012). Why might countries divide the value added chain if they share similar technolo- gies and similar relative factor endowments? The answer we give is economies of scale. A country may become especially proficient at particular tasks that are performed there often. We study a model in which economies of scale at the task level are the only reason for offshoring, and show how country size interacts with costs of offshoring to determine the international pattern of specialisation. In a recent paper, Baldwin and Venables (2010) distinguish two types of pro- duction processes that affect the economics of fragmentation and globalisa- tion in the presence of shipping costs. If the engineering of a product dictates a ‘spider’ production process, multiple ‘limbs’ (parts) can be produced sepa- rately and then brought together to form a ‘body’ (assembly). If the product requires a ‘snake’ production process, the good moves in a linear fashion from 44 Trade in Value Added upstream to downstream, with value added at each stage. Baldwin and Ven- ables show how offshoring costs bind related stages (or tasks) together in the face of international factor price differences. They examine a stylised model of the spider and the snake to show how the forces that shape the location of different parts of the value chain differ in the alternative configurations of production. Costinot et al (forthcoming) provide an elegant model of what Baldwin and Venables term the snake. In their model, goods move from stage to stage with more value added at each one. At each stage of production, some fraction of output is lost due to production ‘errors’. Countries differ only in their proclivi- ties to error, perhaps reflecting the quality of their legal and other institutions. Costinot et al use the model to describe the equilibrium organisation of the value chain and the spillover effects of changes in production technologies (error rates) in some country. Yet another recent trade theoretic paper that addresses the global fragmen- tation of production is Garetto (forthcoming). She adopts an Eaton–Kortum framework to capture heterogeneity in the ability to produce intermediate inputs. Firms choose whether to outsource each of the many intermediate goods that are needed to produce a final good or to produce the input them- selves, and whether to source a good locally or from some foreign country. Organisational choices are governed by the trade-off between mark-up pricing for outsourced parts and the use of a possibly inferior in-house technology, and by the distribution of technologies and factor prices around the globe. Garetto examines the pattern of outsourcing that results and the implications of this globalised production for the gains from trade. The common feature of this recent trade theory is its emphasis on the deter- mination of the location of value added in a multi-stage or multi-task global production process. The theory addresses traditional questions, like ‘where does production take place?’ and ‘what effect does trade have on factor prices and the distribution of income?’, but the realities of world trade have shifted attention from the industry as the subject of analysis to a much finer level of economic activity. 4 DOES IT MATTER? The new theory focuses on the realities of modern-day global value chains. Does this theory lend any new insights? After all, as Mankiw and Swagel (2006) famously noted, [s]ervices offshoring … fits comfortably within the intellectual framework of comparative advantage built on the insights of Adam Smith and David Ricardo. Surely, the theory suggests a new set of factors that affects the location of economic activity. Grossman and Rossi-Hansberg (2008) emphasise the ease The Importance of Measuring Trade in Value Added 45 or difficulty with which a task can be performed at a distance, citing Blin- der (2006), Autor et al (2003) and Leamer and Storper (2001) for discussion of services that must be delivered personally or can be delivered electroni- cally, tasks that are more or less ‘routine’ and instructions that are ‘codifiable’ or not. Baldwin and Venables focus on the complementarity between tasks, some of which must be co-located for efficiency, while others can more read- ily be separated. Grossman and Rossi-Hansberg (2012) argue that tasks that are more costly to offshore may locate in larger countries, while Costinot et al (forthcoming) identify a force that drives downstream tasks to the more pro- ductive economies. All of these hypotheses about the location of activity are relatively new to trade theory, as these issues do not arise when a complete production process must be carried out in one place. Perhaps more interesting are the possible implications for the gains from trade. Mankiw and Swagel (2006) conjecture that it is obvious to economists that outsourcing simply represents a new form of international trade, which as usual creates winners and losers but involves gains to overall productivity and incomes. But Grossman and Rossi-Hansberg (2008) point to a ‘productivity effect’ from offshoring that can mitigate the distributional conflicts from trade. Whereas trade in final goods inevitably creates winners and losers in a world with factor-endowments à la Stolper and Samuelson, Grossman and Rossi- Hansberg show that, in a similar environment, offshoring possibly can gener- ate benefits for all. If domestic factors readily can move from performing tasks that are easy to offshore to other tasks that cannot so readily be performed at a distance, then improvements in communication technologies that facilitate greater offshoring generate aggregate productivity gains that are shared by domestic workers. This feature of task trade makes it different from goods trade in an important respect. The new theory also suggests some subtle forces that might influence national policy. In the face of learning externalities, a country might wish to focus more on tasks and occupations, and less on industries, than has been true in the past. Education policy that targets human capital development per- haps should take into account the ease of offshoring of the tasks that would be performed by workers with certain skills. In general, trade and industrial policy should be focused less on the industry and more on occupations and tasks. 5 HOW CAN WE TEST AND CALIBRATE? Some predictions of the new theories have been tested using labour market data. For example, Harrison and McMillan (2011), Ebenstein et al (2011) and Change OK? Hummels et al (2011) have examined the distributional effects of offshoring 46 Trade in Value Added and have investigated whether there might be a ‘productivity effect’ of the sort suggested by Grossman and Rossi-Hansberg. But efforts to test the new the- ories, and to investigate their quantitative implications, have been hampered by the lack of appropriate trade data, as we argued in our 2008 paper. As the papers presented at this workshop 1 make abundantly clear, trade data based on gross flows is increasingly inadequate as a basis for under- standing modern trade. The existence of task trade and global supply chains implies that contributions to the value of a final good come from many places. In order to understand the forces that shape the allocation of activity and also the effects of international specialisation on prices, incomes etc, it is crit- ical that we know where production is taking place. But, as national income accountants have known for decade, economic activity is best measured by value added not gross output. Who contributed to the production of a Toy- ota car or an Boeing jet? What part of the value chain was performed in each country? What was the pattern of task trade? It is simply impossible to know the answers to these questions with information about how many finished cars crossed international borders, or how many sales Boeing made to air- lines outside the USA. Take, for example, the hunt for a productivity effect of offshoring. The theory suggests a link between the pattern and extent of task trade and domestic factor rewards. Researchers investigating this link have been forced to rely on proxies for the amount and sources of foreign value added. Often approximations are used, such as the well known ‘proportionality assump- tion’. Global trade data on a value-added basis would obviate the need for proxies and approximations. Before the second unbundling, gross trade flows accurately measured much of world trade, because goods were predominantly produced in one place. Today, that is no longer true, and the gross flows mask the patterns of specialisation that we need to understand. In short, as Gross- man and Rossi-Hansberg concluded in 2008, the globalisation of production processes mandates a new approach to trade data collection, one that records international transactions, much like domes- tic transactions have been recorded for many years. REFERENCES Autor, D., F. Levy, and R. Murnane (2003). The Skill Content of Recent Technological Change: An Empirical Exploration. Quarterly Journal of Economics 118(4), 1279– 1334. Baldwin, R. E. (2006). Globalisation: The Great Unbundling(s). In Globalisation Chal- lenges for Europe. Helsinki: Office of the Prime Minister of Finland. 1 The World Bank Workshop on Fragmentation of Global Production and Trade in Value Added. The Importance of Measuring Trade in Value Added 47 Baldwin, R. E., and A. Venables (2010). Relocating the Value Chain: Offshoring and Agglomeration in the World Economy, NBER Working Paper 16111. Blinder, A. S. (2006). Offshoring: The Next Industrial Revolution? Foreign Affairs 85(2), 113–128. Costinot, A., J. Vogel and S. Wang (forthcoming). An Elementary Theory of Global Supply Chains. Review of Economic Studies. Dixit, A. K., and G. M. Grossman (1982). Trade and Protection with Multistage Produc- tion. Review of Economic Studies 49(4), 583–594. Ebenstin, A., A. Harrison, M. McMillan and S. Phillips (2009). Estimating the Impact of Trade and Offshoring on American Workers Using the Current Population Surveys. NBER Working Paper 15107. Feenstra, R. C., and G. H. Hanson (1999). The Impact of Outsourcing and High- Technology Capital on Wages: Estimates for the United States. Quarterly Journal of Economics 114(3), 907–941. Garetto, S. (forthcoming). Input Sourcing and Multinational Production. American Eco- nomic Journal: Macroeconomics. Grossman, G. M., and E. Rossi-Hansberg (2008). Trading Tasks: A Simple Theory of Offshoring. American Economic Review 98(5), 1978–1997. Grossman, G. M., and E. Rossi-Hansberg (2012). Task Trade Between Similar Countries. Econometrica 80(2), 593–629. Harrison, A., and M. McMillan (2011). Offshoring Jobs? Multinationals and US Manu- facturing Employment. The Review of Economics and Statistics 93(3), 857–875. Hummels, D., R. Jorgenson, J. Munch and X. Chong (2010). The Wage Effects of Outsourcing: Evidence from Danish Matched Worker–Firm Data. NBER Working Paper 17496. Leamer, E. E., and M. Storper (2001). The Economic Geography of the Internet Age. Journal of International Business Studies 32(4), 641–665. Mankiw, G., and P. Swagel (2006). The Politics and Economics of Offshore Outsourcing. Journal of Monetary Economics 53(5), 1027–1056. B: Why Measuring Value-Added Trade Matters for Developing Countries Judith M. Dean By allowing each country that is a member of the supply chain to special- ize in the part or component in which it has a comparative advantage, the internationalisation of supply chains creates enormous economic benefits. Lamy (2010) In the new millennium, trade economists have argued that the international fragmentation of production should bring significant benefits to developing countries. Unlike intra-industry trade,which is driven by similar incomes and preferences, the fragmentation of production internationally depends upon 48 Trade in Value Added differences in comparative advantage. The more production can be split glob- ally, and tasks dispersed based on comparative advantage, the more lower- income countries might be able to participate in these chains (Jones et al 2005; Arndt and Kierzkowski 2001). Thus, international production fragmen- tation should encourage trade between industrial and developing countries. The gains from trade should expand for all countries, since stages of produc- tion are allocated more efficiently. In addition, developing countries should now be able to expand their activities to include tasks within the production of high-tech or skill-intensive products, instead of waiting until they can effi- ciently produce the complete product. Have developing countries received some of the economic benefits from global supply chains? Value-added (VA) trade measures can help us answer this question. We know that the gains from trade (gains from specialisation, exchange, variety, etc ) cannot be measured directly by the value added in a country’s exports. But VA trade measures can contribute greatly to our under- standing of global supply-chain trade. In particular, they can help us answer four key questions. First, do developing countries actually participate in these supply chains? There is some evidence of global production networks in East Asia, and some recent evidence on China’s growing role in those chains. But data on partici- pation of other developing countries is still scarce. Second, even if they do participate, what role do firms in developing coun- tries play in these chains? The oft-cited work of Linden et al (2009) analysing the iPod supply chain suggests that China is involved in assembly activities that generate only a tiny part of the value added in the product. Recently, Chinese researchers presented evidence that China was nearly always at the ‘end of the value chain’, engaged in low-skilled labour-intensive activities in high-tech industries, such as pharmaceuticals and electronics (USITC 2011). Does this mean that China and other developing countries benefit little from global supply-chain trade? Third, will developing country participation in supply chains generate pro- tectionist sentiment in industrial countries? The fears that China was now competing in high-tech exports with the OECD (The Economist 2007) suggest that the benefits of supply-chain trade might be choked off by new trade barriers. Fourth, will the supply-chain trade come at the cost of the environment? After all, China’s rapid growth in supply-chain trade appears to have occurred concurrently with ever-worsening environmental degradation. Even if devel- oping countries do participate and benefit from global supply chains, might protectionism and environmental damage reduce or eliminate benefits in the longer run? Answering these questions about global supply-chain trade requires detail on the structure and nature of these chains. VA trade measures can potentially provide that detail. This chapter explores some of the recent evidence we have on developing country participation and position in global The Importance of Measuring Trade in Value Added 49 supply chains, and then discusses how VA trade measures could advance that discussion greatly. The chapter then examines how global supply-chain trade may help explain China’s ‘export similarity’ to the OECD, and discusses how more data on VA trade could not only help avoid protectionist responses, but promote more open markets. Finally, the chapter explores some recent evidence that supply-chain trade may have a beneficial impact on the envi- ronment in developing countries, and discusses how VA trade data could contribute to that debate. 1 ARE DEVELOPING COUNTRIES PARTICIPATING IN GLOBAL SUPPLY CHAINS? Until recently, evidence on developing country participation in global produc- tion networks has been scarce. A few studies have measured the importance of trade in parts and components in global, East Asian and Chinese trade, or China’s growing prominence in such trade. Jones et al (2005) found that world trade in parts and components grew by about 9% per year from 1990 to 2000, outstripping total world trade growth of 6.5% per year. Ng and Yeats (2003) found evidence of a strong network of Asian suppliers in the parts and compo- nents trade. Estimates for 1984–96 showed that Asian global exports of parts and components grew by more than 500%, compared with Asian total export growth of 300%. Using a similar approach, Athukorala (2009) and Athukorala and Yamashita (2006) found that the East Asian share of global exports of parts and components grew from 29.3% in 1992 to 39.2% in 2003. In fact, the share of components in East Asian intra-regional trade was far higher than its share in extra-regional trade. While this evidence is suggestive of Asian participation in supply-chain trade, it does not reveal which countries participate in which global chains, nor how tasks within a specific production supply chain are split up across countries. The pioneering study by Hummels et al (2001) took a step closer to accomplishing this. These authors combined input–output tables with trade data to measure vertical specialisation (VS), or foreign content, in a number of countries’ exports. A high VS share indicates that imported intermedi- ate goods make up a large proportion of the value of a country’s exports, potentially indicating a country’s greater degree of involvement in global production chains. Hummels et al measured not only the imported inputs used directly in producing an export, but also the indirect use of imported inputs in domestic intermediate goods used to produce that export. Their evidence showed that the foreign content of OECD exports grew significantly between the 1970s and 1990s. But their analysis focused mostly on industrial countries. According to Chinese Official Customs data, about half of China’s remark- able trade growth between 1995 and 2008 is attributable to Chinese pro- cessing trade–imports of intermediates that are further processed solely for 50 Trade in Value Added 80 Split 2002 70 Non-split 2002 60 Proc. exp. share 2002 50 % 40 30 20 10 0 Hong Kong USA Singapore Taiwan World Japan EU15 Canada Rep. Korea Australia/NZ Mexico Rest of Europe Rest of Southeast Asia Rest of World EU25 (excl. EU15) Brazil Rest of East Asia India Eastern Europe/Central Asia Rest of Latin Amer./Caribbean Middle Easr/North Africa Sub-Saharan Africa Rest of South Asia Former Soviet Union Figure 3.1: Foreign content share (%) of Chinese exports, 2002. Source: Dean et al (2011). export. Dean et al (2011) thus focused on measuring the VS share in Chinese exports, building on Hummels et al. They developed an improved method of identifying intermediates using both Chinese processing trade data and the UN Broad Economic Classification. Using this method, they found evidence of an extensive Asian network of input suppliers to China. In 2002, for example, Japan and the Tigers 1 accounted for half of China’s total imported intermedi- ates, with an additional 10% from other East and Southeast Asian countries. A similar pattern emerged for processing intermediate imports, with nearly 80% of imported intermediates coming from this Asian network. 2 Dean et al then used the official Chinese input–output table, and also a split Chinese input–output table developed by Koopman et al (2008) to calculate VS shares for Chinese exports by destination and by industry. The split table allows for the relatively high imported intermediate intensity of processing 1 Hong Kong, Singapore, South Korea and Chinese Taipei. 2 Deanet al (2009) describe in more detail the types of imported intermediates sourced from different supplier countries. The Importance of Measuring Trade in Value Added 51 exports compared with normal exports or domestic sales. As Figure 3.1 shows, using either the official or the split IO tables, China’s exports to industrial countries were found to have high foreign content, in contrast to its exports to developing countries. Together these findings suggest a picture of global supply chains in which intermediates are produced in Japan and the Four Tigers, then exported to China for processing, and ultimately exported by China to the USA and Europe. They also provide some evidence that supply-chain trade may indeed be larger between industrial and developing countries. But these VS measures only begin to tell us broadly about one part of the global supply chain for one country. VA trade measures can add much detail to this picture. Measuring VA trade in a specific product or industry could show which countries are participating in the production process and the stage at which they enter into the process. The structure of different industry chains could be traced more clearly, from the innovation stages of a product to its completion. This would allow a clearer view of the interdependence between specific industrial and developing coun- tries. Data over time could reveal when specific countries first become part of a specific global chain. Although VA trade does not measure the gains from trade, it could provide some indication of whether, and to what extent, devel- oping countries are able to participate in the new trade opportunities that international fragmentation offers. More disaggregated VA trade data might also allow analysis of the role of global chains in promoting indirect exports from developing countries. It is often argued that small- and medium-scale enterprises (SMEs) in developing countries are not able to obtain financing for exporting directly, or to sur- mount other informational obstacles to participate in global markets. How- ever, global supply chains might help promote SME participation in exporting, by opening up opportunities to contract as suppliers to global chains. SMEs could then participate in global trade indirectly, and allow the lead firms in the supply chain to handle the management, information and financing issues (OECD 2008). 2 WHAT ROLE DO DEVELOPING COUNTRIES PLAY IN GLOBAL SUPPLY CHAINS? What determines the position of a country within a supply chain? How can China and other developing countries ‘move up’ within a global supply chain? Trade theory would suggest that differences in comparative advantage should explain the allocation of tasks across countries. Thus, changing factor endow- ments should play a key role in any shift in a country’s firms to different activities within a supply chain. Research by Antras (2005), Feenstra and Hanson (2005) and others sug- gests that improved property rights and better quality control may also help 52 Trade in Value Added developing countries move up the supply chain. 3 When a product embod- ies extensive research and development (R&D) or intellectual property, and is new, firms may be less likely to offshore tasks, or to offshore them only through affiliates. This is due to the risk that intermediate goods may not be made to exact specification if contracted to independent firms, and/or that contracts and property rights may not be enforced. Once a product is more standardised, firms are both more likely to offshore tasks, and more likely to do so using independent contracts. Positive spillovers from participation in supply chains might also help developing countries move up the chain. Firms initially performing the least skilled tasks may learn through interaction with other firms in the chain, and be able to move to higher value activities. Indian software firms in the 1990s, for example, were largely in the middle to lower end of the software devel- opment chain, engaged in contract programming, coding and testing (Lateef 1997). Yet now some Indian firms engage in business and technology consult- ing, systems integration, product engineering, custom software development and other more skill-intensive activities. 4 The vertical specialisation data from Dean et al (2011) offer some support for the role of comparative advantage. Their VS share data show wide vari- ation in foreign content across industries. With the split input–output table, for example, foreign content estimates for 2002 Chinese exports were over 90% for computers and telecommunications equipment, suggesting that China was at the end of the value chain in IT-related sectors. In contrast, foreign content in Chinese metal products, general industrial machinery and paper (more capital-intensive sectors) was about 40–50%, and in textile production (a relatively labour-intensive sector) was only about 25%. Recent work by Dean and Fung (2009) offers some evidence on whether variation in vertical specialisation across China’s industries can be explained by R&D intensity. Using the Dean et al VS measures and a two-step estima- tion process, they analyse the amount of processing trade in a sector and the foreign content of that processing trade. Results show a strong negative cor- relation between R&D intensity and the share of Chinese processing exports in an industry’s exports. Given the level of processing exports, Dean and Fung find that R&D-intensive industries have relatively high foreign content in their processing exports. They also find that the possibility of producing via a for- eign affiliate increases the share of processing exports, even for relatively R&D-intensive industries. Together these results show some support for the idea that R&D intensive industries are more likely to retain control over most stages of the production chain, by either producing most of them domestically or producing them via a foreign affiliate. 3 For a survey of the literature, see Spencer (2005). 4 One example would be InfoSys, http://www.infosys.com/about/what-we-do/Pages/ index.aspx. The Importance of Measuring Trade in Value Added 53 0.6 China–OECD ESI to destination 1997 2002 0.4 SNG TWN HK PAK INDON KOR THAI SNG PAK PHIL HK CYP VN PHIL MYS 0.2 MYS THAI TWN US KOR US VN 0.2 0.4 0.6 China's VS share to destination Figure 3.2: Export similarity and vertical specialisation, 1997 and 2002. Source: Dean et al (2011). ESI, Export Similarity Index. Once again, measuring VA trade at the country–industry level could provide a much more comprehensive assessment of roles within production chains. These measures could help reveal which countries’ firms are engaged in which stages or tasks within a production chain. Over time, these data could help trace out changes in specialisation within a chain, or entrance into new supply chains. This might help us see whether developing countries are moving to higher value activities within specific global supply chains, and if so which ones. Such data could allow tests for the role of factor accumulation, property rights improvements and spillovers in explaining the position of developing countries at a point in time, and changes in their position over time. 3 HAS DEVELOPING COUNTRY PARTICIPATION IN SUPPLY CHAINS TRIGGERED PROTECTIONISM? During the last decade, international controversy and protectionist senti- ment arose in response to the perception that China was suddenly competing directly with the USA and other industrial countries in high-tech, sophisti- cated exports. Provocative research by Rodrik (2006) and Schott (2008) sug- gested that the bundle of goods exported by China to the USA closely resem- bled the export bundles of higher income, OECD countries and not developing countries at similar income levels. One interpretation of these results was that China has somehow leapfrogged over its traditional comparative advantage. A closer look suggests that international production fragmentation is a key factor in understanding this dramatic increase in the ‘sophistication’ of Chi- nese exports to the USA. The study by Dean, et al (2011) found that Chi- 54 Trade in Value Added nese exports to richer countries had a higher foreign content than Chinese exports to poorer countries. In addition, they found that a large share of Chi- nese imported inputs were sourced from Japan, with additional smaller shares sourced from the EU and the USA. Thus, Chinese exports to the USA might resemble those of other OECD countries because much of their value origi- nated in the OECD. Examining exports to nearly 200 destinations in 1997 and 2002, Dean et al found that Chinese and OECD exports differed dramatically across destinations. Where Chinese exports were similar to those of the OECD, they had high foreign content (Figure 3.2). This suggested that ‘sophistication’ arose from being part of a global supply chain. Econometric testing revealed that a higher share of foreign content in Chinese exports had a significant, positive impact on the similarity between Chinese and OECD exports. VA trade measures could help generate light instead of heat regarding global competition. With detailed data by sector, one could trace the sources of intermediates and semi-finished goods imported into a developing country like China. This would allow an assessment of how much domestic content is actually reimported by industrial countries in the form of finished goods from developing countries operating at the final stage in a production chain. The pattern of VA exports and imports would reveal much more about compara- tive advantage differences. Thus, some of what looks like export similarity in the gross export data would be revealed as differences in specialisation across tasks within a production chain. VA trade measures would also make more clear the interconnectedness of global production, and the importance of firms in each country in the supply chain. By providing information on these kinds of interdependence, VA trade data might encourage more open trade policy and more international effort towards trade facilitation. 4 WILL GROWTH IN SUPPLY-CHAIN TRADE BE HARMFUL TO THE ENVIRONMENT? China’s enormous trade and income growth since the mid-1990s has been concurrent with severe and growing environmental problems. One notable article described China as ‘choking on growth’ (Kahn and Yardley 2007). While major improvements have been made in pollution regulation during this time (OECD 2005), and some progress has been made in achieving cleaner water and air, ‘[r]elative shortage of resources, a fragile ecological environment and insufficient environmental capacity [have become] critical problems hindering China’s development’ (Ministry of Environmental Protection 2006). Thus, it is no surprise that China’s experience has fuelled the popular view that trade growth is harmful to the environment (Gardner 2008). Yet recent work by Dean and Lovely (2010) argues that global supply-chain trade may have had beneficial effects on China’s environment. A close look Table 3.1: Export shares, processing trade and pollution intensity by Chinese industrial sector, 2006. Share of total Proc. exports ISIC Rev. 3 Two Digit Sector mfg exports (% of sector exports) COD∗ SO∗ 2 Smoke∗ Dust∗ Top 5 manufacturing Communications equipment 16.95 83.55 0.03 0.02 0.01 0.01 industries by Office and computing machinery 14.69 95.94 0.03 0.02 0.01 0.01 export shares Textiles 8.06 20.28 0.53 0.50 0.21 0.01 Wearing apparel 7.83 24.76 0.25 0.27 0.13 0.01 Machinery 7.22 32.56 0.04 0.07 0.05 0.03 Top 5 manufacturing Non-metallic minerals 1.79 10.88 0.12 3.79 2.49 10.30 industries by Paper 0.53 56.82 5.34 1.47 0.72 0.04 pollution intensities Food products and beverages 2.26 24.12 1.09 0.41 0.33 0.02 Wood 1.07 17.72 0.44 0.95 0.76 0.32 Basic metals 4.83 13.87 0.08 0.85 0.33 0.48 ∗ kilograms per thousand yuan output (1995 yuan values). COD, chemical oxygen demand. Source: derived from Dean and Lovely (2010), and updated by the authors. The Importance of Measuring Trade in Value Added 55 56 Trade in Value Added at the data reveals that Chinese industries with the largest share of manu- facturing exports are not highly polluting. Meanwhile, those industries that are highly polluting account for relatively small shares of Chinese exports (Table 3.1). In fact, Chinese exports have been shifting over time towards highly fragmented sectors (office and computing machinery and communi- cations equipment) and away from traditional exports that are less frag- mented. Dean and Lovely find strong support for the fact that sectors heavily involved in processing exports are less polluting than those involved in ordi- nary exports. Dean and Lovely argue that the growth of processing trade could be benefi- cial for China’s environment in two ways. First, China has the opportunity to shift some resources into tasks within these relatively clean, relatively high- tech industries, and out of relatively dirtier industries. Second, as foreign investment grows within these fragmented industries, the costs of carrying out tasks within China should fall. This should expand the ranges of tasks undertaken in China. If the relatively dirtier tasks were originally done there, this expansion would bring in relatively cleaner tasks, lowering the average pollution intensity of Chinese activities within the fragmented industry. In addition, if the foreign-invested enterprises responsible for most of this trade bring greener technologies than those used by domestic producers, this will tend to make trade even cleaner. Dean and Lovely’s econometric analysis sug- gests that the amount of Chinese involvement in global supply-chain trade— proxied by the extent of processing exports—has played a key role in explain- ing the drop in the pollution intensity of Chinese exports over time. They find foreign direct investment inflows have contributed significantly to this decline, both indirectly through expanding processing exports and directly, presumably through cleaner technologies. More detailed analysis of these hypotheses at the industry level is hampered by the lack of good data on the actual range of activities or tasks done in China within an industry, and a measure of how that range of tasks expands. In the absence of these data, proxies such as the extent of processing exports are used. But this proxy captures neither foreign content relative to domestic content nor information on the position of the country’s firms within the production chain. VA trade flows could begin to fill this gap, by more clearly showing the foreign content of imported intermediates and the value added in a developing country’s exports within an industry or a product. Tracing this over time would provide a better proxy of changes in the range of activities/ tasks undertaken, and allow a more direct assessment of changes in average pollution intensity of those tasks. 5 CONCLUSION Developing countries have the potential for large benefits from the inter- national fragmentation of production. Participation in global supply chains The Importance of Measuring Trade in Value Added 57 opens up opportunities for diversification of productive activities into goods which would normally be outside a country’s comparative advantage. The ability to produce stages or tasks within these production chains expands the scope of a country’s comparative advantage, widening the gains from special- isation. Participation may also generate spillovers, through interaction with other members of the production chain or through learning by doing, that raise productivity. Increasing involvement in global supply chains may also mean two new channels through which trade might benefit the environment: shifting the composition of production and exports towards the cleaner, frag- mented industries; taking on cleaner tasks within an industry over time. VA trade measures do not directly capture the gains to developing countries from global supply-chain trade. But they can increase our ability to measure how much developing countries are participating in these chains, what tasks they undertake and how those tasks change over time. By tracing out the changing trade patterns between industrial and developing countries, and underscoring the interdependence of firms, they can also help to promote more open markets and better trade facilitation. Finally, by providing better measures of the range of activities carried out in specific countries, they can help in testing the potential environmental benefits of supply-chain trade. REFERENCES Antras, P. (2005). Incomplete Contracts and the Product Cycle. American Economic Review 95, 1054–1073. Arndt, S., and H. Kierzkowski (eds) (2001). Fragmentation. Oxford: Oxford University Press. Athukorala, P. (2009). The Rise of China and East Asian Export Performance: Is the Crowding-Out Fear Warranted? The World Economy 32, 234-266. Athukorala, P., and N. Yamashita (2006). Production Fragmentation and Trade Inte- gration: East Asia in a Global Context. North American Journal of Economics and Finance 17, 233–256. Dean, J. M., and K. C. Fung (2009). Explaining China’s Position in the Global Supply Chain, Joint Symposium on US–China Advanced Technology Trade and Industrial Development, 23–24 October 2009, Tsinghua University. Manuscript. Dean, J. M., and M. E. Lovely (2010). Trade Growth, Production Fragmentation, and China’s Environment, in China’s Growing Role in World Trade (ed. R. Feenstra and S. Wei). Chicago: NBER and University of Chicago Press. Dean, J. M., M. E. Lovely and J. Mora (2009). Decomposing China–Japan–US Trade: Verti- cal Specialization, Ownership, and Organizational Form. Journal of Asian Economics 20, 596–610. Dean, J. M., K. C. Fung and Z. Wang (2011). Measuring Vertical Specialization: The Case of China. Review of International Economics 19, 609–625. The Economist (2007). Leapfrogging or Piggybacking? The Economies of India and China Are Not As Sophisticated as They Appear. 8 November. URL: http://www .economist.com/node/10053145. 58 Trade in Value Added Feenstra, R., and G. Hanson (2005). Ownership and Control in Outsourcing to China: Estimating the Property-Rights Theory of the Firm. Quarterly Journal of Economics 120(2), 729–761. Gardner, T. (2008). Rich World Behind Much of Global Pollution. Reuters, 21 October. URL: http://www.reuters.com/article/2008/10/21/us-pollution-global -rich-idUSTRE49K9BB20081021. Hummels, D., J. Ishii and K.-M. Yi (2001). The Nature and Growth of Vertical Special- ization in World Trade. Journal of International Economics 54, 75–96. Jones, R. W., H. Kierzkowski and C. Lurong (2005). What Does Evidence Tell Us about Fragmentation and Outsourcing? International Review of Economics and Finance 14, 305–316. Kahn, J., and J. Yardley (2007). Choking on Growth, Part I. New York Times, 26 August. URL: http://www.nytimes.com/2007/08/26/world/asia/26china.html. Koopman, R., Z. Wang and S.-J. Wei (2008). How Much of Chinese Exports Is Really Made in China? Assessing the Domestic Value-Added When Processing Trade is Pervasive”, NBER Working Paper 14109. Lateef, A. (1997). Linking Up With The Global Economy: A Case Study of the Ban- galore Software Industry. New Industrial Organization Programme, ILO Discussion Paper 96/1997. Linden, G., K. Kraemer and J. Dedrick (2009). Who Captures Value in a Global Innova- tion Network? The Case of Apple’s iPod. Communications of the ACM 52 140–144. Ministry of Environmental Protection, 2006. Environmental Protection in China 1996– 2005. Information Office of the State Council, Beijing, People’s Republic of China. URL: http://www.china.org.cn/english/MATERIAL/170257.htm. Ng, F., and A. Yeats (2003). Major Trade Trends in East Asia: What Are Their Implica- tions for Regional Cooperation and Growth? World Bank, Policy Research Working Paper 3084. OECD (2005). Governance in China. Paris: OECD Publishing. OECD (2008). Enhancing the Role of SMEs in Global Value Chains. Paris: OECD Publish- ing. Rodrik, D. (2006). What’s So Special about China’s Exports? China and the World Econ- omy 14, 1–19. Schott, P. (2008). The Relative Sophistication of Chinese Exports. Economic Policy 53, 5–40. Spencer, B. (2005). International Outsourcing and Incomplete Contracts. Canadian Journal of Economics 38, 1107–1135. USITC (2011). Proceedings of the Joint Symposium on US–China Advanced Technol- ogy Trade and Industrial Development. Journal of International Commerce and Eco- nomics 3, 1–239. C: Implications for Macroeconomic Policy Mika Saito and Ranil Salgado 1 1 This part of the chapter was prepared for the World Bank Workshop on ‘Fragmentation of Global Production and Trade in Value Added: Developing New Measures of Cross Border Trade’, 9–10 June 2011. The Importance of Measuring Trade in Value Added 59 1 WHY IMPROVING VALUE-ADDED TRADE MEASUREMENT IS IMPORTANT Using accurate value-added trade data would improve exchange rate assess- ments. Real effective exchange rates based on value-added trade weights would reveal more accurate measures of competitiveness of a country than those based on gross trade weights. Switching to value-added trade weights could have potentially important implications; for example, some exchange rates that might be considered ‘misaligned’ using gross trade weights may no longer be so using value-added trade weights (or vice versa). Real effective exchange rates based on value-added trade would improve our estimates of the impact of changes in relative prices, including that on global rebalancing. For instance, the International Monetary Fund (2011) finds that a downstream (as opposed to upstream) position in a supply chain cushions the impact of relative price changes on both exports and imports. This reflects the higher foreign content in the downstream country’s exports, which miti- gates the impact of exchange rate changes (more detail is given below). Decomposing foreign value added (FVA) in exports by source country would help us understand how disruptions to supply chains can have spillover effects. Disruptions of trade flows could be either policy induced, such as preferen- tial/regional trade agreements, or naturally caused, such as the recent earth- quake in Japan. In either case, being able to track FVA by source would help us to understand the impact of disruptions in supply chains. Disruption of imports from a trading partner (eg Japan) does not necessarily mean that gross exports of a country (eg China) will fall by the share of that trading partner’s value added in the country’s exports (eg by Japan’s value-added share in China’s gross exports). The extent of the impact would depend on the nature of the shock and the availability of substitutes. Hence, the analysis needs to be supplemented by more disaggregated and higher frequency data than input–output data. Nevertheless, using value-added trade data would be a good starting point. Bilateral balances, if discussed for political economy considerations, are better measured with value-added, rather than gross, trade data. For example, the US trade balance with China has received a lot of attention in recent years. Indeed, the US bilateral trade deficit with China accounts for a large fraction of the overall US trade deficit: about 35% (Table 3.2). Many countries, however, export intermediate goods to China that are then processed and exported as goods to the USA (similarly, US exports contain FVA from China and other countries). Excluding FVA contained in exports to and from China reduces the size of US bilateral trade deficit to close to 20% of overall deficit. 60 Trade in Value Added Table 3.2: US trade balance (percent of GDP). 1995 2008 World −2.3 −5.8 Asia −1.8 −3.0 of which China −0.5 −2.0 of which China excl. FVA1 −0.4 −1.4 of which China excl. FVA from Asia2 −0.4 −1.7 Memorandum: US bilateral trade balance with China US data −0.5 −2.0 imports data only −0.4 −1.9 Chinese data −0.1 −1.2 1 FVA contained in exports to and from China is excluded. 2 FVA from Asia contained in exports to and from China is excluded. Source: Direction of Trade Statistics (DOTS); World Economic Outlook (WEO); OECD IO tables; and IMF staff estimates. Table 3.3: China’s external balance, 2008 (percent of GDP). Gross VA1 Trade balance 8.0 8.0 exports 31.7 23.0 imports −23.8 −15.1 Current account balance 9.6 9.6 Memorandum: FVA contained in exports 8.7 0.1 1 FVA contained in exports are excluded and imports are assumed to contain no domestic value added. Source: DOTS; WEO; OECD IO tables; and IMF staff estimates. 2 WHY IT IS NOT IMPORTANT Macroeconomists focus on overall balances, not bilateral ones: for the former, gross trade data already give us the correct information. For example, FVA contained in China’s exports is almost equivalent to China’s trade surplus in 2008, about 8% of GDP (Table 3.3). Does this mean that the trade balance, the current account balance and hence the saving-investment gap should be adjusted downwards, for example, implying that discouraging net savings in China is the wrong policy recommendation? No! If we were to record exports excluding FVA, then we should also make corresponding adjustments on the import side, leaving external balances the same. Similarly, if we were to record imports excluding domestic value added (DVA), then corresponding adjust- ments must be made on the export side. With these changes, however, the balance of payments statistics will record value-added flows that are differ- ent from the actual gross transactions. Is this feasible or even desirable, espe- The Importance of Measuring Trade in Value Added 61 30 25 Total exports Non-commodity exports 20 % 15 10 5 0 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 Figure 3.3: World exports relative to production (percent of GDP). Sources: DOTS, WEO and UN Comtrade. The ratio for 1949–61 is calculated based on 15 major exporters. cially when gross trade data already provide us with correct overall external balances? In summary, value-added trade data can be used to complement, and not necessarily substitute for, the gross trade data. The reminder of this chapter summarises relevant findings of International Monetary Fund (2011). 3 CHANGING PATTERNS OF GLOBAL TRADE The growth of trade relative to output in the last few decades was in large part driven by the emergence and growth of global supply chains. As a share of global output, trade is now more than four times its level in the early 1950s (Figure 3.3). This partly reflects trade liberalisation since then, which led to significantly lower trade barriers in advanced economies, followed more recently by developing countries. Along with lower trade barriers, technology- led declines in transportation and communication costs also facilitated the fragmentation of production beyond national borders. These developments led supply chains to become regional, as in the case of ‘Factory Asia’ (Baldwin 2008) or even global, as in the case of the iPod (Dedrick et al 2010). A conver- gence in income levels and factor endowments across countries also played a role in the growth of trade (relative to output), especially that of intra-industry trade. The FVA share in gross exports has almost doubled since 1970, and the growth in FVA share has accelerated in recent years. Updates by IMF staff on the 62 Trade in Value Added Table 3.4: Share of foreign value added in gross exports. Hummels et al (2001)1 Update2 1970 1990 1995 2005 FVA share of gross exports 0.18 0.24 0.27 0.33 Growth in FVA share 31.3 21.5 Contribution of FVA exports 32.5 55.9 to growth in exports/GDP 1 Twenty-eight countries are included in Hummels et al (2001): Australia, Canada, China, EU15, Hong Kong SAR, Indonesia, Japan, Korea, Mexico, Taiwan Province of China (Chinese Taipei), Malaysia, Singapore, Thailand and the USA. 2 The 34 countries included in the update are EU15, Australia, Brazil, Canada, Switzerland, China, Czech Republic, Hungary, Indonesia, India, Israel, Norway, New Zealand, Poland, Russian Federation, Slovak Republic, Slovenia, Taiwan Province of China (Chinese Taipei), Turkey and the USA. Sources: Hummels et al (2001); IMF staff estimates using OECD input–output tables. work by Hummels et al (2001) show that the foreign content embedded in gross exports have increased on average from 18% in 1970 to 33% in 2005 (Table 3.4). 2 Growth in FVA share also accelerated in recent years; the growth rate was 10% per decade during 1970–90, but was 20% per decade during 1995–2005. Advanced economies and emerging market economies (EMEs) play different roles in global supply chains. Advanced economies tend to be upstream in the supply chain. This position is reflected in relatively small FVA in exports and relatively large contributions to value added in exports of downstream countries (Koopman et al 2010). By contrast, EMEs tend to be downstream in the supply chain, with relatively large shares of imported content in their exports (Figure 3.4). The different positions in the supply chain lead to dif- fering implications for the sensitivity of trade patterns. For example, at the aggregate level, the impact of a exchange rate fluctuations on trade is more cushioned for downstream countries than upstream ones (more details are given below). The Asia supply chain is more integrated than those in North America or Europe. In the Asian supply chain, goods in process cross borders several times, including through the hub (Japan), before reaching their final destina- tion (Table 3.5). For instance, about 15% of Japanese value added embodied in Chinese products goes through other countries in Asia before reaching China. In contrast, in other regions, almost all foreign input is imported directly from the hub: the USA in the North American Free Trade Agreement (NAFTA) coun- 2 The update (last two columns) is based on 34 countries, while the original figures by Hummels et al (2001) for 1970–90 (first two columns) are based on 28 countries. The Importance of Measuring Trade in Value Added 63 1200 0.278 0.123 (0.135) 1000 Service Gross exports (US$ billion) 0.274 ROW 800 USA 0.095 0.208 0.152 (0.135) EA 600 0.082 Non-EA OEA 400 JPN 200 0.155 CHN DVA 0 1995 2005 1995 2005 1995 2005 1995 2005 CHN JPN USA DEU Figure 3.4: Foreign contents in gross exports. Sources: IMF staff estimates using OECD input–output tables, Comtrade and OECD STAN data. Shares above the bar chart indicate FVA share in gross exports. Shares in parentheses exclude FVA from the euro area. Table 3.5: Hub’s VA contained in gross exports. In imports from In imports from Total the hub1 the neighbours2 China 8.0 6.8 1.2 Mexico 31.3 31.0 0.3 EU accession 17.5 17.3 0.2 1 ForChina, Mexico and EU accession countries, hubs are Japan, the USA and the EU, respectively. 2 For China: Australia, Hong Kong, India, Indonesia, Korea, Malaysia, the Philippines, Singapore, Taiwan (Chinese Taipei), Thailand, Vietnam and the rest of East Asia are included. For Mexico: Canada, Brazil and Latin America are included. For EU accession countries: EFTA, and Russia are included. Source: IMF staff estimates using Koopman et al (2010). tries and EU15 in Europe. 3 The greater integration of production in the Asia renders it potentially more vulnerable to disruptions of trade flows, whether policy induced, such as preferential trade agreements, or naturally caused, such as the recent Japan earthquake. The emergence of global supply chains has allowed EMEs to enhance the tech- nology content of their exports, including inputs embedded in high-tech exports 3 NAFTA refers to the North American Free Trade Agreement countries: Canada, Mexico and the USA. EU15 is the 15 member states of the European Union prior to 2004. 64 Trade in Value Added 180 0.485 160 Gross exports (US$ billion) Service 140 0.100 ROW 120 0.174 0.312 USA 100 EA 80 0.166 Non-EA 60 0.215 OEA 0.241 40 JPN 0.201 CHN 20 DVA 0 1995 2005 1995 2005 1995 2005 1995 2005 CHN JPN USA DEU Figure 3.5: Foreign contents in gross exports: high-tech sectors. Sources: IMF staff estimates using OECD input–output tables, Comtrade and OECD STAN data. Shares above the bar chart indicate FVA share in gross exports. Change in gross exports (US$ billion) 900 800 0.368 Service 700 ROW USA 600 0.331 EA 500 Non-EA 400 OEA 300 JPN 200 0.443 0.256 CHN 100 0.994 DVA 0 Mfg LT MLT MHT NT Figure 3.6: Source of change in exports of advanced countries (1995–2005). Sources: IMF staff estimates using OECD input–output tables, Comtrade and OECD STAN data. Shares above the bar chart indicate the change in FVA exports in the overall change in exports. Mfg, manufacturing; LT, low technology; MLT, medium-low technology; MHT, medium-high technology; HT, high technology; EA, EU accession countries. of advanced countries. 4 The share of high-tech exports such as computers and office equipment has increased remarkably in China since 1995, boosted 4 The classification is based on the OECD measure of trade by technology intensity (Organisation for Economic Co-operation and Development 2005). The Importance of Measuring Trade in Value Added 65 Table 3.6: Simulated long-run impacts of relative price shocks on external balances: base year = 2008 (percent of national GDP, unless otherwise noted). 10% appreciation 10% depreciation China Euro Area1 Japan USA Pre- Post- Pre- Post- Pre- Post- Pre- Post- shock shock2 shock shock2 shock shock2 shock shock2 Current account balance 9.6 5.9 −1.7 −4.7 3.2 6.4 −4.7 −2.4 of which trade balance 8.0 4.2 −0.6 −3.6 0.8 3.9 −5.8 −3.5 exports 31.7 28.9 17.0 15.1 15.3 17.5 9.1 10.2 imports −23.8 −24.7 −17.5 −18.6 −14.5 −13.5 −14.9 −13.8 Memorandum items (percent change from pre-level): exports (%) −10.9 −12.7 17.0 13.7 imports (%) 1.7 4.5 −4.5 −6.7 nominal GDP (%) −2.3 −1.8 2.2 1.0 1 Euro Area trade data was obtained from the IMF Direction of Trade Statistics. 2 Trade levels implied in the long run by simulated relative international price shocks are in absence of other shocks. Sources: WEO, DOTS and IMF staff estimates. by processing trade and with significant imported contributions from Japan and other Asian countries (Figure 3.5). China is also moving upstream in the value-added chain, with imports from China contributing significantly to high-tech exports of advanced countries (Figure 3.6). With China and other EMEs increasing their presence in sectors traditionally dominated by advanced economies, the similarity in export structures has increased over time and so has competitive pressure. Changes in relative prices would result in non-symmetric rebalancing effects between downstream and upstream countries, as different sizes of FVA shares at the sectoral (and aggregate) level lead to different adjustment patterns. We have examined the impact of relative price changes on trade structures of four key players in global trade, namely China (downstream country), the Euro Area, Japan and the USA (upstream countries). At the aggregate level, a downstream (as opposed to upstream) position in a supply chain cushions the impact of a relative price change on both exports and imports (Table 3.6). This reflects the higher FVA in the exports of the downstream country miti- gating the impact of exchange rate changes. At the sectoral level, the impact on technology intensity of exports is different for each country (Figure 3.7). In China, the impact is most prevalent in both high-tech and medium-tech exports, while elsewhere medium-high-tech exports change the most. Adjust- ments are smaller for sectors with larger FVA shares, though the size of the sector in each country also matters. Finally, relatively more adjustment in the trade balance seems to take place outside of the supply chain, as exports to supply chain partners are more resilient to relative price changes (Figure 3.8). 66 Trade in Value Added 18 16 Simulated impact (percent change) 14 12 10 HT 8 MHT MLT 6 LT 4 2 0 CHN JPN USA Euro Figure 3.7: Simulated impact of exports by sector. Sources: UN Comtrade and IMF staff estimates. Euro area China Japan USA 0 10 20 30 40 50 Share of TB/GDP adjustment within supply chain Supply chain share in total trade Figure 3.8: Contribution to adjustment in trade balance. Sources: UN Comtrade and IMF staff estimates. Supply chain contribution to adjust- ment in trade balance is smaller than its importance as a trading partner would sug- gest. REFERENCES Baldwin, R. (2008). Managing the Noodle Bowl: The Fragility of East Asian Regionalism. Singapore Economic Review 53(3), 449–478. Dedrick, J., K. L. Kraemer and G. Linden (2010). Who Profits from Innovations in Global The Importance of Measuring Trade in Value Added 67 Value Chains: A Study of the iPod and Notebook PCs. Industrial and Corporate Change 19(1), 81–116. Hummels, D, J. Ishii and K.-M. Yi (2001). The Nature and Growth of Vertical Special- ization in World Trade. Journal of International Economics 54, 75–96. International Monetary Fund (2011). Changing Patterns of Global Trade. IMF Policy Paper. URL: http://www.imf.org/external/pubs/cat/longres.aspx?sk=25104.0. Koopman, R., W. Powers, Z. Wang and S.-J. Wei (2010). Give Credit Where Credit Is Due: Tracing Value Added in Global Production Chains. NBER Working Paper 16426, September. Organisation for Economic Co-operation and Development (2005). Measuring Global- isation: OECD Handbook on Economic Globalisation Indicators (Paris: OECD). 4 Accounting for Intermediates: Production Sharing and Trade in Value Added ROBERT C. JOHNSON AND GUILLERMO NOGUERA 1 Trade in intermediate inputs accounts for as much as two-thirds of interna- tional trade. By linking production processes across borders, this input trade creates two distinct measurement challenges. First, conventional gross trade statistics tally the gross value of goods at each border crossing, rather than the net value added between border crossings. This well-known ‘double-counting’ problem means that conventional data overstate the domestic (value-added) content of exports. Second, multi-country production networks imply that intermediate goods can travel to their final destination by an indirect route. For example, if Japanese intermediates are assembled in China into final goods exported to the US, then Chinese bilateral gross exports embody third party (Japanese) content. Together, ‘double-counting’ and multi-country production chains imply that there is a hidden structure of trade in value added under- lying gross trade flows. In this chapter, we compute and analyse the value-added content of trade. To do so, we require a global bilateral input–output table that describes how particular sectors in each destination country purchase intermediates from both home and individual foreign sources, as well as how each country sources final goods. Because these bilateral final and intermediate goods linkages are not directly observed in standard trade and national accounts data sources, we construct a synthetic table by combining input–output tables and bilateral 1 We thank the editor, Daniel Trefler and two anonymous referees for comments that improved the chapter. We thank Rudolfs Bems, Judith Dean, Stefania Garetto, Pierre-Olivier Gourinchas, Russell Hillberry, David Hummels, Brent Neiman, Nina Pavcnik, Esteban Rossi- Hansberg, Zhi Wang and Kei-Mu Yi, as well as participants in presentations at the Federal Reserve Board, Hamilton College, Harvard University, the International Monetary Fund, the Philadelphia Federal Reserve, Princeton University, the US International Trade Com- mission, University of California (Berkeley), University of Cape Town, Wesleyan University, the 2009 FREIT Empirical Investigations in International Trade conference and the 2010 AEA Meetings, for helpful conversations. Noguera gratefully acknowledges financial sup- port from UC Berkeley’s Institute for Business and Economic Research. This chapter is based on Johnson and Noguera (2012). 70 Trade in Value Added trade data for many countries. Using this table, we split each country’s gross output according to the destination in which it is ultimately absorbed in final demand. We then use value added to output ratios from the source country to compute the value added associated with the implicit output transfer to each destination. The end result is a data set of ‘value-added exports’ that describes the destination where the value added produced in each source country is absorbed. These data on the value-added content of trade have many potential uses. Most directly, we compare them to gross bilateral trade flows to quantify the scope of production sharing. This approach to measuring production shar- ing yields comparable figures for many countries and sectors and respects the multilateral structure of production sharing. Further, because we use the national accounts definition of intermediates, our measures are easily trans- lated into models. 2 This is important because the value-added content of trade is a key theoretical object and calibration target in many trade and macro- economic models. For example, value-added exports can be used to calibrate ‘openness’ and bilateral exposure to foreign shocks in international business cycle research. 3 For trade research, value-added flows could be used to cal- ibrate gravity-style trade models to allow for differences in trade patterns for final and intermediate goods. 4 They could also be employed to calibrate many-country models of multi-stage production and vertical specialisation, as in Yi (2003, 2010). And these applications only scratch the surface. Our approach to measuring the value-added content of trade draws on an older literature on input–output accounting with multiple regions. Our method of tracking the flow of intermediate inputs across borders was initially developed by Trefler and Zhu (2010), who in turn built on the older multi- regional input–output literature (see Isard 1951; Moses 1955, 1960; Miller 1966). Trefler and Zhu use their procedure to track the movement of each intermediate input across each border and then use this information to cal- culate the factor content of trade, ie the amount of primary factors such as labour that are embodied in the trade of intermediate and final goods. In con- trast, we use their tracking procedure as a first stage in calculating the value- added content of trade, ie the value of primary factors that are embodied in the trade of intermediate and final goods. 5 2 This contrasts with alternative approaches, such as using data on trade in parts and components (see, for example, Yeats 2001) or trade between multinational parents and affiliates (see, for example, Hanson et al 2005). 3 See Bems et al (2010) for elaboration of this argument. 4 SeeNoguera (2011) for an analysis of estimated trade elasticities in gravity models with and without intermediate goods. 5 Belke and Wang (2006) and Daudin et al (2011) also develop value-added trade com- putations along the lines of those used in this chapter. See also Powers et al (2009) on splitting up the value chain within Asia. Accounting for Intermediates 71 Our work is also related to an active literature on measuring vertical special- isation and the domestic content of exports. 6 Aggregating across sectors and export destinations for each source country, the ratio of value added to gross exports can be interpreted as a metric of the domestic content of exports. 7 Our domestic content metric generalises the work by Hummels et al (2001), who compute the value-added content of exports under the restrictive assumption that a country’s exports (whether composed of final or intermediate goods) are entirely absorbed in final demand abroad. That is, it rules out scenar- ios in which a country exports intermediates that are used to produce final goods absorbed at home. By using input–output data for source and destina- tion countries simultaneously, we are able to relax this assumption. While this generalisation results in only minor adjustments in aggregate domestic con- tent measurements in our data, we demonstrate that relaxing this assumption is critically important for generating accurate bilateral value-added flows. Turning to our empirical results, we find that the ratio of value added to gross exports (VAX ratio) varies substantially across countries and sectors. Across sectors, we show that VAX ratios are substantially higher in agricul- ture, natural resources, and services than in manufactures. This is mostly due to the fact that the manufacturing sector purchases inputs from non- manufacturing sectors, and therefore contains value added generated in those sectors. Across countries, the composition of trade drives aggregate VAX ratios, with countries that export Manufactures having lower aggregate VAX ratios. Aggregate VAX ratios do not covary strongly with income per capita, however, due to two offsetting effects. While richer countries tend to export manufactures, which lowers their aggregate VAX ratios, they also export at higher VAX ratios within the manufacturing sector. 8 Moving from aggregate to bilateral data, VAX ratios differ widely across partners for individual countries. For example, US exports to Canada are about 40% smaller measured in value-added terms than gross terms, whereas US exports to France are essentially identical in gross and value-added terms. These gaps arise for two main reasons. First, bilateral (‘back-and-forth’) pro- duction sharing implies that value-added trade is scaled down relative to gross trade. And these scaling factors differ greatly across bilateral partners. Sec- ond, multilateral (‘triangular’) production sharing gives rise to indirect trade that occurs via countries that process intermediate goods. For some country pairs, bilateral VAX ratios are larger than 1, as bilateral value-added exports exceed gross exports. 6 See NRC (2006) for the US; Dean et al (2007), Chen et al (2008) and Koopman et al (2008) for China. See also Hummels et al (2001) and Miroudot and Ragoussis (2009) for changes in domestic content over time for mainly OECD countries. 7 Bilateralor sector level ratios of value added to exports do not have this domestic content interpretation. 8 VAX ratios within manufactures are correlated with income because richer countries tend to export in sub-sectors with relatively high VAX ratios. 72 Trade in Value Added These adjustments imply that bilateral trade imbalances often differ in value-added and gross terms. For example, the US–China imbalance is approx- imately 30–40% smaller when measured on a value-added basis, while the US– Japan imbalance is approximately 33% higher. These adjustments point to the importance of triangular production chains within Asia. To illustrate the mechanisms at work in generating these results, we present two decompositions. In the first decomposition, we show that most of the vari- ation in bilateral value added to export ratios arises due to production sharing, not variation in the composition of goods exported to different destinations. The second decomposition splits bilateral exports according to whether they are absorbed in the destination, embedded as intermediates in goods that are reflected back to the source country or redirected to third countries embed- ded as intermediates in goods ultimately consumed there. Variation in the degree of absorption, reflection and redirection across partners is an impor- tant driver of variation in bilateral value added to export ratios. The rest of the chapter is structured as follows. Section 1 presents the gen- eral accounting framework, defines our value-added trade measures, and dis- cusses the interpretation of value added to export ratios. Section 2 describes the data sources and assumptions we use to implement the accounting exer- cise. Section 3 presents our empirical results and Section 4 concludes. 1 THE VALUE-ADDED CONTENT OF TRADE In this section, we introduce the accounting framework and demonstrate how intermediate goods trade generates differences between gross and value- added trade flows. We begin the section by presenting a general formulation of the framework with many goods and countries that we use in the calcula- tions below. To aid intuition, we then exposit several results in stripped-down versions of this general framework. Results from these simple models carry over to the general model. We close by discussing the relationship between our framework and two related lines of work on regional input–output link- ages and measurement of the factor content of trade. 1.1 The Value-Added Content of Trade Assume there are S sectors and N countries. Each country produces a single differentiated tradeable good within each sector, and we define the quantity of output produced in sector s of country i to be qi (s). This good is produced by combining local factor inputs with domestic and imported intermediate goods. It is then either used to satisfy final demand (equivalently, ‘consumed’) or used as an intermediate input in production. The key feature of the global input–output framework is that it tracks bilateral shipments of this output for final and intermediate use separately. Accounting for Intermediates 73 Tracking these flows requires four-dimensional notation denoting source and destination country as well as source and destination sectors for shipments of intermediates. Let the quantity of final goods from sector s in country i c absorbed in destination j be qij (s) and the quantity of intermediates from sec- m tor s in country i used to produce output in sector t in country j be qij (s, t). The global input–output framework organises these flows via market clear- c m ing conditions. Markets clear in quantities: qi (s) = j qij (s) + j t qij (s, t). If we evaluate these quantity flows at a common price, say pi (s), then we can rewrite the market clearing condition in value terms as yi (s) = cij (s) + mij (s, t), (4.1) j j t c m where yi (s) ≡ pi (s)qi (s), cij (s) ≡ pi (s)qij (s) and mij (s, t) ≡ pi (s)qij (s, t) are the value of production, final demand and intermediate goods shipments. Gross bilateral exports, denoted xij (s), include goods destined for both final and intermediate use abroad: xij (s) = cij (s) + t mij (s, t). Then (4.1) equiv- alently says that output is divided between domestic final use, domestic inter- mediate use and gross exports. To express market clearing conditions for many countries and sectors in a compact form, we define a series of matrices and vectors. Collect the total value of production in each sector in the S × 1 vector yi and allocate this output to final and intermediate use. Denote country i’s final demand for its own goods by S × 1 vector cii , and shipments of final goods from i to country j by the S × 1 vector cij . Further, denote use of intermediate inputs from i by country j by Aij yj , where Aij is an S × S input–output matrix with elements Aij (s, t) = mij (s, t)/yj (t). A typical element describes, for example, the value of steel (s = steel) imported by Canada (j = Canada) from the USA (i = USA) used in the production of automobiles (t = autos) as a share of total output of automobiles in Canada. Gross exports from i to j (i ≠ j ) are then xij = cij + Aij yj . With this notation in hand, we collect information on intermediate goods sourcing and final goods flows in vector/matrix form: ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ A11 A12 ... A1N y1 c1j ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ A21 A22 · · · A2N ⎟ ⎜ y2 ⎟ ⎜ c2j ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ A≡⎜ . . .. . ⎟ , y ≡ ⎜ . ⎟ , cj ≡ ⎜ . ⎟. ⎜ . . . . ⎟ ⎜ . ⎟ ⎜ . ⎟ ⎝ . . . ⎠ ⎝ . ⎠ ⎝ . ⎠ A N 1 AN 2 . . . ANN yN cNj Then, we write the S × N goods market clearing conditions as y = Ay + cj . (4.2) j This is the classic representation of an input–output system, where total out- put is split between intermediate and final use. Whereas a typical input–output 74 Trade in Value Added system focuses on sectoral linkages within a single economy, this system is expanded to trace intermediate goods linkages across countries and sectors. We therefore refer to A as the global bilateral input–output matrix. Using this system, we can write output as y= (I − A)−1 cj . (4.3) j To interpret this expression, (I − A)−1 is the ‘Leontief inverse’ of the input– output matrix. The Leontief inverse can be expressed as a geometric series: ∞ (I − A)−1 = Ak . k=0 Multiplying by the final demand vector, the zero-order term cj is the direct output absorbed as final goods, the first-order term [I + A]cj is the direct output absorbed plus the intermediates used to produce that output, the second-order term [I + A + A2 ]cj includes the additional intermediates used to produce the first round of intermediates (Acj ) and the sequence continues as such. Therefore, (I − A)−1 cj is the vector of output used both directly and indirectly to produce final goods absorbed in country j . Equation (4.3) thus decomposes output from each source country i into the amount of output from the source used to produce final goods absorbed in country j . To make this explicit, we define ⎛ ⎞ y1j ⎜ ⎟ ⎜ y2j ⎟ ⎜ ⎟ ⎜ . ⎟ ≡ (I − A)−1 cj , (4.4) ⎜ . ⎟ ⎝ . ⎠ yNj where yij is the S × 1 vector of output from i used to produce final goods absorbed in j . These output transfers are conceptually distinct from gross exports. Gross exports xij (s) are directly observed as a bilateral shipment from sector s in country i to country j . In contrast, bilateral output transfers are not directly observed, but rather constructed using information on the global input requirements for final goods absorbed in each country. Importantly, as inputs from a particular country and sector travel through the produc- tion chain, they may be embodied in final goods of any sector or country. For example, inputs exported from country i to country j may be embedded in country j ’s final goods that are absorbed in a third country k, or inputs produced by sector s may be embodied in final goods from sector t . These possibilities give rise to important differences in the structure of bilateral output transfers versus bilateral trade. Our system of equations (4.1)–(4.4) tracks the flow of each intermediate input across each border. These equations and the resulting tracking method Accounting for Intermediates 75 are identical to what appears in Trefler and Zhu (2010). Having developed the method, they then applied it to calculating the factor content of trade. We explain this application at the end of Section 1.2. Our interest here is different: we wish to calculate the value-added content of international trade. To calculate the value added associated with these implicit output transfers, define the ratio of value added to output for each sector within country i as ri (t) = 1 − j s Aji (s, t). This value-added ratio, expressed here as 1 minus the share of domestic plus imported intermediates in total output, is equal to payments to domestic factors as a share of gross output. In other words, this is the ratio of GDP to gross output at the sector level. With this notation in hand, we can now define value-added exports and the value added to export ratio (‘VAX ratio’) as a measure of the value-added content of trade. Definition 4.1 (value-added exports). The total value added produced in sec- tor s in source country i and absorbed in destination country j is vaij (s) = ri (s)yij (s). Total value added produced in i and absorbed in j is then vaij = s vaij (s). Definition 4.2 (VAX ratio). The sector-level bilateral value added to export ratio is given by vaij (s)/xij (s). The aggregate bilateral value added to export ratio is vaij /ιxij , where ι is a 1 × S vector of ones. 1.2 Discussion We turn to special cases to interpret value-added trade flows and the value- added content of trade. We use a two-country model to develop intuition for the value-added content of trade calculations and link our analysis to previ- ous work on the domestic content of exports (equivalently, vertical speciali- sation) by Hummels et al (2001). We then use a stylised three-country model to demonstrate how the framework tracks value added through the multi- country production chain, even if that value added travels to its final desti- nation via third countries. We also discuss the interpretation of VAX ratios in multi-sector models. We conclude by setting our framework in the context of related literature on regional input–output linkages and the measurement of the factor content of trade. Two Countries, One Sector per Country Suppose that there are now only two countries, and each country produces a single differentiated aggregate good. Then the analogue to the output decom- position (4.3) is −1 −1 y1 α11 α12 c11 α11 α12 c12 = I− + I− . (4.5) y2 α21 α22 c21 α21 α22 c22 76 Trade in Value Added This system describes how the gross output of each country is embodied in final consumption in each of the two countries. To unpack this result, we solve for the breakdown of country 1’s production: y1 = y11 + y12 , (4.6) with α12 α12 y11 = M1 c11 + c21 and y12 = M1 c22 + c12 , 1 − α22 1 − α22 where α12 α21 −1 M1 ≡ 1 − α11 − 1 1 − α22 is an intermediate goods multiplier that describes the total amount of gross output from country 1 required to produce one unit of country 1’s net output. 9 The first term (y11 ) is the total amount of country 1’s output that is required to produce final goods absorbed in country 1. This term includes both output dedicated to satisfy country 1’s demand for its own final goods (M1 c11 ), as well as output needed to satisfy country 1’s demand for country 2 final goods (M1 (α12 /(1 − α22 )c21 )). 10 The second term (y12 ) has a similar interpretation in terms of country 2’s demand. 11 Because (4.6) geographically decomposes country 1’s output, we can translate this into a decomposition of value added: va1 = va11 + va12 , where vaij = [1 − α11 − α21 ]yij is value added generated by country i that is absorbed in country j . There are four output concepts underlying flows from country 1 to coun- try 2: final goods c12 ; gross exports x12 ; implicit output transfers y12 ; and value-added exports va12 . We pause here to clarify the relationship between them. To begin, note that x12 = c12 + α12 y2 , so c12 x12 when there are exported intermediates. Further, using the output decomposition for coun- try 2 (y2 = y22 + y21 ), we decompose gross exports as x12 = α12 y21 + (c12 + α12 y22 ). Multiplying both sides of the expression by (1 − α11 )−1 then translates exports into the gross output required to produce them. 12 It is 9 This multiplier is greater than 1 because output is ‘used up’ in the production pro- cess. Without exported intermediates (α12 = 0), this multiplier would be (1 − α11 )−1 . The additional term reflects the fact that intermediate goods sourced from country 2 contain output produced by country 1. 10 Exporting final goods c −1 21 requires producing (1 − α22 ) c21 units of country 2 output, which itself requires α12 (1 − α22 )−1 c21 units of country 1’s output as intermediates. To produce this country 1 output requires M1 times α12 (1 − α22 )−1 c21 units of country 1’s output overall, because some output is used up in the production process. 11 To highlight how the output decomposition depends on cross-border intermediate linkages, note that if α12 = 0, the output decomposition would be y11 = (1 − α11 )−1 c11 and y12 = (1 − α11 )−1 c12 . In this counterfactual case, output of country 1 is only used to produce final goods originating in country 1. 12 This follows from manipulation of the market clearing condition for country 1: y = 1 (1 − α11 )−1 (c11 + x12 ). Accounting for Intermediates 77 straightforward to show that y12 = (1 − α11 )−1 (c12 + α12 y22 ). Therefore, y12 = (1 − α11 )−1 x12 − (1 − α11 )−1 α12 y21 . So the implicit output transferred from country 1 to country 2 is equal to the gross output required to produce exports minus the gross output that is reflected back by being embedded in country 2 goods that are absorbed by country 1. 13 Finally, we note that va12 y12 , because the value added to output ratio is bounded above by 1. To directly compare value-added exports to gross exports, we compute the VAX ratio: va12 (1 − α11 − α21 )y12 = x12 x12 1 − α11 − α21 x12 − α12 y21 = , (4.7) 1 − α11 x12 where the second line follows from the discussion in the previous paragraph. The difference x12 − α12 y21 is exports minus reflected intermediates, or equiv- alently the portion of exports genuinely consumed abroad. The VAX ratio will always be less than 1, so value-added exports are scaled down relative to gross exports. The VAX ratio for a country can be thought of as a metric of the ‘domestic content of exports’. Indeed, it is closely related to previous approaches to mea- suring domestic content in the literature. To see this, note that the VAX ratio has two components. The first component, (1 − α11 − α21 )/(1 − α11 ), is equiva- lent to a metric of domestic content developed in Hummels et al (2001). 14 This metric captures the value added associated with the gross output needed to produce exports as a fraction of total exports. The Hummels–Ishii–Yi metric is equal to the VAX ratio only when country 2 does not use imported inter- mediates (α12 = 0), and therefore country 1 exports final goods alone. 15 In contrast, with two-way trade in intermediates the Hummels–Ishii–Yi metric overstates the amount of domestic value added that is generated per unit of exports. 16 The second component of the VAX ratio allows some exports to be dedicated to producing goods that are ultimately consumed at home. That 13 Notethat if α12 = 0, then y12 = (1 − α11 )−1 x12 , so the gross output required to pro- duce exports equals the actual amount of output transferred from country 1 to country 2. 14 Hummels et al focus their discussion on measuring vertical specialisation or the ‘import content of exports’, which is given by α21 (1 − α11 )−1 . Domestic content is then 1 minus the import content of exports. Though we discuss these concepts here in a scalar case, they generalise in a straightforward way to models with many sectors. 15 The condition α 12 = 0 is necessary and sufficient for equality between the two metrics when there is one aggregate sector, except in pathological cases. With more than one sector, restricting country 1 to export only final goods (α12 (s, t) = 0 for all s, t ) is sufficient, but not necessary. 16 Footnote 18 in Trefler and Zhu (2010) provides a related discussion of how the factor content of trade differs depending on whether one assumes intermediates are traded or not. 78 Trade in Value Added is, it allows for a portion of exports to be reflected back to the source rather than absorbed abroad. Three Countries, One Sector per Country While the two-country framework illustrates the basic discrepancy between value-added and gross trade flows, additional insights emerge as one intro- duces a third country to the mix. We focus on a special, algebraically straight- forward case that illustrates how the accounting framework tracks the final destination at which value added by a given country is consumed, even if this value circulates through a multi-country production chain en route to its final destination. We construct the special case to approximate a stylised account of production chains between the USA and Asia. 17 Let country 1 be the USA, country 2 be China and country 3 be Japan. Fur- ther, assume that China imports intermediates from the USA and Japan and exports only final consumption goods only to the USA. For simplicity, we assume that the USA and Japan do not export any final goods, and only export intermediates to China. This configuration of production can be represented as ⎛ ⎞ ⎛ ⎞⎛ ⎞ ⎛ ⎞ y1 α11 α12 0 y1 c11 ⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎜ ⎟ ⎝y 2 ⎠ = ⎝ 0 α22 0 ⎠ ⎝y2 ⎠ + ⎝c22 + c21 ⎠ . (4.8) y3 0 α32 α33 y3 c33 This then can be solved to yield the following three-equation system: 1 α12 α12 y1 = c11 + c21 + c22 , 1 − α11 (1 − α11 )(1 − α22 ) (1 − α11 )(1 − α22 ) y11 y12 1 1 y2 = c21 + c22 , 1 − α22 1 − α22 y21 y22 α32 α32 1 y3 = c21 + c22 + c33 . (1 − α33 )(1 − α22 ) (1 − α33 )(1 − α22 ) 1 − α33 y31 y32 y33 This system provides the implicit output transfers needed to calculate value- added flows. Two points are interesting to note. First, as in the two-country case above, US demand for US output has both a direct component (1/(1 − α11 ))c11 and an indirect component (α12 /(1 − α11 )(1 − α22 ))c21 that accounts for the fact that US imports of final goods from China include embedded US content. Thus, a 17 This example was inspired by Linden et al (2007), who trace the iPod production chain. The iPod combines US intellectual property from Apple with a Japanese display and disk drive, which is manufactured in China. These components are assembled in China and the iPod is shipped to the USA. Accounting for Intermediates 79 larger share of US output is ultimately absorbed at home than bilateral trade statistics would indicate. Correspondingly, Chinese bilateral exports overstate the true Chinese content shipped to the USA due to bilateral US–China pro- duction sharing. The second point is that, although Japan does not export directly to the USA, the USA does import Japanese content embedded in Chinese exports to the USA. This effect is the result of multi-country production chains, and was absent in the two-country case analysed above. In the equation for Japan (country 3), this effect appears as (α32 /(1 − α33 )(1 − α22 ))c21 . Because Chinese exports to the USA contain both US and Japanese content, the bilateral VAX ratio of China–US trade is va21 va31 +α12 y21 =1− < 1. (4.9) x21 x21 This illustrates that the bilateral VAX ratio removes both the Japanese value added (va31 ) and US intermediate goods (α12 y21 ) from Chinese exports to the USA. 18 Turning to Japan, it has positive value-added exports to the USA and zero direct bilateral exports. Therefore, the bilateral VAX ratio for Japan– US trade is undefined, or practically infinite for small bilateral exports. This extreme ratio illustrates another general lesson. Though the aggregate VAX ratio is bounded by 1 for each country, bilateral VAX ratios may be greater than 1 when an exporter sends intermediates abroad to be processed and delivered to a third country. Thus, bilateral VAX ratios pick up the influence of both bilateral and multilateral production sharing relationships. When bilateral VAX ratios vary across partners, bilateral value-added bal- ances do not equal bilateral trade imbalances. To illustrate this, we define tb12 ≡ x12 − x21 and va b12 ≡ va12 − va21 to be bilateral US–China trade and value-added balances. In this special case, where the configuration of produc- tion is given by (4.8), these balances are related as follows: tb12 + α32 y21 = va b12 . (4.10) That is, tb12 < va b12 . So, assuming the USA runs a trade deficit with China in this example, it will run a smaller deficit with China in value-added terms due to the fact that Chinese bilateral trade contains Japanese content (α32 y21 ). As a corollary, the USA’s bilateral balance with Japan will be distorted in the opposite direction. 18 US imports from China contain US content because the US exports intermediates to China and imports final goods from China. Thus, US intermediates are reflected back to the USA and constitute a portion of the value added that the USA purchases from itself. 80 Trade in Value Added To generalise this result, we can write any given bilateral value-added bal- ance as vaij vaji va bij = xij − xji xij xji vaij vaji 1 vaij vaji = 1 2 (xij + xji ) x − + + [xij − xji ]. (4.11) ij xji 2 xij xji The first term adjusts the value-added balance due to differences in VAX ratios between exports and imports. When the VAX ratio for exports is high rel- ative to imports, the value-added balance is naturally pushed in a positive direction. Note here that this is true even if gross trade is balanced. The sec- ond term adjusts the value-added balance based on the average level of VAX ratios. Starting from an initial imbalance, the value-added balance is scaled up or down relative to the trade balance, depending on whether VAX ratios are greater than or less than 1 (on average). So, differences in VAX ratios between partners within a bilateral relationship and the absolute level of the VAX ratios between partners both influence the size of the adjustment in converting gross imbalances to value-added terms. Two Countries, Many Sectors The interpretation of aggregate value-added exports and VAX ratios devel- oped in the one-sector examples in previous sections carries over to the many- country, multi-sector framework. One important distinction between the one- sector and multi-sector frameworks is that the VAX ratio at the sector level cannot be interpreted as the domestic content of exports. To explain its inter- pretation, we turn to an example with two countries and many sectors. 19 With two countries (i, j = {1, 2}) and many sectors, the VAX ratio for sector s in country 1 can be written as va12 (s) r1 (s)y12 (s) = . x12 (s) x12 (s) Then the sectoral VAX ratio depends on the value added to output ratio within a given sector (r1 (s)) and the ratio of gross output produced in a sector that is absorbed abroad (y12 (s)) to gross exports from that sector (x12 (s)). The role of the value added to output ratio is straightforward: all else being equal, sectors with low value added to output ratios (eg manufacturing) will have low VAX ratios relative to other sectors. The role of differences in y12 (s) versus x12 (s) across sectors is more subtle. To sort this out, we note that we can link y12 and the export vector x12 as 19 The many-country version of the framework can always be collapsed to an equivalent two-country framework, in which input–output linkages among countries in the rest of the world are subsumed into the ‘domestic’ input–output structure of the rest-of-the-world composite. Accounting for Intermediates 81 in Section 1.2. Specifically, x12 = (I − A11 )y12 + A12 y21 . Rearranging this expression yields y12 = (I − A11 )−1 [x12 − A12 y21 ]. This is the many sector, matrix analogue to computations embedded in Equation 4.7, wherein y12 is the gross output needed to produce exports less reflected intermediates. This decomposition points to two ways in y12 could differ from x12 . First, suppose that A12 y21 is a vector of zeros, so that exports are 100% absorbed abroad. 20 This implies that y12 = (I − A11 )−1 x12 . All that remains here separating exports and gross output for individual sectors is the domes- tic input–output structure. Generically, y12 (s) ≠ x12 , so variation in this ratio across sectors influences sector-level value added. One important implication of this is that the sectoral VAX ratio captures information on how individual sectors engage in trade. For example, consider a situation in which producers in one sector sell intermediates to purchasers in another sector, who in turn produce goods for export. 21 In this case, the intermediate goods suppliers engage in trade indirectly. Hence, we observe no direct exports from the intermediate goods supplier, but do observe value- added exports because value added from that sector is embedded in the purchaser’s goods. Thus, value-added exports from a particular sector may be physically embodied in goods exported from that sector or embodied in exports of other sectors. High ratios of value-added exports to gross trade (possibly above 1) at the sector level are evidence of indirect participation in trade. Low ratios instead indicate that a given sector’s gross exports embody value added produced outside that sector. Second, suppose now that A12 is not composed of zeros, but rather that country 1 exports intermediates to country 2 that are used to produce goods that are absorbed in country 1, captured by the term A12 y21 > 0. In this case, the sectoral VAX ratio is influenced by how individual sectors fit into cross- border production chains. For example, if we shut down all domestic input– output linkages, setting A11 to zero, then y12 = x12 −A12 y21 . Then the sectoral VAX ratio depends on the sector’s connection to foreign production chains. Specifically, the VAX ratio will be depend on what share of output is absorbed abroad versus the share used to produce foreign goods that are ultimately absorbed at home. If exports are largely absorbed abroad (ie y12 (s)/x12 (s) ≈ 1), one would see a relatively high VAX ratio. 20 IfA12 is a matrix of zeros, so that country 1 exports only final goods, this obviously holds. This can also hold for cases in which elements of A12 are positive, so long as the corresponding elements y21 are zero. For example, country 1 could export intermediates to country 2, so long as the sector purchasing those intermediates only produces output for consumption in country 2. 21 For example, the ‘raw milk’ sector in our data has near zero exports, but raw milk is sold to the ‘dairy products’ sector, which does export. With two sectors, where 1 is the dairy products and 2 is the milk sector, this could be represented as an A11 matrix with one non-zero element α11 (2, 1) and export vector with x12 (1) > 0 and x12 (2) = 0. This structure implies y12 (1)/x12 (1) = 1 and y12 (2)/x12 (2) = ∞. 82 Trade in Value Added Though these influences are difficult to separate empirically in general cases, we discuss evidence below that sheds light on the relative importance of these channels. Regional Input–Output Models and the Factor Content of Trade The framework above is intimately related to two strands of literature in regional science and factor content of trade. First, we draw on an extensive literature on regional input–output models. These models, outlined in seminal work by Isard (1951), Moses (1955, 1960) and Miller (1966), provide frameworks for analysing linkages across regions within countries that can be extended across borders (as above). Among this literature, Moses (1955) is the closest antecedent, as he uses proportionality assumptions to allocate inputs purchased from other regions, as we do, to build a multi-region model of the USA. 22 One shortcoming of this line of work is that it typically assumes that the regional system is ‘open’ vis-à-vis the rest- of-the-world, in the sense that shipments to regions not included in the model are entirely absorbed there. This assumption is a multi-region analogue of the assumptions under which the Hummels–Ishii–Yi domestic content calculation is equal to the value-added content of trade. 23 Second, the value-added framework above shares a common structure with a recent parallel literature on measuring the factor content of trade. Reimer (2006) and Trefler and Zhu (2010) both outline procedures to compute the net factor content of trade when inputs are traded, and use these factor content measures to study the Vanek prediction. To draw out the similarities, note that one can think of computing both factor contents and value-added contents using a two-step procedure. First, one needs to compute the output transfers, specified above, that indicate how much output from each source country and sector are absorbed in final demand in a given destination. Second, one needs to use source country information on either factor contents (eg quantities of factors used to produce one dollar of output) or value added to output ratios to compute the factors or value added that is implicitly being traded. 24 22 Isard (1951) suggests this technique as well, but does not pursue an empirical appli- cation himself. 23 Powers et al (2009) work with a model of this type for Asia. 24 Let us trace out the calculation explicitly. Trefler and Zhu define Ti to be an (NS × T T T T 1) vector of trade flows arranged as follows: Ti = [. . . , −xi −1,i , xi , −xi+1,i , . . . ] , where xi = j ≠i xij is an (S × 1) vector of total exports from country i to the rest of the world and xj,i is an (S × 1) vector of bilateral trade flows from j ≠ i to i. Further, they define B to be an F × SN matrix of factor requirements for each good: B ≡ [B1 , . . . , Bi , . . . , BN ], where Bi is the F × S matrix of factor requirements for country i, with F denoting the number of factors. The factor content of trade for country i is then B(I − A)−1 Ti . To link this to our framework, we note that the calculation (I − A)−1 Ti returns a vector of (signed) output transfers. In particular, (I − A)−1 Ti = [. . . , −yi T T T T −1,i , yxi , −yi+1,i , . . . ] , where yxi ≡ j ≠i yij is the total output produced in country i that is absorbed abroad, and yj,i is the output produced in Accounting for Intermediates 83 Despite this similarity in the underlying structure of value-added and fac- tor content calculations, we emphasise that there are important conceptual differences between factor contents and value added. For one, the theoretical driving forces of trade in value added may be very different from trade in fac- tors. Costinot et al (2011) point out that differences in absolute endowments across countries influence where countries are located in the value chain, so absolute (as opposed to relative) factor endowments are a source of compara- tive advantage underlying trade in value added. 25 This is just one example of a general point: the empirical shift from factor content to value-added content embodies a deeper conceptual shift in how we think about trade. 2 DATA Our data source is the GTAP 7.1 database assembled by the Global Trade Analysis Project at Purdue University. This data is compiled based on three main sources: World Bank and IMF macroeconomic and Balance of Payments statistics; United Nations Commodity Trade Statistics (Comtrade) Database; and input–output tables based on national statistical sources. To reconcile data from these different sources, GTAP researchers adjust the input–output tables to be consistent with international data sources. 26 The GTAP data includes bilateral trade statistics and input–output tables for 94 countries plus 19 composite regions covering 57 sectors in 2004. 27 Regarding sector definitions, there are 18 Agriculture and natural resources sectors, 24 manu- factures sectors and 15 services sectors. In the data, we have information on six objects for each country: 1. yi is a 57 × 1 vector of total gross production; country j ≠ i that is absorbed in country i. Thus, as suggested above, one can think of first computing output transfers embedded in trade flows, and then computing the factor requirements needed to produce those output transfers. See Johnson (2008) for an extended discussion of these calculations. 25 Like absolute endowments, absolute productivity differences are also a source of com- parative advantage in the Costinot et al model. 26 See the GTAP website at http://www.gtap.agecon.purdue.edu/ for documentation of the source data. Since raw input–output tables are based on national statistical sources, they inherit all the shortcomings of those sources. For example, import tables are often constructed using a ‘proportionality’ assumption whereby the imported input table is assumed to be proportional to the overall aggregate input–output table. 27 GTAP assigns composite regions ‘representative’ input–output tables, constructed from input–output tables of similar countries. Composite regions do not play an impor- tant role in our results, accounting for 5% of world trade and 3% of world value added. To measure bilateral services trade, GTAP uses OECD data where available and imputes bilat- eral services trade elsewhere. Because services account for less than 18% of exports for the median country, our results are likely to be insensitive to moderate mismeasurement of services trade. 84 Trade in Value Added 2. cDi is a 57 × 1 vector of domestic final demand; 3. cIi is a 57 × 1 vector of domestic final import demand; 4. Aii is a 57 × 57 domestic input–output matrix, with elements Aii (s, t); 5. AIi is a 57 × 57 import input–output matrix, with elements AIi (s, t) = j ≠i Aji (s, t); 6. {xij } is a collection of 57 × 1 bilateral export vectors for exports from i to j . The definition of ‘final demand’ is based on the national accounts, including consumption, investment and government purchases. We value each coun- try’s output at a single set of prices, regardless of where that output is shipped or how it is used. This ensures that the value of production rev- enue equals expenditure. 28 Following input–output conventions, we use ‘basic prices’, defined as price received by a producer (minus tax payable or plus subsidy receivable by the producer). 29 Note that we do not directly observe the bilateral input–output matrices Aji and final demand vectors cji that are needed to assemble the global input– output matrix. Rather, we need to allocate total imported intermediate use AIi and imported final demand cIi to individual country sources. To do so, we use bilateral trade data and a proportionality assumption. Specifically, we assume that, within each sector, imports from each source country are split between final and intermediate in proportion to the overall split of imports between final and intermediate use in the destination. Further, conditional on being allocated to intermediate use, we assume that imported intermediates from each source are split across purchasing sectors in proportion to overall imported intermediate use in the destination. Formally, for goods from sector s used by sector t , we define bilateral input– output matrices and consumption import vectors: xji (s) xji (s) Aji (s, t) = AIi (s, t) and cji (s) = cIi (s) . j xji (s) j xji (s) These assumptions imply that all variation in total bilateral intermediate and final goods flows arises due to variation in the composition of imports across partners. For example, we would find that US imports from Canada are inter- mediate goods intensive because most imports from Canada are goods that are on average used as intermediates (eg automobile parts). 28 In other words, while quantity choices may reflect price differences across destina- tions or uses that arise due to transport costs, tariffs and markups, we value the resulting quantity flows at a single set of prices. 29 In our framework, the level of value added differs from the one used in national accounts. We calculate value added as output at basic prices minus intermediates at basic prices, whereas the national accounts calculate value added as output at basic prices minus intermediates at purchaser’s prices. Accounting for Intermediates 85 The proportionality assumptions above are the standard approach to deal- ing with the fact that data on Aji and cji are not collected in national accounts. 30 Initially adopted in early work on regional input–output accounts by Moses (1955), they have also been used by Belke and Wang (2006), Daudin et al (2011) and Trefler and Zhu (2010) to construct global input–output tables as in this chapter. Several recent papers have explored the consequences of relaxing some proportionality assumptions using alternative data sources, and appear to find that relaxing these assumptions has small effects on aggre- gate VAX ratios or factor contents. 31 In the main calculation, we also assume that production techniques and input requirements are the same for exports and domestically absorbed final goods. This assumption is problematic for countries that have large export- processing sectors. These processing sectors (almost by definition) produce distinct goods for foreign markets with different input requirements and lower value added to output ratios than the rest of the economy. Ignoring this fact tends to overstate the value-added content of exports. As an alternative calculation, we relax this assumption for China and Mex- ico, two prominent countries with large export-processing sectors (roughly two-thirds of exported Manufactures originates in these sectors) and key trad- ing partners with the USA. 32 We present supplementary calculations below that adjust the value-added content of exports using an adaptation of a pro- cedure from Koopman et al (2008). The basic idea is to measure the share of exports and imports that flow through the export-processing sector, and then impute separate input–output coefficients for the processing sector so as to be consistent with these flows. Details of the procedure are presented in the appendix in Section 5. We then compute the value-added content of trade using a new input–output system that includes these amended tables. 33 30 Proportionality assumptions are so common in input–output accounting that many countries, including the USA, even construct the import matrix (AIi ) itself using a propor- tionality assumption in which imported inputs are allocated across sectors in the same proportion as total input use (aggregating over imported and domestic inputs). Some coun- tries augment this data with direct surveys of input use in constructing imported input use tables. However, no countries (to our knowledge) directly collect information on bilateral sources of inputs used in particular sectors. 31 Puzzello (2012) compares factor content calculations with and without the propor- tionality assumption using IDE-JETRO regional input–output tables for Asia. Koopman et al (2010) compute value-added content using disaggregate data classified under the BEC system to estimate bilateral intermediate goods flows. While relaxing proportionality seems to have small aggregate consequences, it may simultaneously have large effects on value-added trade at the sector level. This remains to be explored. 32 For Mexico, we classify exports originating from maquiladoras as processing exports. For China, we use estimates from Koopman et al (2008) constructed from Chinese trade statistics, obtained from Zhi Wang (personal communication). 33 We perform this calculation at a higher level of aggregation than our baseline calcula- 86 Trade in Value Added 3 EMPIRICAL RESULTS 3.1 Multilateral Value-Added Exports Table 4.1 reports aggregate VAX ratios for each country, grouped by region. 34 Across countries, value-added exports represent about 73% of gross exports. The magnitude of the adjustment varies both across and within regions. At the regional level, VAX ratios are lowest for Europe (broadly defined) and East Asia, and higher in the Americas, South Asia and Oceania, and the Middle East and Africa. Looking within regions, the new EU members (eg Estonia, Hun- gary, Slovakia and the Czech Republic) stand out as having low VAX ratios in Central-Eastern Europe, while Japan stands out with a high VAX ratio relative to East Asia. For China and Mexico, we report two separate calculations of the VAX ratio in the table, one computed without adjusting for processing trade and a sec- ond adjusted for processing trade. 35 VAX ratios for both China and Mexico fall substantially when we adjust for export processing trade, from 0.70 to 0.59 for China and from 0.67 to 0.52 for Mexico. This brings the ratios for China and Mexico in line with other emerging markets, such as South Korea or Hungary, and is evidence of the low value added to export ratios within each country’s processing sector. 36 Moving down a level of disaggregation, we report VAX ratios for three com- posite sectors by country in Table 4.1 as well. The three sectors are agriculture and natural resources, manufacturing and services. VAX ratios are typically greater than or equal to 1 in the agriculture and natural resources and services sectors, and markedly less than 1 in manufacturing. This cross-sector varia- tion is primarily due to differences in the manner in which each sector engages in trade, rather than differences across sectors in the degree of participation in cross-border production sharing. Further, differences in value added to output ratios across sectors are also an important source of variation. tion, with three composite sectors. We believe the results are not very sensitive to aggre- gation, as aggregate value-added flows are nearly identical in the original, unadjusted data whether computed using 57 sectors or 3 composite sectors. 34 We omit ratios for composite regions from the table. 35 In the calculation adjusted for processing trade in China and Mexico, VAX ratios in all countries change relative to the unadjusted benchmark calculation. The absolute size of the changes in aggregate VAX ratios is very small, with a median of 0.016 and 90% of changes less than 0.053. Therefore, we report only one set of ratios for all countries other than China and Mexico. 36 For the processing sector, we estimate that China’s VAX ratios is 0.13, while Mexico’s VAX ratio is 0.08. These ratios measure the value added produced within the process- ing sector as a share of processing exports. These ratios represent a lower bound on the domestic content of processing exports, since the processing sector purchases intermedi- ates from other domestic sectors. Accounting for Intermediates 87 To sort through these influences, we refer back to Section 1.2. Recall that sectoral VAX ratios tend to be low when exports are used to produce foreign goods that are ultimately absorbed at home. If we assume that all output is absorbed abroad, then the output needed to produce exports would be ˜ix = (I − Aii )−1 y xij , j ≠i ˜ is used to signify that this is a counterfactual value and where y ˜ix = y ˜ij . y j ≠i Then the counterfactual sectoral value added to export ratios would be ˜ix ri (s)y , with xi (s) = xij . xi (s) j ≠i In our data, this counterfactual calculation yields ratios that are very close to the actual VAX ratios. As such, differences across sectors in the degree of for- eign absorption of exports does not appear to drive the VAX ratios. Further, we note that differences in value added to output ratios also cannot explain the full variation in VAX ratios across sectors. In the data, the value added to out- put ratio in manufactures is roughly 0.25 lower than in agriculture and natural resources and services sectors. This goes part of the way towards explaining differences in VAX ratios across sectors, but falls substantially short. The remaining driver of variation in VAX ratios across sectors is cross- sector variation in the extent to which sectoral output is directly exported versus indirectly exported, embodied in other sectors’ goods that are then exported. Recall that we observe gross exports from a given sector (ie j xij (s) > 0) only if output from that sector crosses an international border with no further processing. With this in mind, it is obvious that sector-level VAX ratios are greater than 1 when a sector exports value added embodied in another sector’s gross output and exports. In the data, it appears that manu- factures, which are directly exported, embody substantial value added from the other sectors. One implication of this fact is that the composition of aggre- gate value-added flows differs from that of gross trade. Figure 4.1 summarises this fact by plotting the share of manufactures and services in both types of trade for the ten largest exporters. The role of manufactures in value-added trade is diminished, while that of services is increased by a roughly equivalent amount. 37 The upshot is that services are far more exposed to international commerce than one would think based on gross trade statistics. 37 ‘Agriculture and natural resources’ constitutes a roughly equal share of value added and gross trade. 88 Trade in Value Added To organise the inter-country variation in the data, we construct a ‘between- within’ decomposition of the aggregate VAX ratio. The decomposition is con- structed relative to a reference country as follows: ωi (s) + ω(s) ¯ VAXi −VAX = [VAXi (s) − VAX(s)] s 2 within term VAXi (s) + VAX(s) + [ωi (s) − ω(s)] ¯ , (4.12) s 2 between term where s denotes sector, i denotes country and ω(s) and VAX(s) are the export share and VAX ratio in sector s . Bars denote reference country variables, which are constructed based on global composites. 38 In this decomposition, the within term varies primarily due to differences in VAX ratios within sectors across countries, while the between term is influenced mainly by differences in the sector composition of trade. To isolate compositional shifts between man- ufactures and non-manufactures, we calculate the decomposition using two composite sectors, pooling services plus agriculture and natural resources into a single composite non-manufacturing sector. Cross-country variation in aggregate VAX ratios is to a large extent driven by variation in the composition of exports. To illustrate this, we plot VAX deviations (VAXi −VAX) against the between and within terms separately in Figure 4.2. 39 In part (a), the between term is a strong and tight predictor of a country’s aggregate VAX ratio. In contrast, the within term is actually weakly negatively correlated with the aggregate VAX ratio in part (b), and this rela- tionship is relatively noisy. This visual impression is naturally confirmed by a simple variance decomposition. If we split the covariance of the between and within terms equally, the between term ‘accounts for’ nearly all the variation in the aggregate VAX ratio. 40 The between term is dominant because of the large differences in VAX ratios across sectors. Countries that export predom- inantly manufactures, the sector with the lowest VAX ratio, tend to have low aggregate VAX ratios as well. 38 Reference country VAX ratios for each sector are the ratios of value-added exports to gross exports for the world as a whole. Export shares are the share of each sector in total world exports. 39 The regression line in part (a) is VAXi −VAX = 0.26 × between term, with robust standard error 0.04 and R 2 = 0.36. The regression line in part (b) is VAXi −VAX = −0.11 × within term, with robust standard error 0.06 and R 2 = 0.04. 40 Specifically, the variance breaks down as follows: var(agg. VAX) = 0.01, var(within) = 0.03, var(between) = 0.04, and cov(within, between) = −0.03. Due to the negative covari- ance between the two terms, the variance decomposition is sensitive to how one chooses to assign the covariance. The scatter plots in Figure 4.2 can be thought of as representing a situation in which one assigns the covariance equally to the two terms. Accounting for Intermediates 89 (a) 1.0 0.8 0.6 Share 0.4 0.2 0 bel can chn deu fra gbr ita jpn kor usa (b) 1.0 0.8 0.6 Share 0.4 0.2 0 bel can chn deu fra gbr ita jpn kor usa Export shares Value-added export shares Figure 4.1: Composite sector shares of gross exports and value-added exports, by coun- try (2004): (a) manufactures; (b) services. Despite this strong composition effect, aggregate VAX ratios are only weakly related to the overall level of economic development. Part (a) of Table 4.2 shows that a one log point increase in income per capita is asso- ciated with a fall in domestic content of 0.8 percentage points, though this correlation is not significantly different from zero at conventional significance levels. 41 This weak aggregate correlation is a manifestation of two offsetting effects. First, richer countries tend to have exports concentrated in manufac- tures, which has a relatively low VAX ratio. Second, richer countries tend to export with higher VAX ratios than poorer countries within composite sectors, particularly within manufactures. To illustrate these offsetting effects, we project the between term and the within term separately on exporter income to quantify the relative contri- bution of each to the overall correlation. In part (a) of Table 4.2, we see that there is a strong negative correlation of the between term with exporter income. That is, countries systematically shift towards manufacturing (which has lower value added to output on average) as they grow richer and this 41 The p -value for a two-sided test that the correlation does not equal zero is 14%. In this regression, we omit outliers Belgium, Luxembourg and Singapore. If these three countries are included, the correlation roughly doubles in size and becomes highly significant. 90 Trade in Value Added (a) 0.2 VAX ratio − world VAX 0.1 0 −0.1 −0.2 −0.5 0 0.5 (b) 0.2 VAX ratio − world VAX 0.1 0 −0.1 −0.2 −0.5 0 0.5 Figure 4.2: Between–within decomposition of aggregate VAX ratios, by country (2004). (a) Between term; (b) within term. depresses the aggregate VAX ratios. The effect of this on overall VAX ratios is obscured because the within term is significantly positively correlated with exporter income. This positive correlation is mostly due to the fact that rich countries have higher VAX ratios within Manufactures. Part (b) of Table 4.2 shows the correlation of VAX ratios for manufactures with income per capita and splits this into between and within terms as above. 42 The positive correla- tion between manufactures VAX ratios and income is itself driven by a positive composition (‘between’) effect, wherein richer countries tend to specialise in manufacturing sectors with high VAX ratios. 3.2 Bilateral Value-Added Exports and Balances For a particular exporter, bilateral VAX ratios differ widely across destina- tions. For concreteness, we graphically present bilateral value added to trade ratios for the two largest exporters, the USA and Germany, in Figure 4.3. In the 42 VAX ratios for the non-manufactures composite are positively correlated with income per capita, but the correlation is not significant. Therefore, we do not report these results separately. Accounting for Intermediates 91 (a) Ratio of VA to gross trade 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0 mex_adj sgp mys irl can mex bel chn_adj twn kor chn bra deu gbr fra jpn ita aus esp rus (b) Ratio of VA to gross trade 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0 mex_adj sgp mys irl can mex bel chn_adj twn kor chn bra deu gbr fra jpn ita aus esp rus Exports Imports Figure 4.3: Value added to gross trade ratios for the USA and Germany, by partner (2004). (a) US bilateral trade; (b) Germany bilateral trade. figure, value added to import ratios are VAX ratios for each country export- ing to the USA/Germany, while value added to export ratios are recorded for US/German exports to each country. 43 Looking at the USA, there is wide variation in VAX ratios. For some partners, value-added exports are quite close to gross exports. For example, the differ- ence between gross and value-added exports to the United Kingdom amounts to only 3% of gross exports. For others, gross trade either overstates or under- states the bilateral exchange of value added. Value-added exports to Canada 43 We display data for the 15 largest trade partners for each country plus additional countries selected for illustration purposes, including adjusted and unadjusted bilateral VAX ratios for China and Mexico. In line with the aggregate results, adjusting for processing trade lowers bilateral VAX ratios vis-à-vis these countries but has only modest effects on ratios for other countries. 92 Trade in Value Added 10 0 −10 −20 −30 −40 −50 −60 −70 −80 −90 −100 −110 Trade deficit Value-added deficit −120 Adjusted VA deficit −130 chn jpn can deu mys ita twn gbr mex kor irl bra rus fra bel esp sgp aus Figure 4.4: Bilateral trade and value-added balances for the USA, by partner (2004). are US$77 billion (40%) smaller than gross exports, and value-added exports to Mexico are US$40–50 billion (36–44%) smaller. Value-added trade falls by a similar proportional amount, between 30% and 50%, relative to gross trade for countries like Ireland, Korea and Taiwan (Chinese Taipei), which are well- cited examples of production sharing partners. At the other end of the spec- trum, several countries have VAX ratios towards the USA above 1. For exam- ple, countries on Europe’s eastern periphery (see Russia) have bilateral VAX ratios above 1 mainly because they supply intermediates to Western European countries that then end up being consumed in the USA. Further, commodity producers (see Australia) also often have ratios above 1. The US data are representative of general patterns in the data. 44 Looking at Germany, discrepancies between value added and gross trade also vary in meaningful ways across partners. Value-added trade is scaled down quite substantially for the vast majority of its large European partners, in contrast to the USA. This surely is an indication of the integrated structure of produc- tion within the European Union and its neighbours. Consistent with anecdo- tal evidence, this is most pronounced for the Czech Republic and Hungary. Geography appears to play a substantial role, as trade with partners of similar income levels, such as the USA and Japan, is relatively less distorted. 44 The median bilateral VAX ratio in the data is 0.91, and the 10th-to-90th percentile range is 0.59 to 2.07. Approximately 40% of the bilateral VAX ratios are greater than 1. Accounting for Intermediates 93 One consequence of these trade adjustments is that bilateral trade bal- ances differ when measured in gross versus value-added terms. Figure 4.4 displays three measures of bilateral balances for the USA: the bilateral trade balance, the bilateral value-added balance and the bilateral value-added bal- ance adjusted for processing exports in China and Mexico. In interpreting this figure, it is important to keep in mind that multilateral trade balances equal the multilateral value-added balance for each country. Therefore, a decline in the bilateral value-added balance relative to the gross trade balance for one country necessarily implies an increase for some other country. Comparing these alternate measures, there are large shifts in bilateral bal- ances in Asia. Most prominently, the US deficit with China falls by roughly 30–40% (US$35–50 billion), while the deficit with Japan rises by around 33% (US$17–18 billion). The end result is that the value-added balances (adjusted for processing trade) are nearly equal for Japan and China. Looking elsewhere within Emerging Asia, US deficits with Taiwan (Chinese Taipei) and South Korea also rise and US surpluses with Australia and Singapore fall. Together, adjustments in these five countries (Australia, Japan, Singapore, South Korea and Taiwan) nearly exactly add up to the fall in the US–China deficit, which points to triangular production sharing within Asia, with these countries feed- ing intermediates to China that are then embodied in Chinese exports to the USA. To understand these adjustments, we focus on the US–China and US–Japan balances with reference to the decomposition of the value-added balance in Equation (4.11). First, looking at China, the VAX ratio for US exports to China exceeds the VAX ratio for imports by about 8% in the unadjusted calculation and 4% in the adjusted calculation. This tends to raise the value-added bal- ance relative to the trade balance, though only modestly (by US$10 billion without adjustment and US$5 billion with adjustment). 45 Second, the value- added content of both bilateral US exports and imports to/from China are well below 1. The simple average VAX ratio across exports and imports is 0.80 without adjustment and 0.66 with adjustment. If VAX ratios for both exports and imports were equal to this average level, this would imply value-added deficits 20% or 34% smaller than the gross deficits. This second ‘level effect’ accounts for most of the adjustment from gross to value-added balances for China (between US$25 billion and US$44 billion of the total change). In con- trast, for Japan, this level effect is virtually nil, as the simple average VAX ratio is near 1 (literally, 0.98 without adjustment and 1.00 with adjustment). The US deficit with Japan rises in value-added terms mainly because the ratio of value-added imports to gross imports is high relative to the ratio of value- added exports to gross exports (the VAX ratio for imports is 0.16 higher than for exports in both calculations). 45 If gross trade were (counterfactually) balanced between the USA and China, the value- added balance would show a surplus due to this force alone. 94 Trade in Value Added 3.3 Inspecting the Mechanism: Bilateral Decompositions To demonstrate that production sharing drives variation in bilateral VAX ratios, we construct two decompositions in the data. The first decomposi- tion splits variation in bilateral VAX ratios into components arising from dif- ferences in the composition of exports across destinations and differences in bilateral production sharing relations. The second decomposition looks directly at how output circulates within cross-border production chains by (approximately) splitting bilateral exports into components absorbed and consumed in the destination, reflected back and ultimately consumed in the source, and redirected and ultimately consumed in a third destination. To construct the first decomposition, we express the bilateral VAX ratio as vaij ι(I − Aii − AIi )yij = ιxij ιxij ι(I − Aii − AIi )(I − Aii )−1 xij ι(I − Aii − AIi )(yij − (I − Aii )−1 xij ) = + . ιxij ιxij bilateral HIY (BHIY) production sharing adjustment (PSA) (4.13) The first term is equivalent to the Hummels–Ishii–Yi measure of the domes- tic content of exports calculated using bilateral exports. For a given source country, it varies only due to variation in the composition of the export basket across destinations. The second term is a production sharing adjustment. This adjustment depends on the difference between the amount of country i output consumed in j , yij , and the gross output from i required to produce bilateral exports to j , (I − Aii )−1 xij . When yij < (I − Aii )−1 xij , the VAX ratio is smaller than the bilateral HIY benchmark. This situation arises when country i’s intermediate goods shipped to country j are either reflected back to country i embedded in foreign produced final goods or intermediate goods used to produce domes- tic final goods, or redirected to third destinations embedded in country j ’s goods. When yij > (I − Aii )−1 xij , the VAX ratio is larger than the HIY bench- mark. This situation arises when country i ships intermediates to some third country that then (directly or indirectly) embeds those goods in final goods absorbed in country j . To quantify the role of each term in explaining bilateral VAX ratios, we decompose the variance of the bilateral VAX ratio for each exporter across destinations, vari (vaij /ιxij ), into variation due to the BHIY term versus the PSA Term. Table 4.3 reports the share of the total variance accounted for by the BHIY and PSA terms for representative exporters. 46 The production sharing adjustment (PSA term) evidently dominates the decomposition. This 46 In the table, we split the covariance equally between the BHIY and PSA terms. Because the covariance is small, our conclusions are not sensitive to how we split the covariance. Accounting for Intermediates 95 implies that variation in production sharing relations across partners, not export composition across destinations, drives the bilateral VAX ratio. In other words, bilateral VAX ratios are determined not by what an exporter sends to any given destination, but rather by how those goods are used abroad. In con- crete terms, even though the USA sends automobile parts to both Canada and Germany, the US VAX ratio with Canada is lower than with Germany because Canada is part of a cross-border production chain with the USA. To look at production chains more directly, we construct a second decom- position that splits bilateral exports according to whether they are absorbed, reflected or redirected by the destination to which they are sent. We con- struct the decomposition using the division of bilateral exports into final and intermediate goods along with the output decomposition for the foreign des- tination: ιxij = ι(cij + Aij yj ) = ι(cij + Aij yjj ) + ιAij yji + ιAij yjk . (4.14) k≠j,i absorption reflection redirection The first term captures the portion of bilateral exports absorbed and con- sumed in destination j , including both final goods from country i and inter- mediates from i embodied in country j ’s consumption of its own goods. The second term captures the reflection of country i’s intermediates back to coun- try i embodied in country j goods. The third term is the summation of coun- try i’s intermediates embodied in j ’s goods that are consumed in all other destinations, ie redirected to third destinations. 47 We report the results of this decomposition for informative bilateral pairs in Table 4.4. Looking at the upper left portion of the table, we see that Japan’s exports to China are primarily either absorbed in China or redirected to the USA. Comparing Japan’s trade with China to that with the USA, we see that Japanese exports to the USA are nearly exclusively absorbed by the USA, indi- cating minimal bilateral US–Japan production sharing. In contrast, looking at the upper right panel, we see that large portions of US exports to Canada and Mexico are reflected back to the USA for final consumption. Looking at the lower left panel, we see that sharing a common border with two differ- ent countries does not necessarily imply tight bilateral production sharing relationships. German exports to France are primarily absorbed there, while 47 This decomposition is only approximate, because the output split used in constructing the decomposition is influenced by the entire structure of cross-border linkages. Nonethe- less, this decomposition is informative as it returns shares that are consistent with the zero order and first round effects of the Leontief matrix inversion (ie [I + A]) describing how final goods absorbed in each destination are produced. We prefer the decomposi- tion in the text to this alternative ‘first-order approximation’ of the production structure because it adds up to bilateral exports. 96 Trade in Value Added nearly half of exports to the Czech Republic are reflected or redirected. Finally, in the lower right corner, we see that Korea is engaged in triangular trade with the USA and other destinations via China. In contrast, a larger share of Korean exports to Japan are eventually consumed there. These results are consistent with our priors regarding the role of China as a production sharing hub in Asia. 48 4 CONCLUDING REMARKS Intermediate goods trade is a large and growing feature of the international economy. Quantification of cross-border production linkages is therefore cen- tral to answering a range of important empirical questions in international trade and international macroeconomics. This requires going beyond specific examples or country/regional studies to develop a complete, global portrait of production sharing patterns. This chapter provides such a portrait using input–output and trade data to compute bilateral trade in value added. We document significant differences between value added and gross trade flows, differences that reflect heterogeneity in production sharing relationships. We look forward to applying this data in future work to deepen our understand- ing of the consequences of production sharing. 5 APPENDIX The basic idea behind the adjustment for processing trade is to split the aggre- gate economy into separate processing and non-processing units, each with its own input–output structure. Both sectors use domestic and imported interme- diates, but they differ in terms of intermediate input intensity and the source (domestic versus imported) of intermediates. Furthermore, all output in the export-processing sector is exported. From the input–output data, we observe the domestic intermediate use matrix mii and import use matrix as mIi for the economy as a whole. From trade data, we observe total exports originating from, and imported interme- P P diates used by, the processing sector, denoted xi and m ¯ Ii , respectively. Out- N P put in the non-processing sector, denoted yi , is calculated by subtracting xi from total output in the input–output accounts. We seek separate intermedi- N P N P ate use matrices for the two sectors {mii , mii , mIi , mIi } and value added by 48 These decompositions are computed without adjusting for processing trade in China. Adjusting for processing trade tends to amplify reflection and redirection effects. Thus, our table understates the amount of redirection within Asia and reflection in US–Mexico trade. Accounting for Intermediates 97 sector {vaN P i , vai } that satisfy N P mii = mii + mii , (4.15) N P mIi = mIi + mIi , (4.16) N N N N yi = vai +ι[mii + mIi ], (4.17) P P P P xi = vai +ι[mii + mIi ], (4.18) P P T ¯ Ii m = mIi ι , (4.19) T 49 where ι is a conformable row vector of ones and ι is its transpose. If there are N sectors, then there are 4(N × N) + 2N unknowns and only 2(N × N) + 3N constraints, so we cannot solve directly for the unknown coeffi- cients. We therefore follow Koopman et al (2008) and use a constrained min- imization routine to impute the unknown coefficients, where the objective function minimises squared deviations between imputed values and target values. Target values are set by splitting intermediate use and value added across processing and non-processing sectors according to their shares in total output. With the resulting split tables, we use bilateral trade data as in the main text to construct bilateral sourcing matrices and the global input–output table. 50 In performing the calculation, we use processing trade shares from Koopman et al (2008) for China. For Mexico, we obtain trade data for the maquiladora sector from the Bank of Mexico. 51 Due to concerns about the quality of dis- aggregate data and the accuracy of the imputation procedure for individual sectors, we aggregate the data to three composite sectors prior to imputing coefficients. Because bilateral value added trade results are essentially iden- tical in the main data when computed with fifty-seven sectors or three com- posite sectors, we believe aggregation does not result in diminished accuracy. REFERENCES Belke, A., and L. Wang (2006). The Degree of Openness to Intra-regional Trade: Towards Value-Added Based Openness Measures. Jahrbucher fur Nationalokonomie und Statistik 226(2), 115–138. Bems, R., R. C. Johnson and K.-M. Yi (2010). Demand Spillovers and the Collapse of Trade in the Global Recession. IMF Economic Review 58(2), 295–326. Chen, X., L. Cheng, K C Fung and L. Lau (2008). The Estimation of Domestic Value- Added and Employment Induced by Exports: An Application to Chinese Exports to 49 These constraints differ from those used by Koopman et al (2008) in that we use the domestic and import intermediate use matrices separately, whereas they pool this infor- mation into a single overall use matrix. 50 In the resulting system, China and Mexico effectively have 2N sectors, where each of the N sectors is separated into processing and non-processing sub-sectors. 51 Data are available at http://www.banxico.org.mx/. 98 Trade in Value Added the United States. In China and Asia: Economic and Financial Interactions, edited by Yin Wong Cheung and Kar Yiu Wong. Routledge. Costinot, A., J. Vogel and S. Wang (2011). An Elementary Theory of Global Supply Chains. NBER Working Paper 16936. Daudin, G., C. Rifflart and D. Schweisguth (2011). Who Produces for Whom in the World Economy? Canadian Journal of Economics 44(4), 1403–1437. Dean, J. M., K. C. Fung and Z. Wang (2007). Measuring the Vertical Specialization in Chinese Trade. USITC Office of Economics Working Paper 2007-01-A. Hanson, G. H., R. J. Mataloni and M. J. Slaughter (2005). Vertical Production Networks in Multinational Firms. The Review of Economics and Statistics 87(4), 664–678. Hummels, D., J. Ishii and K.-M. Yi (2001). The Nature and Growth of Vertical Special- ization in World Trade. Journal of International Economics 54, 75–96. Isard, W. (1951). Interregional and Regional Input–Output Analysis: A Model of a Space- Economy. The Review of Economics and Statistics 33, 318–328. Johnson, R. C. (2008). Factor Trade Forensics: Intermediate Goods and the Factor Con- tent of Trade. Mimeo, Dartmouth College. Johnson, R. C., and G. Noguera (2012). Accounting for Intermediates: Production Shar- ing and Trade in Value Added. Journal of International Economics 86(2), 224–236. Koopman, R., Z. Wang and S.-J. Wei (2008). How Much of Chinese Exports is Really Made in China? Assessing Domestic Value-Added when Processing Trade Is Pervasive. NBER Working Paper 14109. Koopman, R., W. Powers, Z. Wang and S.-J. Wei (2010). Give Credit Where Credit Is Due: Tracing Value Added in Global Production Chains. NBER Working Paper 16426. Linden, G., K. Kraemer and J. Dedrick (2007). Who Captures Value in a Global Innova- tion System? The Case of Apple’s iPod. Mimeo, Personal Computing Industry Center, UC Irvine. Miller, R. (1966). Interregional Feedback Effects in Input–Output Models: Some Prelim- inary Results. Papers in Regional Science 17(1), 105–125. Miroudot, S., and A. Ragoussis (2009). Vertical Trade, Trade Costs and FDI. OECD Trade Policy Working Paper 89. Moses, L. (1955). The Stability of Interregional Trading Patterns and Input–Output Analysis. The American Economic Review 45, 803–826. Moses, L. (1960). A General Equilibrium Model of Production, Interregional Trade, and Location of Industry. The Review of Economics and Statistics 42, 373–397. Noguera, G. (2011). Augmented Gravity: Accounting for Production Sharing. Mimeo, Columbia Business School. NRC (2006). Analyzing the US Content of Imports and the Foreign Content of Exports. Washington, DC: National Research Council, The National Academies Press. Powers, W., Z. Wang and S.-J. Wei (2009). Value Chains in East Asian Production Networks: An International Input–Output Model Based Analysis. USTIC Working Paper 2009-10-C. Puzzello, L. (2012). A Proportionality Assumption and Measurement Biases in the Fac- tor Content of Trade. Journal of International Economics 87(1), 105–111. Reimer, J. (2006). Global Production Sharing and Trade in the Services of Factors. Journal of International Economics 68, 384–708. Trefler, D., and S. C. Zhu (2010). The Structure of Factor Content Predictions. Journal of International Economics 82, 195–207. Yeats, A. J. (2001). Just How Big Is Global Production Sharing? In Fragmentation: New Production Patterns in the World Economy (ed. S. Arndt and H. Kierzkowski), Chap- ter 7. Oxford University Press. Accounting for Intermediates 99 Yi, K.-M. (2003). Can Vertical Specialization Explain the Growth of World Trade? Jour- nal of Political Economy 111, 52–102. Yi, K.-M. (2010). Can Multi-Stage Production Explain the Home Bias in Trade? American Economic Review 100(1), 364–393. 100 Trade in Value Added Table 4.1: VAX ratios by country and sector. Composite Sector Country Code Aggregate Ag.& Nat.R. Manuf. Services Central & Eastern Europe Albania alb 0.79 2.10 0.44 0.97 Armenia arm 0.67 1.21 0.46 1.12 Azerbaijan aze 0.86 1.14 0.18 1.08 Belarus blr 0.69 5.69 0.35 4.25 Bulgaria bgr 0.63 0.85 0.38 1.17 Croatia hrv 0.71 1.04 0.52 0.92 Czech Republic cze 0.59 1.52 0.43 1.51 Estonia est 0.53 1.07 0.34 0.94 Georgia geo 0.77 1.23 0.38 1.44 Hungary hun 0.54 0.96 0.38 1.39 Kazakhstan kaz 0.78 0.53 0.50 3.26 Kyrgyzstan kgz 0.70 0.78 0.49 1.01 Latvia lva 0.64 0.84 0.51 0.96 Lithuania ltu 0.63 0.95 0.46 1.23 Poland pol 0.70 1.34 0.52 1.57 Romania rou 0.70 2.58 0.48 1.95 Russian Federation rus 0.87 0.99 0.41 2.49 Slovakia svk 0.55 1.29 0.39 1.77 Slovenia svn 0.64 2.26 0.44 1.59 Ukraine ukr 0.67 0.92 0.27 2.67 North & South America Argentina arg 0.84 1.27 0.40 2.26 Bolivia bol 0.85 1.08 0.24 1.79 Brazil bra 0.86 0.95 0.51 3.27 Canada can 0.70 1.00 0.44 1.97 Chile chl 0.80 0.92 0.46 2.31 Colombia col 0.86 0.92 0.51 2.16 Costa Rica cri 0.69 0.68 0.37 2.23 Ecuador ecu 0.90 0.90 0.37 3.30 Guatemala gtm 0.79 0.82 0.43 1.83 Mexico mex 0.67 0.69 0.65 0.93 Mexico (adjusted) mex_adj 0.52 0.88 0.41 1.27 Nicaragua nic 0.74 1.12 0.38 2.04 Panama pan 0.84 1.06 0.36 0.91 Paraguay pry 0.84 0.91 0.28 1.07 Peru per 0.93 0.99 0.72 1.78 USA usa 0.77 0.86 0.49 1.58 Uruguay ury 0.71 1.31 0.42 1.30 Venezuela ven 0.89 1.06 0.29 5.54 Source: authors’ calculations based on GTAP Database Version 7.1. Data is for 2004. Accounting for Intermediates 101 Table 4.1: Continued. Composite Sector Country Code Aggregate Ag.& Nat.R. Manuf. Services South Asia & Oceania Australia aus 0.86 0.87 0.50 1.64 Bangladesh bgd 0.75 5.06 0.43 2.66 India ind 0.81 1.80 0.46 1.68 New Zealand nzl 0.82 1.56 0.43 1.60 Pakistan pak 0.82 4.70 0.39 2.18 Sri Lanka lka 0.66 1.10 0.42 1.31 Western Europe Austria aut 0.67 2.09 0.49 1.01 Belgium bel 0.48 0.54 0.32 1.29 Cyprus cyp 0.77 1.18 0.64 0.79 Denmark dnk 0.73 1.27 0.53 1.01 Finland fin 0.72 3.83 0.50 1.52 France fra 0.73 1.17 0.47 1.79 Germany deu 0.74 1.56 0.47 2.52 Greece grc 0.77 1.44 0.56 0.82 Ireland irl 0.66 2.05 0.46 1.11 Italy ita 0.77 2.18 0.53 1.77 Luxembourg lux 0.40 0.83 0.43 0.39 Malta mlt 0.63 0.71 0.62 0.64 Netherlands nld 0.69 0.96 0.43 1.29 Norway nor 0.87 0.91 0.47 1.41 Portugal prt 0.68 2.25 0.46 1.17 Spain esp 0.75 1.19 0.46 1.32 Sweden swe 0.72 1.94 0.43 1.84 Switzerland che 0.67 0.74 0.44 1.43 United Kingdom gbr 0.79 1.05 0.51 1.24 East Asia Cambodia khm 0.62 3.86 0.40 1.26 China chn 0.70 4.11 0.46 2.75 China (adjusted) chn_adj 0.59 3.90 0.40 1.97 Hong Kong hkg 0.73 49.74 0.38 0.84 Indonesia idn 0.79 1.47 0.45 2.39 Japan jpn 0.85 2.70 0.53 3.93 Korea kor 0.63 2.53 0.46 2.62 Lao lao 0.74 1.97 0.33 0.91 Malaysia mys 0.59 1.53 0.41 1.87 Philippines phl 0.58 1.55 0.44 2.15 Singapore sgp 0.37 0.40 0.25 0.80 Taiwan (Chinese Taipei) twn 0.58 1.36 0.39 3.18 Thailand tha 0.60 3.64 0.38 1.52 Vietnam vnm 0.58 1.04 0.35 1.26 Source: authors’ calculations based on GTAP Database Version 7.1. Data is for 2004. 102 Trade in Value Added Table 4.1: Continued. Composite Sector Country Code Aggregate Ag.& Nat.R. Manuf. Services Middle East & Africa Botswana bwa 0.88 0.91 0.57 1.17 Egypt egy 0.81 2.69 0.43 0.79 Ethiopia eth 0.76 1.03 0.18 0.80 Iran irn 0.95 1.09 0.26 1.74 Madagascar mdg 0.75 0.91 0.50 1.02 Malawi mwi 0.72 0.56 0.49 3.70 Mauritius mus 0.72 0.87 0.59 0.86 Morocco mar 0.78 1.26 0.50 1.12 Mozambique moz 0.76 1.25 0.35 1.49 Nigeria nga 0.94 0.95 0.59 0.92 Senegal sen 0.73 1.04 0.48 1.02 South Africa zaf 0.80 0.62 0.45 2.96 Tanzania tza 0.81 1.07 0.26 1.19 Tunisia tun 0.69 1.43 0.38 1.45 Turkey tur 0.76 1.25 0.51 1.46 Uganda uga 0.83 0.89 0.35 1.24 Zambia zmb 0.78 1.02 0.25 9.29 Zimbabwe zwe 0.69 0.58 0.44 2.69 Medians by Region Central & Eastern Europe 0.68 1.10 0.43 1.42 East Asia 0.62 1.97 0.40 1.87 Middle East & Africa 0.77 1.03 0.45 1.21 North & South America 0.84 0.95 0.42 1.97 South Asia & Oceania 0.81 1.68 0.43 1.66 Western Europe 0.72 1.19 0.47 1.29 Overall 0.73 1.09 0.44 1.46 Source: authors’ calculations based on GTAP Database Version 7.1. Data is for 2004. Accounting for Intermediates 103 Table 4.2: Aggregate and manufacturing VAX decompositions. (a) Aggregate VAX decomposition VAXi −VAX Within term Between term Log income per capita −0.008 0.028∗∗ −0.036∗∗∗ (0.005) (0.011) (0.013) R2 0.02 0.07 0.08 N 90 90 90 (b) Manufacturing VAX decomposition VAXi −VAX Within term Between term Log income per capita 0.018∗∗∗ −0.007 0.025∗∗∗ (0.006) (0.009) (0.008) R2 0.11 0.01 0.12 N 89 89 89 Robust standard errors are given in parentheses. Significance levels: ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Constants are included in all regressions. Income per capita equals exporter value added per capita, where value added is calculated using our data and population is from the GTAP 7.1 database. Belgium, Luxembourg and Singapore are excluded from the data in part (a) and Botswana, Hong Kong, Paraguay and Peru are excluded from the data in part (b) as outliers. Table 4.3: Bilateral VAX ratio: bilateral HIY versus production sharing adjustment. Variance decomposition Exporter BHIY term (%) PSA term (%) USA 5 95 Germany 5 95 Japan 1 99 China 9 91 Argentina 1 99 France 8 92 Hungary 5 95 India 7 93 Portugal 9 91 Median country 3 97 See the text for details regarding the decomposition. The median country is the median statistic for all 93 countries in the data. 104 Trade in Value Added Table 4.4: Decomposing trade: absorption, reflection and redirection. Japan exports to: US exports to: China USA Mexico Canada China 64.5 USA 92.7 Mexico 72.3 Canada 68.9 USA 11.1 Canada 1.4 USA 22.1 USA 24.1 Japan 4.3 Mexico 0.7 Canada 0.9 UK 0.7 Germany 2.5 Japan 0.6 Germany 0.4 Japan 0.7 Germany exports to: Korea exports to: France Czech Rep. China Japan France 74.8 Czech Rep. 57.7 China 61.3 Japan 83.1 Germany 3.6 Germany 11.7 USA 12.1 USA 4.7 UK 2.8 UK 3.0 Japan 4.7 China 2.3 USA 2.6 USA 2.6 Germany 2.7 Germany 1.0 All figures are given as percentages. See the text for details regarding the decomposition. The entries in the table describe the approximate share of bilateral exports to each destination that are ulti- mately consumed in that destination. Shares do not sum to 1 because we include only the top four destinations for each bilateral pair. Data is for 2004. 5 Estimating Domestic Content in Exports when Processing Trade Is Pervasive ROBERT KOOPMAN, ZHI WANG AND SHANG-JIN WEI 1 For many questions, it is crucial to know the extent of domestic value added (DVA) in a country’s exports, but the computation is more complicated when processing trade is pervasive. We propose a method for computing domestic and foreign contents that allows for processing trade. By applying our frame- work to Chinese data, we estimate that the share of domestic content in its manufactured exports was about 50% before China’s WTO membership, and has risen to nearly 60% since then. There are also variations across sectors. Those sectors that are likely to be labelled as relatively sophisticated, such as electronic devices, have particularly low domestic content (about 30% or less). 1 INTRODUCTION This chapter proposes an accounting framework for estimating the domestic/ foreign content share in a country’s exports when processing trade is preva- lent. We then apply the framework to the People’s Republic of China (PRC), one of the world’s best known processing exporters; however, the underlying methodology is relevant for all countries (eg Mexico and Vietnam) that use a processing trade scheme. Indeed, the World Trade Organization has identified more than 130 countries that use some form of processing exports (WTO and IDE-JETRO 2011). Processing trade can take on other names in some coun- tries, such as a duty drawback scheme, which means a rebate of tariffs paid on imported inputs if they are used for exports. Of course, the choice of China as an illustration of the general methodology is not random. ‘Made in China’ is one of the most common labels one encoun- ters in a shopping mall in the USA and Europe. Increasingly, many products 1 This chapter was originally published as Koopman et al (2012). It has been modified slightly to conform with the style of this book. We are grateful for the helpful comments by participants of numerous conferences and seminars, and to the editor of this book. The views expressed in this chapter are those of the authors alone. They do not neces- sarily reflect the views of the US International Trade Commission, or any of its individual Commissioners. We are solely responsible for any errors in the chapter. 106 Trade in Value Added that are supposed to be technically sophisticated and therefore likely to be associated with exports from high-income countries, such as digital cameras and computers, also carry that label. Since the most salient characteristic of the factor endowment in China is a vast supply of unskilled labour relative to either physical or human capital, is the country’s actual export structure inconsistent with the predictions from the international trade theory based on its endowment? A possible resolution to the puzzle is that China is simply the last section of a long global production chain that ends up assembling components from various countries into a final product before it is exported to the USA and EU markets. Indeed, a MacBook computer carries a label on its reverse (in small type) that reads ‘Designed by Apple in California; assembled in China’. This label is likely to be oversimplified already, as it reports only the head and the tail of a global production chain and skips many other countries that supply other components that go into the product. China is the archetype of a national economy that is well integrated into a global production chain. It imports raw material, equipment and manufac- tured intermediate inputs, and then exports a big fraction of its output to the world market. The PRC is not the only country whose production and exports are a part of a global chain; Japan, Korea, Singapore and Malaysia, for exam- ple, participate actively in the international divisions of labour. However, the PRC is noteworthy due to its sheer size. In addition, its export/GDP ratio, at 35% or higher in recent years (compared with about 8% for the USA and 13% for India) is extraordinarily high for a large economy. With a reputation as a ‘world factory’, China is a top supplier of manufacturing outsourcing for many global companies. For many policy issues, it is important to assess the extent of domestic con- tent in exports. For example, what is the effect of a currency appreciation on a country’s exports? The answer depends crucially on the share of domestic content in the exports. All other things being equal, the lower the share of domestic content in the exports, the smaller the effect a given exchange rate appreciation would have on trade volume. As another example, what is the effect of trading with the PRC on US income inequality? The answer depends in part on whether the PRC simply exports products that are intensive in low- skilled labour or whether its exports are more sophisticated. Rodrik (2006) notes that the per capita income typically associated with the kind of goods bundle that the PRC exports is much higher than the country’s actual income. He interprets this as evidence that the skill content of its exports is likely to be much higher than its endowment may imply. Schott (2008) documents an apparent rapid increase in the similarity between the PRC’s export structure and that of high-income countries, and interprets it as evidence of a rise in the level of sophistication embedded in the country’s exports. Wang and Wei (2008) use disaggregated regional data to investigate the determinants of the rise in export sophistication. Indeed, many other observers have expressed fear that the PRC is increasingly producing and exporting sophisticated prod- Estimating Domestic Content in Exports 107 ucts and may be providing wage competition for mid- to high-skilled workers in the USA and Europe. However, Xu (2007) points out that the calculation of Rodrik (2006) and Schott (2008) did not take into account possible quality differences between Chinese varieties and those of other countries, and also did not take into account diverse production capabilities and income level in different Chinese regions. Our study further indicates that the calculations by Rodrik (2006) and Schott (2008) do not take into account the imported con- tent in the country’s exports. Therefore, Rodrik’s and Schott’s assessments on the sophistication of China’s exports are very likely to be exaggerated. If the domestic content in exports from the PRC is low, especially in sectors that would have been considered sophisticated or high-skilled in the USA, then imports from the PRC may still generate a large downwards pressure on the wage of low-skilled Americans after all (as pointed out by Krugman 2008). These are important policy questions and have implications for both developing and developed countries. A good understanding of the nature and extent of global supply chains can provide important insights for economists and policymakers. How would one assess foreign versus domestic content in a country’s exports? Hummels et al (2001) (henceforth denoted HIY) propose a method to decompose a country’s exports into domestic and foreign value-added share based on a country’s input–output (IO) table. They make a key assumption that the intensity in the use of imported inputs is the same between production for exports and production for domestic sales. This assumption is violated in the presence of processing exports. Processing exports are characterized by imports for exports with favourable tariff treatment: firms import parts and other intermediate materials from abroad, with exemptions on the imported inputs and tax preferences from local or central governments and, after pro- cessing or assembling, the finished products. It is important to stress that pro- cessing exporters may also use different technologies from normal exporters; these call for different usages of imported inputs. They usually lead to a signif- icant difference in the intensity of imported intermediate inputs in production of processing exports and that in other demand sources (for domestic final sales and normal exports). Since processing exports have accounted for more than 50% of China’s exports every year since 1996, the HIY formula is likely to lead to a significant underestimation of the share of foreign value added in its exports. Since processing exports are widespread, 2 ignoring processing exports is likely to lead to estimation errors, especially for economies that engage in a massive amount of processing trade. In this chapter, we aim to make two contributions to the literature. First, we develop a formula for computing shares of foreign and domestic value added 2 About 3500 export processing zones (EPZs) operated in 130 countries (WTO and IDE- JETRO 2011). 108 Trade in Value Added in a country’s exports when processing exports are pervasive. The formula allows for potential differences in the use of imported inputs between normal and processing exports. We illustrate mathematically that the HIY formula is a special case of this general formula. The differences between the two types of exports could come from differences in the technology used, responses to different tariff or tax treatments, or some other reasons. This chapter does not formally investigate the sources of these differences, and our formula is invariant to the relative importance of the underlying factors. Second, we apply our methodology to China using data for 1997, 2002 and 2007. We esti- mate that the share of foreign value added in PRC’s manufactured exports was about 50% in both 1997 and 2002, almost twice as high as that implied by the HIY formula, but fell to about 40% in 2007 after five years of its WTO member- ship. There also variations across sectors. Those sectors that are likely to be labelled as sophisticated, such as computers, telecommunications equipment and electronic devices, have particularly low domestic content (about 30% or less). By design, this chapter presents an accounting framework and conducts an accounting exercise. As such, it does not examine determinants and conse- quences of changes in the domestic content share in China’s gross exports. However, a solid methodology to estimate foreign value-added share in a country’s exports is a necessary first step towards a better understanding of these issues. In addition to the papers on vertical specialisation in the international trade literature, this chapter is also related to the IO literature. In particular, Chen et al (2004) and Lau et al (2007) were the first to develop a ‘non-competitive’- type IO model for China (ie one in which imported and domestically pro- duced inputs are accounted for separately) and to incorporate processing exports explicitly. However, these papers do not describe a systematic way to infer separate input–output coefficients for production of processing exports versus those for other final demands. It is therefore difficult for others to replicate their estimates or apply their methodology to other countries. They focus on estimating US–China bilateral trade balance and make no connec- tion with vertical specialisation in the international trade literature. In addi- tion, they use an aggregated version of China’s 1995 and 2002 input–output tables, respectively, to perform their analysis, with only 21 goods-producing industries. We provide a more up-to-date and more disaggregated assess- ment of foreign and domestic value added in Chinese exports, with more than 80 goods-producing industries. Finally, they impose an assumption in estimating the import use matrix from the competitive type IO table pub- lished by China’s National Statistical Bureau: within each industry, the mix of the imported and domestic inputs is the same in capital formation, inter- mediate inputs and final consumption. We relax this assumption by refining a method proposed in Dean et al (2007) that combines China’s processing Estimating Domestic Content in Exports 109 imports statistics with United Nations Broad Economic Categories (UNBEC) classification. The rest of the chapter is organised as follows. Section 2 presents a concep- tual framework for estimating shares of domestic and foreign value added in a country’s exports when processing exports are pervasive. It also describes a mathematical programming procedure to systematically infer a set of IO coefficients called for by the new formula but not typically available from a conventional IO table. Section 3 presents the estimation results for Chinese exports. Section 5 concludes. 2 CONCEPTUAL FRAMEWORK AND ESTIMATION METHOD 2.1 When Special Features of Processing Exports Are Not Taken into Account We first discuss how domestic and foreign contents in a country’s exports can be computed when it does not engage in any processing trade. The discussion follows the input–output literature, and is the approach adopted (implicitly) by Hummels et al (2001) and Yi (2003). Along the way, we will point out a clear connection between the domestic content concept and the concept of vertical specialisation. 3 When imported and domestically produced intermediate inputs are accounted separately, value-based input–output table can be specified as fol- lows: 4 AD X + Y D = X, (5.1) M M A X +Y = M, (5.2) D M uA + uA + Av = u, (5.3) where AD = [aD ij ] is an n × n matrix of direct input coefficients for domestic products, AM = [aM ij ] is an n × n matrix of direct inputs of imported goods, Y D is an n × 1 vector of final demands for domestically produced products, including usage in gross capital formation, private and public final consump- tion and gross exports, Y M is an n × 1 vector of final demands for imported products, including usage in gross capital formation, private and public final consumption, X is an n × 1 vector of gross output, M is an n × 1 vector of 3 We use the terms ‘domestic value added’ and ‘domestic content’ interchangeably. Sim- ilarly, we use the terms ‘foreign value added’, ‘foreign content’ and ‘vertical specialisation’ to mean the same thing. 4 Such a model is called a ‘non-competitive’ model in the IO literature. HIY do not this system explicitly but go straight to the implied Leontief inverse, while Chen et al (2004) specify only the first two equations. A fully specified model facilitates better understanding of the connection between vertical specialisation and domestic content, and facilitates a comparison with the model in the next subsection that features processing exports. 110 Trade in Value Added imports, Av = [av j ] is a 1 × n vector of each sector j ’s ratio of value added to gross output and u is a 1 × n unity vector; i and j indicate sectors, and super- scripts ‘D’ and ‘M’ represent domestically produced and imported products, respectively. Equations (5.1) and (5.2) define horizontal balance conditions for domesti- cally produced and imported products, respectively. A typical row k in Equa- tion (5.1) specifies that total domestic production of product k should be equal to the sum of the sales of product k to all intermediate and final users in the economy (the final sales include domestic consumption and capital formation, plus exports of product k). A typical row h in Equation (5.2) specifies that the total imports of product h should be equal to the sum of the sales of product h to all users in the economy, including intermediate inputs for all sectors, plus final domestic consumption and capital formation. Equation (5.3) is both a vertical balance condition and an adding-up constraint for the input–output coefficient. It implies that the total output X in any sector k has to be equal to the sum of direct value added in sector k and the cost of intermediate inputs from all domestically produced and imported products. From Equation (5.1) we have 5 X = (I − AD )−1 Y D . (5.4) D −1 (I − A ) is the well-known ‘Leontief inverse’, a matrix of coefficients for total domestic intermediate product requirement. Define a vector of share of domestic content, or domestic value added, in a unit of domestically pro- duced products, DVS = {dvs}j , a 1 × n vector, as the additional domestic added generated by one additional unit of final demand of domestic prod- ucts (∆Y D = uT ): ˆv ∆X A DVS = ˆv (I − AD )−1 = Av (I − AD )−1 , =A (5.5) ∆Y D where A ˆv is an n × n diagonal matrix with av as its diagonal elements. Equa- j tion (5.5) indicates that the domestic content for an IO industry is the cor- responding column sum of the coefficient matrix for total domestic interme- diate goods requirement, weighted by the direct value-added coefficient of each industry. Because the standard model assumes that exports and domes- tic sales are produced by the same technology, the share of domestic content in final demand and the share of domestic content in total exports are the same. So Equation (5.5) is also the formula for the share of domestic content in total exports for each industry. Define a vector of share of foreign content (or foreign value added) in final demand for domestically produced products by FVS = u − DVS. By making use of Equation (5.3), it can be verified that FVS = u − Av (I − AD )−1 = uAM (I − AD )−1 . (5.6) 5 (I − AD ) has to be full rank. Estimating Domestic Content in Exports 111 Intermediate use Production for Production Final use Gross output domestic use & of processing (C + I + G + E) or imports normal exports exports DIM 1, 2, …, N 1, 2, …, N 1 1 Production for 1 domestic use … & normal ZDD ZDP YD − EP X − EP Domestic exports (D) N intermediate 1 inputs Processing … exports (P) 0 0 EP EP N 1 Intermediate inputs ZMD ZMP YM M … from imports N Value added 1 VD VP Gross output 1 X − EP EP Figure 5.1: Input–output table with separate production account for processing trade. For each industry, this is the column sum of the coefficient matrix for total intermediate import requirement. This turns out to be the same formula used to compute vertical specialisation by Hummels et al (2001). In other words, the concepts of vertical specialisation and of foreign content are identical. 2.2 Domestic Content in Exports when Processing Trade Is Prevalent We now turn to the case in which processing exports are prevalent and, impor- tantly, these exports could have a different intensity in the use of imported inputs than do domestic final sales (and normal exports). Conceptually, we wish to keep track separately of the IO coefficients of the processing exports and those of domestic final sales and normal exports. For now, we ignore the fact that these IO coefficients may not be directly available, and shall discuss a formal approach to estimate them in the next subsection. The IO table with a separate account for processing exports is represented by Figure 5.1. We use superscript ‘P’ and ‘D’, respectively, to represent processing exports on the one hand, domestic sales and normal exports on the other. This expanded IO model can be formally described by the following system of 112 Trade in Value Added equations: I − ADD −ADP X − EP Y D − EP = , (5.7) 0 I EP EP AMD (X − E P ) + AMP E P + Y M = M, (5.8) DD MD uA + uA + AD v = u, (5.9) DP MP uA + uA + AP v = u. (5.10) This is a generalisation of the model discussed in the previous subsection. Equations (5.7) and (5.8) are a generalisation of Equations (5.1) and (5.2), and Equations (5.9) and (5.10) are a generalisation of Equation (5.3), with a separate account for processing exports. Equations (5.9) and (5.10) are also the new adding-up constraint for the IO coefficients. The analytical solution of the system is −1 X − EP I − ADD −ADP Y D − EP = . (5.11) EP 0 I EP The generalised Leontief inverse for this expanded model can be computed follows: −1 I − ADD −ADP B DD B DP (I − ADD )−1 (I − ADD )−1 ADP B= = = . 0 I B PD B PP 0 I (5.12) Substituting Equation (5.12) into Equation (5.11), we have X − E P = (I − ADD )−1 (Y D − E P ) + (1 − ADD )−1 ADP E P . (5.13) Substituting Equation (5.13) into Equation (5.8), the total demand for imported intermediate inputs is M − Y M = AMD (I − ADD )−1 (Y D − E P ) + AMD (1 − ADD )−1 ADP E P + AMP E P . (5.14) It has three components: the first is total imported content in domestic sale and normal exports, and the second and the third are indirect and direct imported content in processing exports, respectively. We can compute vertical specialisation (VS) or foreign share processing and normal exports in each industry separately: T T VSSD uAMD (I − ADD )−1 = . (5.15) VSSP uAMD (1 − ADD )−1 ADP + uAMP The total foreign content share in a particular industry is the sum of the two weighted by the share of processing and non-processing exports s P and uT s P , where both s and u are 1 × n vectors: VSSD VSS = (u − s P , s P ) . (5.16) VSSP Estimating Domestic Content in Exports 113 The foreign (or foreign value added) share in country’s total exports is: E − EP EP TVSS = uAMD (I − ADD )−1 + u(AMD (1 − ADD )−1 ADP + AMP ) , (5.17) te te where te is a scalar, the country’s total exports. Equation (5.16) is a generalisa- tion of Equation (5.7), the formula to compute industry-level share of vertical specialisation. Equation (5.17) is a generalisation of the formula for country- level share of vertical specialisation proposed by Hummels et al (2001, p. 80). In particular, either when ADD = ADP and AMD = AMP , or when E P / te = 0, Equation (5.18) reduces to the HIY formula for VS. Similarly, the domestic content share for processing and normal exports at the industry level can be computed separately: T DVSD (I − ADD )−1 (I − ADD )−1 ADP ¯v B = AD =A v AP v DVSP 0 I DD −1 T ADv (I − A ) = DD −1 DP . (5.18) Av (I − A ) A + AP D v The total domestic content share in a particular industry is weighted sum of the two: DVSD DVS = (u − s P , s P ) . (5.19) DVSP The domestic content share in a country’s total exports is: DD −1 E − EP DD −1 DP EP TDVS = AD V (I − A ) + (AD V (1 − A ) A + AP V) . (5.20) te te When either ADD = ADP and AD v = Av or E / te = 0, Equation (5.20) reduces to P P the HIY formula in Equation (5.5). Note we can easily verify that for both pro- cessing and normal exports the sum of domestic and foreign content shares is unity. 2.3 Estimation Issues Equations (5.18)–(5.20) allow us to compute the shares of domestic content in processing and normal exports for each industry as well as in a country’s total exports. However, statistical agencies typically only report a traditional IO matrix, A, and sometimes AM , but not ADP , ADD , AMP and AMD separately. Therefore, a method to estimate these matrices, based on available informa- tion, has to be developed. In this subsection, we propose to do this via a quadratic programming model by combining information from trade statis- tics and conventional IO tables. The basic idea is to use information from the standard IO table to determine sector-level total imports/exports, and information from trade statistics to determine the relative proportion of processing and normal exports within 114 Trade in Value Added each sector, and thus use all available data to split the national economy into processing and non-processing blocks, each with its own IO structure. The first step (using the data from the IO table to determine sector-level total imports/exports) helps to ensure that the balance conditions in the official IO account are always satisfied, and that the IO table with separate processing and non-processing accounts are consistent with the published official table. The second step (using data from trade statistics to determine the relative proportion of processing and normal exports within each sector) helps to ensure that the estimated new IO table is consistent with the trade structures implied by official trade statistics. The following data are observable from a standard IO table and enter the model as constants. xi : gross output of sector i. zij : goods i used as intermediate inputs in sector j . vj : value added in sector j . mi : total imports of sector i goods. yi : total final demand except for exports of goods i. We combine those observed data from the IO table and processing trade shares 6 observed from trade statistics to determine the values for the follow- ing. p mi : imports of sector i good used as intermediate inputs to produce processing exports. d mi : imports of sector i goods used as intermediate inputs for domestic production and normal exports. n ei : normal exports of sector i. p ei : processing exports of sector i. The partition of imports into intermediate and final use is based on a com- bination of China Customs import statistics and UN BEC classification, as described in Dean et al (2007). The results of such partition and the actual numbers used in our empirical estimation are reported and discussed in Section 3.1 on the data source. Parameters on domestic and imported final demand can be inferred from the observed data discussed above. m d p yi : final demand of goods i from imports (residuals of mi − mi − mi ). d yi : final demand of goods i provided by domestic production m (residual of yi − yi ). 6 China Customs officially report processing and normal exports at the HS eight-digit- level. Processing trades include trade regime ‘Process & assembling’ (14) and ‘Process with imported materials’ (15) in China Customs statistics. These statistics are relatively accu- rate because they involve duty exemption and value-added tax rebates, which come under intensive Customs monitoring. Estimating Domestic Content in Exports 115 Define the following. dd zij : domestically produced intermediate good i used by sector j for domestic sales and normal exports. dp zij : domestically produced intermediate good i used by sector j for processing exports. md zij : imported intermediate good i used by sector j for domestic sales and normal exports. mp zij : imported intermediate good i used by sector j for processing exports. d vj : direct value added by domestic and normal export production in industry j . p vj : direct value added by processing export production in industry j . Then the IO coefficients for the expanded IO model can be written as: dd md zij zij A DD = [add ij ] = p , A MD = [amd ij ] = p , xj − e j xj − e j d dp vd vj dp zij AD v = [aj ] = p , ADP = [aij ] = p , xj − e j ej mp p mp zij vp vj AMP = [aij ] = p , AP v = [aj ] = p . ej ej To obtain unobservable IO coefficients, we need to estimate with-industry dd dp md mp transactions [zij ], [zij ], [zij ] and [zij ], as well as sector-level value added d p [vj ] and [vj ], subject to the flowing IO accounting identities and adding-up constraints: K dd dp n d p (zij + zij ) = xi − ei − ei − yi , (5.21) j =1 K md mp m (zij + zij ) = mi − yi , (5.22) j =1 K dd md d p (zij + zij ) + vj = xj − e j , (5.23) j =1 K dp mp p p (zij + zij ) + vj = ej , (5.24) i=1 K md d zij = mi , (5.25) j =1 K mp p zij = mi , (5.26) j =1 116 Trade in Value Added K K dd dp d p (zij + zij ) = zij − (mi + mi ), (5.27) j =1 j =1 dd mddp mp zij + zij + zij + zij = zij , (5.28) d p vj + vj = vj . (5.29) The economic meanings of the nine groups of constraints are straightfor- ward. Equations (5.21) and (5.22) are row sum identities for the expanded IO account. They state that total gross output of sector i has to equal the sum of domestic intermediaries, final demand and exports (both processing and normal exports) in that sector. Similarly, total imports have to equal imported intermediate inputs plus imports delivered to final users. Equations (5.23) and (5.24) are column sum identities for the expanded IO account. They define the value of processing exports in sector j as the sum of domestic and imported intermediate inputs as well as primary factors used in producing processing exports; these four groups of constraints correspond to Equations (5.7)–(5.10) in the extended IO model, respectively. Equations (5.25) to (5.29) are a set of adding-up constraints to ensure that the solution from the model is consistent with official statistics on sector-level trade and within-industry transactions. We can make initial guesses about the values of the unobserved within- industry transactions and sector-level value added using a combination of offi- cial statistics and some proportional assumptions (to be made precise later). These initial values may not satisfy all the adding-up constraints and need to be modified. We cast the estimation problem as a constrained optimisation procedure to minimise following objective functions: dd dd 2 dp dp md md 2 K K (zij − z0ij ) K K (zij − z0ij )2 K K (zij − z0 ij ) min S = dd + dp + md i=1 j =1 z0 ij i=1 j =1 z0ij i=1 j =1 zij mp mp d d 2 p p K K (zij − z0ij )2 K (vj − v0j) K (vj − v0j )2 + mp + d + p . (5.30) i=1 i=1 z0ij j =1 v0j j =1 v0j Here z and v are variables to be estimated, and those variables with a 0 in the suffix denote initial values. Because all parameters in the nine groups of lin- ear constraints (the right-hand sides of Equations (5.21)–(5.29)) were directly or indirectly obtained from observable official statistical sources, model solu- tions are thus restricted into a convex set and will be relatively stable with respect to variations in these initial values as long as all the parameters in these linear constraints are kept as constants. md md d p The initial values of zij and zij are generated by allocating mi and mi in proportion to input i’s usage in sector j as in Equation (5.31): p p mp zij (ej /xj ) p md zij (xj − ej )/xj d z0ij = N p mi , z0 ij = N p mi . (5.31) k zik (ek /xk ) k zik (xk − ek )/xk Estimating Domestic Content in Exports 117 The split of total inter-sector intermediate inputs flowing from sector i to sector j between normal and processing use are based on their proportion in gross output. The residuals of the total intermediate inputs and the imported intermediate inputs estimated from Equation (5.31) are taken as the initial values for domestically produced intermediate inputs as in Equations (5.32) and (5.33): p dd (xj − ej ) md z0 ij = zij − z0 ij , (5.32) xj p dp ej mp z0ij = zij − z0ij . (5.33) xj The initial values for direct value added in the production for processing p exports in sector j (v0j ), are generally set to be the residuals implied by Equa- tion (5.24). However, we set a minimum value at the sum of labour compensa- tion depreciation in a sector multiplied by the share of processing exports in p that sector’s total output. In other words, the initial value v0j is set equal to the greater of the residuals from Equation (5.24) or the minimum value. The initial value for direct value added in the production for domestic sales and d normal exports (v0 j ) is set as the difference between vj (from the IO table) p and v0j . We conduct some sensitivity checks using alternative initial values. It turns out that they do not materially alter our basic conclusions. We implement this programming model in Gams (Brooke et al 2005); related computer pro- grams and data files are available at the authors’ and the USITC websites for downloading. 3 ESTIMATION RESULTS After describing the data sources, we report and discuss the estimation results for shares of domestic and foreign content in Chinese exports at the aggregate level, and by sector, firm ownership and major destination countries. 3.1 Data Inter-industry transaction and (direct) value-added data are from China’s 1997, 2002 and 2007 benchmark IO tables published by the National Bureau of Statistics of China (NBS), while detailed export and import data of 1997, 2002 and 2007 are from the General Customs Administration of China. The trade statistics are first aggregated from the eight-digit HS level to China’s IO industry, and then used to compute the share of processing exports in each IO industry. Modifying a method from Dean et al (2007), we partition all imports in a given commodity classification into three parts, based on the distinction 118 Trade in Value Added Table 5.1: Major trade share parameters used in estimation, 1997–2008. Imported intermediates (%) Imported capital goods (%) Processing Imported exports as for for for for final % of processing normal processing normal consumption total exports use exports use (%) exports Year 1 2 3 4 5 6 1996 46.2 26.8 16.7 8.1 2.2 56.0 1997 51.2 28.2 12.1 7.3 1.3 55.1 1998 50.7 28.2 9.7 10.0 1.4 57.4 1999 43.6 35.0 8.2 11.2 2.0 57.3 2000 39.4 41.2 8.5 9.1 1.8 55.7 2001 36.6 41.2 8.7 11.6 1.9 55.9 2002 38.0 39.1 10.2 11.0 1.8 55.9 2003 35.0 41.8 10.7 10.8 1.6 56.0 2004 34.7 43.0 11.8 8.9 1.5 56.3 2005 36.1 43.6 10.6 8.1 1.5 55.6 2006 35.3 44.2 9.8 8.9 1.7 53.6 2007 32.7 47.3 9.0 7.6 3.3 50.1 2008 27.5 53.5 8.1 7.2 3.7 48.1 Source: authors’ calculations based on official China Customs trade statistics and the United Nations Broad Economic Categories (UNBEC) classification scheme. ‘Normal use’ refers to ‘normal exports and domestic sales’. The UNBEC scheme classifies each HS six-digit product into one of three categories: ‘intermediate inputs’, ‘capital goods’ and ‘final con- sumption’. For the first two categories, we further decompose the imports into two subcategories: ‘processing imports’ by customs declaration are classified as used for producing processing exports and cannot be sold to any domestic users by regulation, and the remaining imports are classified as for normal use. Capital goods are part of the final demand in a conventional IO model (columns 1–5 sum to 100%). However, this classification may underestimate the import content of exports. We therefore also experiment with classifying a fraction of the capital goods as inputs used in current year of production. This is discussed in Section 3.2. between processing and normal imports in the trade statistics, and on the UN BEC classification scheme: (i) intermediate inputs in producing processing exports; (ii) intermediate inputs for normal exports and other domestic final sales; and (iii) those used in gross capital formation and final consumption. A summary of these trade statistics as a percentage of China’s total imports along with share of processing exports during the period 1996–2008 is reported in Table 5.1, which shows a downwards trend for the use of imported inputs in producing processing exports and an upwards trend in their use in producing normal trade and domestic final sales. 7 7 Sector level counterparts of the data in Table 5.1 are used to determine the parameters in Equations (5.21)–(5.26). Additional parameters in Equations (5.27)–(5.29) are obtained directly from China’s official benchmark IO tables. Estimating Domestic Content in Exports 119 We report detailed trade share parameters for each IO industry in the three benchmark year (1997, 2002 and 2007) in the online appendix. These data computed directly from detailed Chinese official trade statistics (at eight-digit HS) are important to understand our estimates of domestic and imported content in Chinese gross exports, especially cross-sector heterogeneity and their changes over time. Our estimation results reflect these parameters. 3.2 Domestic and Foreign Content in Total Exports Table 5.2 presents the results for the decomposition of aggregate foreign and domestic value-added shares in 1997, 2002 and 2007. For comparison, the results the HIY method that ignores processing trade are also reported. The estimated aggregate domestic value-added share in China’s merchandise exports was 54% in 1997 and 60.6% in 2007. For manufacturing products, these estimated shares are slightly lower but trend upwards more significantly from 50.0% in 1997 to 59.7% in 2007. In general, the estimated direct domestic value-added shares are less than half of the total domestic value-added shares. However, the estimated indirect foreign value-added share is relatively small; most of the foreign content comes from directly imported foreign inputs, particularly in 1997 and 2002. The indirect foreign value added increases over time, and reaches about a quarter of China’s directly imported foreign inputs in 2007, indicating that the share of simple processing and assembling of foreign parts is declining, while more imported intermediates are being used in the production of other intermediate inputs that are then used in the production process of exported goods. Relative to the estimates from the HIY method, our procedure produces esti- mates of a much higher share of foreign value added in Chinese gross exports and with a different trend over time. To be more precise, estimates from the HIY method show that the foreign content share (total VS share) increased from 17.6% in 1997 to 28.7% in 2007 for all merchandise exports, and from 19.0% to 27.1% for manufacturing only during the same period. In contrast, our estimates suggest a trend in the opposite direction, with the share of foreign value added in all merchandise exports falling from 46% in 1997 to 39.4% in 2007, and a somewhat more dramatic decline for the share in man- ufacturing exports from 50% in 1997 to 40.3% in 2007. The decline occurred mainly during the period 2002–7, which corresponds to the first five years of China’s inclusion in the WTO. Our estimates indicate that the HIY method appears to incorrectly estimate both the level and the trend in domestic versus foreign content in the PRC’s exports. These striking differences indicate the importance of taking account of differences between processing and normal exports. What accounts for the difference between our approach and the HIY approach? There are at least three factors that drive the change of foreign content share in the country’s gross exports, including: 120 Trade in Value Added Table 5.2: Shares of domestic and foreign value added in total exports (%). The HIY method The KWW method 1997 2002 2007 1997 2002 2007 All merchandise Total foreign value added 17.6 25.1 28.7 46.0 46.1 39.4 Direct foreign value added 8.9 14.7 13.7 44.4 42.5 31.6 Total domestic value added 82.4 74.9 71.3 54.0 53.9 60.6 Direct domestic value added 29.4 26.0 20.3 22.2 19.7 17.1 Manufacturing goods only Total foreign value added 19.0 26.4 27.1 50.0 48.7 40.3 Direct foreign value added 9.7 15.6 16.3 48.3 45.1 32.4 Total domestic value added 81.1 73.6 72.9 50.0 51.3 59.7 Direct domestic value added 27.5 24.6 24.6 19.6 18.1 16.5 Source: authors’ estimates based on China’s 1997, 2002 and 2007 Benchmark Input–Output Table published by the Bureau of National Statistics, and official China trade statistics from China Customs. The HIY method refers to estimates from using the approach in Hummels et al (2001). The KWW method refers to estimates from using the approach developed in this chapter that takes into account special features of processing exports. 1. the relative proportions of imported intermediate inputs in producing processing exports and normal exports and domestic sales; 2. the share of processing exports in its total exports; and 3. the sector composition of its exports. Because processing exports tend to use substantially more imported inputs, and processing exports account for a major share of China’s total exports, the HIY indicator substantially underestimates the true degree of foreign content in China’s exports. This explains why the level of domestic content by our measure is much lower than that of the HIY indicator. On the other hand, as exporting firms (both those producing for normal exports and those pro- ducing for processing exports) gradually increase their intermediate inputs sourcing from firms within China, including multinationals that have moved their upstream production to China, the extent of domestic content in exports rises over time. This process is likely to be aided by China’s accession to the WTO. However, because exports from industries with relatively lower domes- tic content often grow faster, the composition of a country’s total exports may play as an offsetting role to slow down the increase of domestic value-added share in the country’s total exports. As the Chinese government started to narrow the gap in policy treatments for both foreign-invested firms relative to domestic firms and processing exports relative to normal exports since the end of 2006, the domestic content share of Chinese exports could continue its rise in the future. Our interpretation is confirmed by DVA shares for processing and normal exports estimated separately (Table 5.3). There is an increase by more than Estimating Domestic Content in Exports 121 Table 5.3: Domestic and foreign value added: processing versus normal exports (as percentage of total exports). Normal exports Processing exports 1997 2002 2007 1997 2002 2007 All merchandise Total foreign value added 5.2 10.4 16.0 79.0 74.6 62.7 Direct foreign value added 2.0 4.2 5.0 78.6 73.0 58.0 Total domestic value added 94.8 89.6 84.0 21.0 25.4 37.3 Direct domestic value added 35.1 31.9 23.4 11.7 10.1 10.9 Manufacturing goods only Total foreign value added 5.5 11.0 16.4 79.4 75.2 63.0 Direct foreign value added 2.1 4.5 5.2 79.0 73.6 58.3 Total domestic Value added 94.5 89.0 83.6 20.7 24.8 37.0 Direct domestic value added 31.5 29.5 22.4 11.7 10.0 10.9 Source: authors’ estimates based on China’s 1997, 2002 and 2007 Benchmark Input–Output Table published by the Bureau of National Statistics and official China trade statistics from China Customs. 10 percentage points in the total foreign value-added share for domestic sales and normal exports between 1997 and 2007. However, in processing exports, we see that as more domestically produced inputs were used, the domestic value-added share increased from 20.7% in 1997 to 37.0% in 2007, up by more than 16 percentage points. Because processing exports still constitute more than 50% of China’s total exports in 2007, the domestic value-added share in total exports climbed up during the decades. Because the gap in the domestic content shares is large between the two types of exports, it is unlikely to disappear any time soon. We perform a number of robustness checks on the sensitivity of our main results to alternative ways set the initial values of variables and the share p d parameters of import use. First, we initialise v0j and v0 j by apportioning the observed direct value added in a sector to processing exports and other final demands based on their respective portions in the sector’s total output. Sec- p ond, we initialise v0j either at the residuals implied by Equation (5.24) if the residuals are positive, or by following the previous alternative if the residu- als are non-positive. Third, when we partition imports into different users, we use the average of a three-year period (previous, current and following years) rather than just one year’s statistics. Fourth, we experiment with 0% versus 10% annual depreciation rate for capital goods. These variations pro- duce relatively little change in the main results. In particular, the pattern of a trend increase in the domestic content share in total exports is robust to these variations. 122 Trade in Value Added Table 5.4: Shares of domestic value added in exports by firm ownership (%), 2002 and 2007. Non- Weighted processing Processing sum Direct Total Direct Total Direct Total SPET DVA DVA DVA DVA DVA DVA SFT 2002 Wholly foreign owned 87.5 34.9 90.1 9.8 25.3 13.0 33.4 28.9 Joint venture firms 70.5 31.2 89.4 9.9 24.5 16.2 43.6 22.9 State-owned firms 32.2 32.1 89.6 10.7 26.4 25.2 69.3 38.1 Collectively owned firms 27.4 29.9 89.6 10.8 28.2 24.7 72.8 5.8 Private firms 9.0 30.7 89.6 10.7 26.3 28.9 83.9 4.3 All firms 55.7 31.8 89.3 10.1 26.1 19.7 53.9 100.0 2007 Wholly foreign owned 83.0 23.8 83.8 11.4 36.0 13.5 44.1 38.1 Joint venture firms 59.5 23.0 83.6 10.4 38.7 15.5 56.9 17.7 State-owned firms 25.8 23.4 83.4 10.0 39.5 20.0 72.1 18.9 Collectively owned firms 24.0 22.4 83.1 8.9 42.0 19.1 73.3 4.0 Private firms 9.6 23.5 84.9 9.8 42.0 22.2 80.8 21.3 All firms 50.0 23.5 83.9 10.5 38.7 17.1 60.6 100.0 Source: authors’ estimates based on China’s 2002 and 2007 Benchmark Input–Output Table published by Bureau of National Statistics and official China trade statistics from China Customs. SPET, share of processing exports in total exports; SFT, share of exports by firm ownership in China’s total exports. The IO structure is assumed to be the same for a given export regime within a sector across all type firms. The variation of domestic value added by firm types is due solely to variation in sector composition and the relative reliance on processing exports. 3.3 Domestic Content in Exports by Firm Ownership Since foreign-invested firms account for over half of China’s exports, one may be interested in the domestic content share in their exports. However, since there is no information on separate input–output coefficients by firm owner- ship, we cannot meaningfully distinguish foreign firms from local firms within a sector and trade regime (processing or normal exports). Instead, we provide an estimate of the domestic content share of aggregate exports by foreign- invested firms. By construction, the differences across firms of different own- ership are driven entirely by different degrees of their reliance on processing exports within a sector and differences in the sector composition of their total exports (both are observed directly from the customs trade statistics). Estimates of the domestic content shares by firm ownership are in Table 5.4. The results show that exports by wholly foreign-owned enterprises exhibit the lowest share of domestic value added, but rose relatively quickly (from 33.4% in 2002 to 44.1% in 2007), followed by Sino-foreign joint venture companies (at about 44% in both 2002 and 2007). Exports from Chinese private enterprises embodied the highest domestic content shares (83.9% and 80.8% in 2002 and 2007, respectively), while those from state-owned firms were in the middle Estimating Domestic Content in Exports 123 (about 70% in both years). Note that these estimates represent the best guesses based on currently available information; better estimates can be derived once information on IO coefficients by firm ownership becomes available. The most noticeable feature of this table is the rising domestic content shares in exports produced by foreign-invested firms by more than 10 per- centage points from 2002 to 2007. This suggests that the increase in the domestic content share is mainly due to foreign-invested processing exporters sourcing more of their intermediate inputs from within China. This is presum- ably also linked to more multinationals moving their upstream production to China. 3.4 Domestic Content by Sector To see if there are interesting patterns at the sector level, Tables 5.5 and 5.6 report, in ascending order of the domestic content share, the value-added decomposition in Chinese exports by industry in 2002 and 2007, respectively, together with the shares of processing trade and foreign-invested firms in each sector’s exports and the sector’s share in China’s merchandise exports. Because the sector classifications are consistent between 2002 and 2007 (but less so between 1997 and 2002), we choose to report the sector-level results for 2002 and 2007. Among the 57 manufacturing industries in the table, 15 have a share of domestic value added in their exports less than 50% in 2002; they account for nearly 35% of China’s merchandise exports that year. It is interesting to note that many low-DVA industries are likely to be labelled as sophisticated, such as telecommunication equipment, electronic computers, measuring instru- ments or electronic devices. A common feature of these industries is that processing exports account for over two-thirds of their exports (and foreign- invested enterprises played an overwhelming role). In 2007, the number of industries with less than 50% domestic contents in their exports declined to 10, and their collective share in China’s total exports also declined to 32%. The next 18 industries in Table 5.6 have their shares of domestic value added in the range 51–65%; they collectively accounted for 28% of China’s total merchandise exports in 2002. Several labour-intensive sectors are in this group, as furniture, toys and sports products, leather, fur, down and related products. The remaining 24 industries have high shares of domestic value added. They as a group produced slightly less than 30% of China’s total merchan- dise exports in 2002. Apparel, the country’s largest labour-intensive export- ing industry, which by itself was responsible for 7% of the country’s total mer- chandise exports in 2002, is at the top of this group, with a share of domestic content of 66%. The 12 industries at the bottom of Table 5.6 with DVA share more than 75% collectively produced only 10% of China’s total merchandise exports in 2002. 124 Trade in Value Added Table 5.5: Domestic value added share in manufacturing exports by sector, 2002. VA decomposition (%) % of % of % of IO industry Non- Weighted process. FIE merch. description processing Processing sum exports exports exports Telecommunication 87.5 5.3 12.5 91.2 88.4 3.2 equipment Shipbuilding 82.3 14.7 17.5 95.8 21.0 0.6 Electronic computer 83.6 18.7 19.3 99.1 89.7 7.0 equipment Cultural and 79.7 19.3 23.3 93.4 71.6 4.3 office equipment Household electric 88.2 6.8 23.9 79.1 56.9 1.9 appliances Household 82.5 21.3 27.0 90.6 62.3 5.2 audiovisual apparatus Printing, reproduction 91.1 19.7 31.9 83.0 62.7 0.3 or recording media Plastic 84.4 10.3 36.6 64.5 51.2 2.4 Electronic components 84.6 32.8 38.1 89.7 87.5 3.4 Steelmaking 89.0 12.8 44.3 58.8 86.1 0.0 Generators 85.2 32.0 44.3 76.8 55.8 0.9 Other electronic and 97.8 36.0 45.3 84.9 84.9 1.8 communication equipment Rubber 90.6 12.2 48.9 53.1 44.4 1.6 Non-ferrous metal 86.2 7.5 49.3 46.9 48.7 0.4 pressing Measuring instruments 85.8 32.9 49.5 68.6 51.8 1.8 Paper and paper products 90.8 12.4 51.1 50.7 57 0.5 Furniture 88.3 12.5 52.5 47.2 56.8 1.7 Articles for culture, 87.5 38.2 52.7 70.6 56.3 3.3 education and sports activities Non-ferrous metal smelting 88.9 10.6 53.6 45.0 17.4 0.8 Smelting of ferroalloy 83.6 13.0 54.8 40.8 13.1 0.2 Source: authors’ estimates. China 2002 and 2007 Benchmark Input–Output Tables have 84 and 90 goods-producing sectors, respectively. They both concord to China’s four-digit classification of eco- nomic activities (GB/T 4754-2002). This concordance enables us to aggregate both year’s estimates to 77 consistent goods-producing industries reported in this table. The high-DVA industries saw their weights in the country’s total exports to rise significantly from 2002 to 2007. The number of industries with DVA share of more than 75% increased from 12 in 2002 to 25 in 2007 (compar- ing the bottoms of Tables 5.5 and 5.6), and their exports as a share of the country’s total exports also rose from 10% in 2002 to more than 30% in 2007. Among these high-DVA industries, besides the traditional labour-intensive Estimating Domestic Content in Exports 125 Table 5.5: Continued. VA decomposition (%) % of % of % of IO industry Non- Weighted process. FIE merch. description processing Processing sum exports exports exports Synthetic materials 80.5 37.1 55.2 58.3 65.4 0.3 Petroleum refining and 79.4 5.5 55.7 32.1 24.9 0.8 nuclear fuel Metal products 90.3 10.2 55.7 43.2 45.6 4.4 Other transport 86.0 12.7 55.8 41.2 50.5 1.2 equipment Other electric machinery 88.4 40.1 56.2 66.8 60.1 5.6 and equipment Special 82.9 31.4 58.7 46.9 48.4 0.8 chemical products Other 89.2 31.3 59.0 52.2 37.6 1.7 manufacturing products Woollen textiles 91.1 8.8 60.1 37.8 42.6 0.3 Paints, printing inks, 83.5 8.3 61.6 29.1 44.4 0.4 pigments and similar products Motor vehicles 89.6 10.0 61.6 35.2 48.2 0.8 Glass and its products 86.8 16.5 63.6 33.0 48.8 0.5 Leather, fur, down and 91.9 40.4 63.9 54.3 50.3 4.5 related products Chemical products for 85.3 26.8 64.1 36.3 43.6 0.4 daily use Wearing apparel 91.3 34.3 65.6 45.1 39.2 7.0 Chemical fibre 80.2 9.2 65.7 20.5 29.2 0.0 Other special 89.3 32.0 66.4 39.9 44.0 1.3 industrial equipment Boiler, engines 85.9 13.1 66.5 26.7 28.4 0.4 and turbine Other industrial 90.1 38.6 67.6 43.7 43.7 3.5 machinery Iron-smelting 86.8 11.0 68.8 23.7 3.0 0.1 Source: authors’ estimates. China 2002 and 2007 Benchmark Input–Output Tables have 84 and 90 goods-producing sectors, respectively. They both concord to China’s four-digit classification of eco- nomic activities (GB/T 4754-2002). This concordance enables us to aggregate both year’s estimates to 77 consistent goods-producing industries reported in this table. industries such as furniture, textiles and apparel, we start to see capital- and skill-intensive industries such as automobile, industrial machinery and rolling steel (accounting for nearly one third of these high-DVA sector’s exports). This is likely to reflect industrial upgrading in the Chinese economy. 126 Trade in Value Added Table 5.5: Continued. VA decomposition (%) % of % of % of IO industry Non- Weighted process. FIE merch. description processing Processing sum exports exports exports Railroad transport 83.9 14.6 70.1 19.9 5.9 0.1 equipment Wood, bamboo, rattan, 87.8 11.3 72.8 19.6 45.6 1.0 palm and straw products Knitted and crocheted 90.6 34.7 72.9 31.6 34.2 5.8 fabrics and articles Agriculture, forestry, 85.7 13.9 72.9 17.8 20.8 0.1 animal husbandry and fishing machinery Pesticides 77.0 11.5 72.9 6.3 14.4 0.2 Hemp textiles 89.5 11.7 74.3 19.5 19.5 0.3 Textiles productions 90.1 28.9 75.5 24.0 31.8 1.4 Cotton textiles 91.8 35.6 75.7 28.7 28.8 3.3 Fire-resistant 90.5 15.4 76.2 19.1 49.8 0.1 materials Metalworking 87.2 18.8 78.1 13.3 27.0 0.2 machinery Medicines 90.2 24.3 79.1 16.9 28.7 0.7 Pottery and porcelain 88.2 14.8 79.8 11.4 33.1 0.7 Other non-metallic 90.4 16.7 80.1 14.0 35.7 0.4 mineral products Fertilisers 84.4 9.7 81.1 4.5 21.7 0.1 Basic 87.1 43.7 82.0 11.7 18.8 2.0 chemical raw materials Rolling of steel 90.2 40.5 82.3 16.0 16.8 0.3 Cement, lime 91.0 20.3 86.0 7.0 77.7 0.1 and plaster Coking 91.4 13.2 89.4 2.6 5.3 0.3 Total merchandise 89.6 25.4 53.9 55.7 51.8 92.5 Source: authors’ estimates. China 2002 and 2007 Benchmark Input–Output Tables have 84 and 90 goods-producing sectors, respectively. They both concord to China’s four-digit classification of eco- nomic activities (GB/T 4754-2002). This concordance enables us to aggregate both year’s estimates to 77 consistent goods-producing industries reported in this table. 4 DVA SHARES IN CHINESE EXPORTS BY TRADING PARTNERS By assuming that domestic value-added shares within a given sector and export regime are the same for all destination countries, we can further estimate the domestic value-added share in China’s exports to each of its major trading partners. Note, however, that the variation by destination in Estimating Domestic Content in Exports 127 Table 5.6: Domestic value-added share in manufacturing exports by sector, 2007. VA decomposition (%) % of % of % of IO industry Non- Weighted process. FIE merch. description processing Processing sum exports exports exports Household 75.9 29.6 32.6 93.4 79.1 2.5 audiovisual apparatus Electronic computer 75.7 33.0 33.9 97.9 93.3 11.3 equipment Cultural and 74.1 33.1 36.5 91.7 86.4 1.6 office equipment Other electronic and 68.0 34.7 39.7 84.8 81.6 1.4 communication equipment Telecommunication 75.2 35.3 43.6 79.3 83.6 5.9 equipment Ship building 83.9 39.1 43.8 89.4 16.5 1.1 Petroleum refining 68.7 20.1 44.4 50.1 27.3 0.7 and nuclear fuel Measuring instruments 80.0 37.8 45.8 81.2 73.3 2.5 Synthetic materials 76.4 34.0 47.7 67.7 66.1 0.6 Household electric 82.0 35.6 51.8 65.1 61.7 2.7 appliances Other electric 80.3 33.7 52.1 60.5 65.9 4.9 machinery and equipment Rubber 81.8 27.0 53.4 51.8 41.9 1.7 Plastic 80.8 31.1 55.1 51.7 54.7 1.7 Articles for culture, 83.0 45.6 58.4 66.0 64.9 2.1 education and sports activities Special chemical 76.7 34.0 61.6 35.3 51.2 0.8 products Source: authors’ estimates. China 2002 and 2007 Benchmark Input–Output Tables have 84 and 90 goods-producing sectors, respectively. They both concord to China’s four-digit classification of economic activities (GB/T 4754-2002). This concordance enable us to aggregate both year’s estimates to 77 consistent goods-producing industries reported in this table. this method is driven solely by China’s export structure (sector composition) to each of its trading partners. The decomposition results for China’s total merchandise exports to each of its major trading partners are reported in Table 5.7 in increasing order of the estimated domestic value-added share in 2002. Hong Kong, the USA, Singapore, Taiwan (Chinese Taipei) and Malaysia are at the top of the table in both 2002 and 2007, with less than or about 60% of China’s domestic value added embodied in their exports. The notewor- thy pattern is that China’s exports to developing countries tend to embody 128 Trade in Value Added Table 5.6: Continued. VA decomposition (%) % of % of % of IO industry Non- Weighted process. FIE merch. description processing Processing sum exports exports exports Chemical fibre 76.4 51.9 62.6 56.2 48.7 0.3 Other special 82.5 43.0 65.2 43.8 54.7 2.7 industrial equipment Generators 80.3 51.2 66.6 47.2 50.3 0.7 Railroad transport 77.7 54.1 69.0 37.0 12.2 0.1 equipment Leather, fur, down and 90.4 40.4 69.2 42.5 46.0 2.4 related products Paper and paper 85.5 57.6 69.2 58.4 62.8 0.4 products Metal products 85.1 39.7 70.1 32.9 49.5 4.4 Boiler, engines 81.6 38.7 70.6 25.6 37.8 0.5 and turbines Non-ferrous 78.6 56.1 71.2 32.7 41.4 1.0 metal pressing Other manufacturing 86.5 48.1 72.3 36.8 41.5 1.6 products Paints, printing inks, 76.5 56.8 72.6 20.1 47.3 0.3 pigments and similar products Pesticides 73.9 53.6 72.9 4.8 19.5 0.1 Chemical 80.8 58.4 73.3 33.5 55.5 0.3 products for daily use Non-ferrous 76.2 56.4 73.3 14.6 19.6 0.8 metal smelting Other transport 81.0 54.9 73.8 27.8 46.5 0.9 equipment Basic chemical 80.8 42.5 74.9 15.6 26.4 1.9 raw materials Motor vehicles 84.0 47.4 75.3 23.7 42.0 2.0 Source: authors’ estimates. China 2002 and 2007 Benchmark Input–Output Tables have 84 and 90 goods-producing sectors, respectively. They both concord to China’s four-digit classification of economic activities (GB/T 4754-2002). This concordance enable us to aggregate both year’s estimates to 77 consistent goods-producing industries reported in this table. much higher domestic value added than its exports to OECD countries. While this pattern appears to mirror the finding by Manova and Zhang (2012) that China’s export prices tend to be lower in lower income countries, our data and method do not allow us to estimate destination-specific domestic value share of a product. Interestingly, the domestic value-added share in China’s exports to high- income countries increased between 2002 and 2007, while it declined for Estimating Domestic Content in Exports 129 Table 5.6: Continued. VA decomposition (%) % of % of % of IO industry Non- Weighted process. FIE merch. description processing Processing sum exports exports exports Agriculture, forestry, 80.6 57.7 75.6 21.9 32.7 0.1 animal husbandry and fishing machinery Other industrial 83.6 56.2 75.6 29.0 49.9 3.4 machinery Iron-smelting 75.9 50.6 75.6 1.1 24.3 0.1 Smelting of 75.7 53.3 75.6 0.4 8.8 0.4 ferroalloy Furniture 86.7 56.1 76.2 34.2 56.0 2.0 Printing, reproduction 86.4 61.0 76.5 39.0 44.4 0.2 or recording media Glass and 83.3 59.0 76.7 27.2 46.4 0.6 its products Woollen textiles 89.4 57.9 76.9 39.8 46.8 0.2 Metalworking 81.2 56.8 77.3 16.0 36.4 0.3 machinery Rolling of steel 80.0 52.9 77.8 8.3 22.6 3.8 Fertilisers 81.0 57.3 77.9 13.2 9.5 0.3 Cotton textiles 88.0 45.8 78.9 21.5 26.1 2.1 Wearing apparel 89.5 53.9 79.0 29.7 36.9 4.6 Medicines 87.6 37.5 80.3 14.5 32.3 0.8 Wood, bamboo, rattan, 84.6 58.4 80.4 16.1 33.1 1.0 palm and straw products Steelmaking 80.8 51.7 80.8 0.2 7.1 0.3 Pottery and porcelain 83.4 58.2 82 5.2 29.9 0.5 Textiles productions 88.4 54.9 82.4 18.1 35.1 1.8 Knitted and crocheted 88.2 51.6 82.5 15.6 25.7 5.7 fabrics and articles Other non-metallic 86 56.6 83 10.1 25.1 0.5 mineral products Hemp textiles 86.6 56.8 83.9 9.0 14.7 0.2 Fire-resistant 86.6 55.1 84.7 5.8 51.6 0.1 materials Cement, lime and plaster 89.0 52.9 88.4 1.7 29.6 0.1 Coking 89.6 — 89.6 0.0 11.4 0.3 Total merchandise 84.0 37.3 60.6 50.1 55.7 96.0 Source: authors’ estimates. China 2002 and 2007 Benchmark Input–Output Tables have 84 and 90 goods-producing sectors, respectively. They both concord to China’s four-digit classification of economic activities (GB/T 4754-2002). This concordance enable us to aggregate both year’s estimates to 77 consistent goods-producing industries reported in this table. exports to developing countries. This suggests that progressively more locally supplied inputs were used in making exports to high-income countries, while the opposite may be true for exports to developing countries. 130 Trade in Value Added Table 5.7: Total domestic value-added share in Chinese gross merchandise exports to its major trading partners (%), 2002 and 2007. Share of Share in processing total exports Non- Weighted exports to in total exports processing Processing sum the world Region description 2002 2007 2002 2007 2002 2007 2002 2007 2002 2007 Hong Kong 74.0 77.4 89.8 83.0 26.3 35.3 42.8 46.0 17.5 14.3 USA 67.2 61.7 89.2 84.6 24.3 38.2 45.5 56.0 21.6 19.1 Singapore 62.7 59.7 88.7 83.4 24.3 33.0 48.3 53.3 2.1 2.4 Taiwan province 59.6 50.7 89.3 81.9 27.1 34.9 52.2 58.0 2.0 1.9 (Chinese Taipei) Malaysia 57.6 52.0 90.4 84.0 25.5 33.5 53.0 57.7 1.5 1.5 Japan 59.2 56.4 90.7 85.4 27.6 40.5 53.3 60.1 15.0 8.4 EU15 54.8 50.9 89.4 84.0 23.6 37.2 53.4 60.2 14.9 18.3 Thailand 48.1 38.8 88.3 82.0 22.9 38.7 56.8 65.2 0.9 1.0 Rest of OECD 46.9 38.5 89.7 85.4 25.4 40.3 59.5 68.0 1.7 2.1 Rep. of Korea 45.4 43.2 90.4 83.5 27.1 37.0 61.6 63.4 4.8 4.7 Australia/NZ 41.6 42.8 89.3 84.4 23.0 38.6 61.7 64.8 1.6 1.7 Mexico 42.1 49.1 89.6 84.2 26.6 35.8 63.1 60.4 0.9 0.9 Philippines 37.6 38.2 89.1 83.5 25.2 33.8 65.1 64.5 0.6 0.6 EU12 36.5 50.8 90.2 83.4 22.9 35.8 65.7 59.2 1.5 1.9 Brazil 35.0 36.7 89.4 83.2 27.1 37.7 67.6 66.5 0.5 0.9 India 24.0 27.0 89.3 81.7 21.5 38.6 73.1 70.1 0.8 2.0 Rest of 20.3 24.2 89.2 83.4 23.1 38.1 75.8 72.5 1.6 2.4 Latin America/ Caribbean Indonesia 20.7 23.4 89.4 83.3 25.8 36.1 76.2 72.2 1.1 1.1 Middle East/ 19.4 18.2 89.3 83.9 21.9 38.8 76.3 75.6 3.6 4.8 N. Africa Eastern Europe/ 18.9 16.6 89.4 85.0 26.3 39.2 77.5 77.4 0.9 2.8 Central Asia Rest of Asia 17.2 18.9 88.6 83.5 27.0 41.6 77.9 75.6 2.2 2.6 Sub-Saharan Africa 15.5 16.1 89.6 83.9 22.1 38.8 79.2 76.6 1.4 2.1 Russia 15.5 16.9 90.9 85.6 30.4 39.3 81.5 77.8 1.1 2.4 World 55.7 50.0 89.6 84.0 25.4 37.3 53.9 60.6 100.0 99.9 Source: authors’ estimates based on China’s 2002 and 2007 Benchmark Input–Output Table published by Bureau of National Statistics and Official China trade statistics from China Customs. IO structure is assumed to be the same for a given export regime within a sector across all trading partners. The variation of domestic value added by destination is due solely to variations in sector composition and the relative reliance on processing exports. 5 CONCLUDING REMARKS Segmentation of production across countries allows for reductions in produc- tion costs and more efficient allocation of resources, but also creates a wedge between the gross export value and the domestic value added that is embed- ded in the exports. Because processing exports may have a different tendency to use imported inputs from normal exports, it is important to account for Estimating Domestic Content in Exports 131 such differences in estimating the share of domestic value added in a coun- try’s exports. In this chapter, we present a general framework in assessing the shares of domestic and foreign value added in a country’s exports when process- ing exports are explicitly accounted for. This formula nests the existing best known approach (Hummels et al 2001) as a special case. If separate input– output coefficients for processing and normal exports are available, our for- mula can be applied in a straightforward way. Because some of the IO coefficients called for by the new formula are not readily available from conventional IO tables, we propose an easy-to-replicate mathematical programming procedure to estimate these coefficients by com- bining information from detailed trade statistics (which records process- ing and normal exports/imports separately) with conventional input–output tables. This methodology should be applicable to Vietnam, Mexico and many other developing countries that engage in a significant amount of processing exports. By applying our methodology to the Chinese data, we find several inter- esting patterns. First, the share of foreign content in China’s manufactur- ing exports was close to 50% during 1997–2002, almost twice as high as that calculated using the HIY formula. Second, the share of domestic con- tent increased from 51% to 60% during 2002–7, which corresponds to the first five years of China’s membership of the WTO. We also report interesting heterogeneity across sectors: sectors that are likely to be labelled as sophisti- cated or high-skilled, such as computers, electronic devices and telecommu- nication equipment, tend to have notably lower shares of domestic content. Conversely, many sectors that are relatively intensive in low-skilled labour, such as apparel, are likely to exhibit a high share of domestic content in the country’s exports. Finally, we find that foreign-invested firms (including both wholly owned foreign firms and Sino-foreign joint venture firms) tend to have a relatively low share of domestic content in their exports, as they tend to use more processing exports and take large shares in sectors that have a relatively low domestic value-added share. There are several areas in which future research can improve upon the esti- mation in this chapter. First, we assign initial values of the direct domestic value added for processing exports at industry level based on information in conventional IO table and proportion assumptions. If firm-level survey data becomes available that track separately the direct value added for process- ing and normal exports, and that provide information on how the imported intermediate inputs are allocated across sector users, we can improve the accuracy of our estimates. Second, as an inherent limitation of an IO table, the input–output coefficients are assumed to be fixed (which is the nature of the assumed Leontief technology) rather than be allowed to respond to price changes. If the relevant IO tables are available every year, then the vari- ations in the IO coefficients would be recorded. If IO tables are available only 132 Trade in Value Added sparsely (eg once every five years), which tends to be the case for developing countries, then estimating domestic value shares in exports based on past IO tables could be problematic, especially in years when large shocks could induce large (but unobserved) changes in the IO coefficients. This chapter does not directly investigate causes and consequences of changes in the domestic content share in exports. These can be fruitful areas for future research. REFERENCES Banister, J. (2005). Manufacturing Employment in China. BLS Monthly Labor Review (July), 11–29. Brooke, A., D. Kendrick, A. Meeraus and R. Raman (2005). Gams: User’s Guide. GAMS Development Cooperation, Washington, DC. Chen, X., L. Cheng, K. C. Fung and L. J. Lau (2004). The Estimation of Domestic Value- Added and Employment Induced by Exports: An Application to Chinese Exports to the United States. Working Paper. Stanford University. Chinn, M. D. (2005). Supply Capacity, Vertical Specialization and Tariff Rates: The Implications for Aggregate US Trade Flow Equations. NBER Working Paper 11719. Dean, J. M., K. C. Fung and Z. Wang (2007). Measuring the Vertical Specialization in Chi- nese Trade. Office of Economics Working Paper 2007-01-A. US International Trade Commission. Goh, A.-T., and J. Olivier (2004). International Vertical Specialization, Imperfect Com- petition and Welfare. Working Paper, HEC School of Management (France). Hummels, D., J. Ishii and K. Yi (2001). The Nature and Growth of Vertical Specialization in World Trade. Journal of International Economics 54, 75–96. Koopman, R., Z. Wang and S.-J. Wei (2008). How Much Chinese Exports Is Really Made in China: Assessing Foreign and Domestic Value Added in Gross Exports. NBER Working Paper 14109. Koopman, R., Z. Wang and S.-J. Wei (2012). Estimating Domestic Content in Exports when Processing Trade Is Pervasive. Journal of Development Economics 99(1), 178– 189. Krugman, P. (2008). Trade and Wages, Reconsidered. Brookings Papers on Economic Activity. URL: http://www.brookings.edu/. Lau, L. J., X. Chen, L. K. Cheng, K. C. Fung, Y. Sung, C. Yang, K. Zhu, J. Pei and Z. Tang. (2007). Non-Competitive Input–Output Model and Its Application: An Examination of the China–US Trade Surplus. Social Sciences in China 2007(5), 91–103 (in Chinese). Lawrence, R. (2008). Blue Collar Blues: Is Trade to Blame for Rising US Income Inequal- ity?, Policy Analyses in International Economics, Volume 85. Peter G. Peterson Insti- tute for International Economics, Washington, DC. Linden, G., K. L, Kraemer and J. Dedrick (2007). What Captures Value in a Global Inno- vation System? The Paul Merage School of Business, UC Irvine, Working Paper. Manova, K., and Z. Zhang (2012). Export Prices across Firms and Destinations. Quar- terly Journal of Economics 127(1), 376–436. National Research Council (2006). Analyzing the US Content of Imports and the Foreign Content of Exports. Committee on Analyzing the US Content of Imports and the Foreign Content of Exports. Center for Economics, Governance, and International Studies, Division of Behavioral and Social Sciences and Education. Washington, DC: The National Academies Press. Estimating Domestic Content in Exports 133 Rodrik, D. (2006). What’s So Special about China’s Exports?’ China & World Economy 14(5), 1–19. Schott, P. (2008). The Relative Sophistication of Chinese Exports. Economic Policy 53, 5–49. Varian, H. R. (2007). An iPod Has Global Value. Ask the (Many) Countries That Make It. New York Times, 28 June. Wang, Z., and S.-J. Wei. (2008). What Accounts for the Rising Sophistication of China’s Exports? NBER Working Paper 13771, February. WTO and IDE-JETRO (2011). Trade Patterns and Global Value Chains in East Asia: From Trade in Goods to Trade in Tasks. Report. World Trade Organization, Geneva. Xu, B. (2007). Measuring China’s Export Sophistication. Working Paper, China Europe International Business School. Yi, K.-M. (2003). Can Vertical Specialization Explain the Growth of World Trade? Jour- nal of Political Economy 111(1), 52–102. 6 Foreign and Domestic Content in Mexico’s Manufacturing Exports JUSTINO DE LA CRUZ, ROBERT B. KOOPMAN, ZHI WANG AND SHANG-JIN WEI 1 This chapter provides estimates of foreign and domestic content in Mexico’s manufacturing exports that take into account the import content in produc- tion under the maquiladora and Programa de Importación Temporal para Pro- ducir Artículos de Exportación (PITEX) programmes. We applied a modified version of the methodology developed in Koopman et al (2011) by using a recently available input–output table for the maquiladora industry. We also applied the original method suggested in Koopman et al (2011) and compare the results obtained under both methodologies. This is the first study for Mexico that measures vertical specialisation using a recently available input– output table for the maquiladora industry in addition to trade data from both export promotion programmes. On average, Mexico’s manufacturing exports have a foreign content share of approximately 66%. Those industries that have a foreign content share of 50% or more account for 80% of the country’s man- ufacturing exports. These include computer and peripheral equipment, audio and video equipment, communications equipment, semiconductor and other electronic components and electrical equipment. 1 INTRODUCTION Mexico’s international trade (exports plus imports of goods) grew from US$82.3 billion in 1990 to US$700.5 billion in 2011, an increase of 751.2%. The North American Free Trade Agreement (NAFTA), which took effect on 1 January 1994, played an instrumental role. Total bilateral trade between the USA and Mexico increased by 435.4% from US$78.9 billion in 1993, the year 1 The authors are grateful to Hubert Escaith, Ted H. Moran, Ralph Watkins, Ruben Mata, Hugh Arce, Christine McDaniel and Ricardo Rojas for helpful comments, and Eric Cardenas and Natalia Buniewicz for research assistance. They are especially grateful to José Arturo Blancas Espejo, Rodolfo Daude Balmer, Ernesto Garcia Zuñiga and Jaime A. de la Llata from INEGI for providing data and input–output tables. The views in the chapter are those of the authors and are not the official views of the USITC or of any other organisation that the authors are affiliated with. 136 Trade in Value Added US imports from Mexico US exports to Mexico 450 400 350 Pre-NAFTA Post-NAFTA US$ (billion) 300 250 200 150 100 50 0 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 Figure 6.1: US–Mexico goods trade. Source: US Department of Commerce. prior to NAFTA entering into force, to US$422.6 billion in 2011 (Figure 6.1). In relative terms, Mexico’s share of US imports has also increased from 6.7% in 1993 to 12.0 in 2011. Mexico together with Canada accounted for 26.5% of US imports of goods in 2011. The USA is Mexico’s largest trading partner, and Mexico is the third largest trade partner the USA after Canada and China. In 2011, the USA accounted for 49.7% of Mexico’s total imports, and 78.6% of its total exports. While the trade volume has exploded, the relative dominance of the USA in Mexico’s trade has not changed much; these ratios were 69.3% and 82.7%, respectively, in 1993. 1.1 Production Fragmentation and Its Economic Effects Cross-border production sharing or vertical specialisation has increased its relative importance in world trade and is thought to be responsible for the faster rate of growth in the trade share of GDP (Yi 2003). As a measure of for- eign value added or foreign content in exports, vertical specialisation distorts trade data in terms of export content to GDP, as noted by Feenstra (1998), Feenstra and Hanson (2004) and Johnson and Noguera (2012). Recent litera- ture in international economics shows that vertical specialisation may have important economic effects on wage inequality, employment and business cycles, and on the pass-through effects of changes in tariffs and exchange rates. In addition, it may also have policy implications for the relationships between trade, trade facilitation, investment and intellectual property policy, and the relationship between trade and competition policy (Nordås 2005). Regarding wage inequality, Feenstra (1998, 2008), Feenstra and Hanson (1999, 2004), Krugman (2008) and Ebenstein et al (2009) note that global production sharing, outsourcing or trade in intermediate inputs are poten- tially important in explaining wage differentials between skilled and unskilled Foreign and Domestic Content in Mexico’s Manufacturing Exports 137 20 16 Montly changes (%) year to year 12 8 4 0 −4 −8 −12 USA Mexico −16 −20 Jan Jan Jan Jan Jan Jan Jan Jan Jan Jan Jan Jan 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 Figure 6.2: US and Mexico manufacturing production, 2000–10. Source: Board of Govenors of the Federal Reserve System and Banco de México. The correlation coefficient between these indexes is 0.92. workers in the USA and elsewhere. Specifically, Feenstra and Hanson (1999) found that outsourcing explains 15% of the increase in the US relative wage of nonproduction workers during the period 1979 to 1990. Trade in inputs or vertical specialisation depresses the demand for less-skilled workers while raising the relative demand and wages of the higher skilled. Evidence on Mex- ico also suggests that outsourcing by multinationals has contributed to the increase in the relative wage of skilled workers in the country (Feenstra and Hanson 1997). 2 Production sharing has the potential to synchronise business cycles as well as to increase the volatility and severity of economic fluctuations. Burstein et al (2008) (in a multi-country setting) and López (2007) (for a small open economy) show that production sharing can generate business cycle syn- chronisation. The Lopéz model of business cycle, in which the transmission mechanism is production sharing, successfully replicated real business statis- tics of the Mexican maquiladora, or production sharing manufacturing sec- tor. Empirically, Herrera-Hernandez (2004) and Chiquiar and Ramos-Francia (2005) show that the US and Mexican manufacturing sectors became syn- chronised after NAFTA was enacted. This also seems to be the case dur- ing the period from 2000 to 2011 (Figure 6.2). Furthermore, Bergin et al 2 Rising wage inequality in Mexico may also be explained by trade and quality upgrad- ing noted by Verhoogen (2008), and by trade liberalisation, as suggested by Hanson and Harrison (1999) and Chiquiar (2008). 138 Trade in Value Added (2009a,b) provide theoretical and empirical evidence suggesting that the Mex- ican maquiladora industry associated with US production sharing experiences fluctuations in employment that are twice as volatile as those of their coun- terpart industries in the USA. Feenstra (2008, p. 87) adds: That fact that the maquiladora industries are more volatile means that the US is essentially exporting some of its business cycle, or more precisely, export- ing the cyclical fluctuations due to demand shocks. Regarding vertical specialisation and the severity of business cycles, Yi (2009) analysed the recent collapse of global trade, suggests that vertical specialisa- tion can amplify trade effects so that the collapse in global trade in the fourth quarter of 2008 was sudden, severe and synchronised. Yi’s explanation is based on the linkage between US exports and US imports, ie when US imports decline so do US exports of intermediate goods used in the manufacturing of US imports of final goods. In this instance, we have a multiplicative effect as vertical specialisation links a country’s imports to its exports. With respect to tariffs, in an earlier paper, Yi (2003) theorised that, because of vertical specialisation, tariff reductions can have magnifying effects on imports prices. Empirically, Feenstra (2008) confirmed this with evidence from the Information Technology Agreement (ITA) of the WTO, under which tariffs on high-tech goods were eliminated from 1997 to 1999. Feenstra estimated a tariff pass-through coefficient of 22.6, suggesting that the multilateral tariff reductions under ITA had magnified effects on decreasing US import prices, as prices declined many times more than the tariff decreases. In contrast, the pass-through effect of exchange rates under production sharing seems to be relatively small both empirically and theoretically, which has contributed to keeping prices low. 3 Bergin and Feenstra (2008) estimated that the pass- through effect of exchange rates would fall by about one-fifth of its size as a result of the growing share of US trade with China, a major source of off- shoring. Additionally, Ghosh (2008) presents a theoretical model in which the exchange rate pass-through is lower with production sharing trade compared with the situation of standard trade. The pass-through symmetry of tariffs and exchange rates was tested by Feenstra (1989), but not under production sharing. 1.2 The Maquiladora Programme The maquiladora programme started in the mid-1960s with plants and a few employees manufacturing televisions and plastics (INEGI 2007; Truett and Truett 1984). However, this industry did not grow substantially until the Mex- ican government relaxed its restrictions on foreign direct investment (FDI) 3 Without accounting for the presence of vertical specialisation, most of the current literature asserts that the pass-through effect of exchange rates has been declining from 0.5 to 0.2 (Campa and Goldberg 2006). Foreign and Domestic Content in Mexico’s Manufacturing Exports 139 in the 1980s (Bergin et al 2009a; OECD 1996; Truett and Truett 1984, 1993, 2007). Now, the maquiladora industry appears to be highly integrated with the US manufacturing sector, and most maquiladoras are US owned, but com- panies based in Japan, South Korea and Germany are also important partic- ipants. Maquiladoras received preferential treatment under both the US and Mexican laws by which US firms paid duties on foreign value added only, while Mexico allowed for duty-free imports as long as the maquiladora output was exported back to the USA. However, with the implementation of NAFTA, the preferential tariff treatment afforded to maquiladoras ended. Given the importance of the maquiladora regime as a generator of jobs, exports and foreign exchange in Mexico for more than 35 years, in 2002 the Mexican government established sectoral development programmes (Pro- gramas de Promoción Sectorial, or PROSECs) to maintain competitiveness of the manufacturing sector in Mexico, irrespective of whether or not products were exported (WTO 2008). The PROSECs allowed participating companies to import eligible non-NAFTA inputs and capital equipment at a rate of either 0% or 5% (Gantz 2004). The maquiladoras’ finished products were not contin- gent to subsequent exportation and were permitted to be sold in Mexico or exported. In addition, maquiladoras’ exports were exempted from the value- added tax and, upon complying with certain rules, income tax and asset tax were done away with (Baker & McKenzie 2006). Thus, in spite of NAFTA’s Arti- cle 303, growth in the maquiladora industry accelerated, and by 2006 there were 2810 maquiladora plants, with 1.2 million employees. In addition, Bergin et al (2009a) point out that the industry’s real value added approximately tripled between 1994 and 2005. 1.3 PITEX, IMMEX and Other Programmes Mexico’s second major export promotion programme, the ‘Program of Tempo- rary Imports to Produce Export Goods’ (Programa de Importación Temporal para Producir Artículos de Exportación, or PITEX) was established in 1990. This programme, designed for firms already established in Mexico and pro- ducing for the domestic and export markets, also granted fiscal and admin- istrative benefits, eg importing intermediates and machinery free of duty as long as the final product was exported (USITC 1998b). One benefit of PITEX was to allow foreign investors to register as a national supplier to the auto- motive industry (USITC 1998b). Also, the programme included duty drawback for firms that had a significant share of imported inputs in their exports, in addition to special administrative, fiscal and financial benefits (OECD 1996). However, firms under PITEX were subject to taxes for which maquiladora firms were exempt (USITC 1998b). In 2006, PITEX firms numbered 3620 and included all motor vehicle assembly plants and most of their parts suppliers. They tended to locate in the interior of Mexico because a significant portion of their sales was destined to the domestic market, while maquiladora firms 140 Trade in Value Added tended to locate in the border states. PITEX and maquiladora firms together employed approximately 60% of Mexico’s total manufacturing employment in 2006. On 23 November 2006, the Mexican government merged the maquiladora and PITEX programmes into a new regime to promote exports, named the ‘Manufacturing Industry, Maquiladora and Export Services Program’ (Indus- tria Manufacturera, Maquiladora y de Servicios de Exportación, or IMMEX), administered by the Secretariat of Economy. The new programme simplified procedures and requirements for firms’ import inputs, raw materials, parts and components, and machinery and equipment free of duty as long as the fin- ished product was exported. Firms under the IMMEX programme also enjoyed certain tax exemptions. In addition to the programme, Mexico has other programmes to promote export through tariff and tax concessions and administrative facilities. These include the ‘High-Volume Exporting Companies’ (Empresas Altamente Expor- tadoras, ALTEX) programme and the ‘Foreign Trade Companies’ (Empresas de Comercio Exterior, ECEX) programme. At the end of 2006, there were 2644 firms in the ALTEX programme and 340 firms in the ECEX programme. Between 2002 and 2006, the government approved 46,989 Mexican exporters under the duty-drawback programme (WTO 2008). Mexico’s processing exports through its maquiladora, PITEX, and other pro- grammes underscore the importance of estimating the true domestic and for- eign value added in its exports. We estimate these value-added measures by applying a variation of the methodology developed by Koopman et al (2011), which takes into account an actual input–output (IO) table for the maquiladora industry. In contrast, in estimating the domestic value added in China’s exports, Koopman et al (2011) use an optimising algorithm to estimate the structure of processing export sectors. For comparison purposes, we also perform the calculations with their original methodology. In both instances, we assume that other export-promoting programmes, including PITEX, have the same IO coefficients as those of the maquiladora industry. This chapter, to the best of our knowledge, is the first study for Mexico that measures vertical specialisation using a recently available input–output table for the maquiladora industry in addition to using trade data from both export pro- motion programmes, the maquiladora and PITEX; to date most studies on pro- cessing exports for Mexico have used trade data only from the maquiladora industry. Our results suggest that Mexico’s industrial strategy has resulted, although modestly and only in some industries, in its insertion into the global supply chains as the domestic value added share in Mexico’s manufacturing exports increased in recent years. The estimated measures indicate that on average Mexico’s domestic value added in its manufacturing exports is about 34%. Accounting for 80% of the country’s manufacturing exports, 41 industries (out of a total 75 three-digit NAICS), have a domestic content of less than 50%. These industries include Foreign and Domestic Content in Mexico’s Manufacturing Exports 141 computer and peripheral equipment, audio and video equipment, commu- nications equipment, semiconductor and other electronic components, and electrical equipment among others. The remainder of this chapter explains the data and the methodology in Section 2, the estimation results in Section 3 and the conclusion in Section 4. 2 DATA AND ESTIMATION METHOD 2.1 Mexico’s Input–Output Table for 2003 and Trade Data The most up to date input–output table for Mexico was the one for 2003 developed by Mexico’s statistical agency, the Instituto Nacional de Estadís- tica, Geografía e Informática (INEGI), which has 255 four-digit North American Industry Classification System (NAICS) sectors. A notable feature is a specific IO table for the maquiladora industry. 4 This table includes national produc- tion of goods and services classified under Mexico’s NAICS for 2002, inputs purchased in the domestic economy and imports from the rest of the world. Mexico’s trade data at the Harmonized System (HS) eight-digit level for 1996–2006 were obtained from the World Trade Atlas. The data were available for both the maquiladora and PITEX firms’ imports and exports by country source and destination. INEGI reports trade data for the maquiladora indus- try but not PITEX. This is important because excluding PITEX data from an analysis of the processing industry in Mexico would omit important infor- mation. Moreover, US data on production sharing or US imports under HS Chapter 98 are likely to be underestimated as a result of the implementation of NAFTA and other preferential agreements (Burstein et al 2008). The World Trade Atlas trade data are from the Mexican government but the values are greater than those reported by the US Department of Commerce by about 10–12% (US Department of Commerce 2000, 2001). 2.2 Trade Statistics Exports of manufactured goods under the maquiladora and PITEX pro- grammes accounted for 85.4% of the total manufactured exports of US$195.6 billion in 2006, but in previous years this share was larger; for instance, in 2000 it was 93.5% (Table 6.1). Maquiladora and PITEX firms’ imports accounted for 69.8% of their exports in 2006, ie out of one US dollar of exports from these firms, 69.8 cents consisted of imported parts and com- ponents. In 2006, the leading suppliers of these imports were the USA (51%), China (12.2%) and Japan (8.2%) (Table 6.2). Historically, the USA was the pre- dominate supplier, but China, Japan, South Korea, Taiwan (Chinese Taipei), Malaysia and Singapore have gained market shares in recent years. The main 4 We are grateful to INEGI for providing us with the input–output table. 142 Trade in Value Added Table 6.1: Mexico’s processing manufacturing exports, 1996–2006. Year A B C D 1996 86.7 61.9 71.6 28.4 1997 89.0 58.9 69.2 30.8 1998 91.3 58.9 69.6 30.4 1999 93.0 59.6 68.6 31.4 2000 93.5 59.9 70.3 29.7 2001 92.7 57.1 68.0 32.0 2002 91.5 56.3 67.8 32.2 2003 89.9 55.1 68.0 32.0 2004 87.9 54.7 70.3 29.7 2005 85.7 53.2 70.8 29.2 2006 85.4 52.7 69.8 30.2 Source: World Trade Atlas. A: share of processing exports (PE) in total exports (TE) (100 × (PE/TE)). B: share of processing imports (PM) in total imports (TM) (100 × (PM/TM)). C: ratio of processing imports to processing exports (100 × (PM/PE)). D: processing trade surplus as a share of processing exports (100 × (PE − PM)/PE). Processing manufacturing refers to exports and imports under the maquiladora and PITEX programmes. Data include HS Chapters 28–97 only. Table 6.2: Mexico’s total imports for processing exports, by leading markets, 2000–6. Market 2000 2001 2002 2003 2004 2005 2006 USA 80.8 74.5 69.6 68.7 60.3 55.7 51.0 China 1.1 2.0 3.7 6.6 9.3 10.0 12.2 Japan 3.7 5.9 6.9 5.4 6.6 7.8 8.2 Germany 2.8 2.6 2.2 2.3 2.3 2.7 2.8 Canada 1.4 1.6 1.5 1.3 1.6 1.7 1.8 Sum 89.8 86.6 83.9 84.3 80.1 77.9 76.0 Rest 10.2 13.4 16.1 15.7 19.9 22.1 24.0 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Source: World Trade Atlas. Imports for processing exports refer to imports under the maquiladora and PITEX programmes. Data include HS Chapters 1–99. destination of Mexico’s processing exports is the USA, to which Mexico exports about 90%, followed by Canada, with about 2% (Table 6.3). Mexico’s maquiladora processing exports amounted to US$111.9 billion in 2006, including, at the HS two-digit level, electrical machinery (49.0%), machin- ery (18.4%), automobiles and automobile parts (6.2%), medical instruments (6.1%), furniture and bedding (4.2%), knitted and non-knitted apparel (4.2%) and plastics (1.8%). These products combined represent about 90.0% of the total. Similarly, in the same year, Mexican firms under the PITEX programme exported US$62.3 billion, including automobiles and automobile parts (48.7%), machinery (12.3%), electrical machinery (6.4%), iron and steel (3.2%), beverages (3.1%), iron and steel products (3.0%), vegetables (2.9%) and medical instru- ments (2.1%); combined, these represent about 82.0% of the total. Foreign and Domestic Content in Mexico’s Manufacturing Exports 143 Table 6.3: Mexico’s total processing exports, by leading markets, 2000–6. Market 2000 2001 2002 2003 2004 2005 2006 USA 92.4 92.3 92.4 92.8 92.8 90.2 89.1 Canada 2.1 2.0 1.9 1.8 1.4 1.9 2.1 Germany 1.0 1.0 0.7 1.0 0.9 1.3 1.4 Colombia 0.1 0.2 0.2 0.2 0.2 0.6 0.8 Netherlands 0.3 0.3 0.4 0.4 0.3 0.4 0.5 Sum 95.8 95.7 95.6 96.2 95.7 94.5 93.9 Rest 4.2 4.3 4.4 3.8 4.3 5.5 6.1 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Source: World Trade Atlas. Processing exports refer to exports under the maquiladora and PITEX programmes. Data include HS Chapters 1–99. 2.3 Estimation Methods Hummels et al (2001) (henceforth denoted HIY) proposed the concept of ver- tical specialisation (VS) or foreign content or foreign value added in a coun- try’s trade as ‘the imported input content of exports, or equivalently, foreign value added embodied in exports’. They provided a formula to compute shares based exclusively on a country’s IO table. A key assumption needed for the HIY formula to work is that the intensity in the use of imported inputs is the same between production for exports and production for domestic sales. However, Koopman et al (2011) noted that such an assumption is violated in the presence of processing exports and indicated that the HIY formula is likely to lead to a significant underestimation of the share of foreign value added in a country’s exports. This is particularly important when policy preferences for processing trade lead to a significant difference in the intensity of import intermediate inputs in production for processing exports and the production for domestic final sales and normal exports. They developed a formula that can be used to estimate domestic and foreign content for economies that engage in a massive amount of tariff or tax-favoured processing trade, such as that of China, Mexico and Vietnam. They also demonstrated that there is a clear connection between the domestic content concept and the concept of vertical specialisation proposed by HIY. The methodology, applied here, that uses an IO table for the maquiladora industry is discussed in De La Cruz et al (2011). That methodology is based on Koopman et al (2011) (henceforth denoted KWW). It implies that with a one- year single-country IO table and detailed bilateral export data for different years and with different trading partners, one is able to compute the domestic and foreign value-added shares at the aggregate level for different years and trading partners separately. The variation in such a computation will come only from the variations in export composition changes over time and across 144 Trade in Value Added Table 6.4: Domestic and foreign value added in Mexico’s manufacturing exports: three-digit NAICS versus four-digit NAICS (in percent of total manufacturing exports). The KWW formula The HIY formula Lower bounda Upper boundb 2000∗ 2003 2006∗ 2000∗ 2003 2006∗ 2000∗ 2003 2006∗ Based on three-digit NAICS IO table Total foreign VA 47.1 48.9 48.2 54.2 55.0 55.1 72.1 70.5 68.1 Direct foreign VA 42.1 44.0 43.3 51.0 51.8 51.9 70.5 68.9 66.3 Total domestic VA 52.9 51.1 51.8 45.8 45.0 44.9 27.9 29.5 31.9 Direct domestic VA 28.7 28.0 28.0 24.3 24.1 23.5 15.6 16.7 17.6 Based on four-digit NAICS IO table Total foreign VA 46.9 46.6 46.3 54.5 52.4 52.5 70 66.2 63.8 Direct foreign VA 41.9 42.4 42.1 51.5 49.9 49.9 68.4 64.5 61.9 Total domestic VA 53.1 53.4 53.7 45.5 47.6 47.5 30.0 33.8 36.2 Direct domestic VA 28.4 32.4 32.1 23.7 28.8 28.2 17.2 20.3 21.1 Source: authors’ estimates. a Only exports under Maquila are counted as processing exports, while exports under PITEX are counted as normal exports. b Both Maquila and PITEX are counted as pro- cessing trade. The HIY method refers to estimates from using the approach in Hummels et al (2001). The KWW method refers to estimates using the method in Koopman et al (2008). ∗ The estimates for 2000 and 2006 are preliminary as they use 2000 and 2006 exports as weights but sector domestic/foreign value added computed from the 2003 IO table, which is the latest available. different trading partners, since the domestic and foreign content shares are the same at sector level. 3 ESTIMATION RESULTS Decomposition results for foreign and domestic value-added shares in 2000, 2003 and 2006 for Mexico’s manufacturing exports, with the exception of food, aggregated from both the three-digit and four-digit Mexican NAICS IO tables are reported in Table 6.4. Because exports under the PITEX programme may have a different intensity in using imported intermediates from those of exports under the maquiladora programme, we report two estimates: one in which exports under the PITEX programme are treated as normal exports, and the other when they are treated as processing exports. For comparison, the results from the HIY formula that ignore processing trade are also reported. The KWW estimates indicate that, aggregated from the three-digit NAICS IO table, the total domestic value-added share in Mexico’s manufacturing exports was 45.8% in 2000, 45% in 2003 and 44.9% in 2006 when only exports under the maquiladora programme were counted as processing exports. When exports under the PITEX programme are also counted as processing exports, the share declines to 28%, 30% and 32% in the same years (Table 6.4). If aggregated from Foreign and Domestic Content in Mexico’s Manufacturing Exports 145 the four-digit NAICS IO table, the values are slightly higher: 30%, 34% and 36% when exports under both Marquiladora and PITEX are counted as processing exports; 45.5%, 47.6% and 47.5% when exports under the PITEX programme were treated as normal exports. In general, the direct domestic value-added shares are less than two-thirds of the total domestic value-added shares. How- ever, the indirect foreign value-added share (equal to the total foreign value- added share minus the direct foreign value-added share) was relatively small, suggesting that most of the foreign content comes from directly imported foreign inputs that are used for further processing and assembling, which are then exported back to the world market (mostly to the USA) as final products. The share of indirect foreign value added under the upper-bound estimates is smaller than that in the lower-bound estimate when only Maquila counted as processing trade, suggesting that it is reasonable to classify both Maquila and PITEX as processing exports. Therefore, we will focus our discussion of the results on the upper bound KWW estimates, but we will refer to the lower- bound estimates when necessary. Relative to the HIY’s estimates, the KWW calculations resulted in much higher shares of foreign value added in Mexico’s gross exports and showed a different trend over time. To be more precise, considering aggregation from the four-digit NAICS IO table, estimates of the HIY method show that there is almost no trend in foreign content share (total VS share) in the data (47%, 47% and 46% in 2000, 2003 and 2006, respectively). However, when both maquiladora and PITEX are counted as processing exports, KWW estimates reveal that the foreign content in Mexican manufacturing exports declined steadily from 70% in 2000 to 64% in 2006 (or from 72% to 68% if aggregated from the three-digit NAICS IO table). This indicates that the domestic value added in Mexico’s manufacturing exports is relatively low, but increased over the period 2000–6. Overall, the HIY method appears to incorrectly estimate both the level and the trend in domestic versus foreign content in Mexican manufacturing exports (Table 6.4). The results also reveal another interesting fact: the differ- ence (or bias) from trade regime aggregation (whether differentiate processing and normal trade) is much larger than the difference from aggregation based on more detailed sector classifications. There is only about a 2 percentage point difference in domestic or foreign content share estimates between the three-digit and four-digit NAICS classification using the HIY formula, while such a difference doubled when the KWW formula was applied (comparing the upper and lower panels of Table 6.4). But that difference is still less than 4 percentage points smaller than the difference between such estimates based on the HIY formula and the KWW formula (comparing the first, second and third panels in Table 6.4). Treating PITEX as processing exports also makes a difference in the estimation results. This shows that it matters whether or not to take processing trade into account: a finding consistent with what Koop- man et al (2011) found using Chinese data. 146 Trade in Value Added 3.1 Estimates for Major Manufacturing Sectors On average, domestic value added in Mexico’s manufacturing exports is 29.5% at the NAICS three-digit level and 33.8% at the NAICS four-digit level (Tables 6.5 and 6.6). Among the 19 manufacturing industries in Table 6.5, 12 industries have domestic content of less than 50%, comprising 89.3% of Mexico’s manufacturing exports in 2003. Similarly, of the 75 industries reported in Table 6.6, 41 industries have domestic content of less than 50% and together represent 79.5% of the coun- try’s manufacturing exports. The industries with the lowest shares of domes- tic value added are: computer and peripheral equipment; audio and video equipment; communications equipment; semiconductor and other electronic components; commercial and service industry machinery component man- ufacturing; hardware and electrical equipment. The following 21 industries have their shares of domestic content or domestic value added higher than 50% but lower than 65% and account for 15.3% of total manufacturing exports. These medium domestic value-added industries include motor vehicle body and trailer, fiber, yarn and tread mills, railroad rolling stock manufactur- ing, nonferrous metal production, fabric mills, and metalworking machin- ery manufacturing. The remaining 13 industries have shares higher than 65% but account for only 5.1% of Mexico’s total manufacturing exports. Leading these high domestic value-added group of industries are: petroleum and coal products, with a share of 90.0%; lime and gypsum products, with a share of 88.2%; and pesticide, fertilizer and other agricultural chemicals, with a share of 79.9%. Counting Mexican manufacturing exports under the PITEX programme as processing trade makes a difference in our calculations across industries. This is particularly important for transportation equipment industries (NAICS 336), but it has relatively less impact on electronic sectors (NAICS 334 and 335). Given the dominance of production sharing arrangements with the USA in Mexico’s automobile sector, this should not be a surprise (PITEX made up more than 60% of Mexico’s exports of transportation equipment, while those under the Maquila programme were only about 34%). These top three NAICS industries with the lowest domestic value added together made up about 70% of Mexico’s total manufacturing exports in 2003. This suggests that Mexi- can manufacturing trade is highly concentrated in a few industries with an extremely high proportion of processing exports: between 72% and 85% and low domestic content of less than 27% (Table 6.5). Similarly, there are some marked differences within industries. For instance, in two sectors within the transportation industry, at the four-digit NAICS clas- sification, exports of motor vehicles and motor vehicle body and trailer (with PITEX exports of 100% and 96%) show very different domestic content: domes- tic value added in motor vehicle body and trailer is 63%, while that of motor vehicle is 35% (Table 6.6). Also, within the computer and electronic product Table 6.5: Domestic value-added share in Mexico’s manufacturing exports by three-digit NAICS, 2003 (sorted by total foreign value added (weighted sum 2) in descending order). Non- Weighted Weighted Three- Total processing Processing sum 1a sum 2b digit manuf. NCICS Industry description exports A C D C D C D B C D E 334 Computer and electronic products 35,103 21.4 28.8 71.2 86.0 14.0 77.4 22.6 84.9 85.0 15.0 98.4 336 Transportation equipment 43,393 26.5 31.2 68.8 75.3 24.7 46.2 53.8 34.1 73.8 26.2 96.6 335 Electrical equipment, appliances 15,804 9.6 23.5 76.5 75.7 24.3 66.5 33.5 82.4 72.4 27.6 93.7 and components 339 Misc. manufacturing 7,809 4.8 16.1 84.0 71.7 28.3 60.3 39.7 79.6 67.0 33.0 91.5 333 Machinery 5,068 3.1 23.1 76.9 76.7 23.4 44.6 55.4 40.1 65.6 34.4 79.4 315 Apparel 6,784 4.1 21.5 78.5 65.3 34.7 52.9 47.1 71.6 63.6 36.4 96.1 314 Textile product mills 676 0.4 24.9 75.1 72.5 27.5 44.3 55.7 40.9 61.9 38.1 77.7 332 Fabricated metal products 3,502 2.1 20.9 79.1 72.1 27.9 45.9 54.1 48.9 61.3 38.7 78.9 337 Furniture and related products 1,652 1.0 16.2 83.8 67.2 32.8 50.7 49.3 67.7 59.9 40.1 85.7 323 Printing and related activities 289 0.2 20.7 79.3 64.9 35.1 55.6 44.4 79.0 57.6 42.4 83.5 326 Plastics and rubber products 2,074 1.3 27.6 72.4 66.2 33.8 47.0 53.0 50.3 56.1 43.9 73.8 316 Leather and allied products 512 0.3 20.2 79.8 72.1 27.9 35.7 64.3 29.9 53.9 46.1 65.0 331 Primary metal 3,239 2.0 19.4 80.6 64.4 35.6 22.4 77.6 6.7 45.4 54.6 57.8 322 Paper 790 0.5 26.3 73.7 67.3 32.7 40.6 59.4 34.9 45.0 55.0 45.6 327 Non-metallic mineral products 1,929 1.2 9.7 90.3 64.3 35.7 21.1 78.9 20.8 43.2 56.8 61.3 313 Textile mills 729 0.4 29.9 70.1 54.8 45.2 39.2 60.8 37.5 43.0 57.0 52.8 321 Wood products 212 0.1 7.9 92.1 58.1 41.9 24.8 75.2 33.7 40.3 59.7 64.6 325 Chemical 6,891 4.2 15.6 84.4 66.4 33.6 17.8 82.2 4.4 33.8 66.2 35.8 324 Petroleum and coal products 855 0.5 8.1 91.9 79.1 20.9 8.8 91.2 1.0 10.1 89.9 2.9 TOT Total manufacturing goods except food 137,312 83.7 19.9 80.1 76.8 23.2 55.0 45.0 56.0 70.5 29.5 89.0 Foreign and Domestic Content in Mexico’s Manufacturing Exports Source: authors’ estimates. A, percentage of Mexico’s total merchandise exports; B, Maquila exports as a percentage of industry exports; C, total FVA; D, total DVA; E, Maquila and PTEX exports as a percentage of industry exports. a Only exports under Maquila counted as processing exports, while exports under PITEX counted as normal exports. b Both Maquila and PITEX counted as processing trade. 147 148 Table 6.6: Domestic value-added share in Mexico’s manufacturing exports by four-digit NAICS, 2003 (sorted by total foreign value added (weighted sum 2) in descending order). Non- Weighted Weighted Four- Total processing Processing sum 1a sum 2b digit manuf. NCICS Industry description exports A C D C D C D B C D E 3341 Computer and peripheral equipment 11,261 6.9 36.1 63.9 91.5 8.5 77.0 23.0 73.9 90.9 9.1 98.9 3343 Audio and video equipment 8,962 5.5 31.0 69.0 86.9 13.2 84.3 15.7 95.4 86.5 13.5 99.3 3342 Communications equipment 4,460 2.7 20.7 79.3 85.1 14.9 83.2 16.8 97.1 84.0 16.0 98.3 3344 Semiconductor and other electronic component 7,276 4.4 19.7 80.3 84.8 15.3 75.0 25.0 85.0 83.6 16.4 98.3 manufacturing 3333 Commercial and service industry 580 0.4 32.0 68.0 84.7 15.3 46.7 53.3 27.8 81.4 18.7 93.6 machinery manufacturing 3325 Hardware 747 0.5 18.0 82.0 79.1 20.9 68.6 31.4 82.8 77.2 22.9 96.9 3353 Electrical equipment 5,820 3.5 15.9 84.1 76.9 23.1 66.9 33.1 83.6 75.3 24.7 97.4 3345 Navigational, measuring, electronic medical and 2,600 1.6 23.6 76.4 77.2 22.8 63.8 36.2 75.0 74.6 25.4 95.1 control instruments 3359 Other electrical equipment and component 6,278 3.8 25.9 74.1 78.0 22.0 68.7 31.3 82.2 74.1 25.9 92.5 manufacturing 3346 Magnetic and optical media 544 0.3 16.2 83.8 80.2 19.8 58.3 41.7 65.8 73.6 26.4 89.7 3363 Motor vehicle parts 21,708 13.2 26.8 73.2 76.1 23.9 57.5 42.5 62.3 73.4 26.7 94.5 Trade in Value Added 3391 Medical equipment and supplies 3,561 2.2 18.0 82.0 74.4 25.6 69.1 31.0 90.5 73.0 27.0 97.5 3366 Ship and boat building 107 0.1 4.0 96.0 72.8 27.2 37.0 63.0 47.9 72.0 28.0 98.9 3379 Other furniture related products 515 0.3 25.9 74.1 73.0 27.0 66.1 33.9 85.4 71.3 28.8 96.3 3351 Electric lighting equipment 1,413 0.9 16.2 83.8 73.7 26.4 64.8 35.2 84.7 66.9 33.1 88.3 3313 Alumina and aluminum production and 82 0.0 20.1 79.9 73.3 26.7 41.2 58.8 39.6 66.6 33.4 87.5 processing 3352 Household appliance 2,293 1.4 29.7 70.3 69.3 30.8 60.8 39.2 78.7 65.7 34.3 91.1 3151 Apparel knitting mills 32 0.0 18.3 81.7 71.5 28.5 52.9 47.1 65.0 65.1 34.9 88.0 Source: authors’ estimates. A, percentage of Mexico’s total merchandise exports; B, Maquila exports as a percentage of industry exports; C, total FVA; D, total DVA; E, Maquila and PTEX exports as a percentage of industry exports. a Only exports under Maquila are counted as processing exports, while exports under PITEX counted as normal exports. Table 6.6: Continued. Non- Weighted Weighted Four- Total processing Processing sum 1a sum 2b digit manuf. NCICS Industry description exports A C D C D C D B C D E 3361 Motor vehicles 6,657 4.1 33.2 66.8 64.8 35.2 33.2 66.8 0.0 64.8 35.2 99.9 3152 Cut and sew apparel 6,633 4.0 22.4 77.6 64.6 35.4 52.7 47.3 71.9 63.1 36.9 96.5 3331 Agriculture, construction, and mining 426 0.3 20.4 79.6 76.7 23.3 48.4 51.6 49.7 63.1 36.9 75.8 machinery 3339 Other general purpose machinery 1,685 1.0 21.1 78.9 72.2 27.8 48.7 51.3 54.0 63.1 36.9 82.2 3336 Engine, turbine, and power transmission 1,308 0.8 25.7 74.3 72.1 27.9 37.0 63.0 24.4 62.7 37.3 79.7 equipment 3149 Other textile product mills 484 0.3 25.9 74.1 71.5 28.5 44.1 55.9 40.0 62.4 37.6 80.0 3364 Aerospace products and parts 1,176 0.7 9.6 90.4 74.2 25.8 33.1 66.9 36.3 62.4 37.6 81.8 3272 Glass and glass products 852 0.5 13.4 86.6 71.9 28.1 25.5 74.5 20.6 62.1 38.0 83.1 3329 Other fabricated metal products 1,485 0.9 22.3 77.8 74.5 25.5 51.9 48.1 56.7 62.1 37.9 76.4 3399 Other misc. manufacturing 4,248 2.6 15.7 84.3 68.6 31.4 52.9 47.1 70.4 61.4 38.6 86.5 3334 Ventilation, heating, air-conditioning and 669 0.4 26.6 73.4 71.7 28.3 50.8 49.3 53.6 61.3 38.7 77.0 commercial refrigeration equipment 3322 Cutlery and hand tools 222 0.1 17.3 82.7 73.1 26.9 31.3 68.7 25.2 60.0 40.1 76.5 3141 Textile furnishings mills 192 0.1 24.2 75.9 73.1 26.9 45.2 54.8 43.0 59.4 40.6 71.9 3261 Plastics products 1,586 1.0 28.5 71.5 66.6 33.4 49.4 50.6 55.0 58.6 41.4 79.1 3231 Printing and related support activities 289 0.2 21.1 78.9 64.8 35.2 55.9 44.1 79.6 57.6 42.4 83.5 3372 Office furniture 923 0.6 19.6 80.4 62.1 37.9 46.0 54.0 62.2 54.9 45.1 83.2 3311 Iron and steel mills and ferroalloy 1,239 0.8 19.4 80.7 65.7 34.3 19.8 80.2 1.0 54.1 45.9 75.0 3159 Apparel accessories and other apparel 119 0.1 16.4 83.6 64.5 35.5 45.8 54.2 61.1 53.5 46.5 77.2 3161 Leather and hide tanning and finishing 109 0.1 16.4 83.6 77.0 23.0 20.2 79.8 6.4 53.3 46.7 60.8 Foreign and Domestic Content in Mexico’s Manufacturing Exports 3169 Other leather and allied products 140 0.1 19.9 80.1 60.6 39.4 38.8 61.2 46.5 53.3 46.7 82.1 149 150 Table 6.6: Continued. Non- Weighted Weighted Four- Total processing Processing sum 1a sum 2b digit manuf. NCICS Industry description exports A C D C D C D B C D E 3162 Footwear 263 0.2 20.7 79.3 76.4 23.6 37.7 62.3 30.7 52.7 47.3 57.5 3371 Household and institutional furniture and 214 0.1 14.6 85.4 65.9 34.1 39.6 60.5 48.7 51.1 48.9 71.3 kitchen cabinets 3327 Machine shops, turned products and screws, 61 0.0 16.4 83.6 63.6 36.4 40.9 59.1 51.9 50.9 49.1 73.1 nuts and bolts 3324 Boiler, tank and shipping container 126 0.1 23.6 76.4 66.5 33.5 34.2 65.8 24.7 49.9 50.1 61.2 3133 Textile and fabric finishing and fabric coating 100 0.1 26.3 73.7 71.7 28.4 47.1 52.9 45.9 49.3 50.7 50.8 mills 3212 Veneer, plywood and engineered wood products 55 0.0 13.9 86.1 69.0 31.0 20.6 79.4 12.2 48.5 51.5 62.8 3259 Other chemical products and preparation 835 0.5 22.4 77.6 70.6 29.4 28.4 71.6 12.4 48.0 52.0 53.1 3326 Spring and wire product 509 0.3 21.0 79.0 54.6 45.4 23.5 76.5 7.4 47.9 52.1 80.3 3211 Sawmills and wood preservation 3 0.0 4.4 95.6 65.9 34.1 13.1 86.9 14.1 47.4 52.6 70.0 3262 Rubber products 487 0.3 26.9 73.1 62.8 37.2 38.8 61.2 33.1 47.1 52.9 56.4 3222 Converted paper products 695 0.4 25.2 74.8 67.2 32.8 41.7 58.3 39.2 46.4 53.7 50.3 3369 Other transportation equipment 31 0.0 32.8 67.2 56.2 43.8 38.8 61.3 25.5 45.8 54.2 55.7 Trade in Value Added 3332 Industrial machinery 146 0.1 14.6 85.4 62.0 38.0 32.7 67.3 38.1 43.0 57.0 59.8 3312 Steel product using purchased steel 620 0.4 22.3 77.7 54.3 45.7 26.1 73.9 11.9 41.9 58.1 61.3 3219 Other wood product 154 0.1 13.7 86.3 56.8 43.2 31.7 68.3 41.9 41.7 58.3 65.1 3323 Architectural and structural metals 250 0.2 22.1 77.9 48.4 51.6 30.6 69.4 32.5 41.5 58.5 73.8 3335 Metalworking machinery manufacturing 255 0.2 18.2 81.8 63.2 36.8 21.2 78.8 6.6 40.6 59.4 49.8 3252 Resin, synthetic rubber, and artificial 1,145 0.7 25.9 74.1 58.2 41.9 26.3 73.7 1.1 40.5 59.6 45.1 synthetic fibers and filaments 3315 Foundries 30 0.0 15.1 84.9 60.1 39.9 18.4 81.6 7.3 38.9 61.1 52.9 3132 Fabric mills 514 0.3 29.1 70.9 44.8 55.2 35.8 64.2 42.4 38.8 61.2 61.5 3314 Non-ferrous metal (except aluminum) 1,267 0.8 16.2 83.8 74.4 25.6 20.7 79.3 7.7 38.1 61.9 37.6 production and processing Table 6.6: Continued. Non- Weighted Weighted Four- Total processing Processing sum 1a sum 2b digit manuf. NCICS Industry description exports A C D C D C D B C D E 3365 Railroad rolling stock manufacturing 202 0.1 40.1 59.9 37.1 63.0 39.5 60.5 19.3 37.5 62.5 85.6 3131 Fiber, yarn, and thread mills 115 0.1 32.6 67.4 62.5 37.5 35.1 64.9 8.4 37.2 62.8 15.5 3362 Motor vehicle body and trailer 13,512 8.2 6.4 93.6 36.7 63.3 7.5 92.5 3.5 36.7 63.3 99.8 3251 Basic chemical 1,561 1.0 12.0 88.0 53.5 46.5 13.8 86.3 4.2 33.8 66.2 52.5 3221 Pulp, paper and paperboard mills 94 0.1 29.4 70.6 67.0 33.0 30.6 69.4 3.2 33.5 66.5 10.8 3273 Cement and concrete product 121 0.1 7.1 92.9 63.0 37.1 22.5 77.5 27.6 33.1 66.9 46.6 3279 Other non-metallic mineral product 313 0.2 16.7 83.3 60.1 39.9 28.8 71.2 28.0 32.3 67.7 36.1 3271 Clay product and refractory 609 0.4 9.1 91.0 52.7 47.3 16.6 83.5 17.2 30.9 69.1 50.2 3254 Pharmaceutical and medicine 1,510 0.9 11.8 88.3 60.7 39.3 13.4 86.6 3.3 28.7 71.4 34.5 3321 Forging and stamping 103 0.1 19.4 80.6 57.6 42.4 24.1 75.9 12.4 27.0 73.0 19.8 3255 Paint, coating and adhesive 902 0.6 24.4 75.6 60.6 39.4 25.5 74.5 3.3 25.7 74.3 3.8 3256 Soap, cleaning compound and toilet preparation 841 0.5 18.4 81.6 74.2 25.8 20.9 79.1 4.4 25.2 74.8 12.2 3328 Coating, engraving, heat treating and 0 0.0 20.4 79.6 57.1 42.9 — — 0.0 21.1 78.9 2.0 allied activities 3253 Pesticide, fertilizer and other 95 0.1 21.2 78.8 17.6 82.4 21.1 78.9 3.7 20.2 79.9 29.6 agricultural chemical 3274 Lime and gypsum product 35 0.0 11.7 88.3 36.0 64.0 11.7 88.3 0.2 11.8 88.2 0.5 3241 Petroleum and coal products 855 0.5 8.0 92.0 79.1 20.9 8.7 91.3 1.0 10.0 90.0 2.9 TOT Total manufacturing goods except food 137,312 83.7 19.9 80.1 72.0 28.0 52.4 47.6 55.7 66.2 33.8 89.0 Foreign and Domestic Content in Mexico’s Manufacturing Exports 151 152 Trade in Value Added industry (whose exports are mostly under the Maquila programme) exports of communications equipment, audio and video equipment, semiconductor and other electronic component manufacturing, and computer and periph- eral equipment show an average domestic content of 14%. In contrast, also within the computer and electronic product industry navigational, measuring, electromedical and control instruments show a domestic value added of 25%. Differences in the electrical equipment, appliance, and component industry (also mostly maquiladora exports) are less prominent. For instance, exports of electrical equipment and other electrical equipment and component man- ufacturing average a domestic value added of 25%, while those of electric lighting equipment and household appliances average a value added of 34%. This indicates that exporting industries that tend to use the maquiladora pro- gramme the most, eg electronics, have low domestic value added, while those industries that export under PITEX, eg automobile and machinery industries, have relatively higher domestic content. 3.2 Exports to Major Markets The USA is the leading market for Mexican manufacturing exports, to which Mexico exported 86.4% of its total in 2006 (De La Cruz et al 2011, Table 9). Although this share declined from 2003 to 2006, the USA continues to play a dominant role as a market for Mexico’s manufacturing exports. Canada fol- lows with approximately 2% of Mexico’s total manufacturing exports. Most of Mexico’s manufacturing exports to the USA and Canada are processing exports in excess of 87% of such exports. Although the share of domestic value added in Mexico’s processing exports is increasing, it remained relatively low, at about 34.3% for the USA and 36.8% for Canada, in 2006. Mexico’s trad- ing partners and its manufacturing exports under both the maquiladora and PITEX programmes indicate that in 2006 both programmes were important for the USA and Canada, but PITEX was particularly important for Brazil, the European Union and Japan. The share of Maquila exports to the USA remained at 60%, while that of PITEX declined from 35% to 27% from 2000 to 2006. 3.3 Comparing Mexico and China On average, Mexico’s domestic value added in manufacturing exports is about 34% (Table 6.6), a share that is relatively lower than that of 51% for China (Koopman et al 2011, Table 2). Low domestic content industries in both coun- tries include computers and accessories and telecommunications equipment. Some higher domestic value-added industries that are similar in both coun- tries include motor vehicles and cement. Mexico’s domestic content in processing trade for computers (8.5%; see Table 6.6) is lower than that of China (18.7%; see Koopman et al (2011, Table 5)), suggesting some integration in Mexico’s information and commu- Foreign and Domestic Content in Mexico’s Manufacturing Exports 153 nications technology but not as much as in China. Mexico has promoted part- nerships among domestic firms, foreign firms and the university system in the city of Guadalajara, to create the country’s ‘Silicon Valley’. 5 In addition, the country has also moved, although modestly, in the global supply chain in the areas of software development and information technology services. Mexico’s domestic value added in communication equipment (14.9%) and elec- tronic components (15.3%) are almost half of China’s (36.0% and 32.8%, respec- tively). High domestic value-added processing industries in Mexico are rail- road rolling stock manufacturing (63.0%) and pesticide, fertilizer, and other agricultural chemicals (82.4%), which are more than twice as high as those of China (14.6 and 16.5%, respectively). Estimates of domestic value added in manufacturing exports by coun- try or region of destination indicate that domestic content in both Mexico and China’s exports to the USA is similar but less that 50–44.7% for Mexico (De La Cruz et al 2011, Table 9) and 45.5% for China (Koopman et al 2011, Table 7). Moreover, Mexico’s domestic content in exports to Japan and Brazil is, on average, higher (68.9%) than for China (60.5%). Notably, both countries’ domestic value added in manufacturing exports to the rest of Latin America and the Caribbean is relatively high: 77.7% for Mexico and 75.8% for China. 3.4 Comparing Content Shares Estimates As described above, the estimation method in this chapter uses a ‘true’ IO account that separately traces processing exports and other production trans- actions in the Mexican economy but which rarely exists for other countries. Mexico’s statistical agency, the Instituto Nacional de Estadística, Geografía e Informática (INEGI), compiled a 2003 benchmark IO table based on an economic census, which has separate accounts for Mexico’s domestic and maquiladora industries. The IO table includes national production of goods and services classified under Mexico’s 2002 three- and four-digit NAICS, inputs purchased in the domestic and maquiladora industries and imports from the rest of the world by both economies. The ‘true’ domestic and for- eign content shares computed directly from this special Mexico IO table at the three-digit NAICS are summarised in Table 6.7 for convenience. It provides a reference benchmark to test the performance of the estimation method proposed in Koopman et al (2011). Thus, with exports and import data for the maquiladora industries from the World Trade Atlas and Mexico’s aggre- gate 2003 IO table, we implemented the same quadratic programming model that generates estimates for domestic and foreign content of exports as described in Koopman et al (2011). The computed estimates of domestic and foreign value-added shares for Mexico manufacturing exports are reported in 5 We thank Ted H. Moran for making this important remark linked to the formation of backward linkages and supplier networks for multinational investors. 154 Table 6.7: Domestic and foreign content in Mexico’s gross exports, 2003, computed directly from the Mexico IO table with a separate maquiladora economy account. Normal exports Maquiladora exports Weighted sum NAICS Direct Direct Total Total Direct Direct Total Total Direct Direct Total Total code NAICS description FVA DVA FVA DVA FVA DVA FVA DVA FVA DVA FVA DVA 311 Food manufacturing 7.6 38.5 13.3 86.7 48.9 23.3 52.0 48.0 16.7 35.1 21.8 78.2 312 Beverage and tobacco product manufacturing 7.2 42.4 13.0 87.0 8.8 19.2 19.6 80.4 7.3 41.2 13.4 86.6 313 Textile mills 25.0 34.6 29.9 70.1 50.5 19.0 54.8 45.2 40.5 25.1 45.0 55.0 314 Textile Product Mills 18.1 39.5 24.9 75.1 71.4 18.6 72.5 27.5 59.1 23.4 61.6 38.4 315 Apparel manufacturing 15.3 48.7 21.5 78.5 63.3 21.5 65.3 34.7 53.3 27.2 56.2 43.8 316 Leather and allied product manufacturing 12.8 37.3 20.2 79.8 70.7 17.5 72.1 27.9 48.0 25.3 51.8 48.2 321 Wood product manufacturing 5.1 43.8 7.9 92.1 55.8 24.1 58.1 41.9 33.0 32.9 35.6 64.4 322 Paper manufacturing 19.0 33.3 26.3 73.7 65.6 20.0 67.3 32.7 45.3 25.8 49.4 50.6 323 Printing and related support activities 14.2 40.3 20.7 79.3 63.4 19.6 64.9 35.1 48.6 25.8 51.6 48.4 324 Petroleum and coal products manufacturing 4.5 14.3 8.1 91.9 78.4 14.6 79.1 20.9 4.5 14.3 8.1 91.9 325 Chemical manufacturing 11.2 30.7 15.6 84.4 64.3 18.0 66.4 33.6 17.3 29.2 21.5 78.5 326 Plastics and rubber products manufacturing 22.7 34.7 27.6 72.4 64.1 19.0 66.2 33.8 52.8 23.3 55.7 44.3 327 Non-metallic mineral product manufacturing 5.9 54.5 9.7 90.3 62.2 20.3 64.3 35.7 27.8 41.2 31.0 69.0 Trade in Value Added 331 Primary metal manufacturing 12.8 37.0 19.4 80.6 61.9 17.8 64.4 35.6 22.6 33.2 28.4 71.6 332 Fabricated metal product manufacturing 14.7 39.7 20.9 79.1 70.6 16.4 72.1 27.9 45.6 26.8 49.2 50.8 333 Machinery manufacturing 18.2 43.7 23.1 76.9 75.3 11.9 76.7 23.4 43.7 29.5 47.0 53.0 334 Computer and electronic product manufacturing 24.2 43.9 28.8 71.2 85.2 8.2 86.0 14.0 77.7 12.6 78.9 21.1 335 Electrical equipment and component manufacturing 17.8 41.0 23.5 76.5 74.2 13.7 75.7 24.3 63.5 18.9 65.8 34.2 336 Transportation equipment manufacturing 24.8 35.6 31.2 68.8 74.3 16.5 75.3 24.7 45.8 27.5 49.9 50.1 337 Furniture and related product manufacturing 11.5 49.2 16.2 83.8 65.3 18.4 67.2 32.8 52.3 25.9 54.8 45.2 339 Miscellaneous manufacturing 11.7 52.6 16.1 84.0 70.4 18.5 71.7 28.3 61.4 23.7 63.1 36.9 Total 19.0 37.7 24.7 75.3 76.3 13.4 77.5 22.5 54.5 22.6 57.5 42.6 Source: authors’ estimates. Table 6.8: Domestic and foreign content in Mexico’s gross exports, 2003, estimated from aggregated Mexico IO table by our mathe- matical programming model. Normal exports Maquiladora exports Weighted sum NAICS Direct Direct Total Total Direct Direct Total Total Direct Direct Total Total code NAICS description FVA DVA FVA DVA FVA DVA FVA DVA FVA DVA FVA DVA 311 Food manufacturing 6.2 38.4 12.3 87.7 46.7 20.2 50.3 49.7 10.1 36.7 16.0 84.0 312 Beverage and tobacco product manufacturing 12.4 42.4 17.5 82.5 54.9 22.6 57.7 42.3 14.1 41.6 19.1 81.0 313 Textile mills 20.3 33.6 27.7 72.3 46.1 16.4 51.3 48.7 30.0 27.2 36.6 63.4 314 Textile product mills 24.9 35.4 32.4 67.6 66.2 16.4 67.5 32.5 41.8 27.7 46.7 53.3 315 Apparel manufacturing 17.8 49.2 23.7 76.3 67.5 18.6 68.5 31.5 54.1 26.8 56.4 43.6 316 Leather and allied product manufacturing 11.0 36.4 19.3 80.7 64.3 18.7 65.7 34.3 27.0 31.1 33.3 66.8 321 Wood product manufacturing 7.2 43.6 10.7 89.4 59.9 23.0 62.1 37.9 25.3 36.5 28.3 71.7 322 Paper manufacturing 11.8 33.1 19.9 80.1 57.6 17.0 59.6 40.4 27.8 27.5 33.7 66.3 323 Printing and related support activities 14.8 40.5 20.9 79.2 54.9 20.2 56.6 43.4 46.5 24.5 49.1 50.9 324 Petroleum and coal products manufacturing 4.9 14.3 10.0 90.0 56.3 8.1 60.8 39.2 5.2 14.3 10.4 89.7 325 Chemical manufacturing 12.0 30.6 17.1 82.9 48.6 15.8 50.7 49.3 13.6 29.9 18.7 81.4 326 Plastics and rubber products manufacturing 21.9 34.0 28.1 72.0 60.0 16.3 61.8 38.2 41.3 25.0 45.3 54.8 327 Non-metallic mineral product manufacturing 7.6 53.6 11.8 88.2 45.6 27.6 47.9 52.1 15.5 48.2 19.3 80.7 331 Primary metal manufacturing 10.5 36.6 17.5 82.5 74.6 18.9 75.4 24.6 15.0 35.3 21.7 78.4 332 Fabricated metal product manufacturing 18.4 38.6 24.5 75.5 57.2 18.2 59.6 40.5 37.3 28.7 41.6 58.4 333 Machinery manufacturing 18.4 40.9 25.1 74.9 60.1 19.0 61.8 38.2 35.1 32.1 39.9 60.2 334 Computer and electronic 30.6 39.0 38.0 62.0 82.4 8.8 83.1 16.9 74.6 13.4 76.3 23.7 product manufacturing 335 Electrical equipment and 25.2 42.3 31.3 68.7 70.6 13.3 72.1 27.9 62.7 18.3 65.0 35.0 component manufacturing 336 Transportation equipment manufacturing 24.8 34.6 31.6 68.4 83.8 16.1 83.8 16.2 45.1 28.2 49.6 50.4 337 Furniture and related product manufacturing 16.3 47.1 21.2 78.8 57.9 21.4 59.6 40.4 44.5 29.7 47.2 52.8 339 Miscellaneous manufacturing 19.9 50.5 24.9 75.1 65.7 18.5 67.1 32.9 56.2 25.2 58.2 41.8 Total 20.4 36.7 26.9 73.2 75.8 13.2 76.8 23.2 51.2 23.6 54.6 45.4 B1 Error at manufacture aggregate 1.4 −1.0 2.1 −2.1 −0.4 −0.3 −0.7 0.7 −3.3 1.0 −2.8 2.8 compared to true data Foreign and Domestic Content in Mexico’s Manufacturing Exports B2 Mean absolute percentage error 22.7 2.9 17.3 4.3 16.1 14.1 14.2 28.0 18.5 7.8 15.3 12.4 from the true data B3 Correlation with true data 0.829 0.985 0.862 0.862 0.462 0.666 0.474 0.474 0.936 0.972 0.944 0.944 Source: authors’ estimates. 155 156 Trade in Value Added Table 6.8. The three panels in Table 6.8 list direct, total domestic and total foreign value-added shares for normal, processing and total exports, respec- tively. To compare both methodologies and to quantitatively assess how much each set of value-added share estimates differs from the ‘true’ share data (computed directly from the Mexico IO table with a separate processing trade account) we report three metrics in the three bottom rows of Table 6.8. The row labelled B1 in Table 6.8 shows the difference between the estimated shares and the ‘true’ shares computed directly from Mexico IO table with a separate processing trade account for manufactures as a whole. The errors for the various share estimates appear to be less than 3.5 percentage points. A second metric that measures the proportionate errors is the ‘mean absolute percentage error’ (MAPE) with respect to the ‘true’ shares. It is calculated as follows: n 100 i=1 |si − s0i | MAPE = n , i=1 s0i where si is the estimated share and s0i is the reference ‘true’ share for indus- try i. The MAPE index is reported in row labelled B2. The error ranges from 4% to 17% for normal exports, 14% to 28% for processing exports and 12% to 15% for total exports. The third metric, reported in the row labelled B3 (Table 6.8), is the correlation coefficients between the estimated shares and the reference or ‘true’ shares. These suggest that the estimates are highly correlated with the ‘true’ shares computed directly from the Mexico IO table with a separate processing trade account for normal and total exports, though the correla- tions are lower for processing exports. Overall, the estimates calculated with the quadratic programming model and those ‘true’ shares computed directly from the Mexico IO table with a separate processing trade account show close values at the aggregate or total value, but not for some industries, including beverages and tobacco, petroleum and coal products, chemical manufactur- ing, non-metallic mineral products and machinery manufacturing. 4 CONCLUSIONS Vertical specialisation is pervasive in Mexico. In line with global trade, Mex- ico’s trade has increased at impressive rates over the last fifteen years, and more than 85% of its manufacturing exports are production sharing opera- tions. In this chapter, we estimated domestic and foreign value-added shares that are present in Mexico’s manufacturing exports for 2000, 2003 and 2006. The estimation was carried out by applying the methodology developed by Koop- man et al (2011), but with a slight modification. Instead of estimating the structure of the processing export sector via an optimising algorithm, we used an input–output table compiled specifically for the production sharing sector, ie for the maquiladora industry for 2003. This is the first study of its kind in Foreign and Domestic Content in Mexico’s Manufacturing Exports 157 that, for Mexico, it provides measures of vertical specialisation using such an input–output table in addition to using trade data from both export promo- tion programmes, the maquiladora and PITEX programmes. The estimation results suggest that on average Mexico’s manufacturing exports have a domestic value-added share of about 34%. Industries that have a domestic content of less than 50% account for approximately 80% of the country’s manufacturing exports. Low domestic value-added indus- tries include computer and peripheral equipment, audio and video equip- ment, communications equipment, semiconductor and other electronic com- ponents, and electrical equipment. Industries that have domestic content shares higher than 65% account for only 5.1% of Mexico’s total manufacturing exports. Some leading industries in this higher domestic value-added group are: petroleum and coal products, with a share of 90.0%; lime and gypsum products, with a share of 88.2%; pesticide, fertilizer and other agricultural chemicals, with a share of 79.9%. Counting Mexican manufacturing exports under the PITEX programme as processing trade makes a difference in our calculations across industries. In particular, it made a significant difference in the transportation equipment industries, whose exports under PITEX made up more than 60% of Mexico’s exports of that industry, while those under the Maquila programme were only about 34%. This reflects the dominance of production sharing arrangements with the USA in Mexico’s automotive sector. Furthermore, the top three NAICS industries with the lowest domestic value added (transportation equipment and electronic sectors), together made up about 70% of Mexico’s total man- ufacturing exports in 2003. This suggests that Mexican manufacturing trade is highly concentrated in a few industries with an extremely high proportion of processing exports: between 72% and 85% and low domestic content of less than 27%. Our results also indicate that exporting industries that tend to use the maquiladora programme the most, for instance electronics, have low domestic value added, while those industries that export under PITEX (auto- motive and machinery industries) have a relatively higher domestic content. Most of Mexico’s manufacturing exports to the USA and Canada consist of processing exports, and the USA is by far the largest single-country export market, to which Mexico exported 86.4% of its total in 2006. Canada’s share of Mexico’s total manufacturing exports was only approximately 2% in the same year. Although relatively low, the domestic value added in Mexico’s exports has increased in recent years, suggesting that Mexico’s industrial strategy has resulted, although modestly and in some industries only, in its insertion into the global supply chains. 158 Trade in Value Added REFERENCES Baker & McKenzie (2006). Doing Business in Mexico. Report. URL: http://www.bakernet .com/. Bergin, R. P., and R. C. Feenstra (2008). Pass-Through of Exchange Rates and Com- petition between Floaters and Fixers. Journal of Money Credit and Banking 41(s1), 35–70. Bergin, R. P., R. C. Feenstra, and G. H. Hanson (2009a). Offshoring and Volatility: Evidence from Mexico’s Maquiladora Industry. American Economic Review 99(4), doi:10.1257/aer.99.4.1664. Bergin, R. P., R. C. Feenstra, and G. H. Hanson (2009b). Volatility Due to Offshoring: Theory and Evidence. University of California, Davis, Working Paper, January. URL: www.econ.ucdavis.edu/faculty/bergin/research/Offshoring_ and_Volatility_theory_web.pdf (accessed 29 May 2009). Burstein, A., C. Kurz and L. Tesar (2008). Trade, Production Sharing, and the Interna- tional Transmission of Business Cycles. Journal of Monetary Economics 55, 775–795. Campa, J. M., and L. S. Goldberg (2006). Pass Through of Exchange Rates to Consump- tion Prices: What Has Changed and Why? NBER Working Paper 12547. Chiquiar, D. (2008). Globalization, Regional Wage Differentials and the Stolper– Samuelson Theorem: Evidence from Mexico. Journal of International Economics 74, 70–93. Chiquiar, D., and M. Ramos-Francia (2005). Trade and Business-Cycle Synchronization: Evidence from Mexican and US Manufacturing Industries. North American Journal of Economics and Finance 16, 187–216. De La Cruz, J., R. B. Koopman, Z. Wang, and S.-J. Wei (2011). Estimating foreign-Value Added in Mexico’s Manufacturing Exports. US International Trade Commission, Office of Economics Working Paper 2011-04A. Ebenstein, A., A. Harrison, M. McMillan, and S. Phillips (2009). Estimating the Impact of Trade and Offshoring on American Workers Using the Current Population Surveys. NBER Working Paper 15107. Feenstra, R. C. (1989). Symmetric Pass-Through of Tariffs and Exchange Rates under Imperfect Competition: An Empirical Test. Journal of International Economics 27, 24–45. Feenstra, R. C. (1998). Integration of Trade and Disintegration in the Global Economy. Journal of Economic Perspectives 12, 31–50. Feenstra, R. C. (2008). Offshoring in the Global Economy. The Ohlin Lectures, presented at the Stockholm School of Economics, 17–18 September. URL: www.econ.ucdavis .edu/faculty/fzfeens/pdf/Feenstra_Ohlin_Lecture_2008.pdf. Feenstra, R. C., and G. H. Hanson (1997). Foreign Direct Investment and Relative Wages: Evidence from Mexico’s Maquiladoras. Journal of International Economics 42, 371– 394. Feenstra, R. C, and G. H. Hanson (1999). The Impact of Outsourcing and High- Technology Capital on Wages: Estimates for the USA, 1979–1990. The Quarterly Journal of Economics 114, 907–940. Feenstra, R. C. and G. H. Hanson (2004). Global Production Sharing and Rising Inequal- ity: A Survey of Trade and Wages. In Handbook of International Trade (ed. By E. K. Choi and J. Harringan), Volume 1 (Oxford: Blackwell). Feenstra, R. C., G. H. Hanson, and D. Swenson (2000). Offshore Assembly from the United States: Production Characteristics of the 9802 Program. In The Impact of International Trade on Wages (ed. R. C. Feenstra), pp. 85–128. University of Chicago Press and NBER. Foreign and Domestic Content in Mexico’s Manufacturing Exports 159 Gantz, A. D. (2004). NAFTA, Article 303, PROCEC and the New Maquiladora Regime in Mexico. In The Auto Pact: Investment, Labour and the WTO (ed. M. Irish). New York: Kluwer Law International. Ghosh, A. (2008). Implications of Production Sharing on Exchange Rate Pass-Through. International Journal of Finance and Economics 14(4), 334–345. Hanson, G. H., and A. Harrison (1999). Trade, Technology, and Wage Inequality. Indus- trial and Labor Relations Review 52(2), 271–288. Hanson, G. H., R. Mataloni, and M. Slaughter (2005). Vertical Production Networks in Multinational Firms. Review of Economics and Statistics 87(4), 664–678. Herrera-Hernandez, J. (2004). Business Cycles in Mexico and the United States: Do They Share Common Movements? Journal of Applied Economics 7(2), 303–323. Hummels, D., J. Ishii, and K.-M. Yi (2001). The Nature and Growth of Vertical Special- ization in World Trade. Journal of International Economics 54, 75–96. INEGI (2007). Industria maquiladora de Exportación. Estadísticas Económicas INEGI, Febrero. Koopman, B. R., Z. Wang, and S.-J. Wei (2011). Estimating Domestic Content in Exports when Processing Trade Is Pervasive. Journal of Development Economics 99(1), 178– 189. Johnson, R. C., and G. Noguera (2012). Accounting for Intermediates: Production Shar- ing and Trade in Value Added. Journal of International Economics 86(2), 224–236. Kose, M. A., and K.-M. Yi (2001). International Trade and Business Cycles: Is Vertical Specialization the Missing Link? American Economic Review 91(2), 371–375. Kose, M. A., and K.-M. Yi. (2006). Can the Standard Real Business Cycle Model Explain the Relation between Trade and Comovement? Journal of International Economics 68(2), 267–295. Krugman, P. (2008). Trade and Wages, Reconsidered. Brookings Papers on Economic Activity. URL: http://www.brookings.edu/. López, J. J. (2007). Production Sharing and Real Business Cycles in a Small Open Econ- omy. Federal Reserve Bank of Dallas, Globalization and Monetary Policy Institute, Working Paper 5, December. URL: http://www.dallasfed.org/institute/wpapers/ 2007/0005.pdf (accessed 28 May 2009). Nordås, K. H. (2005). International Production Sharing: A Case for a Coherent Policy Framework’, WTO Discussion Papers 11. WTO Geneva, Switzerland. OECD (1996). Trade Liberalisation Policies in Mexico. Paris: Organization for Economic Co-operation and Development. Truett, B. D., and L. J. Truett (1984). The Maquiladoras: Prospects for Mexico. Journal of Economic Development 9(2), 45–66. Truett, B. D., and L. J. Truett (1993). Maquiladora Response to US and Asian Relative Wage Rate Changes. Contemporary Economic Policy 11(1), 18–28. Truett, B. D., and L. J. Truett (2007). NAFTA and the Maquiladoras: Boon or Bane. Contemporary Economic Policy 25(3), 374–386. United States International Trade Commission (1994). Production Sharing: US Imports under Harmonized Tariff Schedule Provisions 9802.00.60 and 9802.00.80, 1989– 1992. USITC Publication 2729, February. US Department of Commerce (2000). Merchandise Trade Reconciliation: United States– Mexico–Canada 1996–1997. Report. Bureau of the Census, Economics and Statis- tics Administration. URL: http://www.census.gov/foreign-trade/aip/recon9697.pdf (accessed 17 June 2009). US Department of Commerce (2001). Merchandise Trade Reconciliation: United States– Mexico–Canada 1998–1999. Bureau of the Census, Economics and Statistics Admin- 160 Trade in Value Added istration. URL: http://www.census.gov/foreign-trade/aip/recon9899.pdf (accessed 17 June 2009). USITC (1998a). The Use and Economic Impact of TSUS Items 806.30 and 807.00. United States International Trade Commission Publication 2053, January. USITC (1998b). Production Sharing: Use of US Components and Materials in Foreign Assembly Operations, 1994–1997. United States International Trade Commission Publication 3146, December. Verhoogen, E. A. (2008). Trade, Quality Upgrading, and Wage Inequality in the Mexican Manufacturing Sector. Quarterly Journal of Economics 123(2), 489–530. WTO (2008). Trade Policy Review: Mexico. Review ET/TPR/S/195/Rev. 1, World Trade Organization. 2 May. URL: http://www.wto.org/english/tratop_e/tpr_e/tpr_e.htm (accessed 16 June 2009). Yi, K.-M. (2003). Can Vertical Specialization Explain the Growth of World Trade? Jour- nal of Political Economy 111(1), 52–102. Yi, K.-M. (2009). The Collapse of Global Trade: the Role of Vertical Specialization. In The Collapse of Global Trade, Murky Protectionism, and the Crisis: Recommendations for the G20 (ed. R. Baldwin and S. Evenett), pp. 45–48. London: VoxEU/CEPR. 7 Gravity Chains: Estimating Bilateral Trade Flows when Trade in Components and Parts Is Important RICHARD E. BALDWIN AND DARIA TAGLIONI Trade is measured on a gross sales basis, while GDP is measured on a net sales basis, ie value added. The rapid internationalisation of production in the last two decades has meant that gross trade flows are increasingly unrep- resentative of the value added flows. This fact has important implications for the estimation of the gravity equation. We present empirical evidence that the standard gravity equation performs poorly by some measures when it is applied to bilateral flows where parts and components trade is important. We also provide a simple theoretical foundation for a modified gravity equa- tion that is suited to explaining trade where international supply chains are important. 1 INTRODUCTION Trade is measured on a gross sales basis, while GDP is measured on a value- added basis. For the first decades of the postwar period, this distinction was relatively unimportant. Trade in intermediates was always important, but it was quite proportional to trade in final goods. The rapid internationalisa- tion of supply chains in the last two decades has changed this (Yi 2003). Indeed, such trade has in recent decades boomed between advanced nations and emerging economies as well as among emerging nations, especially in Asia, where the phenomenon is known as ‘Factory Asia’. There are, however, similar supply chains in Europe and between the USA and Mexico (Kimura et al 2007). As a result, gross trade flows are increasingly unrepresentative of the value-added flows. This fact has important policy implications (Lamy 2010), but it also has important implications for one of trade economists’ standard tools: the gravity equation. The basic point is simple. The standard gravity equation is derived from a consumer expenditure equation with the relative price eliminated using a general equilibrium constraint (Anderson 1979; Bergstrand 1985, 1989, 1990). The corresponding econometrics widely used today is based on this theory 162 Trade in Value Added (Anderson and Van Wincoop 2003). As such the standard formulation— bilateral trade regressed on the two GDPs, bilateral distance and other controls—is best adapted to explaining trade in consumer goods. When con- sumer trade dominates, the GDP of the destination nation is a good proxy for the demand shifter in the consumer expenditure equation; the GDP of the ori- gin nation is a good proxy of its total supply. By contrast, when international trade in intermediate goods dominates, the use of GDPs for the supply and demand proxies is less appropriate. Consider, for instance, the determinants of Thai imports of automobile parts from the Philippines. The standard formulation would use Thai GDP to explain Thailand’s import demand. However, the underlying demand for parts is generated by Thai gross production of automobiles, not its value added in automobiles. As long as the ratio of local to imported content does not change, value added is a reasonable proxy for gross output, so the stan- dard regression is likely to give reasonable results. However, for regions where production networks are emerging, value added can be expected to be a poor proxy. Why do incorrectly specified mass variables matter? A large number of gravity studies focus on variables that vary across country pairs, say, free- trade agreements, cultural ties or immigrant networks. The most recent of these studies employ estimators that control for the mass variables with fixed effects. Such studies do not suffer from mass-variable mis-specification and so are unaffected by our critique. There are, however, a number of recent studies, especially concerning the ‘distance puzzle’, that do proxy for the production and demand variables with GDP. It is these studies that our work addresses. For example, Rauch (1999), Brun et al (2005), Berthelon and Freund (2008) and Jacks et al (2008) use GDP as the mass variable when they decompose the change in the trade flow into the effects of income changes and trade cost changes; Anderson and Van Wincoop (2003) also use GDP as the mass vari- able in one of their estimation techniques. Since most of these studies are concerned with a broad set of nations and commodities, the mis-specification of the mass variable probably has a minor impact on the results, as the find- ings of Bergstrand and Egger (2010) showed. More worrying, however, is GDP use by authors that focus on trade in parts and components, such as Athuko- rala and Yamashita (2006), Kimura et al (2007), Yokota (2008) and Ando and Kimura (2009). These papers all use the consumer good version of the gravity model to describe parts and components trade and thus have mis-specified the mass variable. 1.1 Literature Review There is nothing new about trade in intermediates. Intermediates have long been important in the trade between the USA and Canada; the 1965 US–Canada Auto Pact, for example, explicitly targeted preferential tariff reductions on Gravity Chains 163 cars and cars parts. Intermediates have also long been important within West- ern Europe, as early studies of the European Economic Community demon- strated (see, for example, Dreze 1961; Verdoorn 1960; Balassa 1965, 1966). The famous book by Grubel and Lloyd (1975) made it clear that much of intra-industry trade was in intermediates, not final goods, and the impor- tance of intermediates was reflected in early work by well-known theorists. For example, Vanek (1963) presented an extension of the Heckscher–Ohlin model that allows for intermediates trade, and Ethier (1982) cast his model of intra-industry trade in a world where all trade was in intermediates. As better data and computing technology became available, the importance of intermediates in trade was rediscovered and documented more thoroughly. In the context of efforts to understand the impact of the EU’s Single Mar- ket Programme, European scholars focused on the role of intermediates. For example, Greenaway and Milner (1987) list this as one of the ‘unresolved issues’, writing it is becoming increasingly obvious that a significant proportion of measured [intra-industry trade] is accounted for by trade in parts and components. [Nev- ertheless,] most of the models developed so far assume trade in final goods. The modelling of trade in intermediates needs to be explored further. The issue attracted renewed interest following development of the new trade theory in the 1980s (Helpman and Krugman 1985) 1 and again in the 1990s with Jones and Kierzkowski (1990), and Hummels et al (1998), 2 and more recently Kimura et al (2007) and Grossman and Rossi-Hansberg (2008). The traditional gravity model was developed in the 1960s to explain factory- to-consumer trade (Tinbergen 1962; Poyhonen 1963; Linnemann 1966). This concept is at the heart of the first clear microfoundations of the gravity equation: the seminal paper by Anderson (1979). 3 This proposed a theoret- ical explanation of the gravity equation based on constant elasticity of sub- stitution (CES) preferences when nations make a single differentiated prod- uct. Anderson and Van Wincoop (2003) use the Anderson (1979) theory to develop appropriate econometric techniques. Subsequent theoretical refine- ments have focused on showing that the gravity equation can be derived from many different theoretical frameworks (including monopolistic competition, and Melitz-type trade models with heterogeneous firms). 4 1 As illustrated by Grunwald and Flamm (1985). 2 See Feenstra (1998) for a survey of the 1990s literature. 3 Leamer and Stern (1970) informally discuss three economic mechanisms that might generate the gravity equations, but these were based on rather exotic economic logic; Anderson (1979) was the first to provide clear microfoundations that rely only on assump- tions that would strike present-day readers as absolutely standard. 4 On the monopolistic competition frameworks see Krugman (1980); Bergstrand (1985, 1989); Helpman and Krugman (1985); on the Heckscher–Ohlin model see Deardorff (1998); 164 Trade in Value Added Studies on the gravity equations applicability to intermediate goods trade are more limited. These include Egger and Egger (2004) and Baldone et al (2007). The study that is closest to ours is Bergstrand and Egger (2010). These authors developed a computable general equilibrium model that explains the bilateral flows of final goods, intermediate goods and foreign direct invest- ment (FDI). Calibration and simulation of the model suggests a theoretical rationale for estimating a near-standard gravity model for the three types of bilateral flows. Using a large data set on bilateral flows of final and intermedi- ate goods trade, and a data set on bilateral FDI flows, they estimate the three equations and find that the standard gravity variables all have the expected size and magnitude. The value added of our paper is primarily empirical: showing that the stan- dard gravity specification performs poorly when applied to flows where trade in intermediates is important. Moreover, the failures line up with the predic- tions of our simple theory model that suggests a gravity equation formulation that is appropriate to intermediates trade. Note that when we perform the estimates on data pooled across a wide range of nations, we find the same results, namely that the standard specification performs well. We believe the difference in our results from those of Bergstrand and Egger (2010) is due to the fact that for many trade flows, the pattern of trade in intermediates is quite proportional to trade in final goods. This is especially for trade among developed nations. 1.2 Plan of the Chapter The chapter starts with simple theory that generates a number of testable hypotheses. We then confront these hypotheses with the data and find that the estimated coefficients deviate from standard results in the way that the simple theory says they should. The key results are that the standard economic mass variable, which reflects consumer demand, does not perform well when it comes to bilateral trade flows where intermediates are dominant. Finally, we consider new proxies for the economic mass variables and show that using the wrong mass variable may bias estimates of other coefficients. 2 THEORY To introduce notation and fix ideas, we review the standard gravity derivation following Baldwin and Taglioni (2007). 5 Using the well-known CES preference on Ricardian models see Eaton and Kortum (2001); on Melitz (2003) model applications, see Chaney (2008) and Helpman et al (2008). 5 Another well-known derivation is from Helpman and Krugman (1985); they start from (8.1) and make supply-side assumptions that turn po into a constant, but make nod pro- portional to nation o’s GDP, so the resulting gravity equation is similar, at least in the case of frictionless trade (the case they worked with in 1985). Gravity Chains 165 structure for differentiated varieties, spending in nation d on a variety pro- duced in origination nation o is 1−σ pod vod ≡ Ed , σ > 1, (7.1) Pd where vod is the expenditure in destination country d, pod is the consumer price inside nation d of a variety made in nation o, Pd is the nation d CES price index of all varieties, σ is the elasticity of substitution among varieties, and Ed is the nation d consumer expenditure. From the well-known profit maximisation exercise of producers based in nation o, pod = µod mo τod , where µod is the optimal price mark-up, mo is the marginal costs, and τod is the bilateral trade cost factor, ie 1 plus the ad valorem tariff equivalent of all natural and manmade barriers. The mark-up is identical for all destinations if we assume perfect competition or Dixit–Stiglitz monopolistic competition; in these cases, the price variation is characterised by ‘mill pricing’, ie 100% pass-through of trade costs to consumers in the destination market. 6 Here we work with Dixit–Stiglitz competition exclusively, so the mark-up is always σ /(σ − 1). This means the local consumer price is σ poo = mo τoo , σ −1 where τoo is unity as we assume away internal trade barriers. Using this and summing over all varieties (assuming symmetry of varieties by origin nation for convenience), we have 1−σ 1−σ τod Vod = no poo 1−σ Ed (7.2) Pd where Vod is the aggregate value of the bilateral flow (measured in terms of the numeraire) from nation o to nation d; no is the number (mass) of nation o varieties (all of which are sold in nation d as per the well-known results of the Dixit–Stiglitz–Krugman model). To turn this expenditure function (with optimal prices) into a gravity equa- tion, we impose the market-clearing condition. Supply and demand match when (7.2), summed across all destinations (including nation o’s sales to itself), equals nation o’s output. When there is no international sourcing of parts, the nation’s output is its GDP, denoted here as Yo . Thus, the market- clearing condition is 1−σ 1−σ σ −1 Yo = no poo τod P d Ed . d 6 If one works with the Ottaviano et al (2002) monopolistic competition framework, the mark-up varies bilaterally and so mill pricing is not optimal. 166 Trade in Value Added 1−σ Solving this, we obtain that no poo = Yo /Ωo , where Ωo is the usual market- potential index (namely, the sum of partners’ market sizes weighted by a distance-related weight that places lower weight on more remote destina- tions); specifically, it is 1−σ σ −1 Ωo ≡ τod P d Ed . d Plugging this into (7.2) yields the traditional gravity equation: 1−σ 1 1 Vod = τod Ed Yo 1−σ . (7.3) Pd Ωo Here Pd is the nation d CES price index, while Ωo is the nation o market- 1−σ potential index. It has become common to label the product Pd Ωo as the ‘multilateral trade resistance’ term. However, it is insightful to keep in mind the fact that ‘multilateral trade resistance’ is a combination of two well-known, well-understood and frequently measured components. In the typical gravity estimation, Ed is proxied with nation d’s GDP, Yd is proxied with nation o’s GDP and τ is proxied with bilateral distance. 2.1 Gravity when Trade in Components and Parts Is Important To extend the gravity equation to allow for trade in parts and components among firms, we need a trade model where intermediate goods trade is explic- itly addressed. It proves convenient to work with the Krugman and Venables (1996) ‘vertical linkages’ model, which focuses squarely on the role of inter- mediate goods. Here we present the basic assumptions and the manipulations that produce the modified gravity equation. Krugman and Venables (1996) work with the standard new economic geog- raphy model, where each nation has two sectors (a Walrasian sector, A, and a Dixit–Stiglitz monopolistic competition sector M ) and a single primary factor: labour, L. Production of A requires only L, but production of each variety of X requires L and a CES composite of all varieties as intermediate inputs (ie each variety is purchased both for final consumption and for use as an inter- mediate). Following Krugman and Venables (1996), the CES aggregate on the supply side is isomorphic to the standard CES consumption aggregate. The indirect utility function for the typical consumer is 1/(1−σ ) I 1−α 1−σ V = , P c ≡ pA (P )α , P≡ pi di , (7.4) Pc i∈G where I is consumer income, P c is the ideal consumer price index, pA is the price of A, the parameter α is the Cobb–Douglas expenditure share for M - sector goods, σ is the elasticity of substitution among varieties, P is the CES price index for M varieties, pi is the consumer price of variety i and G is the set of varieties available. Gravity Chains 167 The cost function of a typical firm in a typical country is: C[w, P , x] = (F + aX x)w 1−α P α . (7.5) Here x is the output of a typical variety, F and aX are cost parameters, w is the wage and α is the Cobb–Douglas cost share for intermediate inputs. 7 As noted above, mill pricing is optimal under Dixit–Stiglitz monopolistic competition. This, combined with the identity of the elasticity of substitution, σ , for each good’s use in consumption and production, tells us that the price of each variety will be identical across the two types of customers. Choosing units such that aX = 1 − 1/σ , the landed price will be 1−α α pod = τod wo Po for all o, d. (7.6) Using Shepard’s and Hotelling’s lemmas on (7.4) and (7.5), and adding the total demand for purchasers located in nation d, we have an expression that is isomorphic to (7.2) except the definition of E now includes purchases by customers using the goods as intermediates: 1−σ 1−σ τod Vod = no poo 1−σ Ed , Ed ≡ α(Id + nd Cd ), (7.7) Pd where Id is nation d’s consumer income and Cd is the total cost of a typical nation d variety. 1−σ As before, we solve for the endogenous no poo using the market-clearing condition. In this case, the value that nation o must sell is the full value of its M -sector output (not just its value added). Under monopolistic competition’s free entry assumption, the value of sales equals the value of full costs, so the market clearing equation becomes 1−σ 1−σ 1−σ no Co = no poo Σd τod P d Ed , Co ≡ C[wo , Po , xo ], (7.8) where the cost function C is given in (7.5). Solving (7.8) and plugging the result into (7.7) yields a gravity equation modified to allow for intermediates goods trade, namely 1−σ 1 1 Vod = τod Ed Co 1−σ , (7.9) Pd Ωo where Ed is defined in (7.7) and Co is defined in (7.8), and 1−σ σ −1 Ωo ≡ τod P d Ed . d 7 The assumption that the Cobb–Douglas parameter is identical in the consumer and producer CES price index is one of the strategic implications in the Krugman–Venables model; see their book for a careful examination of what happens when this is relaxed (Fujita et al 2001). The standard conclusion is that it does not qualitatively change results but it does significantly complicate the analysis in a way that requires numerical simulation. 168 Trade in Value Added Expression (7.9) is the gravity equation modified to allow for trade interme- diates. The key differences show up in the definition of the economic ‘mass’ variables, since purchases are now driven by both consumer demand (for which income is the demand shifter) and intermediate demand (for which total production cost is the demand shifter). 3 BREAKDOWN OF THE STANDARD GRAVITY MODEL This theory exercise suggests a key difference that should arise between grav- ity estimates on nations and time periods where most imports are consumer goods versus those where intermediates trade is important. Specifically, the standard practice of using the GDP of origin and destination countries as the ‘mass’ variables in the gravity equations is inappropriate for bilateral flows, where parts and components are important. Of course, if the consumer demand and producer demand move in synch, as they may in a steady-state situation, then GDP may be a reasonable proxy for both consumer and pro- ducer demand shifters. But if the role of vertical specialisation trade is chang- ing over time, GDP should be less good at proxying for the underlying demand shifters. For this reason, we expect that origin country’s GDP and destination country’s GDP will have diminished explanatory power for those countries where value-chain trade is important. These observations generate a number of testable hypotheses. • The estimated coefficient on the GDPs should be lower for nations where parts trade is important, and should fall as the importance of parts trade rises. • As vertical specialisation trade has become more important over time, the GDP point estimates should be lower for more recent years. • In those cases where the GDPs of the trade partners lose explanatory power, bilateral trade should be increasingly well explained by demand in third countries. For example, China’s imports should shift from being explained by China’s GDP to being explained by its exports to, say, the USA and the EU. There are two ways of phrasing this hypothesis. First, China’s imports are a function of its exports rather than its own GDP. Second, China’s imports are a func- tion of US and EU GDP rather than its own, since US and EU GDP are critical determinants of their imports from China. To check these conjectures, we estimate the standard gravity model for different sets of countries and sectors for a panel that spans the years 1967 to 2007. We run standard log-linear gravity equations using pooled cross-section time series data, namely Yot Edt ln(Vodt ) = G + α1 ln 1−σ + α2 ln τodt + εodt . (7.10) Ωot Pdt Gravity Chains 169 A key econometric problem is that the price index Pdt and the market potential index Ωot are unobservable and yet include factors that enter the regressions independently (eg E , Y and τ ). Thus, ignoring them can lead to serious biases. If the econometrician is only interested in estimating the impact of a pair- specific variable, such as distance or tariffs, the standard solution is to put in time-varying country-specific fixed effects. This eliminates all the terms multiplied by α1 in Equation (7.10). Plainly, we cannot use this approach to investigate the impact of using GDPs as the economic mass proxies when trade in parts and components is important. We thus need other means of controlling for Ωot and Pdt . Our baseline specification accounts for the terms Ωot and Pdt explicitly. As precise measures of Ωot and Pdt are hard to construct, we perform robustness checks using fixed effects specifications. To ensure comparability with the fixed effects specification, in the key specifications we enter the importer’s and exporter’s economic mass as a single product-term into the equation, with the shortcoming of forcing the coefficient of the importer and exporter mass variables to be the same. Specifically, the term accounting for the product of the trade partners’ economic mass is the product of importer d’s real GDP (so as to account for Pdt ) and of exporter o’s nominal GDP divided by a proxy for Ωot , constructed adapting a method first introduced by Baier and Bergstrand (2001), namely 1/(1−σ ) Ωot = GDPdt (Distod )1−σ . d The elasticity value in the Ωot relationship has been set as σ = 4, which corresponds to estimates proposed in empirical literature (see, for example, Obstfeld and Rogoff (2001); Carrere 2006). Turning to the trade cost variable, τ , we introduce standard trade frictions, including log of bilateral distance, dummies for contiguity and common lan- guage. Moreover, for robustness we also test for additional time-varying trade frictions measured by the ratio of cost, insurance and freight (CIF) prices to free on board (FOB) prices, as proposed by Bergstrand and Egger (2010). The data used for the bilateral trade flows, and the CIF/FOB ratios are taken from the UN Comtrade database. GDPs are from the World Bank’s World Devel- opment Indicators. Bilateral distances, contiguity and common language are from the CEPII database. Data for Taiwan (Chinese Taipei), which are missing from the UN databases, are from CHELEM (CEPII) and national accounts. Estimation is by simple ordinary least squares with the standard errors clustered by bilateral pairs, since we work in direction-specific trade flows rather than the more traditional average of bilateral flows. 3.1 Empirical Results In Table 7.1 we report the gravity equation estimates for all goods as well as for intermediate and final goods separately. Intermediate and final goods have 170 Table 7.1: Bilateral flows of total, intermediate and final goods, 187 nations, 2000–7. All goods Intermediates only Consumer goods only Variables (1) (2) (3) (4) (5) (6) GDPot GDPdt ln 0.860∗∗∗ 0.865∗∗∗ 0.898∗∗∗ 0.905∗∗∗ 0.791∗∗∗ 0.796∗∗∗ Ωot Pdt (0.006) (0.006) (0.007) (0.007) (0.008) (0.008) ln(CIF/FOB) −0.0833∗∗∗ −0.0798∗∗∗ −0.189∗∗∗ −0.184∗∗∗ −0.341∗∗∗ −0.338∗∗∗ (0.013) (0.013) (0.015) (0.015) (0.017) (0.017) ln distance −0.775∗∗∗ −0.777∗∗∗ −0.851∗∗∗ −0.855∗∗∗ −0.758∗∗∗ −0.760∗∗∗ (0.019) (0.019) (0.022) (0.022) (0.025) (0.025) Contiguity 1.575∗∗∗ 1.565∗∗∗ 1.711∗∗∗ 1.697∗∗∗ 1.356∗∗∗ 1.347∗∗∗ (0.105) (0.105) (0.119) (0.119) (0.127) (0.127) Common language 0.966∗∗∗ 0.972∗∗∗ 0.997∗∗∗ 1.005∗∗∗ 1.186∗∗∗ 1.192∗∗∗ (0.046) (0.046) (0.052) (0.052) (0.059) (0.059) Constant −28.61∗∗∗ −28.74∗∗∗ −30.84∗∗∗ −31.03∗∗∗ −26.87∗∗∗ −27.02∗∗∗ Trade in Value Added (0.359) (0.363) (0.400) (0.404) (0.456) (0.459) Time dummies Yes Yes Yes Observations 62,875 62,875 62,875 62,875 58,468 58,468 R2 0.627 0.628 0.585 0.587 0.479 0.480 Source: authors’ calculations. Dependent variable: imports + reimports. Standard errors are clustered by bilateral pair. Robust standard errors are reported in parentheses: ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. Gravity Chains 171 Table 7.2: Classification for intermediate and final goods. Goods BEC categories Intermediate goods: 111 Primary food and beverages, mainly for industry 121 Processed food and beverages, mainly for industry 21 Primary industrial supplies not elsewhere specified 22 Processed industrial supplies not elsewhere specified 32 Processed fuels and lubricants 42 Parts and accessories of capital goods (except transport equipment) 53 Parts and accessories of transport equipment Consumption goods: 112 Primary food and beverages, mainly for household consumption 122 Processed food and beverages, mainly for industry 51 Passenger motor cars 6 Consumer goods not elsewhere specified Other: 31 Primary fuels and lubricants 41 Capital goods, excluding parts and components 51 Other transport equipment 7 Other Source: Comtrade’s Broad Economic Categories. For details see http://unstats.un.org/unsd/tradekb/ Knowledgebase/Intermediate-Goods-in-Trade-Statistics. been identified according to the UN Broad Economic Categories Classification (see Table 7.2). The sample includes all the nations where data is available, namely 187 nations. Coefficients have the expected signs and are statistically significant. For all six regressions (all goods, only intermediates and only consumer goods with and without time fixed effects) the estimates are broadly similar. The mass variables are all estimated to be close to unity. The bilateral distance variable is negative and falls in the expected range. The additional trade cost measure, the CIF/FOB ratio, is always negative, as expected for the subsamples, but positive for the aggregate sample. Continuity and language always have the expected sign and fall in the usual ranges. These Table 7.1 results confirm the findings of Bergstrand and Egger (2010), namely that the size of the estimated coefficients does not vary for consumer and intermediate goods. As such, it would seem that our concern about mis- estimating the gravity equation is misplaced. However, as noted above, if the consumer and intermediate trade is roughly proportional over time, GDP will be a reasonable proxy for both consumer income and gross value added. The real test of the stability of the parameters would be on a sample where the importance of intermediates trade was rising significantly. To check this, we turn to a subsample of nations where we a priori expect intermediate trade to be both very important and growing more rapidly than consumer trade. Specifically, we estimate a gravity model as in Table 7.1, but on bilateral trade between pairs of Factory Asia countries (ie Japan, Indonesia, 172 Trade in Value Added Table 7.3: Bilateral flows of total goods among Factory Asia nations (1967–2008). No time interactions Variable mass coefficient Variables (1) (2) (3) (4) (5) ln(GDPo GDPd /Ωo Pd ) 0.725∗∗∗ 0.725∗∗∗ 0.764∗∗∗ 0.425∗∗∗ 0.504∗∗∗ (0.009) (0.028) (0.026) (0.055) (0.051) ∗ years 1967–1986 0.318∗∗∗ 0.278∗∗∗ (0.048) (0.048) ∗ years 1987–1996 0.177∗∗∗ 0.164∗∗∗ (0.027) (0.032) ∗ years 1998–2002 0.007 0.00274 (0.015) (0.017) ln(distance) −0.258∗∗∗ −0.258 −0.0414 (0.0570) (0.298) (0.297) Contiguity 0.188∗∗∗ 0.188 0.167 (0.0682) (0.386) (0.367) Colony −0.487∗∗∗ −0.487 0.0695 (0.101) (0.388) (0.405) Common −0.620∗∗∗ −0.620* −0.296 coloniser (0.116) (0.325) (0.324) Constant −7.218∗∗∗ −7.218∗∗∗ −8.825∗∗∗ −1.465 −2.632∗∗ (0.433) (2.281) (0.485) (2.279) (1.178) Observations 1722 1722 1722 1722 1722 R2 0.833 0.833 0.936 0.851 0.948 Time effects Yes Yes Exporter∗ time effects Yes Yes Yes Importer∗ time effects Yes Yes Yes Pair effects Yes Yes Yes Observations 820 820 820 820 820 R2 0.932 0.932 0.978 0.934 0.978 Clustered Yes Yes Yes Yes standard errors Source: authors’ calculations; Note: Standard errors are clustered by bilateral pair. Robust standard errors in parentheses: ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. Factory Asia countries: Japan, Indonesia, Republic of Korea, Malaysia, Thailand and Taiwan (Chinese Taipei). Korea, Malaysia, Philippines, Thailand and Taiwan (Chinese Taipei)). To gauge the stability of parameters, we interact time dummies with the mass variable. The results, shown in Table 7.3, are quite different to those of Bergstrand and Egger (2010) and to those of Table 7.1. The baseline regressions (without time interactions) show the fairly com- mon result that the gravity model does not work well on Factory Asia nations. The estimated mass coefficient is fairly low at about 0.7. The distance esti- mate, however, at −0.26 is much lower than the commonly observed −0.7 to −1.0. When we include time interaction terms for the economic mass vari- able, we find that the coefficient is not stable over time. When the standard Gravity Chains 173 1.2 1.0 0.8 0.6 0.4 1985 1995 0.2 Year coefficient 0 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 Figure 7.1: GDP coefficients for Factory Asia countries, 1967–2008. Estimated coefficients of α1 with year dummies. Base estimation is specified as in (7.9). Fixed effects estimation is specified as in (7.10). Factory Asia countries: Japan, Indone- sia, Republic of Korea, Malaysia, Thailand, and Taiwan (Chinese Taipei). controls are included (see column 4), the base case estimate is 0.4, to which must be added the period coefficients, which are 0.3 for the pre-Factory Asia period (Baldwin 2006), 0.2 for the 1987–96 period and essentially zero (and insignificant) for the post-1998 period. To estimate the mass variable’s instability over time more clearly, we re-do the same regression but allow yearly interaction terms. The results, displayed in Figure 7.1, show the evolution of the GDP coefficients. The mass elasticity fall over time, with two clear breaks in the estimated coefficients, 1985 and 1998. The timing and direction of these structural changes are very much in line with the literature on the internationalisation of production. According to many studies, production unbundling started in the mid-1980s and acceler- ated in the 1990s (see, for example, Hummels et al 1998). The idea is that coordination costs fell with the information and communications technology (ICT) revolution and this permitted the spatial bundling of production stages (Baldwin 2006). The ICT revolution came in two phases. The Internet came online in a massive way in the mid-1980s, and then, in the 1990s, the price of telecommunications plummeted with various ITC-related technical inno- vations and widespread deregulation (Baldwin 2011). The upshot of all these changes was that it became increasingly economical to geographically sep- arate manufacturing stages. Stages of production that previously were per- formed within walking distance to facilitate face-to-face coordination could be dispersed without an enormous drop in efficiency or timeliness. 174 Trade in Value Added Table 7.4: Estimates for EU15, and USA, Canada, Australia and New Zealand, 1967–2008. No time interactions Variable mass coefficient Variables (1) (2) (3) (4) (5) ln(GDPo GDPd /Ωo Pd ) 0.659∗∗∗ 0.659∗∗∗ 0.632∗∗∗ 0.725∗∗∗ 0.703∗∗∗ (0.009) (0.025) (0.027) (0.058) (0.034) ∗ years 1967–1986 −0.0408 −0.0503 (0.051) (0.044) ∗ years 1987–1996 −0.0376 −0.0444 (0.036) (0.032) ∗ years 1998–2002 0.0132 0.005 (0.017) (0.014) ln(distance) −0.843∗∗∗ −0.843∗∗∗ −0.688∗∗ (0.059) (0.233) (0.276) Constant −1.630** −1.630 −8.819∗∗∗ −4.966 −10.72∗∗∗ (0.726) (2.284) (0.657) (3.733) (0.917) Time effects Yes Yes Exporter∗ time effects Yes Yes Yes Importer∗ time effects Yes Yes Yes Pair effects Yes Yes Yes Observations 8020 8020 8020 8020 8020 R2 0.932 0.932 0.978 0.934 0.978 Clustered standard Yes Yes Yes Yes errors Source: authors’ calculations. Standard errors are clustered by bilateral pair. Robust standard errors are reported in parentheses: ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. As far as the Figure 7.1 results are concerned, the notion is that as trade became increasingly focused on intermediates, GDP became an increasingly poor determinant of trade flows, as suggested by our theory. The impacts of the mid-1980s and the mid-1990s changes are clear from the estimated GDP elasticities. More specifically, from 1967 to 1985 the elasticity of these countries’ bilateral imports to GDP was stable, with a coefficient of about 0.77. Between 1985 and 1997, it steadily decreased, to reach a coefficient value of about 0.60, and after 1998 it further dropped, to close to 0.40. The coefficient estimates for the different periods in Factory Asia are summarised in Table 7.3, columns (4) and (5). For comparison we also report results of time-year interactions with GDP for bilateral trade between countries where we a priori expect bilateral trade to be dominated by consumption goods and/or a stable ratio of intermedi- ates to final goods trade. To this end, we re-run the Table 7.3 regressions for bilateral trade between each of the EU15 nations, the USA, Canada, Aus- tralia and New Zealand. Because most of the internationalisation of supply chains is regional rather than global (except for microelectronics), we expect these bilateral trade flows to be less influenced by the second unbundling Gravity Chains 175 that so marked Factory Asia trade. The results, shown in Table 7.4 tend to confirm our view that the gravity model breaks down only for bilateral flows where production sharing is especially important and growing quickly. That is, as predicted by our theory, we find no breaks over time in the trade coef- ficients, while distance coefficients have elasticity levels which are closer to unity. None of the time interaction terms in columns (4) and (5) are significant and the other point estimates fall in the expected ranges. 3.2 More Precise Estimates of the Impact of Components on the Mass Estimate These two sets of results are highly suggestive. On data that is widely recog- nised as being dominated by parts and components trade, we find structural instability in the mass variable coefficient moving in the expected direction. However, on data where this sort of production fragmentation is not widely viewed as having been important, we find that mass point estimates are stable over time. To explore this more systematically, we consider a less granular relation- ship between the importance of components trade and the point estimate on the mass variable on the full sample. Our basic assertion is that the compo- sition of trade flows will influence the point estimates of the economic mass variables, since the standard gravity model is mis-specified when it comes to the mass variable. The most direct test of this hypothesis is to include the ratio of intermediates to total trade as a regressor, both on its own and, more importantly, as an interaction term with the economic mass variable. Of course a mis-specification of one part of the regression has implications for the point estimates of the other regressors, so we also consider the ratio’s interaction with the other main regressors. To this end, we re-estimate the basic equation on the full sample of 187 countries for the years 2000–8, allowing for interactions with a variable that accounts for the share of intermediate goods over total imports in each particular bilateral trade flow. The idea here is that GDP as a measure for economic mass should work less well for those bilateral flows that are marked by relatively high shares of intermediates trade. By estimating the effect on the full sample, we avoid the problem of identifying the exact sources of the variation in the coefficients. We implement the idea in two ways. First we estimate the standard regression but include the share of bilateral interm imports that is in intermediates (denoted as Md /Md ). This new variable is included on its own and interacted with the other right-hand side variables. Table 7.5 reports the estimated results for the coefficients of interest. The regression results tend to confirm our hypothesis. The regression reported in column (1) includes the ratio on its own and interacted only with the mass variable. The coefficients for economic mass and distance are a 176 Trade in Value Added Table 7.5: Interactions with share of intermediates in total imports, full sample. Variables (1) (2) (3) (4) interm Md /Md 6.536∗∗∗ 8.018∗∗∗ 6.954∗∗∗ 7.330∗∗∗ (0.858) (1.015) (0.835) (1.004) ln(GDPo GDPd /Ωo Pd ) 1.031∗∗∗ 1.027∗∗∗ 1.064∗∗∗ 1.058∗∗∗ (0.010) (0.010) (0.010) (0.010) ∗ M interm /M −0.129∗∗∗ −0.118∗∗∗ −0.137∗∗∗ −0.126∗∗∗ d d (0.017) (0.017) (0.017) (0.016) ln(distance) −1.173∗∗∗ −1.051∗∗∗ −1.011∗∗∗ −0.954∗∗∗ (0.018) (0.037) (0.0191 (0.037) ∗ M interm /M −0.232∗∗∗ −0.110* d d (0.059) (0.0601 Contigod 1.350∗∗∗ 0.967∗∗∗ (0.101) (0.246) ∗ M interm /M 0.625* d d (0.369) Common language 1.215∗∗∗ 1.126∗∗∗ (0.044) (0.078) ∗ M interm /M 0.178 d d (0.119) Constant −27.58∗∗∗ −28.40∗∗∗ −30.85∗∗∗ −31.07∗∗∗ (0.551) (0.634) (0.541) (0.625) Observations 121,737 121,737 121,737 121,737 R2 0.604 0.604 0.621 0.621 Mdtinter m /Md is the share of intermediate imports by a country d over its total imports. Robust standard errors are reported in parentheses: ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. very reasonable, at 1.031 and −1.173, respectively (both significant at the 1% level). The ratio on its own comes in positive as expected (bilateral trade links marked by a high share of intermediates tend to have ‘too much’ trade compared with the prediction of the standard gravity equation). The ratio interacted with economic mass also has a negative sign, −0.129, which con- forms with our hypothesis (the higher the ratio of intermediates for the par- ticular trade pair, the lower the estimate of the economic mass variable). All coefficients are significantly different to zero at the 1% level of confidence. The other columns report robustness checks on the main regression. The qualitative results on the variables of interest (the mass coefficient, the ratio coefficient and the mass × ratio interaction coefficient) are robust to inclusion of interaction terms with any or all of the control variables. This confirms the more informal tests based on an a priori separation of the sample. Interestingly, the interaction term is also highly significant and negative for distance in specification (2). That is, distance seems to matter more for components trade: a result that is not in line with our simple model, but is expected from the broader literature on offshoring. For example, transporta- Gravity Chains 177 Table 7.6: All countries, 2000–7, by share of intermediate imports. GDPo GDPd Variables ln(distance) Constant Ωo Pd Base effect 0.985∗∗∗ −1.105∗∗∗ −26.29∗∗∗ (0.018) (0.018) (0.898) Base effect∗ d2 −0.0308 (0.021) Base effect∗ d3 0.0108 (0.021) Base effect∗ d4 −0.0330 (0.020) Base effect∗ d5 −0.0803∗∗∗ (0.020) Base effect∗ d6 −0.103∗∗∗ (0.021) Base effect∗ d7 −0.0903∗∗∗ (0.021) Base effect∗ d8 −0.0723∗∗∗ (0.022) Base effect∗ d9 −0.118∗∗∗ (0.024) Base effect∗ d10 −0.0748∗∗∗ (0.022) Observations 121,712 R2 0.610 Source: authors’ estimations. Deciles categorise countries’ bilateral imports by increasing shares of intermediate imports over total imports. Hence, d10 indicates the 10% bilateral import relationships where the share of intermediate imports in total imports is highest, and the base effect indicates the 10% bilateral import relationships where the share of intermediate imports in total imports is lowest. Common language and contiguity included by not reported. Standard errors are clustered by bilateral pair. Robust standard errors are reported in parentheses: ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. tion costs become more important when trade costs are incurred between each stage of production, while the value added per stage is modest. The second approach is to use decile-dummies to permit a more flexible relationship between the share of imports made up of components and the mass point estimate. The idea is that the inclusion of the intermediates-ratio imposes linearity on the relationship. The deciles approach allows the inter- action terms to be nonlinear, for example, it allows for the possibility of a threshold effect whereby the interaction is significant but only for ratios that are sufficiently large. More specifically, the dummies categorise the share of intermediates in total imports, ie a dummy that selects bilateral flows where the proportion of intermediate imports is below 10%, between 10% and 20%, etc. The results are shown in Table 7.6. All results are robust to the addition of other trade determinants. For the variable of greatest interest, the economic mass variable, the coeffi- cient for the base-case decile is 0.985, which is very close to unity as expected, 178 Trade in Value Added 1.05 1.00 Size coefficient 0.95 0.90 0.85 0 10 20 30 40 50 60 70 80 90 Share (%) of intermediates in total imports 1−σ Figure 7.2: Coefficient for the size variables measured as ln((Yot /Ωot )(Edt /Pdt )). Source: authors’ estimations. and very precisely estimated. The subsequent rows show the additional effects for each decile. What we see is that the interaction terms are insignificant for shares of intermediates below 50% of total imports. However, for high con- centrations of intermediates, the interaction terms are all negative and highly significant – at the 1% level. The additional effects lower the base case point estimate by around 0.10. The distance term is a very reasonable (−1.1) and highly significant. The results in Table 7.6 suggest that there is something of a threshold effect in operation. What we see is that the standard gravity specification works rather well for bilateral trade flows where the ratio of intermediates is not too great. For trade flows where intermediates are more important, however, we get the by now familiar result that the mass coefficient is significantly lower. Since this share is indeed rather low for most bilateral trade flows in the world (since production fragmentation tends to be a regional phenomenon), this may help explain the Bergstrand and Egger (2010) result mentioned above. To illustrate the point graphically, we plot, in Figure 7.2, the point estimates and standard errors using a candle chart. Here the point estimates of the mass coefficients are plotted as the horizontal bar; the associated standard errors are shown with the vertical bar. 4 A SEARCH FOR MASS PROXIES WHEN INTERMEDIATES ARE IMPORTANT The previous section provides clear evidence that the standard gravity equa- tion is ‘broken’ when it comes to bilateral flows where trade in intermediates is important. The theory suggests that the perfect solution would require data on total costs to construct the demand shifter for intermediates imports. If the economy is reasonably competitive, gross sales would be a good proxy for the total costs. Unfortunately, such data are not available for a large number of Gravity Chains 179 nations, especially the developing nations, where production fragmentation is so important. On the mass variable for the origin nation, theory suggests that we use gross output rather than value added. Again such data are not widely available. This section presents the results of our search for a pragmatic ‘repair’ which relies only on data that is available for a wide range of nations. The basic thrust is to use the theory in Section 2 to develop some proxies for economic mass variables that better reflect the fact that the demand for intermediates depends upon gross output, not value added. 4.1 Fixes for Economic Mass Proxies We start with the destination nation’s mass variable. In Section 2 we showed that a bilateral flow of total goods is the sum of goods whose demand depends upon the importing nation’s GDP (ie consumer goods) and goods whose demand depends upon the total costs of the sector buying the relevant inter- mediates. The theory says that our economic mass measure should be a linear combination of two mass measures, not a log-linear combination (see equa- tions (7.9) and (7.7)). This suggests a first measure that adds imports of intermediates to GDP. The idea here is to exploit the direct definition of total costs as the cost of primary inputs plus the value of intermediate inputs. For any given local firm, some of the intermediates it purchases will be from local suppliers, but on summing across all sectors and firms within a single nation, such intermedi- ates will cancel out, leaving only payments to local factors of production and imports of intermediates. Our first pragmatic fix therefore is to measure the destination nation’s demand shifter by using interm E d ≡ Yd + Vd,i , (7.11) i=o where V interm is the value of bilateral imports of intermediates. If we summed across all partners, this measure would include part of the bilateral flow to be explained (namely, intermediates from nation o to nation d). To avoid putting the trade flow to be explained on both sides of the equation, we build the measure for each pair in a way that excludes the pair’s bilateral trade. For the economic mass variable size pertinent to the origin nation, we are trying to capture gross output that must be sold. The proposed measure is a straightforward application of the theory; it uses the origin nation’s value added in manufacturing and its purchases of intermediate inputs from all sources except from itself (due to a lack of data): manuf interm Co ≡ AVo + Vi,o . (7.12) i=0 Note that our specification of the gravity equation uses the exports from nation o to nation d, so the second term in this does not include the bilateral 180 Trade in Value Added Table 7.7: New mass proxies with share of intermediate, all nations, 2000–7. Variables (1) (2) (3) (4) interm Md /Md 1.180 2.644∗∗ 2.044∗∗ 1.907∗ (1.020) (1.142) (0.988) (1.143) ln(Ed Co /Ωo Pd ) 0.898∗∗∗ 0.889∗∗∗ 0.945∗∗∗ 0.932∗∗∗ (0.012) (0.0116) (0.012) (0.012) ∗ M interm /M −0.0322 −0.0132 −0.0289 −0.0247 d d (0.020) (0.020) (0.020) (0.020) ln(distance) −1.080∗∗∗ −0.929∗∗∗ −0.908∗∗∗ −0.838∗∗∗ (0.018) (0.038) (0.019) (0.038) ∗ M interm /M −0.279∗∗∗ −0.131∗ d d (0.065) (0.067) Contigod 1.441∗∗∗ 1.211∗∗∗ (0.092) (0.224) ∗ M interm /M 0.356 d d (0.354) Common language 1.251∗∗∗ 1.047∗∗∗ (0.047) (0.088) ∗ M interm /M 0.385∗∗∗ d d (0.143) Constant −20.05∗∗∗ −20.87∗∗∗ −24.17∗∗∗ −24.08∗∗∗ (0.623) (0.687) (0.610) (0.685) Observations 87,258 87,258 87,258 87,258 R2 0.607 0.607 0.631 0.631 Note: Robust standard errors are reported in parentheses: ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. Pair effects and standard errors are clustered by pair. Mdinterm /Md is the share of intermediate imports by a country d over its total imports. New mass variables are defined in the text. flow to be explained. The second term involves nation o’s imports from all nations, not its exports to nations. 4.2 Empirical Results To test whether these proposed proxies work better than GDP, we run regres- sions like those reported in Table 7.5 but with the new proxies for economic mass replacing the standard proxy (ie GDP). The results are shown in Table 7.7. The results in Table 7.7 (compared with those in Table 7.5) suggest that our proxies work better than GDP. The key piece of evidence can be seen in column (1). This includes the ratio of intermediates in total bilateral trade both on its own and interacted with the mass variable. The lack of significant of the ratio in either role suggests that our new proxy is doing a better job than GDP did in picking up demand and supply of intermediates. Interestingly, the column (2) regression, which allows an interaction between distances on the ratio of intermediates, suggests that the distance coefficient may also be mis-specified. When the interaction effect between the above mentioned ratio and the dummy for distance is computed, results show Gravity Chains 181 Table 7.8: New mass proxies with intermediate deciles, all nations, 2000–7. ln(Ed Co /Ωo Pd ) ln(distance) Constant Base effect 0.877∗∗∗ −1.051∗∗∗ −19.29∗∗∗ (0.022) (0.018) (1.074) Base effect∗ d2 0.0402 (0.024) Base effect∗ d3 0.0365∗∗∗ (0.025) Base effect∗ d4 0.0294 (0.024) Base effect∗ d5 −0.0256 (0.024) Base effect∗ d6 −0.0531∗∗ (0.025) Base effect∗ d7 −0.0390 (0.025) Base effect∗ d8 −0.0306 (0.026) Base effect∗ d9 −0.0652∗∗ (0.028) Base effect∗ d10 0.0102 (0.027) Observations 87,251 R2 0.609 See notes to Table 7.6. that the distance estimate falls somewhat on average, but especially for trade flows where parts and components are especially important (ie the ratio is high). This suggests that distance is more important, not less, for bilateral trade flows dominated by intermediates. The finding may reflect the well-known fact that most production fragmentation arrangements are regional rather than global (components trade is more regionalised that overall trade). This result, however intriguing, does not really stand up to minor changes in the specification. In regression (4), which includes the ratio’s interaction with all variables, the distance result fades; indeed, only the common language effect seems to be magnified for trade flows marked by particularly high ratios of intermediates. Importantly, we note that in all specifications, the ratio’s interaction term on the economic mass is always insignificant. This suggests that our new mass proxies are doing a better job of picking up the true supply and demand variables including intermediates. For symmetry, and to check for nonlinear interaction terms, we use our new mass proxies in a regression akin to Table 7.6. The idea is to use ratio decile dummies instead of the ratio itself in order to allow the interactions to vary 182 Trade in Value Added nonlinearly for bilateral flows marked by different degrees of intermediates trade. The results are shown in Table 7.8. To interpret our findings, recall that the significance of the upper-tier decile interaction terms was taken as evidence that GDP was not working well for trade flows marked by much trade in intermediates. Thus, the results in Table 7.8 suggest that our new proxy is working better than GDP. Specifically, the base effect for our economic mass variable and the distance coefficients are estimated at very reasonable point estimates (0.88 and −1.1, respectively). Critically, only one of the decile interaction terms is significant, and it is positive, not negative as the theory would suggest. Two other inter- action terms are borderline significant and negative: those for the sixth and tenth deciles. 5 WHY DO INCORRECTLY SPECIFIED MASS VARIABLES MATTER? A large number of gravity studies focus on variables that vary across country pairs, say free-trade agreements, cultural ties or immigrant networks. The most recent of these studies employ estimators that control for the mass variables with fixed effects. Such studies do not suffer from mass-variable mis-specification and so are unaffected by our critique. There are, however, as mentioned in Section 1, a number of recent studies, especially concerning the ‘distance puzzle’, that do proxy for the production and demand variables with GDP. It is these studies that our work addresses. 8 However, since most of these studies are concerned with a broad set of nations and commodities, the mis-specification of the mass variable probably has a minor impact on the results, as the findings of Bergstrand and Egger (2010) showed and we confirmed with our Table 7.1 results. More worrying, however, is its use by authors who focus on trade in parts and components. 9 These papers use the consumer-good version of the gravity model and thus mis-specify the mass variable. Once the equation is mis-specified—in particular, if the standard economic mass proxies do not correctly reflect the supply and demand constraints—we are in the realm of omitted variable biases. The first task is to explore the nature of the biases that would arise from this mis-specification. To simplify, we assume away GDPs and distance and focus on a pair-wise policy variable, say, nation d’s tariffs on imports from nation o; we denote this by Tod . The estimated gravity equation will thus have the following structure: ln Vodt = const. + a5 ln Todt + εodt , (7.13) 8 See Rauch (1999), Brun et al (2005), Berthelon and Freund (2008), Jacks et al (2008), and Anderson and Van Wincoop (2003). 9 See Athukorala and Yamashita (2006), Kimura et al (2007), Yokota (2008) and Ando and Kimura (2009). Gravity Chains 183 where the error is assumed to be independent and identically distributed (iid). Because intermediates supply is measured by total costs rather than GDP, and the supply of intermediates that must be sold depends upon gross out- put rather than value added. This means that the true model includes an additional term. That is, ln Vodt = a0 + a5 ln Todt + a6 ln Zodt + εodt , (7.14) where Zodt is the difference between the GDP-based mass variables and the true mass variables as specified in (7.7). We can write Zodt as a function of Todt in an auxiliary regression: ln Zodt = b0 + b1 ln Todt + uodt , (7.15) where u is assumed to be iid. Using this notation for the coefficients of the auxiliary regression, we can see that in estimating (7.3), we are actually esti- mating ln Vodt = (a0 + bo a6 ) + (a5 + a6 b1 ) ln Todt + (εodt + a6 uodt ). (7.16) What this tells us is that the coefficient on the policy variable of interest will almost surely be biased. The point is that the only way it is not biased is if there is no correlation between the mis-specification of the economic mass variables and the policy variable. What sort of correlation should we expect? Recall that the mis-measurement of the economic mass variable goes back to the importance of trade in inter- mediate goods. Since almost all bilateral variables of interest are things that affect bilateral trade flows, it seems extremely likely that the variable of inter- est will also affect the flow of intermediates. As long as it does, then we know that the mis-specification of the mass variable will also lead to a bias in the pair-wise variables. 10 For example, let us suppose that tariffs discourage trade overall, but they especially discourage intermediates trade (for the usual effective rate of pro- tection reasons, ie the tariff is paid on the gross trade value but its incidence falls on the value added only). In this case, we should expect low tariffs to encourage two things: an overall increase in trade and an increase in the ratio of intermediates. In this case, the bias in the mis-specified gravity equation is likely to be negative, since the policy variable is negatively correlated with the omitted variable. Furthermore, the mis-specification also affects the standard errors, which would result in a biased inference (Wooldridge 2003, Chapter 4). 10 As noted above, the modern techniques for controlling for mass with time-varying country-specific dummies eliminates such biases, since they correctly control for the role of intermediates. 184 Trade in Value Added 6 CONCLUDING REMARKS In this chapter we present empirical evidence that the standard gravity model performs poorly by some measures when it is applied to bilateral flows where parts and components trade is important. The chapter also provides a sim- ple theoretical foundation for a modified gravity equation that is suited to explaining trade where international supply chains are important. Finally, we suggest ways in which the theoretical model can be implemented empirically. REFERENCES Anderson, J. (1979). The theoretical foundation for the gravity equation. American Economic Review 69, 106–116. Anderson, J., and E. van Wincoop (2003). Gravity with gravitas: a solution to the border puzzle. American Economic Review 93(1), 170–192, Anderson, J., and Y. V. Yotov (2010). The changing incidence of geography. American Economic Review 100(5), 2157–86. Ando, M., and F. Kimura (2009). Fragmentation in East Asia: Further Evidence. ERIA Discussion Paper Series, DP-2009-20, October. Athukorala, P., and N. Yamashita (2006). Production fragmentation and trade integra- tion: east asia in a global context. The North American Journal of Economics and Finance 17(3), 233–256. Baier, S. L., and J. H. Bergstrand (2001). The Growth of World Trade: Tariffs, Transport Costs, and Income Similarity. Journal of International Economics 53(1), 1–27. Balassa, B. (1965). Economic Development and Integration Report. Centro de Estudios Monetarios Latinoamericanos. Balassa, B. (1966). Tariff reductions and trade in manufacturers among the industrial countries. American Economic Review 56(3), 466–473. Baldone, S., F. Sdogati and L. Tajoli (2007). On some effects of international fragmen- tation of production on comparative advantages, trade flows and the income of countries. The World Economy 30(11), 1726–1769. Baldwin, R. (2006). Globalisation: The Great Unbundling(s). In Globalisation Challenges For Europe, Chapter 1, pp. 5–47. Secretariat of the Economic Council, Finnish Prime Minister’s Office, Helsinki. Baldwin, R. (2011). 21st Century Regionalism: Filling the Gap between 21st Century Trade and 20th Century Trade Governance. CEPR Policy Insight, No. 56. Baldwin, R., and D. Taglioni (2007). Trade effects of the euro: a comparison of estima- tors. Journal of Economic Integration 22(4), 780–818. Bergstrand, J. (1985). The gravity equation in international trade: some microeconomic foundations and empirical evidence. Review of Economics and Statistics 67(3), 474– 481. Bergstrand J. (1989). Generalized Gravity Equation, Monopolistic Competition, and the Factor-Proportions Theory in International Trade. The Review of Economics and Statistics 71(1), 143–153. Bergstrand J. (1990). The Heckscher–Ohlin–Samuelson Model, the Linder Hypothesis and the Determinants of Bilateral Intra-industry Trade. Economic Journal 100(403), 1216–1229. Bergstrand, J., and P. Egger (2010). A general equilibrium theory for estimating gravity equations of bilateral FDI, final goods trade and intermediate goods trade. In The Gravity Chains 185 Gravity Model in International Trade: Advances and Applications (ed. S. Brakman and P. Van Bergeijk). New York: Cambridge University Press. Berthelon, M., and C. Freund (2008). On the conservation of distance in international trade. Journal of International Economics 75(2), 310–320. Brun, J.-F., C. Carrère, P. Guillaumont and J. de Melo (2005). Has distance died? Evidence from a panel gravity model. World Bank Economic Review 19(1), 99–120. Carrere, C. (2006). Revisiting the Effects of Regional Trade Agreements on Trade Flows with Proper Specification of the Gravity Model. European Economic Review 50(2), 223–247. Chaney, T. (2008). Distorted Gravity: The Intensive and Extensive Margins of Interna- tional Trade. American Economic Review 98(4), 1707–1772. Coe, D., A. Subramanian and N. Tamirisa (2007). The missing globalization puzzle: evidence of the declining importance of distance. IMF Staff Papers 1(54), 34–58. Deardorff, A. (1998). Determinants of Bilateral Trade: Does Gravity Work in a Neo- classical World?, In The Regionalization of the World Economy, pp. 7–32. National Bureau of Economic Research. Dreze, J. (1961). Les exportations intra-CEE, en 1958 et la position Beige. Recherches Economiques de Louvain 27, 717–738. Eaton, J., and S. Kortum (2001). Technology, Trade, and Growth: A Unified Framework. European Economic Review 45(7), 1195–1235. Egger, H., and P. Egger (2004). Outsourcing and Trade in a Spatial World. CESifo Work- ing Paper Series 1349. Ethier, W. J. (1982). The General Role of Factor Intensity in the Theorems of Interna- tional Trade. Economics Letters 10, 337–342. Evenett, S. J., and W. Keller (2002). On theories explaining the success of the gravity equation. Journal of Political Economy 110(2), 281–316. Feenstra, R. (1998). Integration of trade and disintegration of production in the global economy. Journal of Economic Perspectives 12(4), 31–50. Fujita, M., P. Krugman and A. J. Venables (2001). The Spatial Economy: Cities, Regions, and International Trade, Volume 1 (1st edition). The MIT Press. Greenaway, D, and C. Milner (1987). Intra-Industry Trade: Current Perspectives and Unresolved Issues. Weltwirtschaftliches Archiv 123(1), 39–57. Grossman, G. M., and E. Rossi-Hansberg (2008). Trading tasks: a simple theory of offshoring. American Economic Review 98(5), 1978–97. Grubel, H. G., and E. J. Lloyd (1975). Intra-industry Trade: The Theory and Measurement of International Trade in Differentiated Products. London: Macmillan. Grunwald, J., and K. Flamm (1985). The Global Factory: Foreign Assembly in Interna- tional Trade. Washington, DC: Brookings Institution. Haddad, M. (2007). Trade Integration in East Asia: The Role of China and Production Networks. World Bank Policy Research Working Paper 4160, March. Harrigan, J. (1996). Openness to trade in manufactures in the OECD. Journal of Inter- national Economics 40, 23–39. Helpman, E., and P. Krugman (1985). Market Structure and Foreign Trade. Boston, MA: MIT Press. Helpman, E., M. Melitz and Y. Rubinstein (2008). Estimating trade flows: trading part- ners and trading volumes. The Quarterly Journal of Economics 123(2), 441–487. Hummels, D., D. Rapoport and K.-M. Yi (1998). Vertical specialization and the changing nature of world trade. Federal Reserve Bank of New York Economic Policy Review June, 79–99. Jacks, D., C. Meissner and D. Novy (2008). Trade costs, 1870–2000. American Economic Review 98(2), 529–534. 186 Trade in Value Added Jones, R., and H. Kierzkowski (1990). The Role of Services in Production and Inter- national Trade: A Theoretical Framework, in The Political Economy of International Trade Jones (ed. R. Jones and A. Krueger), Chapter 3. Oxford: Blackwell. Kimura, F., Y. Takahashi and K. Hayakawa (2007). Fragmentation and parts and compo- nents trade: comparison between East Asia and Europe. The North American Journal of Economics and Finance 18(1), 23–40. Krugman P. (1980). Scale Economies, Product Differentiation, and the Pattern of Trade. The American Economic Review 70(5), 950–959. Krugman, P., and A. Venables (1996). Integration, specialisation, and adjustment. Euro- pean Economic Review 40, 959–967. Lamy, P. (2010). An urban legend about international trade. Speech, 5 June 2010, URL: http://other-news.info/index.php?p=3390. Leamer, E. E., and R. M. Stern (1970). Quantitative International Economics. Boston, MA: Allyn and Bacon. Linnemann, H. (1966). An Econometric Study of International Trade Flows. Amsterdam: North-Holland. Melitz, M. J. (2003). The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity. Econometrica 71(6), 1695–1725. Novy, D. (2010). Trade costs in the first wave of globalization. Explorations in Economic History 47(2), 127–141. Obstfeld, M., and K. Rogoff (2001). The Six Major Puzzles in International Macro- economics: Is There a Common Cause?, In NBER Macroeconomics Annual 2000, Volume 15, pages 339–412. National Bureau of Economic Research. Ottaviano, G. I. P., T. Tabuchi and J.-F. Thisse (2002). Agglomeration and trade revisited. International Economic Review 43, 409–436. Poyhonen, P. (1963). A tentative model for the volume of trade between countries. Weltwirtschaftliches Archiv 90, 93–99. Rauch, J. (1999). Networks versus markets in international trade. Journal of Interna- tional Economics 48, 7–35. Tinbergen, J. (1962). Shaping the World Economy: Suggestions for an International Economics Policy. New York: The Twentieth Century Fund. Vanek, J. (1963). Variable factor proportions and interindustry flows in the theory of international trade. Quarterly Journal of Economics 77(1), 129–142. Verdoorn, R. (1960). The Intra-Block Trade of Benelux. In Economic Consequence of the Size of Nations (ed E. Robinson). London: Macmillan. Woolridge J. (2002). Econometric Analysis Cross Section Panel. Boston, MA: MIT Press. Yi, K.-M. (2003). Can vertical specialization explain the growth of world trade? The Journal of Political Economy 111(1), 52–102. Yokota, K. (2008). Parts and components trade and production networks in East Asia: a panel gravity approach. In Vertical Specialization and Economic Integration in East Asia (ed. D. Hiratsuka and Y. Uchida), Chapter 3. Chosakenkyu-Hokokusho: IDE- JETRO. 8 Using Trade Microdata to Improve Trade in Value-Added Measures: Proof of Concept Using Turkish Data NADIM AHMAD, SÓNIA ARAÚJO, ALESSIA LO TURCO AND DANIELA MAGGIONI 1 1 INTRODUCTION The dynamics of globalisation pose new challenges for economic and pol- icy analysis. The liberalisation of trade policies and capital controls, coupled with reductions in transport, communication and information costs, has led to a significant reduction in trade costs and facilitated a reorientation of firms’ production strategies in recent decades towards increasingly fragmented pro- cesses, with each production stage assigned to the most cost-effective loca- tion: a phenomenon which has became known as international fragmentation of production (Jones and Kierzkowski 2001). 2 Vertical fragmentation of production can occur within the firm, as the firm internalises countries’ and regions’ comparative advantages and estab- lishes subsidiaries abroad. Another option is for the firm to outsource certain parts of the production process to non-affiliated companies located overseas. Whether within the boundaries of the firm or at arm’s length, the vertical fragmentation of production has changed trade patterns in a significant way. Firstly, intermediate goods and services cross borders several times as they incorporate subsequent stages of production. Miroudot et al (2009) estimate that in 2006 trade in intermediate inputs represented 56% and 73% of over- all trade flows in goods and services, respectively, and Yi (2003) noted that 1 The authors are grateful to TurkStat, Turkey’s National Statistical Institute, for granting access to their micro databases which allowed testing the methodology outlined in this chapter. 2 Several terms have been used to coin the international fragmentation of production: global value chains, international supply chains, internationally sliced up value-added chain, segmentation of production across national borders, vertical fragmentation, etc. This chapter uses these different terms interchangeably. 188 Trade in Value Added the increase in trade in intermediates is the single most important factor explaining why world trade has grown much faster than global GDP in the past three decades. Secondly, it gives rise to vertical specialisation, by which countries specialise in very specific stages of the production process. Indeed, as firms or production units located in different countries increasingly collec- tively contribute to the production of a single final product, the usefulness of the concept ‘country of origin’ has also become increasingly questionable. This phenomenon has also led many to question the meaning that can be attributed to conventional estimates of trade statistics, which record the full value of a good or service each time it crosses a border as it passes along the production chain. In other words, this is a form of multiple counting that risks exaggerating the economic importance of trade to an economy. An often cited case study is that of the Apple iPod, undertaken by Linden et al (2009), which concludes that only 10% of the price of an assembled iPod at the Chinese factory-gate is Chinese value added. The bulk of the components (around 70% of the iPod’s value at the factory-gate) are imported from Japan, with much of the rest coming from the USA and Korea. Yet the export figures for China record the full value of the final good. This is not an isolated example and applies to a wide range of goods coming from many countries, as reported by Koopman et al (2008), who estimated that, on average, foreign countries contribute 80% or more of the value added embodied in recorded Chinese exports of information and communications technology equipment. 3 It is clear that the multiple counting masks the contribution that exports make to domestic value added as well as the identification of the products a country truly has a comparative advantage in. These increasingly international production processes call for the development of measures of trade in the underlying value added embodied in a product. 4 Indeed, there are a number of areas where measuring trade in value added can bring a new perspective and is likely to impact on policy choices. Global imbalances. Accounting for trade in intermediate parts and compo- nents, and taking into account ’trade in tasks’, does not change the overall trade balance of a country with the rest of the world: it redistributes the sur- pluses and deficits across partner countries. When bilateral trade balances are measured in gross terms, the deficit with final goods producers (or the surplus of exporters of final products) is exaggerated because it incorporates 3 Similar studies include Apple’s iPhone (Xing and Detert 2010), the Boeing 787 Dream- liner (Newhouse 2007), Mattel’s Barbie doll (Tempest 1996) and Nokia’s N95 Smartphone (Ali-Yrkkö et al 2011). 4 Using input–output tables for Sweden, Isakson and Wajnblom (2011) show that the share of national value-added exports in GDP is 18 percentage points lower than the share of gross exports in GDP, which includes the value of intermediate imports (31% versus 40%, respectively). Using Trade Microdata to Improve Trade in Value-Added Measures 189 the value of foreign inputs. 5 The true imbalance is therefore also with the countries who have supplied inputs to the final producer. As pressure for rebalancing increases in the context of persistent deficits, there is a risk of protectionist responses that target countries at the end of global value chains on the basis of an inaccurate perception of the origin of trade imbalances. Market access and trade disputes. Measuring trade in value added sheds new light on today’s trade reality, where competition is not between nations, but between firms. Competitiveness in a world of global value chains means access to competitive inputs and technology. Outsourcing and offshoring of elaborate parts and components can only take place in situations where the regulatory frameworks are non-discriminatory and intellectual property is respected. The optimum tariff structure in such a situation is flat (little or no escalation) and reliable (contractual arrangements within supply chains, espe- cially between affiliated establishments, tend to be long term). WTO’s World Trade Report 2011 on preferential trade agreements (PTA) reveals that more and more PTAs are going beyond preferential tariffs, with numerous non-tariff areas of a regulatory nature being included in the agreements. According to the report, global production networks may be prompting the emergence of these ‘deep’ PTAs, as good governance on a range of regulatory areas is far more important to these networks than further reductions in already low tar- iffs (WTO 2011). Moreover, in the context of the fragmentation of production and global value chains, mercantilist-styled ‘beggar thy neighbour’ strategies can turn out to be ‘beggar thyself’ miscalculations. As mentioned earlier, domestic value added is found not only in exports but also in imports: some goods and services are intermediates, shipped abroad, whose value is returned to the domestic economy embodied in imports. As a consequence, tariffs, non-tariff barriers and trade measures, such as anti-dumping rights, are likely to impact domestic producers in addition to foreign producers. For example, a study of the Swedish National Board of Trade on the European shoe industry high- lights that shoes ‘manufactured in Asia’ incorporate between 50% and 80% of European Union value added. In 2006, anti-dumping rights were introduced by the European Commission on shoes imported from China and Vietnam. An analysis in value-added terms would have revealed that EU value added was in fact subject to the anti-dumping rights (National Board of Trade 2007). The impact of macroeconomic shocks. The 2008–9 financial crisis was char- acterised by a synchronised trade collapse in all economies. Various authors have discussed the role of global supply chains in the transmission of what was initially a shock on demand in markets affected by a credit shortage. In 5 See Maurer and Degain (2010). Koopman et al (2008) find that the domestic value added of Chinese exports is on average 60%. 190 Trade in Value Added particular, the literature has emphasised the ‘bullwhip effect’ of global value chains. 6 When there is a sudden drop in demand, firms delay orders and run down inventories with the consequence that the fall in demand is ampli- fied along the supply chain and can translate into a standstill for companies located upstream. A better understanding of value-added trade flows would provide tools for policymakers to anticipate the impact of macroeconomic shocks and adopt the right policy responses. Any analysis of the impact of trade on short-term demand is likely to be biased when looking only at gross trade flows. This was recently demonstrated in the aftermath of the natural disaster that hit Japan in March 2011. 7 Trade and employment. Several studies on the impact of trade liberalisation on labour markets try to estimate the ‘job content’ of trade. Such analysis is only relevant if one looks at the value added of trade. What the value- added figures can tell us is where exactly jobs are created. Decomposing the value of imports into the contribution of each economy (including the domes- tic one) can give an idea of who benefits from trade. The EU shoe industry example given above can be interpreted in terms of jobs. Traditional think- ing in gross terms would regard imports of shoes manufactured in China and Vietnam by EU shoe retailers as EU jobs lost and transferred to these countries. But in value-added terms, one would have to account for the EU value added, and while workers may have indeed lost their job in the EU at the assembly stage, value added based measures would have highlighted the important contribution made by those working in the research, development, design and marketing activities that exist because of trade (and the fact that this fragmented production process keeps costs low and EU companies com- petitive). When comparative advantages apply to ‘tasks’ rather than to ‘final products’, the skill composition of labour embedded in the domestic content of exports reflects the relative development level of participating countries. Industrialised countries tend to specialise in high-skill tasks, which are better paid and capture a larger share of the total value added. A WTO and IDE- JETRO study on global value chains in East Asia shows that China specialises in low-skill types of jobs. Japan, on the contrary, has been focusing in export activities intensive in medium and high-skill labour, while importing goods produced by low-skilled workers. The study also shows that the Republic of Korea was adopting a middle-of-the ground position (in 2006), but was also moving closer to the pattern found in Japan (WTO and IDE-JETRO 2011). Trade and the environment. Another area where the measurement of trade flows in value-added terms would support policymaking is in the assessment of the environmental impact of trade. For example, concerns over greenhouse 6 See Escaith et al (2011) and Lee et al (1997). 7 For an application of international IO in this case see Escaith et al (2011). Using Trade Microdata to Improve Trade in Value-Added Measures 191 gas emissions and their role in climate change have triggered research on how trade openness affects CO2 emissions. The unbundling of production and con- sumption and the international fragmentation of production require a value- added view of trade to understand where imported goods are produced (and hence where CO2 is produced as a consequence of trade). Various Organisa- tion for Economic Co-operation and Development (OECD) studies note that the relocation of industrial activities can have a significant impact on differ- ences in consumption-based and production-based measures of CO2 emis- sions (Ahmad and Wyckoff 2003; Nakano et al 2009). Trade, growth and competitiveness. Likewise, indicators of competitiveness such as ‘revealed comparative advantage’ are affected by the measurement of trade in gross terms. Going back to the iPhone example, traditional trade statistics suggest that China has a comparative advantage in produc- ing iPhones, but with value-added measures its comparative advantage is in assembly work. Having in mind growth strategies and the concerns of poli- cymakers in identifying export sectors and promoting industrial policies, the analysis of the export competitiveness of industries cannot ignore the frag- mentation of production and the role of trade in intermediates. The use of input–output (IO) tables to determine the domestic content of exports in value-added terms at the industry level is now widespread and has the great advantage of providing comprehensive estimates, as both direct and indirect imports (embedded in domestic inputs) are included in the calcula- tion of value added. However, IO tables have historically been and are typically constructed by national statistics offices as tools to determine interactions within industries of an economy, with the underlying assumption, when used as an analytical tool, that the production processes of firms within a given sector are homogeneous. However, the advent of global production processes raises questions about this assumption, especially in the context of studies that try to estimate the domestic value added embodied in exports, if the firms producing goods or services for export markets use different production pro- cesses from those firms producing the ‘same’ goods or services for domestic markets. Arguably, therefore, what is needed is an approach that motivates the development of more detailed input–output tables that adequately reflect, by design, this heterogeneity. Motivating such a development will take some time, however, particularly at a time of stretched resources within statistical offices. But other approaches that capitalise on the availability of microdata could provide the basis for simpler solutions. This chapter describes such an approach using Turkish firm-level micro- data. It provides methodological guidelines on how to compute import coeffi- cients at the level of the firm and shows how trade microdata, ie the matching of trade and business activity information at the level of the firm can refine the aggregate nature of the indicators in IO tables, by increasing their granularity. 192 Trade in Value Added Furthermore, the chapter critically assess the results of the implementation of the proposed methodology using Turkish firm-level data, kindly made avail- able by TurkStat, the Turkish National Statistical Institute. This chapter is structured as follows: Section 2 describes the concept of trade in value added (TVA) and how IO tables have been used to measure the contribution a country’s exports make to overall domestic value added. Section 3 explains the limitations of existing aggregate IO tables and the bias which can be introduced when computing trade in value-added measures. Sec- tion 4 proposes a methodology to compile trade microdata indicators that can be produced by statistical offices as standard outputs in their own right but that are also able to be plugged into IO tables. It also presents the data used in the study to test the outlined methodology and documents the main limi- tations found which are directly related with the information available from trade microdata. Section 5 comments on the main findings stemming from the integration of firm-level indicators into Turkey’s IO table, and Section 6 concludes, by proposing a research agenda. 2 TRADE IN VALUE ADDED: CONCEPT OVERVIEW In a perfect world with perfect information it would be possible to decompose each product into a value-added chain that was able to identify where the value added originated by tracing it throughout the production chain. Conceptually (ignoring taxes and subsidies for simplicity), it is possible to decompose any particular product with value V p into the value added VAp generated in country i for the production product p (directly and indirectly), such that the total value of p Vp = VAi . (8.1) i This is relatively clear and simple. However, complications can arise when aggregating up for a whole industry group or for a whole economy, as shown in the following example. Consider an economy i that produces only two products a and b for export, with product a exported to country j for further processing before being reimported into country i for use in the production of b. Let us assume that 100 units of a, with value 200, are produced and exported and then used in the production of 100 units of product c , with value 300, that are in turn used in the production of 100 units of b with value 400. Let us further assume, for simplicity, that each unit of a is produced entirely in country i; in other words, no intermediate inputs are directly or indirectly sourced from abroad. We also assume that, apart from the intermediate imports referred to above, all the value added in b is also generated in country i only. If we consider the global production chain, it is at least, in theory, possi- ble to show that the 100 units of a generated 200 units of domestic value Using Trade Microdata to Improve Trade in Value-Added Measures 193 added, and the 100 units of b generated 300 units of domestic value added (100 directly after processing the 100 units of product c , but 200 indirectly, reflecting the fact that each unit of c reflects two units of value added gener- ated in producing a, an intermediate input into c ). We know that total gross exports in economy i were equal to 600, which overstates the contribution of overall trade to the economy, but simply summing the value-added contribu- tion at the product level (the direct and indirect value added generated by a and the direct and indirect value added generated by b) will also overestimate the significance of trade in this context, as the overall value added generated in the economy through the sale of both a and b is only 300; reflecting the fact that of the 300 units of value added generated through the production of b, 200 units reflect the embodiment of product a, whose value added is separately shown under the production of a. Input–output tables are designed to measure the interrelationships between the producers of goods and services (including imports) within an economy and the users of these same goods and services (including exports). In this context they can be used to estimate the contribution that imports make in the production of any good (or service) for export. For example, if a motor car manufacturer imports certain components (eg the chassis), the direct import contribution will be the ratio of the value of the chassis to the total value of the car. And if the car manufacturer purchases other components from domestic manufacturers, who in turn use imports in their production process, those imports must be included in the car’s value. These indirect imports should be included in any statistic that attempts to measure the contribution of imports to the production of motor cars for export. The total direct and indirect imports are known as ‘embodied imports’. In an input–output framework the relationship between producers and con- sumers can be simply described as follows: g = A · g + y, (8.2) where g is an n × 1 vector of the output of n industries within an economy. A is an n × n matrix describing the interrelationships between industries (known as the Leontief matrix), where aij is the ratio of inputs from domestic industry i used in the output of industry j . y is an n × 1 vector of final demand for domestically produced goods and services, including exports. Assuming that all goods produced by any particular industry are homoge- nous, total imports embodied directly and indirectly within exports and the additional domestic activity induced by this additional production can be cal- culated thus: embodied imports = m(1 − A)−1 × e, (8.3) where m is a 1 × n vector with components mj (the ratio of imports to output in industry j ) and e is an n × 1 vector of exports by industry. In the same way, one can estimate the total indirect and direct contribution of exports to value added by replacing the import vector m above with an 194 Trade in Value Added equivalent vector that shows the ratio of value added to output (v ). So, the contribution of exports to total economy value added is equal to v(I − A)−1 × e. (8.4) At the whole-economy level this works fine, both for imports, if we accept the fact that they are measured gross, and, importantly, for value added. Returning to the example above, the approach would accurately record the 300 contribution exports made to value added. In addition, policymakers are equally interested in understanding the contribution that specific sec- tors make to the domestic content of exports, both directly and indirectly. In advanced industrialised economies, a large share of global GDP (and employ- ment) accrues to services, while international trade remains largely dominated by goods. Yet, identifying backwards linkages from those export-oriented sectors producing tradeable goods (agriculture, manufacture) allows us to map where the domestic value added was created. The break-up of domes- tic content by direct and indirect sectoral value added reveals that a large chunk of the value originates indirectly from service sectors. This breakdown is particularly important when identifying the sources of national competi- tiveness, which may rest in up-stream sectors which are not considered as exporters by traditional statistics, or measuring the employment impact of export production. An additional level of complexity arises because imports may often them- selves embody some domestic value added (reimports). This amount may be significant when economies are closely inserted in global value chains. In order to trace this value, a global input–output table is needed: a table that in effect reallocates imports and exports to intermediate consumption or final domes- tic demand (such as household and government final consumption and capital formation). Let G be a global input–output table with dimensions (nc) × (nc), where c is the number of countries and n is, as before, the number of industries. Fur- thermore, let the table be structured so that rows 1 to n reflect the industries of country 1, and rows n + 1 to 2n reflect the industries of country 2 and so k on, and vi is the direct value added produced by industry i in country k, as a share of its total output. It can be shown that the total direct and indirect domestic value added produced by industry j in country k is equal to k vi L(kn+i)(kn+j) , (8.5) where Lij is the ij th element of the global Leontief inverse (I − G)−1 . Similarly, k vi L(hn+i)(hn+j) (8.6) reflects the total value added generated in country k for unit output of indus- try j in country h, and k vi L(hn+i)(hn+j) (8.7) Using Trade Microdata to Improve Trade in Value-Added Measures 195 reflects value added generated by industry i in country k for unit output of industry j in h, providing a mechanism that shows the contributions made across different sectors of the economy. Therefore, for any given export therefore by an industry, it should be pos- sible to decompose the entire value into (i) the domestic value added generated in its production, both directly from the main producing industry and indirectly via transactions between domestic industries and via transactions between domestic and foreign industries, and (ii) the imported value added generated in producing the imports used in production (not including any part of the import value that reflects domestic value added). A global input–output table will thus allow users and policymakers to decom- pose the entire value of any good in the following way: • direct domestic value added from the final producer; • indirect domestic value added by producing industry; • indirect imported value added by produced country and industry. The ability to generate output such as this is, in itself, beneficial to policy- makers interested in the real contribution that industries make to economic growth, and indeed employment (as the flows above can be reformulated to show employment contributions), since they can be used to assess the domes- tic content of both imports and exports. Overall trade balances, however, will necessarily need to be estimated at a higher level (including all international economy linkages) to remove the double counting that occurs as goods and services criss-cross national boundaries during the production process. But the approach described above will allow more meaningful measures of overall bilateral trade balances, such as the one in a recent WTO report, according to which the US–China trade balance in 2008 would have been about 40% lower if calculated in value-added terms (Maurer and Degain 2010). 3 IMPROVING TRADE IN VALUE-ADDED MEASURES USING TRADE MICRODATA FOR EXPORTING FIRMS A number of efforts have been undertaken in recent years to estimate the value added content of trade, including in the OECD, using linked IO tables. 8 However, improved estimates using microdata could be attained. 8 There are four different recent initiatives to develop global or international IO tables: the Global Trade Analysis Project (GTAP), Asian International Input–Output Tables, OECD Input–Output Database and the World Input–Output Database (WIOD); see Ahmad et al (2011) for an overview. The first studies to estimate the value-added content of interna- tional trade under an explicit international input–output framework all rely on the GTAP database (Daudin et al 2011; Johnson and Noguera 2010; Koopman et al 2011). 196 Trade in Value Added In this context it is important to highlight some of the restrictive assump- tions inherent in the use of IO tables when used to estimate trade in value added. • Domestic sales are assumed to have the same foreign value-added content as exports. This limitation is also a direct consequence of aggre- gating information at the industry level, which can lead to biases in the estimation of the domestic value-added content of exports. If, for instance, the bulk of imported inputs are used in a sub-sector where most of the final production is destined to the domestic market and most of that industry’s exports come from another sub-sector that uses mainly domestic inputs, the foreign content of (aggregate) exports is going to be higher than it is in reality. • Indigenous firms are assumed to source inputs in a similar way as foreign-owned enterprises. As data for China show (Table 8.1), it is likely that foreign-owned firms’ are more engaged in global value chains inputs produced abroad by other parts of the foreign business group of which they are part, which will in turn result in different intensity of imported inputs in intermediate consumption between indigenous and foreign-owned firms. Against this background, it is clear that the use of IO tables that do not ade- quately differentiate between exporting firms and firms producing goods and services for domestic markets may provide an imperfect picture of the domes- tic value added embodied in a country’s exports. Although it is impractical to estimate the domestic value-added content as outlined in Section 2 at a very detailed product level, identifying three simple statistics of exporting firms and those that produce goods for domestic markets only can, as shown below, provide not insignificant improvements to the overall results: value-added-to- output ratios; import-to-intermediate-consumption ratios and share of overall output of exporting firms. 4 DATA The microdata used in this analysis are sourced from the Annual Industry and Service Statistics database (Structural Business Statistics, SBS), the Turk- ish trade register and the Annual Industrial Products Statistics database of TurkStat. Since 2003 SBS has collected information on firm incomes, input costs, employment and investment activity, at the primary four-digit NACE (Rev 1.1) sector of activity and the region of location since 2003. The survey covers the whole population of firms with more than 20 employees operating in Turkey Using Trade Microdata to Improve Trade in Value-Added Measures 197 Table 8.1: Use of imported intermediates and output breakdown by firm type in China. Imported intermediates Export Share of Share of breakdown intermediates intermediates for for Share of Share of processing normal normal processing Year Firm type exports use exports exports Other 2002 Wholly foreign 66.0 10.4 11.9 87.9 0.3 2002 Joint venture 45.3 34.2 27.8 71.0 0.8 2002 State owned 18.2 57.5 64.7 31.8 2.6 2002 Collective 27.1 54.0 70.7 28.1 2.7 2002 Private 8.1 63.2 88.4 8.7 7.6 2002 All 38.3 38.5 42.2 55.9 1.7 2003 Wholly foreign 62.4 12.4 11.8 87.9 0.4 2003 Joint Venture 40.0 38.7 29.4 69.9 1.1 2003 State owned 14.0 62.9 67.2 28.8 2.2 2003 Collective 24.0 56.4 71.2 26.4 1.8 2003 Private 14.3 59.4 78.9 15.9 6.0 2003 All 35.4 41.2 41.9 56.0 1.6 2004 Wholly foreign 60.9 13.2 12.4 87.5 0.4 2004 Joint venture 39.5 37.1 30.1 69.1 1.2 2004 State owned 12.7 68.1 66.7 29.0 1.8 2004 Collective 22.7 61.2 71.8 25.1 2.1 2004 Private 14.9 61.3 81.1 13.8 5.6 2004 All 35.1 42.3 41.6 56.3 1.5 2005 Wholly foreign 63.3 13.3 13.4 86.5 0.7 2005 Joint venture 41.0 38.6 32.0 67.0 1.0 2005 State owned 11.7 70.8 66.5 28.1 1.7 2005 Collective 21.6 64.5 70.4 26.2 1.7 2005 Private 15.4 61.1 82.1 12.0 5.8 2005 All 36.6 42.9 41.9 55.6 1.5 2006 Wholly foreign 61.9 14.9 14.6 85.3 1.1 2006 Joint venture 38.8 40.8 35.2 63.1 1.1 2006 State owned 11.0 71.4 65.8 27.1 1.5 2006 Collective 20.3 67.5 71.8 24.7 1.6 2006 Private 13.8 61.6 84.1 10.3 5.8 2006 All 35.7 43.5 43.5 53.6 1.7 Source: China’s Customs (cited in Wang (2008) and adapted by the authors). and a representative sample of firms with less than 20 employees and whose activity lies in NACE Sections C–K and M–N. 9 The second database used is the trade register which is sourced from 9 The survey excludes firms operating in the following sectors: agriculture and related activities, hunting and forestry, public administration and defence and activities of house- holds and of extra-territorial bodies. 198 Trade in Value Added customs declarations and contains information on merchandise trade only. Hence, exports do not cover services and imported intermediates cover goods only. Also excluded from imports and exports are border and coastal trade, transit, temporary trade and monetary gold transactions under US$100. Import and export flows are collected at 12-digit GTIP 10 classification. Infor- mation on the origin/destination countries of trade flows is also available. The third database used is the Annual Industrial Products Statistics database which contains information on the type and number of produced goods, their volume and value of production together with the total quantity and value of total sales from products produced within the reference year or preceding years. Product data are collected at 10-digit PRODTR level. 11 Pro- duction data are available for firms with more than 20 persons employed and which primary or secondary activity lays either in the C (Mining & Quarrying) or D (Manufacturing) sections of NACE Rev 1.1. This database, available for the period 2005–9, is used to identify the export flows of goods that the firm effectively produces (by matching the codes of the exported products to those of the products produced by the firm) and to exclude from import flows those goods which belong to its product scope (ie, the products that the firm import and that also correspond to products produced by the firm) on the assumption that these are imported goods that are sold without further processing. Merging foreign trade data and produc- tion data at the product level was achieved by establishing a correspondence between the GTIP and PROTR classifications provided by TurkStat. The databases were matched using a single identifier of each enterprise created by TurkStat. The analysis used Turkish enterprise level data for the year of 2006. The IO table for Turkey uses the latest table for 2002 sourced from the OECD’s IO database (see Appendix A on page 204). It is useful at this stage to say a few words on the computation of the indicators mentioned above. • Export intensity: this is the value of the export to output ratio. • Intermediate imports ratio: this is the value of intermediate imports divided by intermediate consumption • Exporting firms’ share of total output: share of sector or total economy production undertaken by exporting firms. The value of output is proxied by firm turnover, available in enterprise sur- veys. It equals all activity incomes plus subsides, fiscal aids and other incomes but excludes other ordinary and extraordinary revenues and profits such as 10 Turkish Customs Tariff and Tariff Classification of Goods. 11 This is national product classification with the first eight digits corresponding to Euro- stat’s Prodcom classification of 2006. Using Trade Microdata to Improve Trade in Value-Added Measures 199 interest and dividends from affiliates and subsidiaries. Also included is the annual change in the stock value of semi-finished and finished products: output = income from sales of goods and services + subsidies, fiscal aids and other incomes + ∆stock value of semi-finished products + ∆stock value of finished products. (8.8) Intermediate consumption comprises all types of expenditures necessary to undertake the economic activity of the enterprise. Hence, it excludes from financing charges (interest) and extraordinary expenses, including non- operating expenses and costs and previous years’ expenditures. 12 From the obtained value of activity expenditures is deducted the annual variation of both changes in the stock value of raw and auxiliary materials, operating and packing goods and changes in the stock value of trading goods: 13 intermediate consumption = total value of equipment, raw and auxiliary materials, operating and packing good purchased to he used in production of goods and services in the reference period + value of goods to be sold without further processing + purchase of electricity + purchase of other fuels + payments made to employment agencies and similar organisations + expenditures on auxiliary activities provided by other enterprises + payments made for production subcontracted to third parties + rental expenses − ∆stock value of raw, auxiliary materials, operating and packing goods − ∆stock value of goods purchased to be sold without further processing. (8.9) Measuring imported intermediates at the firm level has some important caveats, which are worth discussing. Firstly, only direct imports can be cap- tured in customs data. Imported inputs can embody themselves domestic value added which cannot be disentangled from the total import value. Sec- ondly, a firm can buy locally (ie via a wholesaler in the domestic market) inputs 12 Also excluded from the analysis here were advertising, accounting and marketing costs, although these should in theory be included. 13 In theory these should also be adjusted for any stock revaluations (ie holding gains/ losses). 200 Trade in Value Added Table 8.2: Merchandise trade by large economic sectors (as a percentage of total trade in 2009 or latest available year). Total exports (%) Agriculture Industry Trade Services Unspecified Total Canada 1.6 71.6 10.5 16.4 — 100.0 Turkey 0.1 59.9 34.5 5.5 — 100.0 USA 0.4 63.0 25.1 11.5 — 100.0 EU average 0.4 56.8 19.3 11.0 12.5 100.0 Total imports (%) Agriculture Industry Trade Services Unspecified Total Canada 0.5 48.5 40.8 10.3 — 100.0 Turkey 0.0 54.6 31.5 13.9 — 100.0 USA 0.1 47.0 41.6 11.3 — 100.0 EU average 0.2 37.5 38.3 12.8 11.1 100.0 Source: OECD-Eurostat Trade by Enterprise Characteristics (TEC) Database. Rounded figures, which may not sum exactly to 100. that are produced abroad. According to the OECD-Eurostat TEC database, the percentage of wholesalers and retailers in many countries is not insignificant: they account, for example, for 19% of Germany’s extra-EU exporters and 36% of exporting enterprises in the USA, where they are almost 50% of all importing enterprises as well. As shown in Table 8.2, wholesalers and retailers undertake a sizeable share of merchandise trade, which in the case of imports is above 30% of total imports, on average, for the countries covered by the database. 14 Future plans, of both the Eurostat-OECD Trade by Enterprise Characteris- tics expert group and the OECD group working on the measurement of trade in value added, will focus on allocating these imports to using these goods as inputs by separately identifying and treating imports purchased by the whole- saling industry. Research will also focus on creating links between enterprises and any affiliate enterprises they set up as separate wholesale/retail arms. In this study the microdata were based on 2006 results, whereas the IO coef- ficients and the export values were retrieved from the 2002 IO table, meaning that a full reconciliation of data was not possible. But there are other reasons why a complete reconciliation between the two sources would in any case be non-trivial. One of the reasons reflects the fact that the IO tables will, by design, include a number of corrections and adjustments to reflect reporting errors, such as incorrect reporting information from enterprises, and national accounts adjustments to reflect the non-observed (informal, grey, shadow) economy among others. But another equally important reason reflects the 14 For a review of the literature on the role of wholesalers in international trade and its determinants see Crozet et al (2010) and Bernard et al (2011). Using Trade Microdata to Improve Trade in Value-Added Measures 201 allocation of businesses to different industry sectors. The assumption used in the analysis here is that enterprises in the firm-level data are also the basis for constructing the IO tables. In theory, statistical offices are encouraged to construct IO tables using establishments. For most businesses the enterprise and the establishment is one and the same, but this is not always the case, particularly for larger enterprises. Further work will be needed to ensure a reconciliation of allocation methods used in the IO tables with those used in this additional analysis. That said, these caveats are not expected to have a significant impact on the overall results. Firms engaged in the informal sector, for example, are typically small and unlikely to be involved in international trade. And, as noted, most establishments are also enterprises. In any case, to minimise the possibility of these differences having a major impact on the overall results, the approach used here is based on ratios. In other words, in creating a split of any industrial sector into an exporting component and a non-exporting component, the approach has been to split total output in the sector in accordance with the split prevailing in the firm level data (assuming the ratios for 2006 are suitable for 2002). The next step is to create the estimates for each sector using information on the ratio of value added to output and the ratio of import to intermediate consumption for the population of exporting firms in every industry. 5 SUMMARY OF RESULTS Table 8.3 presents the key results by two-digit ISIC industry. The table com- pares the share of imported inputs in total industrial output for exporting firms only (column 3) against the aggregate industry shares (across all firms in each sector) obtained via the aggregate two-digit IO tables (column 5). It shows that estimated imports embodied in exports were, on average, 125% higher combining IO information with firm level information about import shares from exporters compared with the results based on the aggregated IO coefficients. The table also shows that, in every industrial sector where some disaggregation was attempted, the amount of imports embodied in exports was higher and the difference was significant. The only exception was post and telecommunications (IO 64), where the difference was negligible. How- ever, the latter result may be related to the low degree of tradability of that activity. 6 CONCLUSIONS AND RESEARCH AGENDA The experimental results shown above demonstrate that more detailed IO tables which have a greater focus on the structure of exporting firms than 202 Trade in Value Added Table 8.3: Comparison of results, 2002. Import shares Results based on from exporting aggregated firms IO table ISIC Exports A B A B C 1 2,244,050 0.13 299,932 0.06 145,611 106 2 10,896 0.06 662 0.03 311 113 5 65,569 0.14 9,440 0.07 4,585 106 10 1,655 0.26 438 0.12 200 118 11 1,708 0.14 240 0.08 133 81 13 114,967 0.36 40,909 0.16 18,845 117 14 251,485 0.26 64,926 0.13 32,242 101 15 2,457,225 0.29 700,829 0.12 287,108 144 16 122,341 0.56 68,994 0.23 27,552 150 17 7,018,726 0.61 4,260,200 0.27 1,872,164 128 18 8,242,291 0.54 4,447,960 0.24 1,992,294 123 19 287,732 0.63 182,681 0.37 105,676 73 20 146,946 0.70 102,729 0.28 41,604 147 21 344,823 0.63 217,567 0.30 101,801 114 22 46,795 0.44 20,644 0.23 10,562 95 23 606,703 1.11 676,060 0.57 343,648 97 24 1,545,378 0.68 1,045,659 0.28 438,663 138 25 1,327,502 0.82 1,085,012 0.33 444,134 144 26 1,819,875 0.40 729,259 0.18 325,196 124 27 3,700,493 0.79 2,932,985 0.37 1,354,991 116 28 1,116,763 0.64 718,554 0.30 333,180 116 29 2,299,397 0.59 1,348,560 0.28 636,981 112 30 10,242 0.49 5,049 0.27 2,803 80 31 1,239,762 0.73 903,317 0.31 388,618 132 ‘ISIC’ denotes ISIC Rev. 3.1 industry. Export values are given in billion TL. A, direct and indirect imports as share of output. B, value of imports embodied in exports. C, difference in imported input contents, in percent. has hitherto been the case should be pursued and developed by statistical offices. As noted above, this is unlikely to happen soon, but much can be done to motivate this development by exploiting existing microdata to pro- duce indicators that can be integrated into existing IO tables. Moreover, as demonstrated in Appendix B on page 206, the development of these indica- tors is justifiable, as they provide stand-alone inputs for many other forms of analysis. Certainly there remain a number of challenges, some of which have already been mentioned above, such as • the alignment of enterprise and establishment based data, • the inclusion of exporters of services in the exporting sector and imports of services in the calculation of total imports by the non-exporting and exporting sectors, Using Trade Microdata to Improve Trade in Value-Added Measures 203 Table 8.3: Continued. Import shares Results based on from exporting aggregated firms IO table ISIC Exports A B A B C 32 1,889,535 0.95 1,794,366 0.47 890,595 101 33 50,983 0.65 33,033 0.36 18,567 78 34 4,329,850 0.89 3,846,233 0.34 1,459,386 164 35 457,860 0.57 260,670 0.22 100,917 158 36 1,147,834 0.64 739,668 0.41 470,319 57 40 23,590 0.47 11,159 0.24 5,571 100 45 1,258,809 0.43 546,747 0.33 411,653 33 50 1,288,810 0.49 625,801 0.07 95,017 559 51 2,674,465 0.46 1,237,896 0.17 457,308 171 52 3,533,579 0.32 1,138,621 0.15 521,427 118 60 3,949,926 0.22 879,132 0.10 404,293 117 61 1,581,180 0.18 276,993 0.09 138,304 100 62 860,967 0.19 159,600 0.09 75,480 111 63 1,675,591 0.23 393,472 0.11 175,954 124 64 155,113 0.15 23,933 0.15 22,725 5 65 1,481,153 0.30 439,548 0.14 209,600 110 66 225,731 0.18 40,885 0.09 19,296 112 72 23,033 0.28 6,424 0.10 2,263 184 74 153,218 0.15 23,163 0.05 7,854 195 75 226,844 0.11 25,500 0.05 11,942 114 90 90 0.18 16 0.09 8 87 92 97,915 0.12 11,556 0.05 5,193 123 93 102 0.35 36 0.09 10 277 Total 62,109,502 0.52 32,377,056 0.23 14,412,585 125 ‘ISIC’ denotes ISIC Rev. 3.1 industry. Export values are given in billion TL. A, direct and indirect imports as share of output. B, value of imports embodied in exports. C, difference in imported input contents, in percent. • the treatment of imports purchased via non-affiliated or affiliated whole- salers. Perhaps the most pressing area where further work is necessary, however, concerns the further disaggregation of sectors into importing intensity groups and ownership (foreign or domestic) and indeed the possibility of deriving sub-sectors of these groupings (including exporters) based on intensities or other breakdowns, for example, by breaking down exporters’ quartiles based on the proportion of output they export, by size class or more detailed indus- trial classification. But all of these considerations need to be set against confi- dentiality constraints. Appendix B on page 206 provides further information on what is possible here. One other important area of work concerns the nature of importers. Input– output tables in some countries often use limiting assumptions to allocate 204 Trade in Value Added imports to using sectors. Often this is based on a straightforward propor- tionality assumption that allocates imports on the basis of their share within total supply. Some countries tackle this allocation using dedicated surveys, but these are not always conducted systematically. Capitalising on the use of existing microdata, in particular that relating to firms recognised as importers in trade registers, could lead to improvements in this allocation, particularly if this microdata is linked to information regarding the nature of the import (ie whether it is an intermediate good or one destined for final demand; see Appendix B on page 206). This activity forms part of the research agenda that takes this work forward. Ultimately the intention is for the OECD to systematically integrate these new statistics into national IO tables, in conjunction with a number of other initiatives, for example, the creation of a Broad Economic Categories (BEC) data set (Zhu et al 2011). 7 APPENDIX A. IO TABLES OECD’s Science Technology and Industry Directorate has been updating and maintaining harmonised IO tables, splitting intermediate flows into tables of domestic origin and imports, since the mid-1990s, usually following the rhythm of national releases of benchmark IO tables. The process of compiling OECD’s IO database greatly depends on cooperation with national statistical institutes. Ideally, national authorities would provide the latest supply–use tables and benchmark symmetric input–output tables (SIOTs) at the most detailed level of economic activity possible, with a basic price valuation, and, preferably, separating domestically produced and imported intermedi- ate goods and services. However, few countries can meet such requirements. Therefore, in order to maximise country coverage, all relevant partial data is used. It should be noted that one of the main reasons that IO analysis has benefited from renewed attention in recent years is the improved availability and quality of IO tables and related statistics from national sources. The first edition of the OECD IO Database dates back to 1995 and cov- ers 10 OECD countries, with IO tables spanning the period from early 1970 to early 1990. The first updated edition of this database, released in 2002, increased the coverage to 18 OECD countries, plus China and Brazil, and introduced harmonised tables for the mid-1990s. Since 2006, this tradition of growth has continued so that there are now tables available for 46 coun- tries (33 OECD and 13 non-OECD countries) with tables for the mid-2000s (mainly 2005) now available for most of them (Table 8.4). The IO tables show the transactions between domestic industries. The tables break down total imports by user (industry and category of final demand). Some countries provide the latter import tables in conjunction with their IO tables, but in some cases they are derived by the OECD Secretariat Using Trade Microdata to Improve Trade in Value-Added Measures 205 Table 8.4: Country coverage of OECD Input–Output 2009 edition (as of May 2011). OECD Mid-1990s Early 2000s Mid-2000s Australia 1994/95 1998/99 2004/05 Austria 1995 2000 2005 Belgium 1995 2000 2005 Canada 1995 2000 2005 Chile 1996 — 2003 Czech Republic 1995 2000 2005 Denmark 1995 2000 2005 Estonia 1997 2000 2005 Finland 1995 2000 2005 France 1995 2000 2005 Germany 1995 2000 2005 Greece 1995 2000 2005 Hungary 1998 2000 2005 Iceland — — — Ireland 1998 2000 2005 Israel 1995 — 2004 Italy 1995 2000 2005 Japan 1995 2000 2005 Korea 1995 2000 2005 Luxembourg 1995 2000 2005 Mexico — — 2003 Netherlands 1995 2000 2005 New Zealand 1995/96 2002/03 — Norway 1995 2000 2005 Poland 1995 2000 2005 Portugal 1995 2000 2005 Slovak Republic 1995 2000 2005 Slovenia — 2000 2005 Spain 1995 2000 2005 Sweden 1995 2000 2005 Switzerland — 2001 — Turkey 1996 1998 2002 United Kingdom 1995 2000 2005 USA 1995 2000 2005 in producing IO tables directly from supply–use tables, which requires the use of assumptions that will have a significant impact on the results of trade in value-added analysis, particularly at the industry level. The main assump- tion used is the ‘proportionality’ assumption, which assumes that the share of imports in any product consumed directly as intermediate consumption or final demand (except exports) is the same for all users. Indeed, this is also an assumption that is widely used by national statistics offices in constructing IO tables. Improving the way that imports are allocated to users will form a central part of the work-plan going forward. But an important part of the work plan will be the attempt to gain an improved understanding of how countries estimate their import-flow matrices and indeed an attempt to motivate better methods of allocation, at the national level, where possible. The industry classification used in the current version of the IO database 206 Trade in Value Added Table 8.4: Continued. Non-OECD Mid-1990s Early 2000s Mid-2000s Argentina 1997 — — Brazil 1995 2000 2005 China 1995 2000 2005 Chinese Taipei 1996 2001 2006 India 1993/94 1998/99 2006/07 Indonesia 1995 2000 2005 Romania — 2000 2005 Russia 1995 2000 — South Africa 1993 2000 2002 Thailand — — 2005 Vietnam — 2000 — Malaysia∗ 2000 Singapore∗ 1995 2000 2005 A dash means that the available year data is not available. ∗ Not published (internal use only). is based on ISIC Rev. 3 (Table 8.6), meaning that it is compatible with the other OECD industry-based analytical data sets such as the Structural Analy- sis (STAN) database, based on System of National Accounts by activity, and bilateral trade in goods by industry (derived from merchandise trade statis- tics via standard Harmonized System to ISIC conversion keys). By necessity (ie to maximise inter-country comparability), the system is relatively aggregated. 8 APPENDIX B: ANALYSIS OF FIRM-LEVEL HETEROGENEITY 8.1 Firm-Level Heterogeneity This section shows the results of the exploratory work and aims at detailing the level of within-sector heterogeneity found in the key indicators identified above and comparing the values of these indicators at specific points of the distribution with averages computed at the sector level. The analysis is pri- marily centred on establishing a level of detail that could be provided within IO tables without compromising confidentiality constraints, but very clearly the results themselves are useful in understanding firm dynamics, and even without their integration into IO tables they can prove to be powerful policy tools. Table 8.5 depicts correlations between the main variables of interest. It shows that there is a positive and highly significant correlation between the share of output exported by firms and the intermediate import ratio. Also, there is a positive correlation between the share of a firm’s exports in total sector exports (calculated at the two-digit level) and the intermediate import ratio, except for wholesalers and retailers, which raises concerns about aggre- gation bias in TVA measures. The same information is depicted in Figure 8.1, which plots the distribution of the intermediate import ratios and the share Using Trade Microdata to Improve Trade in Value-Added Measures 207 100 Import intensity – BEC (%) 80 60 40 20 0 0 20 40 60 80 100 Export share (%) Figure 8.1: Q–Q plot of intermediate import ratio against export share. Source: authors’ calculations using TurkStat’s databases. Table 8.5: Correlation table between selected indicators. Whole Wholesalers Correlations between: economy Manufacturing & retailers Export intensity and. . . • intermediate import ratio (BEC class.) 0.09 0.24 0.02 • value added per unit of output −0.03 −0.03 −0.02 • value added 0.13 0.27 0.11 • foreign ownership 0.06 0.12 0.02 • firm size 0.15 0.30 0.11 Sector export share and. . . • intermediate import ratio (BEC class.) 0.04 0.11 0.00 • value added per unit of output 0.00 −0.01 0.00 • value added 0.03 0.07 0.03 • foreign ownership 0.07 0.10 0.02 • firm size 0.05 0.09 0.04 Intermediate import ratio (BEC classif.) and. . . • value added per unit of output −0.02 −0.02 −0.01 • foreign ownership 0.10 0.22 0.07 • firm size 0.14 0.33 0.10 Source: authors’ calculations using TurkStat’s databases. All coefficients are significant at 1% unless indicated in bold. of exported output at the firm level. From the figure it emerges that higher export shares correspond to more than proportional increases in the import 208 Trade in Value Added Table 8.6: OECD IO industry classification. NACE Classification – Rev. 1.1. NACE Description 01,02&05 Agriculture, hunting and related service industries 10–12 Mining and quarrying (energy) 13&14 Mining and quarrying (non-energy) 15&16 Food products, beverages and tobacco 17–19 Textiles, textile products, leather and footwear 20 Wood and products of wood and cork 21&22 Pulp, paper, paper products, printing and publishing 23 Coke, refined petroleum products and nuclear fuel 24ex2423 Chemicals excluding pharmaceuticals 2423 Pharmaceuticals 25 Rubber & plastics products 26 Other non-metallic mineral products 271&2731 Iron & steel 272&2732 Non-ferrous metals 28 Fabricated metal products, except machinery & equipment 29 Machinery & equipment, nec 30 Office, accounting & computing machinery 31 Electrical machinery & apparatus, nec 32 Radio, television & communication equipment 33 Medical, precision & optical instruments 34 Motor vehicles, trailers & semi-trailers 351 Building and repairing of ships & boats 352–359 Railroad equipment and transport equipment nec 36&37 Manufacturing nec; recycling (including furniture) intensity measured as the share of BEC intermediates over intermediate con- sumption. 8.2 Export Shares Table 8.7 shows the distribution of sector export shares (calculated as total exports over total output) for the Turkish economy in 2006. Sectors with cells suppressed due to confidentiality are not displayed in the table. The second column reports values for export intensity calculated directly at the sector level (ie by summation of total export and total output values at the two-digit sector and then taking the ratio between the two), while the third column presents average sector values of export intensity calculated at the firm level. The next columns display values for selected the middle and upper part of the distribution, more specifically the 50th, 75th, 90th and 95th percentiles. It is clear from the table that there is a large discrepancy between the export share at the sector level reported in the second column and the average firm- level share displayed in the third column. This is easily explained by the large percentage of firms which do not export. Indeed, the initial idea was to display also values for the lower part of the distribution, but results showed what Using Trade Microdata to Improve Trade in Value-Added Measures 209 Table 8.6: Continued. NACE Description 401 Production, collection and distribution of electricity 402 Manufacture of gas; distribution of gaseous fuels through mains 403 Steam and hot water supply 41 Collection, purification & distribution of water 45 Construction 50–52 Wholesale & retail trade; repairs 55 Hotels & restaurants 60 Land transport; transport via pipelines 61 Water transport 62 Air transport 63 Supporting & auxiliary transport activities; activities of travel agencies 64 Post & telecommunications 65–67 Finance & insurance 70 Real estate activities 71 Renting of machinery & equipment 72 Computer & related activities 73 Research & Development 74 Other business activities 75 Public admin. & defence; compulsory social security 80 Education 85 Health & social work 90–93 Other community, social & personal services 95–99 Private households and extraterritorial organisations is already a stylised fact about export performance: only very few firms in the economy export (Araújo and Gonnard 2011; Ottaviano and Mayer 2008). As such, the values for the lower part of the indicators’ distributions are not displayed, as they are mostly equal to zero, except for sector 16 (manufacture of tobacco products). In the specific case of the Turkish economy, except for ‘manufacture of tobacco products’ (sector 16), and to a much lesser extent ‘mining of metal ores’ (sector 13), all the economy is characterised by the fact that almost 75% of the firms in a sector sell only to the domestic market. Not only is the export base is small, but also only a few firms within sectors have very high export intensities. Focusing on the manufacturing sector, and with the exception of the tobacco industry, the ratio of exports to output is higher than 25% only in sector 27 (manufacture of basic metals) at the 95th percentile. Export shares computed at the sector level convey a different picture: export intensity calculated at this level is typically higher than the average export intensity calculated at the firm level and higher than the 75th percentile value. The only exception is the case of the tobacco industry, where 50% of the firms export around 40% or more of their total output, while the sector aver- age is about half this figure. Conversely, aggregate export intensity is among the highest for motor vehicles, trailers and semi-trailers and other transport 210 Trade in Value Added Table 8.7: Distribution of export shares (%). Average NACE Total across 50th 75th 90th 95th Rev. 1.1 sector firms perc. perc. perc. perc. 13 25.76 13.55 0 0.43 71.25 84.59 14 21.48 5.84 0 0 30.49 47.61 15 10.08 0.68 0 0 0 0 16 23.09 40.62 39.79 78.57 85.63 93.75 17 13.89 1.43 0 0 0 5.50 18 16.66 1.35 0 0 0 0.32 19 9.22 1.86 0 0 0 5.51 20 5.47 0.18 0 0 0 0 21 6.31 1.41 0 0 2.86 9.20 22 1.82 0.21 0 0 0 0 24 10.25 2.64 0 0 5.29 14.77 25 15.60 1.57 0 0 0 8.64 26 6.12 1.17 0 0 0 1.82 27 20.03 3.48 0 0 8.13 28.15 28 11.07 0.61 0 0 0 0 29 17.92 2.20 0 0 0.76 13.58 31 22.43 2.52 0 0 0 16.13 33 9.63 2.52 0 0 2.50 19.96 34 43.98 2.82 0 0 4.97 18.24 35 26.73 2.65 0 0 6.89 15.42 36 10.14 0.97 0 0 0 0 40 0.68 1.27 0 0 0.61 3.14 45 1.41 0.10 0 0 0 0 50 0.66 0.12 0 0 0 0 51 7.49 1.99 0 0 0 2.02 52 0.44 0.10 0 0 0 0 55 1.13 0.02 0 0 0 0 60 0.20 0.02 0 0 0 0 61 3.20 0.17 0 0 0 0 63 0.59 0.03 0 0 0 0 64 0.08 0.03 0 0 0 0 71 0.01 0 0 0 0 0 72 1.75 0.10 0 0 0 0 74 0.42 0.02 0 0 0 0 80 0.01 0.01 0 0 0 0 85 0.05 0.01 0 0 0 0 90 0.10 0.28 0 0 0 0 92 0.09 0.03 0 0 0 0 Source: Source: authors’ calculations using TurkStat’s databases. Data have been made confidential for missing two-digit NACE sectors. equipment (sectors 34 and 35, respectively), while the firm-level ratio shows that at least 95% of the firms operating in these sectors export less than 20% of their output. Excluding from total exports the exports of those products that do not have a code matching the products that each firm have declared they produced Using Trade Microdata to Improve Trade in Value-Added Measures 211 Table 8.8: Distribution of intermediate import ratios (%). Average NACE Total across 50th 75th 90th 95th Rev. 1.1 sector firms perc. perc. perc. perc. (a) All imports 10 5.68 1.37 0 0 0 0.77 13 15.83 2.42 0 0 5.29 13.42 14 5.29 0.78 0 0 0 0.35 15 8.85 0.26 0 0 0 0 16 38.4 15.65 0.59 20.49 65.27 70.92 17 22.78 1.72 0 0 0 6.62 18 11.48 0.54 0 0 0 0 19 17.49 0.99 0 0 0 2.03 20 24.07 0.18 0 0 0 0 21 36.07 4.31 0 0 11.6 34.27 22 9.11 0.5 0 0 0 0 24 55.09 9.88 0 0.77 47.79 66.42 25 36.21 2.09 0 0 0 9.7 26 14.41 0.92 0 0 0 0.11 27 51.91 4.52 0 0 10.66 35.09 28 16.08 0.5 0 0 0 0 29 27.48 1.83 0 0 0 11.25 31 38.25 1.91 0 0 0 9.14 32 79.18 7.9 0 3.49 31.64 49.01 33 34.02 3.88 0 0 7.59 29.77 34 58.66 5.94 0 0 37.48 44.09 35 31.15 2.16 0 0 0.57 15.07 36 9.57 0.34 0 0 0 0 40 1.95 3.29 0 1.15 5.3 15.72 45 1.45 0.18 0 0 0 0 50 7.14 0.21 0 0 0 0 51 10.72 2.91 0 0 0 8.9 52 1.8 0.25 0 0 0 0 55 1.03 0.04 0 0 0 0 60 1.67 0.01 0 0 0 0 61 2.26 0.32 0 0 0 0 63 1.55 0.11 0 0 0 0 64 2.32 0.21 0 0 0 0 70 0.11 0 0 0 0 0 71 0.17 0.14 0 0 0 0 72 5.84 1.25 0 0 0 0.37 74 1.07 0.1 0 0 0 0 80 1.05 0.07 0 0 0 0 85 4.05 0.06 0 0 0 0 90 2.15 21.31 0 8.34 96.86 96.86 92 1.15 0.03 0 0 0 0 93 0.48 0.02 0 0 0 0 Source: authors’ calculations using TurkStat’s databases. Data have been made confidential for miss- ing two-digit NACE sectors. according to the Industrial Production Survey scales down export to output shares by 60% (not shown). One possibility which is advanced in the literature for this disparity is misreporting. The misreporting hypothesis was checked 212 Trade in Value Added Table 8.8: Continued. Average NACE Total across 50th 75th 90th 95th Rev. 1.1 sector firms perc. perc. perc. perc. (b) Only intermediate imports according to the BEC classification 10 5.23 1.07 0 0 0 0 13 13.71 1.54 0 0 2.47 7.02 14 3.68 0.32 0 0 0 0 15 5.78 0.16 0 0 0 0 16 32.2 13.91 0.59 13.94 61.68 65.26 17 19.66 1.36 0 0 0 3.8 18 9.63 0.47 0 0 0 0 19 14.4 0.77 0 0 0 0.39 20 21.48 0.16 0 0 0 0 21 30.96 3.73 0 0 11.48 33.68 22 5.01 0.12 0 0 0 0 24 42.82 7.89 0 0 35.19 54.9 25 32.72 1.69 0 0 0 2.97 26 11.51 0.65 0 0 0 0 27 50.59 3.97 0 0 6.73 29.02 28 13.57 0.36 0 0 0 0 29 15.32 1.2 0 0 0 3.21 31 29.15 1.5 0 0 0 5.24 32 37.95 5.83 0 1.92 18.00 37.38 33 19.61 2.3 0 0 2.73 12.24 34 42.98 2.18 0 0 2.17 11.25 35 19.61 1.46 0 0 0.57 10.04 36 6.31 0.23 0 0 0 0 40 1.61 2.42 0 0.75 4.01 8.81 45 0.81 0.14 0 0 0 0 50 1.22 0.08 0 0 0 0 51 7.4 1.89 0 0 0 1.24 52 0.33 0.1 0 0 0 0 55 0.22 0 0 0 0 0 60 0.52 0 0 0 0 0 61 0.67 0.14 0 0 0 0 63 0.89 0.04 0 0 0 0 64 0.47 0.05 0 0 0 0 70 0.03 0 0 0 0 0 71 0.05 0.12 0 0 0 0 72 1.48 0.29 0 0 0 0 74 0.76 0.09 0 0 0 0 80 0.17 0.01 0 0 0 0 85 0.37 0.01 0 0 0 0 90 0.92 8.79 0 4.05 40.7 40.7 92 0.6 0.01 0 0 0 0 93 0.08 0 0 0 0 0 Source: authors’ calculations using TurkStat’s databases. Data have been made confidential for miss- ing two-digit NACE sectors. by matching customs and product data at a higher level of aggregation at the CPA six-digit level instead of at the PRODCOM ten-digit level. Export shares are scaled down by a smaller amount (40% on average), but there are significant Using Trade Microdata to Improve Trade in Value-Added Measures 213 differences across sectors. However, there are substantial discrepancies in the relative sizes of the reduction of export shares within sectors. 8.3 Intermediate Imports Ratio As discussed above, for the purpose of analysis of the use of imports of inter- mediate goods, three measures of imports were constructed. However, match- ing the codes of imported products with those of the products produced by each firm did not reveal significant discrepancies between import shares, both at the sector level and at the firm level. Table 8.8 consequently displays inter- mediate import ratios according to only two criteria: the first part of the table takes all imports made by firms as imports of intermediate goods used up in the production process, while second part of the table identifies as interme- diates only those products which are so identified by the BEC classification. As with export shares, Table 8.8 only reports non-confidential cells. Table 8.8 shows that, as for exports, the import activity of firms within sec- tors is strongly heterogeneous, with a small share of firms reporting non-zero imports, regardless of the definition of intermediate imports used. Across sectors, imports tend to be more important in manufacturing sectors (corre- sponding to NACE codes 15–37). As for the specific definitions used, as expected, considering all imports as intermediate inputs yields higher intermediate import coefficients, both at the aggregate sector level and in terms of firm-level averages. These discrep- ancies are, however, higher in terms of total sector averages, particularly in sector 32 (radio, television and communication equipment apparatus), and with the exception of sector 90 (sewage and refuse disposal), where the aver- age across firms is higher than the aggregate sector value. 8.4 Firm Size, Ownership and Value Added We have further explored within-sector heterogeneity by looking at the distri- bution of export shares, the intermediate imports ratio and the ratio of value added to output by firm size and ownership status of the firm. 15 Although disaggregated tables with within-sector decompositions were also produced, the small number of foreign firms and their important role in the Turkish economy made it impossible to disclose cells for a number of sectors. Table 8.9 reports a summary of the results instead. Of the disaggregated analysis, it is worth highlighting the following. 15 The figures in this section refer to import shares calculated only on products classified as intermediates by the BEC classification. However, intermediate imports ratios do not change significantly if all imports are considered. Regarding firm size, values reported refer to employment levels calculated in terms of head counts. Results do not change substantially if head counts are replaced by full-time equivalents. 214 Trade in Value Added Table 8.9: Summary of results. Ownership Variable Domestic Foreign Export share (%) 0.34 12.13 imp_all/intermediates (%) 0.45 30.50 imp_bec/intermediates (%) 0.28 19.24 Firm size Variable 0–9 9–49 50–249 250+ Export share (%) 0.22 2.96 5.79 8.96 imp_all/intermediates (%) 0.32 3.19 8.64 15.9 imp_bec/intermediates (%) 0.19 2.08 5.95 11.28 Export intensity Variable Export intensive Non-export intensive Export share (%) 1.77 0.28 imp_all/intermediates (%) 1.53 0.42 imp_bec/intermediates (%) 1.11 0.25 Source: authors’ calculations using TurkStat’s databases. • The export share of foreign-owned enterprises is much larger than the export share of purely domestic ones, except in sectors 28 and 29 (fab- ricated metal products, machinery and equipment); on average, foreign- owned firms account for about 18% of total exports in the economy, but the sector-specific weight of foreign companies is particularly high in the automotive industry, where they account for more than half of exports. A significant share of wholesalers’ exports is also made by for- eign firms (about 40%). • Foreign wholesalers are much more import intensive than indigenous wholesalers, which is consistent with the fact that the former are heavily engaged in intra-firm trade. 16 • Export share increases with firm size, with small firms displaying export values of almost zero and large firms displaying very high export shares. • The intermediate import ratio also increases with firm size. • For the few sectors for which it is possible to disaggregate export and import shares simultaneously by size and ownership, foreign firms have higher import and export shares for firms with more than 49 employees. 16 Indigenous firms are those which are controlled by entities resident in Turkey. Using Trade Microdata to Improve Trade in Value-Added Measures 215 B 3 D A 2 C Density 1 0 0 0.2 0.4 0.6 0.8 1.0 VA/output Figure 8.2: Distribution of value added per unit of output by firm size. Source: authors’ calculations using TurkStat’s databases. Kernel = Epanechnikow. Bandwidth = 0.277. A, 1–9; B, 10–49; C, 50–249; D, 249+. • Differences in ‘import to output’ shares are larger than ‘for export’ shares, which suggests that foreign firms source a higher share of their inputs from abroad, compared with domestic firms. However, the dif- ference is less pronounced for the groups of firms having between 50 and 249 employees. • Domestic firms exhibit, on average, a ratio of value added to output which is about 90% that of foreign-owned firms. However, there are sec- tors where the average firm-level value added per unit of output is higher than that of foreign-owned enterprises by a significant amount: 21% in NACE sector 33 (manufacture of medical, precision and optical instru- ments, watches and clocks), 22% in NACE sector 17 (manufacture of textiles) and 41% in NACE sector 18 (manufacture of wearing apparel). In sectors NACE sectors 29 (manufacture of machinery and equipment, nec) and 34 (motor vehicles) domestic and foreign-owned firms exhibit a similar performance. • Value added increases with firm size. Figure 8.2 shows the distribu- tion of the value added per unit of output for firms in different size segments. It shows that a randomly drawn medium-sized or large firm (with more than 49 employees) is likely to generate a higher value added per unit of output than micro and small firms (those with up to 49 employees). The breakdown of the distribution of value added per unit of output between indigenous and foreign firms (Figure 8.3) shows that foreign-owned firms exhibit a higher share along most of the distribution and are more likely than indigenous firms to display a value-added share of output above 80%. It is quite interesting that the distribution for indigenous firms resembles the 216 Trade in Value Added 3 2 Density 1 0 0 0.2 0.4 0.6 0.8 1.0 VA/output Figure 8.3: Distribution of value added per unit of output by firm ownership. Source: authors’ calculations using TurkStat’s databases. Kernel = Epanechnikow. Bandwidth = 0.0545. , foreign firms; , domestic firms. distribution for smaller firms in Figure 8.2, while the foreign firms’ distribu- tion mimics that of larger firms. There are differences at the sector level, however. In NACE sectors 17 (man- ufacture of textile) and 18 (manufacture of wearing apparel), smaller firms (those with less than 50 employees) have a higher value added per unit of output than medium and large firms (those with 50 or more employees). 17 Within-sector differences in value added per unit of output between foreign and indigenous firms are not as striking as the differences in intermedi- ate import ratios. Indigenous firms account for the majority of value added over sector output in manufacturing, while foreign firms’ value added rep- resents an important share of value added in some service sectors, such as in NACE sectors 64 (post and telecommunications), 71 (renting of machinery and equipment) and 72 (computer and related activities). Disaggregating the ratio of imported inputs over intermediate consumption by firm ownership and firm size reveals that resourcing to foreign inputs dramatically increases with firm size. Table 8.1 shows that there is a positive and highly significant correlation between the ratio of intermediate imports to intermediate consumption and firm size. When firms are split according to their ownership status and firm export share is plot against its input import intensity (Figure 8.4), the general relation- ship found in Figure 8.1 is not verified. It is clear from Figure 8.4 that higher export shares in foreign firms correspond to lower import ratios, regardless of the overall higher import and export orientation of foreign firms. 17 Sector disaggregation is not shown to save space, but it can be provided upon request, within the limits of confidentiality. Using Trade Microdata to Improve Trade in Value-Added Measures 217 (a) 100 Import intensity – BEC (%) 80 60 40 20 0 0 20 40 60 80 100 Export share (%) (b) 100 Import intensity – BEC (%) 80 60 40 20 0 0 20 40 60 80 100 Export share (%) Figure 8.4: Q–Q plot of intermediate import ratio against export share by firm owner- ship. Source: authors’ calculations using TurkStat’s databases. (a) Foreign controlled firms; (b) domestic firms. Finally, Table 8.10 highlights differences in the mean of intermediate import ratios between exporters and non-exporters along firm size class and own- ership status: it confirms that the difference in the means of the import ratios between exporters and non exporters are significantly positive and large regardless of the size class or ownership status considered. REFERENCES Ahmad, N., and A. Wyckoff (2003). Carbon Dioxide Emissions Embodied in Inter- national Trade of Goods. OECD Science, Technology and Industry Working Paper 2003/15. http://dx.doi.org/10.1787/421482436815. Ahmad, N., H. Escaith, S. Miroudot, C. Webb and N. Yamano (2011). Trade in Value- Added: Concepts, Methodologies and Challenges, OECD/WTO Report. 218 Trade in Value Added Table 8.10: Differences in intermediate import ratios between exporters and non-exporters. (a) Total Mean Std. Err. Exporters 8.10 0.10 Non-exporters 0.18 0.00 (b) By firm size 0–9 0–49 50–249 249+ Mean Std. err. Mean Std. err. Mean Std. err. Mean Std. err. Exporters 5.23 0.12 8.61 0.17 14.41 0.32 20.75 0.65 Non-exporters 0.15 0.00 0.95 0.03 1.46 0.09 1.58 0.21 (c) By ownership Domestic Foreign owned Mean Std. err. Mean Std. err. Exporters 7.65 0.10 28.92 1.08 Non-exporters 0.18 0.00 3.68 0.64 Source: authors’ calculations using TurkStat’s databases. Ali-Yrkkö, J., P. Rouvinen, T. Seppälä and P. Ylä-Anttila (2011). Who Captures Value in Global Supply Chains? Case Nokia N95 Smartphone. Discussion Paper 1240. The Research Institute of the Finnish Economy, Helsinki. Araújo, S., and E. Gonnard (2011). Selling to Foreign Markets: a Portrait of OECD Exporters. OECD Statistics Brief 16, February. Bernard, A., J. B. Jensen, S. Redding and P. K. Schott (2011). The Empirics of Firm Heterogeneity and International Trade. NBER Working Paper 17627. Crozet, M., G. Lalanne and S. Poncet (2010). Wholesalers in International Trade. CEPII Working Paper 2010-31. CEPII, Paris. Daudin, G., C. Rifflart and D. Schweisguth (2011). Who Produces for Whom in the World Economy? Canadian Journal of Economics 44(4), 1403–1437. Escaith, H. and F. Gonguet (2011). International Supply Chains as Real Transmission Channels of Financial Shocks. Journal of Financial Transformation 31, 83–97. Escaith, H., R. Teh, A. Keck and C. Nee (2011). Japan’s Earthquake and Tsunami: International Trade and Global Supply Chain Impacts. VoxEU, April. URL: http:// www.voxeu.org/index.php?q=node/6430. Isakson, H., and E. Wajnblom (2011). Made in Sweden? A New Perspective on the Rela- tionship between Sweden’s Exports and Imports. Swedish National Board of Trade, Stockholm. Johnson R. C., and G. Noguera (2010). Accounting for Intermediates: Production Shar- ing and Trade in Value-Added, Paper presented at the World Bank Trade Workshop: The Fragmentation of Global Productions and Trade in Value-Added – Developing New Measures of Cross-Border Trade, June, Washington, DC. Using Trade Microdata to Improve Trade in Value-Added Measures 219 Jones, R., and H. Kierzkowski (2001). A Framework for Fragmentation. In Frag- mentation: New Production Patterns in the World Economy (ed. S. Arndt and H. Kierzkowski), pp. 17–34. New York: Oxford University Press. Koopman R, Z. Wang and S. Wei (2008). How Much of Chinese Exports is Really Made in China? Assessing Foreign and Domestic Value-Added in Gross Exports. Office of Economics Working Paper 2008-03-B. US International Trade Commission, Wash- ington. Koopman, R., W. Powers, Z. Wang and S.-J. Wei (2011). Give Credit to Where Credit is Due: Tracing Value Added in Global Production Chains. NBER Working Paper 16426. Lee, H., V. Padmanabhan and S. Whang (1997). The Bullwhip Effect in Supply Chains. Sloan Management Review 38(3), 93–102. Linden, G., K. L. Kraemer and J. Dedrick (2009). Who Captures Value in a Global Inno- vation Network? The Case of Apple’s iPod’. Communications of the ACM 52(3), 140– 144. Maurer, A. and C. Degain (2010). Globalization and Trade Flows: What You See Is Not What You Get! World Trade Organization Staff Working Paper ERSD-2010-12, Geneva. Miroudot, S., R. Lanz and A. Ragoussis (2009). Trade in Intermediate Goods and Ser- vices, OECD Trade Policy Papers 93. http://dx.doi.org/10.1787/5kmlcxtdlk8r-en. Newhouse, J. (2007). Boeing vs Airbus, The Inside Story of the Greatest International Competition in Business. New York: Random House. Ottaviano, G., and T. Mayer (2008). The Happy Few: The Internationalisation of Euro- pean Firms. Intereconomics: Review of European Economic Policy 43(3), 135–148. Tempest, R. (1996). Barbie and the World Economy. Los Angeles Times, September 22. Wang, Z. (2008). Domestic Value-Added in China’s Exports and Its Impact on China’s National GDP and GNP Accounting, Mimeo. WTO (2011). World Trade Report 2011: The WTO and Preferential Trade Agreements – From Co-Existence to Coherence. Report. World Trade Organization, Geneva. WTO and IDE-JETRO (2011). Trade Patterns and Global Value Chains in East Asia. Report. World Trade Organization and Institute of Developing Economies–Japan External Trade Organization, Geneva. Yi, K. M. (2003). Can Vertical Specialization Explain the Growth of World Trade? Journal of Political Economy 11(1), 52–102. Xing, Y., and N. Detert (2010). How iPhone Widens the US Trade Deficits with PRC. Discussion Paper 10-21, National Graduate Institute for Policy Studies, Tokyo. Zhu, S., N. Yamano and A. Cimper (2011). Compilation of Bilateral Trade Database by Industry and End-Use Category. Science, Technology and Industry Working Paper 6. OECD, Paris. 9 Developing International Input–Output Databases: IDE-JETRO and OECD Experiences SATOSHI INOMATA, NORIHIKO YAMANO AND BO MENG 1 INTRODUCTION In response to the recent development of spatial economics (‘New Economic Geography’), ‘new’ new trade theory and global value chains related issues, it is increasingly recognised that the concepts of ‘space’ and ‘networks’ play an important role in the analysis of economic development and globalisation. Many policymakers and researchers alike, therefore, have come to pay greater attention to the spatial aspect of the economies nowadays. In this regard, the international input–output table as an extension of inter- regional or national input–output (IO) techniques has become a significant analytical tool for the issues of current concern. The Institute of Developing Economies (IDE) has for the last 40 years been making many efforts to con- struct international IO tables in collaboration with the statistical offices and research institutes of East Asian countries. Now, facing the rapid growth of the Chinese economy and the deepening economic interdependency in the Asia- Pacific region, IDE’s Asian International Input–Output (AIO) table is an indis- pensable apparatus for the analysis of Asian economic development from a spatial perspective. On the other hand, the OECD has also been maintaining harmonised non-competitive-type IO tables since the mid-1990s. The latest version of OECD’s IO database includes 48 countries/economies (including 33 OECD members) with industry-by-industry tables covering 48 sectors (max- imum), based on the ISIC Rev. 3 classification, for the mid-1990s, early 2000 and mid-2000s. Based on this database and additional IO tables for reference years (1995, 2000 and 2005), the OECD has been developing an inter-country input–output (ICIO) model that covers 57 economies and 37 industrial sectors for the reference years. The remainder of the chapter is structured as follows. The first part (Sec- tions 2–4) introduces estimation methodologies applied at IDE-JETRO to 222 Trade in Value Added develop international input–output databases. The second part (Sections 5– 8) outlines the methods used for developing an OECD ICIO model with the main data sources, ie national IO tables and bilateral trade statistics. In both parts, we review the availability of underlying source data, summarise the assumptions made and describe the harmonisation techniques used. 2 HISTORY OF THE ASIAN INTERNATIONAL INPUT–OUTPUT TABLES 2.1 Pioneering Work Interregional IO models were pioneered by the prominent economists of the time, including Leontief (Leontief 1953; Leontief and Strout 1963), Isard (1951), Chenery (1953) and Moses (1955). The first international IO model was developed in 1961 by R. J. Wonnacott for the Canadian and US economies. At IDE, Watanabe (1964) proposed the idea of using international IO models as analytical tools for the North–South trade issue. In 1965 IDE developed an international IO model covering six regions: North America, Europe, Oceania, Latin America, Asia and Japan. In 1966 and 1971, IDE constructed interna- tional IO models for ten Asian countries. Yet the IO tables compiled in these studies were subject to a number of limitations, eg estimation techniques were too simple, the industrial sector classification was too crude. These problems notwithstanding, the models were credited with facilitating empirical analy- ses of structural relationships between developed countries and developing countries. In addition to such research-oriented projects, IDE’s IO tables were also used for evaluating the credibility and preciseness of statistical materials of developing countries. From the basic premise of data coherency between demand and supply sectors, the misspecification of estimates by local stat- istical agencies can be logically inferred if any deviation or inconsistency is observed in the table. The ‘targets’ to be scrutinised ranged from production data to consumption data. 2.2 First Phase (1973–77) In 1973 IDE decided to launch development of a comprehensive international IO table to explore the situation of interindustrial transactions among East Asian countries. The Republic of Korea and the five Association of Southeast Asian Nations (ASEAN) countries plus the USA were chosen to be included, as these countries have close economic relationships with Japan. Had all the countries compiled their national IO tables for the same referential year, the project would not have been so laborious. However, Indonesia, Thailand and Singapore had not constructed any IO tables by that time. Also, IDE was not sufficiently experienced in compiling a comprehensive IO table. Thus, the project had to begin with two preliminaries: one was to construct national IO Developing International Input–Output Databases 223 tables for these three countries; the other was to compile bilateral IO tables for the countries in which the national tables were already available. Under this project, three national IO tables (Indonesia for 1971, Singapore for 1973 and Thailand for 1975) and three bilateral IO tables (Korea–Japan for 1970, USA–Japan for 1970 and Philippines–Japan for 1970) were constructed in col- laboration with the national statistical offices and research institutes of the countries concerned. 2.3 Second Phase (1978–82) In 1978, IDE started the second phase of the IO project, with the aim of con- structing the 1975 multilateral IO table for the ASEAN countries, Japan, Korea and the USA. This project proceeded via the following three steps: 1. estimation of national IO tables for the countries that did not have 1975 national tables, 2. construction of 1975 bilateral IO tables for the countries that had already compiled national tables by the time, and 3. construction of the 1975 multilateral IO table. First, existing tables had to be updated to the year 1975 for Malaysia, the Philippines, Singapore and the USA. Next, the 1975 bilateral IO tables for Indonesia–Japan, for Thailand–Japan, and for Korea–Japan were constructed. Finally, these national and bilateral tables were linked together under a single multilateral IO table for 1975, which was completed in 1983. The 1975 mul- tilateral table was used for various analyses of East Asian industrial struc- ture, and the table became the prototype for the subsequent international IO projects. 2.4 Third Phase (1988–) After completion of the 1975 international IO tables, IDE soon launched a new project for construction of the 1985 international IO tables, to cover more Asia-Pacific countries. Since China commenced an Open-Door policy as one of its key development strategies it has rapidly increased its external trade with the USA, Japan, and others. China now plays an important role in the Asia-Pacific region, not only in providing a gigantic market but also in receiving investment from the neighbouring countries. Thus, China and Taiwan (Chinese Taipei) were covered in the 1985 multilateral table, making it even more comprehensive than the previous 1975 version. Since then, IDE has successfully completed the multilateral tables every five years, providing powerful analytical tools for dynamic structural changes in the Asia-Pacific region. 1 1 The 2005 AIO table and 2005 BRICs international IO table will be released by IDE-JETRO in 2013. 224 Trade in Value Added 3 COMPILATION PROCEDURE AND METHODOLOGY OF THE ASIAN INTERNATIONAL IO TABLE Compilation of international IO (IIO) tables is an artistic practice. A number of statistical experts from various countries are involved, exchanging consid- erable amounts of valuable information and technical expertise. 2 Roughly speaking, the compilation process goes through three distinctive phases: 1. adjustment of presentation format; 2. preparation of sector concordance and supplementary data; 3. linking and balancing. What follows is a step-by-step illustration of how the Asian International IO Table is compiled. The first subsection presents a description of the format adjustment for every constituent national table based on the general survey on national tables, which was conducted by IDE in 2003–4 in order to establish a common rule for the format adjustment of the tables. The second subsection briefly explains construction of the system of sector concordance, followed by a brief introduction of estimation methods for supplementary data. Finally, the linking procedure is illustrated, with detailed explanations of the manual balancing/reconciliation work. 3.1 Adjustment of Presentation Format Despite the fact that IO tables constitute the central apparatus of the System of National Accounts, each national table of an individual country exhibits more or less different features and characteristics, reflecting the country’s economic idiosyncrasies and availability of data. Such a variety in form, how- ever, poses a practical difficulty when compiling international IO tables (see Table 9.1), for even though the international table is composed of the seg- ments taken from each national IO table, the interpretation of the data should be mutually consistent and comparable for any part of the whole. Accordingly, one of the most complicated, nerve-racking tasks of compila- tion is the adjustment of national tables to conform to a common format. In general, it is the detailed, information-rich table that has to compromise with less-detailed ones, as the other way round would require the costly (yet often unrewarding) effort of obtaining supporting data. Therefore, there always exists a trade-off between the level of uniformity and the level of informa- tion, and hence careful and thorough consideration is called for in making adjustment rules. 2 IDE’s IIO projects involve many participating organisations from different economies. Most of these organisations are governmental or quasi-governmental statistical institutes. More than 70 experts from 10 different economies have contributed to the compilation of the 2000 AIIO table. Table 9.1: Different features and characteristics in national IO tables across the AIO target economies. China Indonesia Japan Korea Malaysia Taiwan Philippines Singapore Thailand USA 1. Conversion of valuation 1.1 Basic price to producer’s price ✓ 1.2 Private consumption expenditure ✓ ✓ ✓ 1.3 Export vectors ✓ ✓ 1.4 Import matrix/vector ✓ ✓ ✓ ✓ ✓ 2. Negative entries ✓ 3. Dummy sectors ✓ ✓ ✓ ✓ ✓ ✓ 4. Machine-repair ✓ ✓ ✓ ✓ 5. Financial intermediaries ✓ ✓ ✓ ✓ 6. Special treatment of import/export 6.1 Water transportation ✓ 6.2 ‘Pure import’ of gold ✓ 6.3 Re-export ✓ 6.4 Telecommunication ✓ 7. Computer software products ✓ 8. Producers of government services ✓ ✓ Developing International Input–Output Databases 225 226 Trade in Value Added This section reports on the general survey on the characteristic features of national tables of AIO member economies. The survey was conducted in the period 2003–4, in order to construct the basic information reserves for design- ing the AIO common format and adjustment rules. To our knowledge, such an extensive and detailed survey on national tables has never been carried out, and we believe that no institution but the IDE, with a history of significant cooperative relationships with IO experts of various Asian economies, would be able to make such a substantial survey possible and successful. Questionnaire and the Survey Result In the survey, a questionnaire was carefully designed so as to capture every important aspect of an IO table. The questions are grouped under seven broad categories, namely: 1. benchmark year and recording principles; 2. availability of national tables and supporting tables; 3. valuation; 4. form and coverage; 5. special treatment; 6. public/semi-public sectors; 7. response to the 1993 SNA. Major Findings Based on the results of general survey, the major findings can be summarised as follows. Similarity to the Japanese IO table. In Figure 9.1, which is based on the infor- mation in Tables 9.2 and 9.3, the degree of similarity to the Japanese IO table is illustrated. The horizontal axis is the level (number) of industrial classifica- tion, while the vertical axis is concerned with presentation format, giving the percentage rates of the number of questions in the questionnaire to which the country gave the same answers as Japan’s. (The rates are calculated against the sum of valid answers only.) The diagram shows that the most similar table of all is the Korean IO table (the first group), as its industrial classification has just one sector difference with that of the Japanese table, and the rate of the same answer is more than 70%. Then, we can identify the second group, including Indonesia, the Philip- pines, Thailand and Malaysia. Not to mention about the Korean table, it is no wonder that these tables (except Malaysia) show a high degree of similarity to the Japanese table, since their national IO projects are known to have been initiated and conducted under the advice and support of Japanese IO experts. The US table is indicated as having some degree of similarity, but in the survey result it is observed that many answers remain ‘unknown’, so that no Developing International Input–Output Databases 227 0.75 0.70 Korea 0.65 Indonesia Philippines Presentation format 0.60 Malaysia Thailand 0.55 USA 0.50 0.45 Taiwan 0.40 China Singapore 0.35 0 100 200 300 400 500 Industrial classification Japanese IO table column = 405 Figure 9.1: Similarity to the Japanese IO table. Table 9.2: Similarity in the presentation format. Rank Country Rate∗ Classification 1 Korea 0.7077 404 2 Indonesia 0.6462 175 3 Philippines 0.6269 229 4 Thailand 0.5606 180 5 USA 0.5536 491 6 Malaysia 0.5522 106 7 Taiwan 0.4462 160 8 China 0.4063 124 9 Singapore 0.4032 155 ∗ The percentage rates of the number of questions in the questionnaire to which the country gave the same answers as Japan’s. conclusive evaluation can be made against this table (although it is true that the classification difference is the second smallest after the Korean table). The third group, which is least similar to the Japanese table, includes Tai- wan (Chinese Taipei), Singapore and China. In addition to the dissimilarity of the format and of the level of industrial classification, the benchmark years of these national tables differ from that of Japan, ie with ‘0’ or ‘5’ in the last digit 228 Trade in Value Added Table 9.3: Similarity in the industrial classification number. Difference in the number of Rank Country industrial sectors Classification 1 Korea 1 404 2 USA 86 491 3 Philippines 176 229 4 Thailand 225 180 5 Indonesia 230 175 6 Taiwan 245 160 7 Singapore 250 155 8 China 281 124 9 Malaysia 299 106 Table 9.4: Responsiveness to the 1993 SNA. Rank Country Rate∗ 1 Philippines 0.5714 1 USA 0.5714 3 Thailand 0.5385 4 Korea 0.5000 4 Japan 0.5000 6 Singapore 0.4545 7 Indonesia 0.4286 7 Malaysia 0.4286 9 China 0.3077 10 Taiwan 0.2143 ∗ The percentage rates of the number of questions in Section 7 of the questionnaire to which the country gave the answer that follows the SNA recommendation. of the year. So the official tables had to be updated to the year 2000 with the help of some estimation methods, such as the RAS algorithm, which further decreased the accuracy of the tables. The same is true for the national table of the USA. The responsiveness to the 1993 SNA. The System of National Accounts is a comprehensive guideline for compiling national statistical data. If properly followed, the resulting statistics will be mutually consistent and internation- ally comparable. The latest version of the SNA, the 1993 SNA, underwent an extensive revision of its predecessor, the 1968 SNA, to bring the statistical notions and methods up to date. IO tables (or, more precisely, supply and use tables), which constitute a core apparatus of the system, did not remain unaffected, and many countries, including our project partners, have made every effort to make their tables accordant to the new scheme. Developing International Input–Output Databases 229 The survey result (Table 9.4) shows that the most ‘responsive’ countries are the Philippines and the USA. Yet again one must be careful about the result on the US table as it contains a number of ‘unknowns’. The Thai IO table comes next, followed by the Korean and Japanese tables. Although the Korean table and Japanese table ranked the same, the former can be evaluated higher, as it had already succeeded in introducing one of the most challenging schemes in the 1993 SNA, ie the Financial Intermediary Services Indirectly Measured (FISIM). On the other hand, it is rather surprising to observe that Singapore and Malaysia had low rankings, as these national tables are known to have followed the previous, 1967 SNA schemes quite extensively. The areas of conflict. Finally, we briefly examine the areas of conflict, where each country’s treatment is not in line. The most prominent example is the treatment of scraps and by-products. There are normally four adjustment methods for this problem. Each of them has both advantages and disadvan- tages, and the member countries employed the various schemes in quite an uncoordinated fashion. In the absence of supplementary information on the generation and use of scraps/by-products, it is not possible to convert from one scheme to another, making it difficult to reach a common agreement on the adjustment method. The second area of conflict is about the treatment of imputed interest. The previous 1968 SNA recommended that the output of imputed interests (= the difference between the interests receivable and the interests payable) should all go to intermediate transaction, not to final demand. The countries like Japan, Singapore and Malaysia strictly follow this stipulation, while other countries’ tables have output in final demand as well. The introduction of FISIM under the 1993 SNA may provide an integrated guideline for this issue, but so far no member country except Korea is successful in introducing this new scheme. The last prominent area of conflict is the treatment of inventory. The related question in the questionnaire is: Suppose that a car industry (demand-side sector) purchased a set of tyres (supply-side sector) but did not use them this time. How does this input enter in the table? Most of the countries answered that the input should be recorded at the inter- section between ‘tyre’ (supply-side) industry and ‘change in stocks’, but some countries like China, Taiwan (Chinese Taipei) and Singapore answered the opposite, ie at the intersection between ‘tyre’ (supply-side) industry and ‘car’ (demand-side) industry. Singapore gave an explanatory comment on this. It treated this input as a stock of car since ‘tyres are regarded as a work-in- progress of a car’. It is quite surprising to find out that even the very basic economic concept like an ‘inventory’ in fact yields different interpretations among countries. 230 Trade in Value Added 3.2 Preparation of Sector-Concordance and Supplementary Data The Table of Industrial Sector Concordance Each national table has its own industrial classification. In the case of the benchmark tables for the 2000 AIO table, the number of industrial sectors ranges from 98 for the Malaysian table to 517 (row) for the Japanese table. The weight of the industrial category also differs. The countries with large agro-based economies have relatively detailed classification of agricultural sectors, while industrialised economies give more comprehensive coverage to manufacturing sectors. Thus, the sector classification reflects the characteris- tics of the economy concerned, and a precise conversion system that bridges national codes and AIO codes is absolutely essential for the compilation of consistent international IO tables. The system of sector concordance has a treelike image, where AIO classifica- tion (the broadest category) rests on the top, and each AIO code corresponds to one or several national codes. The national codes are subclassified into the Harmonized System of Foreign Trade Statistics (HS), which may be further converted to SITC, another classification system for the trade data. If the concordance system has such a clear-cut tree structure, the aggrega- tion of national tables into AIO classification poses no difficulty. The problem arises when a national code is associated with more than two AIO codes. For example, Singapore’s national code SIO092 ‘Land transport equipment’ corre- sponds to both AIO055 ‘Motor vehicles’ and AIO056 ‘Motorcycles’. Here, the sector splitting of the national IO table is called for before the aggregation procedure. Supplementary Data For the compilation of international tables, the following supplementary data should be prepared by each country at AIO sector classification: 1. import data by commodity and by 11 countries of origin; 3 2. export data by commodity and by 11 countries of destination; 3. import duties and import commodity taxes by commodity; 4. domestic trade and transportation margins (TTM) and domestic freight transport costs on exported goods by commodity; 5. international freight and insurance, by commodity and by 11 countries of origin; 6. other relevant information, such as the distribution ratios of imported goods. 3 The 11 countries are the project member countries plus Hong Kong, EU, the Rest of the World. Developing International Input–Output Databases 231 The import and export data can be directly constructed from the Foreign Trade Statistics, with the help of the HS (or SITC)–national IO–AIO sector con- cordance. The data on import duties and import commodity taxes, on the other hand, are independently presented in the original national IO tables in most cases, but if not (as in the case of the US table), they must be also collected from the Foreign Trade Statistics. The data of TTM on export comes from the supporting tables of the national IO tables. Ideally, those levied on exported goods (for the delivery from facto- ries to ports) should be used, but if they are not available from the table the average figures of the TTM matrices can be used as proxies. Finally, the data on international freight and insurance are collected from the Foreign Trade Statistics, where available. Yet, because not all countries have these data, it is necessary to apply some estimation methods to make up for the missing information. As illustrated below, this is done in two steps: the first step is to obtain the parameter values by creating transport-cost equations for each AIO sector, using the available data; the second step is to project the missing values based on the parameter estimates. In most of the empirical literature on international trade that uses gravity equations, it is a common exercise to use the distance between countries as a proxy for transport costs, owing to the limited availability of direct transport- cost data (see, for example, Anderson 1979). This treatment assumes that the transport cost is a function of geographic distance: Cijk = f (Dij ) (9.1) represents transport costs for country i’s imports from country for sector k, and Dij is the distance between them. The rationale for using distance is that, for a given mode of transport, the greater the distance, the more time and energy are consumed, and hence the transport cost rises. Based on this convention, the following simple variation of transport-cost equations is cre- ated: 4 Cijk = αk + βk Dij + εijk . (9.2) The data for international freight and insurance rates (Cijk ) are available for nine countries (China, Indonesia, Japan, Korea, Malaysia, the Philippines, Sin- gapore, Thailand and the USA), but the quality of data varies across countries, and data for many transactions are missing. For Taiwan (Chinese Taipei), no information on international shipping costs is available. As the distance variable (Dij ), two measures of distance are calculated, ie the shipping-route distance and the straight-line distance. The shipping-route distance is taken from Japan Shipping Exchange (1983), in which the distances 4 Several studies have investigated the appropriateness of the relationship between transport costs and the distance (see Geraci and Prewo 1977; Limao and Venables 2001). However, in our estimation only distance was used as the explanatory variable, owing to data constraints. 232 Trade in Value Added between major ports are reported. The straight-line distance, which can be regarded as an analogue of the air-flight distance, is calculated between com- mercial centres of the countries concerned. Of these two measures, the one that better explains variation in the international freight and insurance rates is employed for projection. By running regressions of Equation (9.2), the parameter estimates α ˆ k and ˆk for each AIO sector are obtained. In cases in which the estimates for βk are β negative, they are replaced by estimates obtained from regressions in more aggregated classifications, ie twenty-four sectors or seven sectors. If the esti- mates in aggregated classifications are still negative, positive estimates for related industries are used for projection (eg estimates for 050 ‘Electronic computing equipment’ are used in lieu of those for 051 ‘Semiconductors and integrated circuits’). Using the parameter estimates α ˆk , projection of the missing values ˆ k and β for international freight and insurance rates (Cijk ) can be done by stacking the distance measures between countries concerned (Dij ) into the transport-cost equation: Cˆijk = α ˆk Dij . ˆk + β (9.3) In addition, the quality of import matrices plays a critical role in determining the accuracy of the international IO table. In order to increase the accuracy of import matrices, a special survey on imported commodities have been done in the current AIO project. The main purposes of the survey are: 1. to identify using industries of the imported commodities by country of origin; 2. to determine the value/rate of the international freight and insurance on each imported commodity; 3. to determine the value/rate of import duties and commodity taxes levied on each import commodity. The respondents of the survey will be the establishments that import the commodities (manufacturers, trading firms, etc ), as they are considered to possess the information on amount imported by country of origin and their distribution amount to domestic industries. The survey is basically carried out as an independent sample survey. Alter- natively, it may be conducted as a rider survey attached to other official sur- veys (which is more efficient and comprehensive). (The sample form of the questionnaire in order to collect the information described above is presented in Figure 9.2). Several problems arise in carrying out the special survey described above. First is the feasibility of the survey. It is difficult for some countries to con- duct the survey, owing to lack of resources (funds, personnel, connections with related authorities and firms, knowledge, etc ). For countries where the Developing International Input–Output Databases 233 Figure 9.2: Sample format of questionnaire. Source: IDE (2008). survey is infeasible, alternative solutions should be sought. One possible alter- native is to modify the import matrices by referring to other countries’ survey results. Second is the sampling issue. Even if the survey can be carried out, it is not easy to collect reliable information. For instance, the samples should be selected in order to represent the characteristics of the industry appropri- ately. However, identifying the typical samples that appropriately reflect the distribution structures is not easy. Third, it may also be difficult to determine the distribution structure, even if samples are chosen appropriately. This problem has two different aspects. The first is the difficulty of determining the final users of imported commodi- ties by country of origin. As discussed above, the imported commodities are usually delivered to the final users through wholesale and retail agents. The respondent to the questionnaire, the importing firm, may not have informa- tion on the final users if they sold their imports to domestic wholesalers or retailers. The second aspect of the problem is that it may be difficult to deter- mine the amount of each imported commodity sold, even though the final users can be identified. This may occur if the survey year is different from the reference year, or the respondent may not be able to trace the transaction records, as they may not keep detailed information. 3.3 Linking and Balancing An international IO table is not just a patchwork of the pieces taken from national tables, but a product of careful utilisation of supplementary data and manual reconciliation/fine-tuning. This section gives a brief description of the linking and balancing work for compiling the AIO table. 234 Trade in Value Added Figure 9.3: Layout of the AIO table. Step 1: So far, all the parts except the highlighted segments have been prepared and are ready for linking. The remaining parts are in fact directly transplanted from the corresponding parts of national tables, after due aggregation into AIO classification. The diagram shows an example of Korea’s case, with arrows indicating the parts’ correspondence between the AIO table and the Korean IO table. All the other member countries should be treated similarly. Step 2: After linking, all the row-wise statistical discrepancies due to the difference in data sources are dumped into a single column vector, QX. (Note that the export vectors to the member countries are not used in the end, to avoid double counting with the corresponding import matrices.) Linking of National Tables All the parts of each member country prepared in the previous steps are linked together in one big table, as shown in Figure 9.3. Figure 9.4 illustrates the process of linking. In this example, the linking of IO tables for countries 1 and 2 is illustrated. As shown in this example, the basic idea of linking is to replace the export vector by the import matrix of the trading partner. At this stage, the valuation of imports in each country’s national IO table is also converted from the cost, insurance and freight (CIF) price to the producer’s price by using the data of international freight and insurance, and domestic transport costs and trade margins compiled in the previous steps. Reconciliation of Data The final step of compilation is the manual balancing and reconciliation work, following the linking of all the parts provided so far. The table is balanced with respect to the input composition, but total demand is not necessarily consistent with total supply for each country at this stage. Such an imbalance stems from the following facts. Developing International Input–Output Databases 235 Figure 9.4: Linking of national IO tables (two-country case). Source: IDE (2011). Here, let us consider the case of Korea. As explained in the previous sec- tion, the blocks AKK , F KK and LKZ (Z = H, O, W ) in Figure 9.3 are calculated from Korea’s input–output table, and they should conform to the transac- tions recorded in the Korean input–output table. However, the other blocks, AKZ and F KK (Z ≠ K), are estimated from the import matrices of other coun- tries, and there is no guarantee that they will be consistent with Korea’s export figures. For example, for the blocks AKM and F KM , at which Korea’s rows and Malaysia’s columns intersect, if the export and import data are to be consis- tent, the following equation must hold true: KM Di = AKM ij + KM Fik − LKM i = 0, (9.4) j k KM where Di represents the difference between Malaysia’s import data and Korea’s export data for the ith industry, the subscripts j and k respectively denote the j th industry and kth final demand, and LKM i represents the exports of Korea’s ith industry to Malaysia (expressed in producer’s prices). In prac- tice, whether or not Equation (9.4) holds true depends on the reliability of the international trade statistics for the two countries and the difference between the import and export figures in the national IO table and international trade KM statistics. As stated above, the results of our linking work show that Di ≠ 0. Of course, the same imbalance occurs with all the other member countries of KM the project. Therefore, we consider that Di denotes the discrepancies in international trade statistics of the two countries, as well as to include the margins of error in estimating blocks AKM and F KM . 236 Trade in Value Added Start (a) Total check by published data sources (b) Compute the CT’s error rates (c) CT’s error rate > specified criterion End Y (d) Calculate the error in trade statistics (e) Identify the reasons for the errors (f) Make adjustment policy/instruction and adjustment card (g) Run adjustment program Figure 9.5: Adjustment procedure. Source: IDE (2006). CT: the figure of output by sector is used as the Control Total in our adjustment procedure. CT error row-wise > 30% CT error row-wise > 5% CT error row-wise > 10% CT error row-wise > 1% 70 60 50 Number 40 30 20 10 0 China Indonesia Japan Korea Malaysia Taiwan Philippines Singapore Thailand USA Figure 9.6: Distribution of CT error. Source: IDE (2006). Developing International Input–Output Databases 237 In order to rationally and efficiently decrease the discrepancies generated through the linking process, the procedure shown in Figure 9.5 is employed in final reconciliation of the AIO table. (a) Initially, we use the linking results to summarise the transactions among the industries of all countries and compile an AIO table that there is only one sector per country. Then it becomes easy to check whether or not the present data in the AIO table at the national level are consistent with the published data sources, such as the GDP statistics for the country, or the IMF statistics. Through the above checking, we gain knowledge of the preliminary linking results. (b) For determining the size of the final adjustment in detail, we calculate the error rates of CT row-wise by sector for each country. Figure 9.6 shows the distribution of the summarised absolute CT error rates for different levels. The vertical axis represents the number of sectors in which CT errors are larger than the specified levels. Obviously, China, Japan and the USA. have relatively smaller numbers, which are counted in each level. On the other hand, Indonesia, Malaysia, the Philippines, Singapore and Thailand have relatively larger numbers. Korea and Tai- wan (Chinese Taipei) exhibit a similar pattern. The distribution shown in Figure 9.6 not only depends on the economic scale but also relates to the statistic system of each country. Considering the large scale of the AIO table and the distribution pattern of error rates, any sector that has a CT error rate over 5% is determined as a target for preliminary adjustment. (c) Though 5% is determined as the criterion for the preliminary adjust- ment, considering that positive errors may offset some negative errors in the row sector, we have to investigate the structure of the error row- wise. As stated in the previous section, the AIO table is based on the import matrices for each country, and the matrices conform to import statistics, but the export statistics are not necessary consistent. In order to discuss the structure of the error in detail, for example, in the case of Korea, we calculate the matrix KM Di = AKZ ij + KZ Fik − LKZ i , j k which represents the difference between country Z ’s imports from Korea and Korea’s exports to country Z for the ith industry. If one refers to this matrix, the structure of Korea’s CT error row-wise becomes easy to understand, and it offers us information about which sectors and which countries should be the main targets for adjustment. (d) The discrepancy is mainly caused by the following three factors. (i) The inconsistency between each country’s sector classifications: though each country is required to make its own code concordance 238 Trade in Value Added from HS code to AIO sector classification, the possibility of differ- ences in statistical concept still exists. (ii) Entrepôt trade is counted in different ways by the trade partners. For example, in the case of China, export via Hong Kong to the USA may be counted by the USA as import from China. In the case of Sin- gapore, where international trade is extremely large compared with the scale of its economy, and there is a large volume of entrepôt trade, there are especially large statistical discrepancies in its inter- national trade matrices. (iii) Other statistical reasons. (e) According to the analysis of ‘matrix D ’ introduced above and careful investigation of the HS–AIO code concordance, most errors can be spec- ified. Then the adjustment policy will be determined. In our project, since the portion for each country has a professional in charge, that person will give instructions to other staff based on the adjustment pol- icy. Then the staff member who is in charge of a country will aggregate all the instructions coming from those who are in charge of other coun- tries into the adjustment card for his or her country. (f) The adjustment cards are used as input files in the adjustment program. Basically, the adjustment is merely executed on the import matrices, and it moves the same amount vertically from one sector to another. This means that CT balance will be maintained columnwise. 5 The above procedure (a)–(f) will be repeated until the results satisfy the specified criteria. Additionally, a spot check is conducted at the end of the adjustment. This is to ‘spot out’ any unnatural entries in the table that might have been brought in during the course of the adjustment. For example, the output of ‘Electricity, Gas and Water Supply’ or other service sector is not supposed to enter any cells along ‘Fixed Capital Formation’ or ‘Change in Stocks’. Any of such mistabulation should be cleared and dealt with properly. It is extremely rare for the international trade statistics of different coun- tries to be consistent with one another. There are usually rather large gaps and errors. While a number of existing studies have analysed the extent and nature of this problem, a standardised methodology for reconciling the international trade statistics of various countries has not yet been established. Even though in our project we require each country to make a code concordance between the AIO’s sector classification and HS code, it is extremely difficult to eliminate the discrepancies completely, because of the large number of codes involved and differences among statistical systems from one country to another. 5 Basically, the remaining CT error row-wise will be moved to the vector QX (Statistical Discrepancy). Developing International Input–Output Databases 239 4 FUTURE CHALLENGES FOR THE ASIAN INTERNATIONAL IO TABLES PROJECT Given the increasing economic interdependence between countries caused by the extension of globalisation and regional integration, international IO tables are considered a very useful data source for the analyses of production net- works, international fragmentation production, global value chains and so on. In response to the increasing attention and requirements from many policy- Changes to this paragraph and makers and researchers, there are a lot of challenges ahead for the project. section heading OK? The first challenge is about the time lag of publication. IDE compiles the AIO tables every five years. However, there is always more than five-year time lag between the benchmark year and reference year. Since most countries construct their national IO table every five years, and also the benchmark years across countries are different, this makes it difficult to speed up the process of linking every country’s data together in time. If the statistic system in many more Asian countries can switch to or follow the Supply and Use Tables (SUTs), national IO tables can be estimated easily. This will help the compilation of international IO table to become speedy. 6 The second challenge is about how to minimise the discrepancy arising from the linking process. As mentioned in the previous sections, the most important reasons for the discrepancy are 1. the inconsistency of export/import figures between national IO table and international trade statistics, 2. the mirror problem in bilateral trade statistics caused by the treatment re-export and reimport, 3. the different treatment of valuation between export statistics (free on board, FOB) and import statistics (CIF). One possible solution to the above problems is to apply the recent UN Broad Economic Categories (BEC) classification to the current trade statistics. Under this classification, trade data can be grouped into different end-use cate- gories, such as intermediate goods, final consumption goods, capital goods. This can improve the precision of allocating bilateral trade data when linking the national IO tables. In addition, according to the new recommendations for International Merchandise Trade Statistics (IMTS) proposed by the United Nations Statistics Division (UNSD), import figures on the FOB basis in addition to the standard CIF valuation are expected to be published in the near future. This may help us make bilateral trade data much more consistent. Finally, the re-export statistics by country of origin and destination should play an important role in solving the mirror problem that occurs in trade statistics. 6 For example, in recent years, many more Asian economies have been considering estab- lishing or improving their SUT systems under an international joint project supported by Asian Development Bank (ADB); see http://www.adb.org/data/icp/reta-6483-activities. 240 Trade in Value Added The third challenge is about the valuation. The AIO tables are in producers’ prices. There is no doubt that the most preferable valuation concerning the requirement of economic analysis is basic price. However, even at present, most of our target countries construct their national IO tables in producers’ prices. In the second part of the chapter we examine the OECD Inter-Country IO Model. 5 INTRODUCTION TO THE OECD INTER-COUNTRY IO MODEL The OECD has been updating and maintaining harmonised IO tables, split- ting intermediate flows into tables of domestic origin and imports, since the mid-1990s, usually following the rhythm of national releases of benchmark IO tables from the national statistical institutes. This harmonised data set has been used for our various country comparative analyses, such as mea- surement of global value chains, vertical specialisation and carbon dioxide emissions embodied in international trade. As countries have increased dependencies on external markets both for inputs (imports of intermediates and final expenditure goods) and outputs (exports), the limitations of single-country-based analytical frameworks have become apparent, ie international feedback and spillover effects are no longer negligible. As outlined in the previous section, the international IO frame- work is an ideal tool for linking national production chains. However, the development of inter-country IO models requires a number of very data intensive steps. Notably, it requires internationally harmonised sources of industry statistics for measuring inter-country economic spillovers. There- fore, maximum statistical cooperation across national statistical institutes is very important in pursuing this avenue of research. 7 6 PROCEDURES FOR DEVELOPING AN INTER-COUNTRY IO (ICIO) MODEL The estimation procedures of an OECD ICIO model are summarised as follows. (a) Preparation of inter-industry tables for reference years using the latest published data sources, eg symmetric IO tables, supply and use tables, other System of National Accounts (SNA) sources and international trade statistics. The OECD ICIO explicitly includes the economic structures of the rest of the world as one endogenous economy in order to close the world economy (See Figure 9.7). The initial input coefficient of the rest of the world economy is based on a proxy country’s structure (Indonesia). 7 See, for example, the World Input–Output Database (http://www.wiod.org/), the EU- KLEMS accounts (http://www.euklems.org/) and IDE-JETRO AIO Project (http://www.ide .go.jp/). Developing International Input–Output Databases 241 Figure 9.7: Format of an OECD inter-country input–output model. (b) Preparation of bilateral merchandise trade data by industry and end-use categories for reference years, which requires aggregation of published trade statistics from product classifications to industries and end-use categories via standard conversion keys. The import flows are primar- ily chosen because the export flows are more biased by the issues of re-exports. In principle, import flows record the country of origin as partner, while the export flows record the country of next consignment as partner. These are further adjusted for analytical purposes to deal with confidential trade flows, trade in waste and scrap products and movements of high volumes of valuable goods, eg diamonds. Trade coefficients of utility services are estimated based on cross-border gas and electricity transmissions. Other trade coefficients of service sectors are based on OECD’s Trade in Services by Partner Country (TISP), Eurostat’s Balance of Payments (EBOPS) and UN Service Trade statistics. The categories of services are classified by EBOPS are converted to industry classifications (ISIC Rev. 3) based on the recommendation in the IMF’s Balance of Payment Manual (BPM5). However, many missing flows need to be estimated using econo- metric modelling techniques, in particular for years before 2000. The estimation steps are summarised as follows (Benz and Miroudot 2012): – collection of data sources; – estimation of predicted zero flows based on gravity model with multilateral resistance (Anderson–van Wincoop model) by Poisson maximum likelihood estimation; – estimated bilateral flows are finally adjusted to balance the trade coefficients with those consistent with the trade columns of IO tables. (c) Conversion of merchandise import figures from a CIF valuation to an FOB valuation to minimise ‘mirror trade’ inconsistencies (import = export) in the international IO system, ie minimise differences between reported imports by A from B and reported exports by B to A. 242 Trade in Value Added Table 9.5: Data sources for OECD inter-country IO model. • National IO tables • Bilateral merchandise trade in goods statistics (OECD’s ITCS and UN’s Comtrade databases) • National Supply–Use Tables (if necessary) • Balance of Payments. Trade in Services by EBOPS categories • National Accounts (SNA93) time series (for output and value added by industry and expenditures by sector) • Electricity transmissions across countries • Transport network information on freight shipment (road and maritime distances, etc ) (d) Separation of import matrices for each country by cleaned trade coeffi- cients. (e) Total adjustments for the remaining discrepancies. It is preferable to prepare as many of the above data sources as possible to minimise the discrepancy between columns and rows of ICIO system, ie increase coun- try coverage and estimate trade coefficients for all industries by end- use categories. However, this statistical approach cannot fully solve the issues of discrepancies generated in international transactions due to, for example, – inconsistent notions of IO’s trade based on the concept of Balance of Payment and merchandise trade statistics based on customs data, – assumptions of the proportionality between sourcing partner shares and intermediate goods and services for all industries in each country; this hypothesis, known as multiregional input– output (MRIO) framework, is widely used to develop various inter- regional models because it is impossible to pursue the ‘alternative’ approach, which is to perform special surveys of all target indus- tries in order to gain the information on origin country of interme- diate supplies. Thus, the rest of the world discrepancies are treated using the mechan- ical biproportional method at the final stage. (f) Merge the inter-country database with regional blocs (optional). 7 OECD DATA SOURCES USED FOR ICIO MODEL The first version of OECD’s ICIO database is based on methodologies previ- ously established for interregional analyses (see, for example, Chenery 1953; Developing International Input–Output Databases 243 Figure 9.8: Estimation procedures for harmonised format IO. i × i, industry by industry. p × p, product by product. Moses 1955; Isard 1951). To link national IO tables by bilateral trade coeffi- cients, we have compiled national data and carried out estimations to produce harmonised intermediary databases such as harmonised IO tables and bilat- eral trade in goods data by industry and end-use (Table 9.5). OECD Input–Output Database Ideally, national authorities would provide the latest Supply–Use Tables and benchmark symmetric input–output tables (SIOTs) at the most detailed level of economic activity possible, with a basic price valuation and, preferably, separated into domestically produced and imported intermediate goods and services. However, few countries can meet such requirements. Therefore, in order to maximise country coverage, all relevant partial data is used. It should be noted that one of the main reasons that IO analysis has benefited from renewed attention in recent years is the improved availability and quality of IO tables and related statistics from national sources. Compilation methodology. The process of compiling OECD’s IO database greatly depends on cooperation with national statistical institutes. Meth- ods used for transformation to the harmonised industry-by-industry tables depend on national data availability. Some countries have been able to pro- vide symmetric industry-by-industry IO tables at basic prices in ISIC Rev. 3 based classification, whereas others have only been able to provide supply and use tables and symmetric product-by-product IO tables. An industry-by-industry format is chosen for various reasons (Yamano and Ahmad 2006). 244 Trade in Value Added • To contribute to harmonised industry analysis with other industry- based data collections, eg OECD STAN, labour statistics, and Research & Development expenditures. • Policy focus: many OECD databases are fundamentally concerned with industrial structures. • Simplicity of the conversion techniques: assuming a fixed product sale structure, no negative numbers appear in the estimated symmetric industry-by-industry tables. The process of transformation is described as in Figure 9.8. Coverage: countries and years. The first edition of the OECD IO Database dates back to 1995, and covered 10 OECD countries, with IO tables spanning the period from early 1970 to early 1990. The first updated edition of this database, released in 2002, increased the country coverage to 18 OECD coun- tries plus China and Brazil, and introduced harmonised tables for the mid- 1990s. Since 2006 this tradition of growth has continued, so that there are now tables available for 48 countries/economies (including 33 OECD mem- bers) with tables for the mid-2000s (mainly 2005) now available for most of them (Table 9.6). 8 Industry classification. The industry classification used in the current ver- sion of the IO database is based on ISIC Rev. 3 (Table 9.7), meaning that it is compatible with the other OECD industry-based analytical data sets such as the Structural Analysis (STAN) database based on SNA by activity, and bilat- eral trade in goods by industry and end-use (derived from merchandise trade statistics via standard Harmonized System to ISIC conversion keys). 8 OECD BILATERAL TRADE DATABASE BY INDUSTRY AND END-USE CATEGORY (BTDIXE) The OECD has recently developed estimates of bilateral trade data by industry and by end-use covering 65 countries/economies for the period 1988–2010 (Zhu et al 2011). The OECD Bilateral Trade Database by Industry and End-Use Category (BTDIxE) is derived from OECD’s International Trade by Commodi- ties Statistics (ITCS) database and UN’s Comtrade database, where values (and quantities) of imports and exports are compiled according to product classi- fications and by partner country. Figure 9.9 shows a summary of the world export structure in 2009, while Figure 9.10 shows the evolution of export structures for selected countries. 8 For more details, and information on how to access the OECD IO tables, go to http:// www.oecd.org/sti/ind/input-outputtables.htm. Developing International Input–Output Databases 245 Table 9.6: Country coverage of OECD Input–Output 2009 edition (as of March 2012). OECD Mid-1990s Early 2000s Mid-2000s Australia 1994/95 1998/99 2004/05 Austria 1995 2000 2005 Belgium 1995 2000 2005 Canada 1995 2000 2005 Chile 1996 — 2003 Czech Republic 1995 2000 2005 Denmark 1995 2000 2005 Estonia 1997 2000 2005 Finland 1995 2000 2005 France 1995 2000 2005 Germany 1995 2000 2005 Greece 1995 2000 2005 Hungary 1998 2000 2005 Iceland — — — Ireland 1998 2000 2005 Israel 1995 — 2004 Italy 1995 2000 2005 Japan 1995 2000 2005 Korea 1995 2000 2005 Luxembourg 1995 2000 2005 Mexico — — 2003 Netherlands 1995 2000 2005 New Zealand 1995/96 2002/03 — Norway 1995 2000 2005 Poland 1995 2000 2005 Portugal 1995 2000 2005 Slovak Republic 1995 2000 2005 Slovenia — 2000 2005 Spain 1995 2000 2005 Sweden 1995 2000 2005 Switzerland — 2001 2006 Turkey 1996 1998 2002 United Kingdom 1995 2000 2005 USA 1995 2000 2005 The OECD International Trade by Commodities Statistics (ITCS) database is updated on the basis of annual data submissions received from OECD member countries as well as from Chinese Taipei and, in some cases, from Eurostat. Due to the convergence of OECD ITCS and UN Comtrade updating processes, data sharing and other related cooperation between the two organisations, tables can also be computed for non-OECD members as declaring countries, notably the countries which belong to the OECD Enhanced Engagement Pro- gram: Brazil, China, India, Indonesia and South Africa. In ITCS and UN Comtrade, data are classified by declaring country (ie the country supplying the information), by partner country (ie origin of imports and destination of exports) and by product (ie according to the HS). In both data sources, trade flows are stored according to the product classification used by the declaring country at the time of data collection. In general, source 246 Trade in Value Added Table 9.6: Continued. Non-OECD Mid-1990s Early 2000s Mid-2000s Argentina 1997 — — Brazil 1995 2000 2005 China 1995 2000 2005 Chinese Taipei 1996 2001 2006 India 1993/94 1998/99 2006/07 Indonesia 1995 2000 2005 Latvia — — 2004 Lithuania — — 2005 Malaysia — 2000 — Malta — 2000 2004 Romania — 2000 2005 Russian Federation 1995 2000 — Singapore∗ 1995 2000 2002 South Africa 1993 2000 2002 Thailand — — 2005 Vietnam — 2000 — EU27 — — 2005 A dash means that the available year data is not available. ∗ Not published (internal use only). data are held according to Standard International Trade Classification (SITC) Rev. 2 for the 1978–87, the Harmonized System (1988) for 1988–95, HS Rev. 1 (1996) for 1996–2001, HS Rev. 2 (2002) for 2002–6 and HS Rev. 3 (2007) from 2007 onwards. To generate estimates of trade in goods by industry and by end-use cate- gory, six-digit product codes from each version of HS from ITCS and Comtrade were assigned to a unique ISIC Rev. 3 industry and a unique end-use category according to the BEC classification, and hence SNA basic classes of goods. Thus, eight sets of conversion keys were estimated by using classification correspondence tables, developed internally by the Directorate for Science Technology and Industry, OECD, and available classification correspondence tables published by UNSD. 9 WISH LIST FOR IMPROVING THE INTERNATIONAL INPUT–OUTPUT TABLE There is no doubt that the international Input–Output database opens up an opportunity to examine the comprehensive economic effects of global value chains. However, further statistical challenges remain for both IO tables and trade statistics. 9.1 Input–Output Tables Firstly, more data sources of symmetric IO and/or Supply–Use Tables are nec- essary in order to develop better models of global production chains. The smaller portion of the rest of the world economy would theoretically reduce Developing International Input–Output Databases 247 Table 9.7: OECD IO industry classification. ISIC Rev. 3 Description 01,02&05 1 Agriculture, hunting and related service industries 10–12 2 Mining and quarrying (energy) 13&14 3 Mining and quarrying (non-energy) 15&16 4 Food products, beverages and tobacco 17–19 5 Textiles, textile products, leather and footwear 20 6 Wood and products of wood and cork 21&22 7 Pulp, paper, paper products, printing and publishing 23 8 Coke, refined petroleum products and nuclear fuel 24ex2423 9 Chemicals excluding pharmaceuticals 2423 10 Pharmaceuticals 25 11 Rubber & plastics products 26 12 Other non-metallic mineral products 271&2731 13 Iron & steel 272&2732 14 Non-ferrous metals 28 15 Fabricated metal products, except machinery & equipment 29 16 Machinery & equipment, nec 30 17 Office, accounting & computing machinery 31 18 Electrical machinery & apparatus, nec 32 19 Radio, television & communication equipment 33 20 Medical, precision & optical instruments 34 21 Motor vehicles, trailers & semi-trailers 351 22 Building and repairing of ships & boats 353 23 Aircraft and spacecraft 352–359 24 Railroad equipment and transport equipment nec 36&37 25 Manufacturing nec; recycling (including furniture) the discrepancies generated by misallocated import transactions by sourcing country. While the national statistical institutes of most countries have been able in recent years to publish the official figures of import matrices, the frequency of update is still about every five years when each country performs special surveys for benchmark year IO tables. Given the import penetration ratios of some industries are not annually stable for economic and social reasons, it is preferable to have more frequent reports on the transaction structures of imported goods and services. In addition, many detailed industrial-level IO tables are required to plan the policy interventions. Only a few countries are currently able to provide over information on transactions of 400 industry levels (eg USA, Japan and Korea). 9.2 Trade Statistics In the current data submission framework, the trade flows of goods and ser- vices are available only for reported flows. In other words, we do not have enough information on unreported flows. Each unreported flow could be zero flow, confidential value or missing information. 248 Trade in Value Added Table 9.7: Continued. ISIC Rev. 3 Description 401 26 Production, collection and distribution of electricity 402 27 Manufacture of gas; distribution of gaseous fuels through mains 403 28 Steam and hot water supply 41 29 Collection, purification & distribution of water 45 30 Construction 50–52 31 Wholesale & retail trade; repairs 55 32 Hotels & restaurants 60 33 Land transport; transport via pipelines 61 34 Water transport 62 35 Air transport 63 36 Supporting & auxiliary transport activities; activities of travel agencies 64 37 Post & telecommunications 65–67 38 Finance & insurance 70 39 Real estate activities 71 40 Renting of machinery & equipment 72 41 Computer & related activities 73 42 Research & Development 74 43 Other business activities 75 44 Public admin. & defence; compulsory social security 80 45 Education 85 46 Health & social work 90–93 47 Other community, social & personal services 95–99 48 Private households and extraterritorial organisations Since confidential values are aggregated in the two-digit chapter level of HS, if a country decided to make one individual flow a confidential entry, at least one other commodity of the same two-digit chapter could be masked as a confidential value. This is why identification of zero flows is an important factor to minimise the unknown trade coefficients. Clarify sentence? The official submission of re-exports of country of origin and destination by commodity and reimports of country of transhipment would be also help- ful in order to identify the deviation of both physical flows of goods and monetary flows (Guo et al 2009). Ideally, the unit quantity used in the trade statistics should be harmonised for each commodity group. Lastly, some industrial waste and by-products are explicitly recorded at the six-digit Harmonized System level of merchandise trade statistics. How- ever, international transactions of second-hand capital goods, eg machinery and vehicles, are not explicitly coded in the current framework of the Har- monized System. It would be ideal to include the international recycling and sorting industries and to separate the production chains of new products and redistribution of used household goods and industrial capital goods. Developing International Input–Output Databases 249 Personal computers and passenger cars Household consumption goods Capital goods Other intermediate goods Mining Intermediate goods for assembly 100 80 60 % 40 20 0 USA ($949 bn) Japan ($560 bn) OECD EU15 ($3785 bn) OECD others ($1529 bn) China ($1529 bn) BRIS ($706 bn) Rest of World ($2540 bn) Figure 9.9: Export share by industry and category (world, 2009). REFERENCES Anderson, J. E. (1979). A Theoretical Foundation for the Gravity Equation. American Economic Review 69(1), 106–116. Benz, S., and S. Miroudot (2012). Building the OECD Estimated Bilateral Trade in Ser- vices by Industry (EBTSI) Dataset. Mimeo. Chenery, H. B. (1953). Regional analysis. In The Structure and Growth of the Italian Economy (ed. H. B. Chenery, P. G. Clark and V. C. Pinna), pp. 91–129 (Rome: US Mutual Security Agency). Eurostat (2008). Eurostat Manual of Supply, Use and Input–Output Tables. Fujita, M., P. Krugman, and A. J. Venables (1999). The Spatial Economy: Cities, Regions, and International Trade. (Cambridge, MA: The MIT Press). Furukawa, S. (1986). International Input–Output Analysis: Compilation and Case Stud- ies of Interaction between ASEAN, Korea, Japan, and the USA, 1975. Occasional Papers Series, No. 21. Furukawa, S., and S. Inomata (1993). Significance of construction and analysis of international input–output table of Asia-Pacific region. In International Industrial 250 Trade in Value Added (a) 100 1600 80 1200 Share (%) 60 800 40 20 400 0 0 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 (b) 100 1600 80 1200 Share (%) 60 800 40 400 20 0 0 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Figure 9.10: Export share by industry and category. (a) China, (b) Germany. ‘Total’ given in US$ billions. Linkages and Economic Interdependency in Asia-Pacific Region, Chapter 1, pp. 2–20 (Tokyo: Institute of Developing Economies). Geraci, V. J., and W. Prewo (1977). Bilateral Trade Flows and Transport Costs, Review of Economics and Statistics 59(1), 67–74. Guo, D., C. Webb and N. Yamano (2009). Towards Harmonised Bilateral Trade Data for Inter-Country Input–Output Analyses: Statistical Issues, OECD Science, Technology and Industry Working Papers 2009/4. IDE (2006). JETRO, Asian International Input–Output Table. Volume 1 Explanatory Notes. Statistical Data Series, No. 89 (Chiba: Institute of Developing Economies). IDE (2008). JETRO, Compilation and Use of the 2005 Asian International Input–Output Table (ed. S. Inomata and H. Kuwamori), Asian International Input–Output Series, No. 70 (Chiba: Institute of Developing Economies). IDE (2011). JETRO, Compilation and Use of the 2005 Asian International Input–Output Table (ed. S. Inomata and H. Kuwamori), Asian International Input–Output Series, No. 76 (Chiba: Institute of Developing Economies). Isard, W. (1951). Interregional and Regional Input–Output Analysis: A model of a Space Economy. Review of Economics and Statistics 33, 318–328. Developing International Input–Output Databases 251 (c) 100 1600 80 1200 Share (%) 60 800 40 (c) 20 400 0 0 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 (d) 100 1600 80 1200 Share (%) 60 800 40 20 400 0 0 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Figure 9.10: Continued. (c) Japan and (d) USA. ‘Total’ given in US$ billions. Japan Shipping Exchange (1983). Distance Tables for World Shipping (8th edition). Japan Shipping Exchange, Inc. Kuroiwa, I. (1995). The Spatial Structure of the Asia-Pacific Economy: Applications of International Input–Output Analysis. PhD Dissertation, University of Pennsylvania. Leontief, W. (1953). Interregional Theory. In Studies in the Structure of the American Economy, Chapter 4, pp. 93–115 (New York: Oxford University Press). Leontief, W., and A. Strout (1963). Multiregional Input–Output Analysis. In Structural Interdependence and Economic Development (ed. T. Barna). (New York: St. Martin’s Press). Limao, N., and A. J. Venables (2001). Infrastructure, Geographical Disadvantage, Trans- port Costs, and Trade. World Bank Economic Reviews 5(3), 451–479. Miller, R. E., and P. D. Blair (1985). Input–Output Analysis: Foundations and Extensions (Prentice-Hall, 1985). Moses, L. N. (1955). The Stability of Interregional Trading Patterns and Input–Output Analysis. American Economic Review 45, 803–832. 252 Trade in Value Added Okamoto, N., T. Sano and S. Inomata (2005). Estimation Technique of international Input–Output Model by Non-Survey Method, IDE Discussion Paper Series, No. 28. Polenske, K. R. (1970). An Empirical Test of Interregional Input–Output Models: Esti- mation of 1963 Japanese Production. American Economic Review 60(1), 76–82. United Nations (1999). Handbook of Input–Output Table Compilation and Analysis. Studies in Methods, Series F, No. 74 (New York: United Nations). Watanabe, T. (1964). International Input–Output Table: Structure and Possibilities, IDE Research Documents, No. 92. (In Japanese.) Wonnacott, R. (1961). Canadian-American Dependency, An Interindustry Analysis of Production and Prices (Amsterdam: North-Holland). Yamano, N., and N. Ahmad (2006). The OECD Input–Output Database: 2006 Edition, OECD STI Working paper 2006/7. Zhu, S., N. Yamano and A. Cimper (2011). Compilation of Bilateral Trade Database by Industry and End-Use Category, OECD STI Working Paper 2011/6. 10 A Three-Stage Reconciliation Method to Construct a Time Series International Input–Output Database NADIM AHMAD, ZHI WANG AND NORIHIKO YAMANO 1 1 INTRODUCTION There is increasing attention on global value chains and what has been described as trade in value added to better understand the importance of trade to economic growth, jobs and material well-being in both academic and policy circles. A global value chain can be characterized as a chain that reveals how, and by how much, each industry involved in the production of a partic- ular good or service contributes to the production of that good or service. Within such a supply chain or production network, each producer purchases inputs and then adds value, which then becomes part of the cost (inputs) of the next stage of production. The sum of the value that is added at every stage in the chain equals the value of the final goods purchased at the end of the chain. Historically these chains were typically constrained within the eco- nomic borders of one country, but in recent decades, driven by cheaper trans- port costs and lower tariffs, there has been an increasing number of chains that cross international borders (international fragmentation of production). This phenomenon has complicated analysis and policymaking. Because goods and services can cross borders many times before they reach their final desti- nation, the value of exports can overstate the importance of a given export to the exporting economy, as the export will embody value that has been added along the supply chain by industries in other countries. Prior to this international fragmentation of production period, a single national input–output (IO) table could be used to give reasonably reliable esti- mates of how different industries within an economy participated in produc- ing final goods, whether for domestic or export markets. But increased frag- mentation has significantly changed the landscape. Imports of manufactured 1 The views in the paper are solely the authors’ own. They are not meant to represent in any way the views of the OECD or US International Trade Commission or any of its indi- vidual Commissioners. The authors thank Li Xin at CCER, Perking University, for efficient data processing and research assistance. 254 Trade in Value Added goods and services are increasingly being used as intermediate inputs in the production of goods and services within global value chains, and in addition the intermediate imports themselves are increasingly embodied with value that was added in an upstream part of the value chain by the importing econ- omy itself. The weaknesses in using a single country’s IO table to analyse and provide evidence on global value chains was recognised by a team of experts contracted by the US National Research Council (NRC) 2 to study how much US content was embodied in its imports and how much foreign con- tent was embodied in its exports. They concluded (National Research Council 2006) that while it was possible to derive proxies of foreign contents in US exports using US input–output statistics, the results themselves, particularly those relating to the US content of imports, were highly dependent on the underlying assumptions. The most serious reservation the team had was the absence of harmonised supply and use (IO) tables that could be linked across countries. Significant progress has been made since the NRC report, however. The 1993 System of National Accounts, for example, recommended the develop- ment of supply–use tables, which has led to widespread use and development of these tables as a tool to balance GDP in most developed economies. Indeed, within the European Union (EU) it is a legal requirement to produce these tables, and the international statistics community has engaged in a number of initiatives to assist developing economies in this area. 3 More recently, the European Commission funded a consortium of 11 Euro- pean research institutions and the Organisation for Economic Co-operation and Development (OECD), to develop a time series of ‘world’ IO tables, the World Input Output Database (WIOD), covering 27 EU countries and 13 other major economies from 1995 to 2009. (Timmer et al 2012). In addition, the OECD has been actively involved in this area since the early 1990s, when it produced a set of harmonised IO tables for 10 countries, expanding the cov- erage to over 20 in the early 2000s and to 58 economies today. There has been widespread recognition within the official international statistics community that international fragmentation requires a new approach to how we measure trade, in particular the need to measure trade in value added (see United Nations Statistics Division 2011). The needs and improvements in national statistics information systems led the OECD and WTO to launch a joint initiative on March 15, 2012, ‘Measuring Trade in Value Added’, 4 which is designed to mainstream the production of trade in value- 2 The committee was chaired by Professor Edward Leamer and consisted of members drawn from the councils of the National Academy of Sciences, the National Academy of Engineering and the Institute of Medicine. 3 The Asian Development Bank organised a project with participation of 17 developing countries (RETA 6483) in Asia Pacific to construct supply and use tables for each partici- pating country. 4 See http://www.oecd.org/trade/valueadded. The Construction of Times Series International Input–Output Database 255 added statistics and make them a permanent part of the statistical landscape. The first official data was released on January 16, 2013. Underpinning this initiative is the creation of a global IO table database (or tables that are as global in their coverage of countries as possible: the 58 countries in the OECD database, for example, reflect 95% of global GDP). But creating these tables is non-trivial and requires the leaping of a number of statistical hurdles. There have been a number of attempts to compile global IO tables in recent years (Lenzen et al 2012; Wang 2011; Wang et al 2012; Johnson and Noguera 2012; and the WIOD project), which has led to important improvements in the qualities of the estimated global IO tables. These include the following: • benchmark to official national accounts estimates of output and final consumption (as not all countries’ supply–use tables are necessarily benchmarked to, nor revised in line with, their GDP by expenditure account); • assumptions used to allocate imports to users have moved away from the traditional crude ‘proportionality’ assumption and now capture het- erogeneities in imports from different sources based on the end-use cat- egory that is available in trade statistics (UN Broad Economic Category classification); • a recognition that shares rather than values per se are what matter in official bilateral trade statistics. Besides these common features, each of these recent works has also provided additional useful experience in the construction of global IO tables, particu- larly in the context of balancing; an important point to note in this context con- cerns deficiencies in official trade statistics which show that global exports dif- fer from global imports. A number of different approaches have thus far been adopted to estimate the balance tables. For example, Wang (2011) introduced estimates of initial data reliability to guide the balancing process, Lenzen et al (2012) proposed a method to estimate the standard error for each cell in the global IO tables to assess their reliability and uncertainty using data of con- straint violation and discrepancies between balanced IO table and unbalanced initial estimates. Another important improvement is the use of Supply and Use Tables (SUTs) as the starting point to integrate trade statistics and derived the final symmetric world IO table, the approach adopted by WIOD. Intuitively, this approach makes sense, as it links trade statistics (which are product based) with the product statistics in the supply–use table on the one hand, and value- added/employment data (that is industry-based) with industry statistics in the supply–use tables on the other hand. It also avoids errors inherent in the assumptions imposed when transferring SUTs to symmetric IO tables before the reconciliation process even start. However, as pointed out by Streicher and Stehrer (2012) the current WIOD method has two major unsolved issues: first, 256 Trade in Value Added its international transportation margins were assumed as being produced in the rest of the world by the ‘Panama assumption’ and not linked back to the world economy. Second, exports to rest of the world were derived as residu- als to balance world exports and imports, which resulted in negative exports from some countries in several products. 5 To overcome these problems, Stre- icher and Stehrer (2012) proposed a method to construct a trade matrix of cost, insurance and freight/free on board (CIF/FOB) margins together with supply and use tables for the Rest of the World. This resulted in a consistent global SUT system with international transportation services also balanced at the global level. Building on the experiences of these recent works this chapter develop mathematical programming model to integrate individual country SUTs with detailed bilateral trade statistics using a three-stage reconciliation procedure to produce a consistent annual global SUT database. The procedure solves the inconsistencies in trade statistics and data from different sources using a system of simultaneous equations that minimise a quadratic penalty func- tion that only allows minimum deviation from both official SUTs and trade statistics. The model deals with the data reconciliation problem at the global level first by reconciling official estimates of each country’s total merchandise and service trade statistics reported in each country’s national accounts with reported total exports to and imports from the world at product level in that country’s SUTs. It results in a set of country product level total exports and imports which satisfy the condition that world total exports (FOB) plus a shipping margin (CIF) equals world total imports (CIF). The use of inter- national margin services is also balanced with its supply from margin pro- ducing industries at the global level simultaneously, similar to Streicher and Stehrer (2012) but achieved in a unified modelling framework. At the second stage, the model reconciles each country’s SUTs with the globally consistent exports and imports data from the first stage. At the third and final stage, the model integrates individual countries’ statistics with international bilat- eral trade statistics by distributing each country’s total exports and imports in every commodity group to its trading partners based on bilateral trade shares computed from bilateral trade in goods and service data, taking each country’s total exports to and imports from the world derived from the first stage as controls and adjusting their distribution among partner countries to produce a consistent annual global SUT. 6 The rest of the chapter is organised as follows: Section 2 specifies the three-stage procedure for accounts and trade statistics reconciliation. Sec- 5 See Timmer et al (2012, p. 38) for details. 6 One important spillover from the model is its ability to produce updated global tables as and when (normal) revisions to GDP and trade statistics occur (ie excluding revisions related to conceptual changes in the accounting framework, such as the capitalisation of R&D in the 2008 System of National Accounts). The Construction of Times Series International Input–Output Database 257 tion 3 describes the major data sources used to implement and test the proce- dure. Section 4 presents preliminary test results and describes how the official statistics were adjusted. Section 5 concludes with a discussion on directions for future work. 2 THE THREE-STAGE RECONCILIATION PROCEDURE 2.1 Stage 1 In the first stage, model reconciles global trade statistics. A key this step is estimate the reconciled value of total global exports and imports and each country’s total imports and exports on goods and services that form part of this global total. The starting point is estimates of trade available in official national accounts statistics 7 of GDP by expenditure. Prior to reconciliation of these national estimates, differences between total exports and imports in FOB price are generally less than 2% of global exports for most of the years in the period covered (see Section 3.1 for a detailed discussion). Using data as controls, we adjust exports and imports in each country’s SUTs provided in WIOD (by product) based on a reliability index of exporters and importers, to obtain a set of country by product exports and imports estimates which satisfies the condition that total global exports equals total global imports for each product. Purchases in the domestic territory by non- residents and direct purchases abroad by residents are treated as a special product in the balancing procedure. This globally consistent trade data set is used as a control to rebalance each country’s SUTs in stage 2, before bilateral trade by product and end use is introduced to obtain the international SUTs in the final stage. The notation used to specify the first stage programming model is as fol- lows. s Ect : exports to the world of commodity group c by country s at year t , FOB prices. r Mct : imports from the world of commodity group c by country r at year t , CIF prices. WEs kt : total exports (k ∈ {G = goods, S = services, T = total}) to the world by country s , FOB price. WMr kt : total imports (k ∈ {G = goods, S = services, T = total}) from the world by country r , FOB price. CIFr ct : cost, insurance and freight for country r ’s total imports of commodity group c from the world at time t . s Eadjkt : purchase in the domestic territory by non-residents. r Madjkt : direct purchases abroad by residents. 7 Sourced from OECD National Accounts database and UN National Accounts. 258 Trade in Value Added RIXs c: reporter reliability index of commodity c by exporter s . RIMrc: reporter reliability index of commodity c by importer r . 8 To be consistent with the official statistics in an individual country’s SUTs and national accounts, the product level exports imports are valued at FOB and CIF price, respectively, but total exports and imports of goods and ser- vices are valued at FOB prices. Product index c is defined over commodity set C ∈ {1, 2, . . . , n} and divided into three subsets: goods (CC), non-margin services (CS) and margin service (CT); country indices s and r are defined over country set G ∈ {1, 2, . . . , g }. Variables without zero suffixes are endogenous in the model, and variables with a zero suffix are parameters, exogenous to the model. Using the above notation, the first stage programming model is specified as follows. Objective Function at Each Year t min S g n s s 1 (Ect − E0 )2 = ct 2 s =1 c =1 (1 − RIXs s ct )E0ct g n r r n (Mct − M0 )2 (CIFct − CIF0ct )2 + ct + r =1 c =1 (1 − RIMr r ct )M0ct c =1 (1 − RIMct )CIF0ct r s r r g (Eadjkt − Eadj0 )2 g (Madjkt − Madj0 )2 + s kt + r kt r =1 k=s,g,t Eadj0 r =1 k=s,g,t Madj0 kt kt g g (WEs s kt − WE0kt ) 2 (WMr r kt − WM0kt ) 2 + 100 + . s =1 k=s,g,t (1 − RIXs s t )WE0kt r =1 k=s,g,t (1 − RIMr r t )WM0kt (10.1) Constraints at Each Year t The country total exports are given by s s Ect + Eadjkt = WEs kt for all s and k. (10.2) c The country total imports are given by r (Mct − CIFr r r ct ) + Madjkt = WMkt for all r and k. (10.3) c The world market equilibrium at the commodity group level for goods trade is given by g g r (Ect + CIFr ct ) = r Mct , c ∈ CC for trade only. (10.4) s =1 r =1 The Construction of Times Series International Input–Output Database 259 The world market equilibrium at commodity group level for non-margin ser- vices trade is given by g g s r Ect = Mct , c ∈ CS for margin services trade only. (10.5) s =1 r =1 The international margin service supply and demand balance is equal to g g s CIFr ct = (Ect r − Mct ), c ∈ CT for margin services trade only. r =1 c ∈CS r =1 c ∈CT (10.6) The world market equilibrium for the goods trade is equal to g g WEs G,t = WMr G,t . (10.7) s =1 r =1 The world market equilibrium for the service trade (including margin trade) is given by g g g WEs S,t − CIFr ct = WMr S,t . (10.8) s =1 r =1 c ∈CT r =1 The total world exports are equal to the total world imports: g g WEs T ,t = WMr T ,t . (10.9) s =1 r =1 The model is used to reconcile official national account data on goods and ser- vices trade statistics (WEs r 0kt WM0kt ) with each country’s reported total exports s r to and imports from the world at commodity group level (E0 ct M0ct ) recorded in each country’s national SUTs. This results in a set of country product level total exports and imports, along with the value of transport costs by country and commodity group, which satisfies the condition that world total exports plus a shipping cost equal world total imports all products and services, including international transportation services. 2.2 Stage 2 To adjust each country’s exports and imports in its SUTs 9 to the globally con- sistent trade data set solved from stage 1, we also use a constrained quadratic programming model that minimises the weighted sum of squares of devia- tions from the benchmark SUTs in value-added, intermediate inputs and gross outputs, and in all final-expenditure categories, over all industries, subject to the following five sets of constraints: 9 And also to estimates in SU tables between benchmark years when annual tables are not available. 260 Trade in Value Added 1. for each industry, total intermediate inputs purchased from all com- modity groups and all sources (domestic and imported) as well as value added generated by the industry sum to the industry’s total gross out- put; 2. for each product group, the amount sold to all industries as domestic intermediate inputs plus the amount sold to final users as domestic final goods and services plus the amount of domestic exports equal the total output produced by the industries; 3. for each product group, the imported intermediates used by all indus- tries plus the amount of imported final goods used by all users plus the amount of goods re-exported minus a re-exports mark-up equal the total imports of that commodity group, which is fixed at the globally consistent level of gross imports solved in stage 1; 4. the domestic exports plus re-exports equals each product groups’ gross exports, which is at the globally consistent level solved in stage 1; 5. the sum of each type of final domestic demand by product group plus net tax on products equals total final domestic demand for each cate- gory as recorded in each country’s GDP by expenditure account. Let us define x , z, v and y as country r ’s output, intermediate inputs, value added and final domestic demands, respectively. mg , mgi, mgy , ntx , ntxi and ntxy are the total, intermediate and final goods transportation margins and net taxes respectively, wx , wz, wv , wy , wg and wt are their corre- sponding reliability weights. We denote products and industries by subscripts (c , i), value-added categories by f , and final domestic demand categories by subscript k, respectively. The variables with suffix ‘0’ stand for the initial esti- mates of the variables. There are n + 1 (adjusted for non-resident purchases in domestic markets and residents’ direct purchases abroad, which are treated as a special product) groups, m industries, l value-added categories (compen- sation for employees, indirect tax, operating surplus and depreciation) and h demand categories (household consumption, government spending, gross fixed capital formation and changes in inventory). All variables are evaluated at basic prices, except net taxes, which are evaluated at purchasers’ prices. Using the notation defined above, the second-stage optimisation model be formally specified as follows. The Construction of Times Series International Input–Output Database 261 The objective function at each year t for country r is given by min S n m r r m n r r 1 (zcit − z0 )2 (xict − x0 )2 = cit r + r cit 2 c =1 i=1 wzcit i=1 c =1 wxict r r t − v0if t ) m l n h r r (vif (yckt − y0 )2 + r + ckt r i=1 f =1 wvif t c =1 k=1 wyckt n h r r m n (mgyckt − mgy0 )2 (mgir r cit − mgi0cit ) 2 + ckt r + r c =1 k=1 wgckt i=1 c =1 wgcit n h r r m n (ntxyckt − ntxy0 )2 (ntxir r cit − ntxi0cit ) 2 + ckt r + r . c =1 k=1 wtckt i=1 c =1 wtcit (10.10) Constraints at Each Year t for Country r The balance condition for industrial gross output and input cost at basic prices is given by n l n r (zcit + ntxir cit ) + r vif t = r xict for all i. (10.11) c =1 f =1 c =1 The balance condition for total product supply and use at basic prices is given by m h m r r r r r zict + yckt + Ect = xict + Mct for all c. (10.12) i=1 k=1 i=1 The balance conditions for margin service supply and use are given by m h mgir ict + r yckt r = mgct for all c, (10.13) i=1 k=1 n mgir ict = 0 for all i, (10.14) c =1 n r mgyckt = 0 for all k. (10.15) c =1 The balance condition for net taxes in use and supply tables is given by m h ntxir ict + r ntxyckt r = ntxct for all c. (10.16) i=1 k=1 262 Trade in Value Added The gross exports and aggregate expenditure components constraints are as follows: r r r dect + r ect = Ect for all i, (10.17) n+1 r r r r (yckt + mgyckt + ntxyckt ) = GDPE0 kt for all k. (10.18) c =1 The GDP from the production side is equal to m l m r r vif t + ntxir ict = GDPt (10.19) i=1 f =1 i=1 and that on the expenditure side is equal to n n+1 r GDPEkt + r (Ect r − Mct ) = GDPr t. (10.20) k=1 c =1 Constraints (10.11) to (10.20) show that the supply and use tables are jointly used to ensure all the national accounting identities hold during the data rec- onciliation process. The adjustment made by the model to initial estimates in individual country’s SUT does not necessarily change a country’s GDP statis- tics nor any of the major aggregates of domestic expenditure components in the National Accounts, although countries total exports and imports, and so their balance of trade with the world may change due to the adjustment needed to reconcile global trade imports and exports. This seems counter- intuitive because a country’s balance of trade (BOT) is part of its GDP account- ing identity, so a change in BOT should result in a change in GDP. However, as noted earlier, SUTs compiled by national statistical institutions are not always frequently revised in line with official GDP statistics. Therefore, GDP statistics computed from national SUTs do not necessarily equal official GDP statistics. In addition, statistical discrepancies often exist in some countries’ GDP by expenditure account. Therefore, when our model eliminates the small dis- crepancy between global exports and imports (1–2% global exports each year) in official trade statistics, depending on the weights used in the reconciliation process, the model returns balance GDP (expenditure) estimates, which typi- cally do not differ from official GDP statistics. Typically, the weighting process means that in cases, where modifications occur, they are most likely to occur in those countries where there are statistical discrepancies between GDP com- puted and published in their SUTs and expenditure-based GDP estimates from the latest national accounts; in other words, the procedure also removes these statistical discrepancies in national accounts (if they exist) together with dis- crepancies between global exports and imports. 10 10 In some ways one can draw analogies here with balancing procedures used in some countries, for example, methods that take an average of GDP income (I), production (O) and expenditure (E) approaches, where a balance is forced by convention. Our approach also forces a balance, but uses an approach that weights initial estimates by their reliability. The Construction of Times Series International Input–Output Database 263 2.3 Final Stage A world supply and use table is a comprehensive account of annual transac- tion and payment flows within and between countries. use the following nota- tion to describe the elements of the world supply and use table (expressed in annual values). r xict : gross output of product c from industry i in country r . r vit : direct value added by production of industry i in country r . sr zcit : product c produced by industry i in country s and used as an intermediate input by sector i in country r . sr yckt : product c produced in country s for final use in final demand type ‘k’ in country r . CIFisr ct,cit : CIF margin by margin service ct for intermediate goods c used in industry i in country r . sr CIFyct,ckt : CIF margin by margin service ct for final goods use in final expenditure category k in country r . sr TFLc : trade flow of product c from country s to country r . Thus, the model used in the final stage of the reconciliation process can be defined as follows. Objective Function at Each Year t sr sr 2 1 g g n m (zcit − z0cit ) + (CIFisr sr ct,cit − CIFi0ct,cit ) 2 min S = sr 2 s =1 r =1 c =1 i=1 wzcit g g n+1 h sr sr 2 sr sr (ycit − y0 cit ) + (CIFyct,cit − CIFy0 ct,cit )2 + sr . s =1 r =1 c =1 k=1 wycit (10.21) Constraints at Each Year t The balance condition for industrial gross output and input cost at basic prices is given by g n g n l n sr (zcit + ntxir ict ) + CIFisr c,ccit + r vif t = r xict , (10.22) s =1 c =1 s =1 c =1 c ∈CT f =1 c =1 and the balance condition for total product supply and use at basic prices is given by m g h g m sr sr s zict + yckt = xict . (10.23) i=1 r =1 k=1 r =1 i=1 Equation (10.22) defines the value of gross i in r as the sum of the values from all of its (domestic plus imported) intermediate and primary factor inputs. 264 Trade in Value Added Equation (10.23) states that total gross output of product group c in country s is equal to the sum of deliveries to intermediate and final users all countries (including itself) in the world. This global SUT account has to be consistent with each individual country’s SUT account and international trade statistics, which requires the following accounting identities also to be satisfied each year: the constraint for intermediate use in the national use tables, which is given by g sr r zcit = z0 cit , (10.24) s =1 the constraint for final demand in the national use tables, which is given by g sr r yckt = y0 ckt , (10.25) s =1 and the constraints for bilateral trade flows at CIF prices, for which, to include international transportation service in a consistent way, the accounting equa- tion for bilateral trade is split over goods and services, m m h m sr zcit + CIFisr c,cit + sr yckt + sr CIFyc,ckt = TFLsr ct for c ∈ CC, i=1 i=1 c ∈CT k=1 i=1 c ∈CT (10.26 a) m h sr sr zcit + yckt = TFLsr ct for c ∈ CS and CT. (10.26 b) i=1 k=1 The range constraints for bilateral trade flows are based on official mirror trade statistics: sr sr min(TFLx0 ct , TFLm0ct ) TFLsr ct sr max (TFLx0 ct sr , TFLm0 ct ), (10.27) sr sr where TFLx0 ct and TFLm0 ct denote country s ’s reported exports to country r and partner country r ’s reported imports from country s . The constraint for exports at FOB prices in national use tables (solved from the first stage) is split over three product sets, goods CC, non-margin services CS and margin service CT: g m g h sr sr s zcit + yckt = Ect for c ∈ CC, (10.28 a) r =s i=1 r =s k=1 g TFLsr s ct = Ect for c ∈ CS, (10.28 b) r =s g m m TFLsr ct + CIFisr ct,cit + sr CIFyct,cckt s = Ect for c ∈ CT. r =s i=1 c ∈CC i=1 c ∈CC (10.28 c ) The Construction of Times Series International Input–Output Database 265 The constraint for imports at CIF prices in national supply tables (solved from the first stage) is given by g TFLsr r ct = Mct . (10.29) s =r Equation (10.28) indicates that a country’s total delivery of final goods and services to other countries for group c must equal its gross exports at the FOB price, which includes both domestic exports and re-exports (if applica- ble) as well as international transportation services from its margin producing industries. Equation (10.29) states each country’s demand for imports of inter- mediate and final goods and services (plus its re-exports if applicable) equals the country’s total gross imports from international markets at CIF prices. The constraint for country-specific CIF margins (solved from the first stage) is given by g m h CIFisr ct,cit + sr CIFyct,ckt = CIFr ct . (10.30) c ∈CT s =r i=1 k=1 The constraint for the margin services product structure is given by g m h g CIFisr c,ccit + sr CIFyc,cckt = s (Ect r − Mct ), c ∈ CT, (10.31) c ∈CC s =r i=1 k=1 r =1 and the GDP and aggregate domestic expenditure constraints are n+1 g sr sr r r r CIFyc,cckt + yckt + mgyckt + ntxyckt = GDPE0 kt . (10.32) c =1 s =1 c ∈CT GDP from the production side equals m l r r r (vif t + ntxiict ) = GDPt , (10.33) i=1 f =1 and the GDP from the expenditure side equals n n+1 r GDPEkt + r (Ect r − Mct )] = GDPr t. (10.34) k=1 c =1 Equations (10.22) to (10.34) must hold for all i ∈ M , k ∈ H and s, r ∈ G in each year. The optimisation problem in the last stage of our data reconciliation proce- dure is formulated to minimise a quadratic penalty function (Equation (10.21)) subject to constraints (10.22) through to (10.34). There are several desirable theoretical properties of such a mathemati- cal programming approach for data reconciliation. As discussed by Harrigan (1990), Canning and Wang (2005) and Wang et al (2010), by imposing valid binding constraints, the optimisation procedure will definitely improve, or at 266 Trade in Value Added sr sr least not worsen, the initial statistics estimates. The weights (wzij , wyic ) in the objective functions play a very important role in the data reconciliation process. By design they minimise the adjustment made to original data known to be of high quality, typically leaving these estimates largely unchanged, but allow changes to be made to data where reliability problems exist. The advantages of such an optimisation framework in data reconciliation are also significant from an empirical perspective. First, it provides consider- able flexibility in achieving global coherence. It encapsulates a wide range of initial information that is used efficiently in the data reconciliation process. Additional constraints can also be easily imposed to allow, for example, upper and lower bonds to be placed on unknown elements (this is very common in mirror trade statistics), or inequality conditions to be added. It is also very flexible regarding to the required known information and accommodates and corrects for missing data in certain blocks of the SUTs, as long as the sum of the elements within the block is known. Such flexibility is important in terms of improving the information content of the final balanced estimates as shown by Robinson et al (2001). Second, the optimisation approach permits alternative measures of the reli- ability of the initial data to be easily included in the reconciliation process, such that it is able to take account of improvements, say, in the statistical information system used in, and so reliability in the statistics of, a given country. The idea of including data reliability weights in data reconciliation can be traced back to Stone (1942) when he explored procedures for compil- ing national income accounts. As noted before, these weights should reflect the relative reliability of the initial statistics. Using properly selected reliabil- ity weights, the optimal solution should yield estimates that deviate less from the initial estimates with higher degrees of reliability than for those with lower degrees of reliability. The three-stage reconciliation procedure described above is solved with an optimisation software package Gams/Cplex. 11 Optimal solutions from this procedure are equivalent to the estimates produced by generalised least square estimations (GLS). 12 3 IMPLEMENTATION AND NUMERICAL TESTING OF THE MODEL The key in implementing the three-stage recompilation procedure to produce a balanced SUT is to carefully link each variable in the model with the best available statistics. This section documents the data sources used to initialise 11 Gams/Cplex is a well-established, versatile, powerful, high-performance optimisation system for solving large linear and quadratic programming models. 12 Since the optimal solutions are equivalent to the GLS estimates, the term ‘optimal solution’ and ‘estimates’ are sometimes used interchangeably here. The Construction of Times Series International Input–Output Database 267 Table 10.1: Countries/regions included in World Input–Output Database. ISO3 Country name ISO3 Country name AUS Australia ITA Italy AUT Austria JPN Japan BEL Belgium KOR Korea BGR Bulgaria LTU Lithuania BRA Brazil LUX Luxembourg CAN Canada LVA Latvia CHN China MEX Mexico CYP Cyprus MLT Malta CZE Czech Republic NLD Netherlands DEU Germany POL Poland DNK Denmark PRT Portugal ESP Spain ROM Romania EST Estonia RUS Russia FIN Finland SVK Slovakia FRA France SVN Slovenia GBR United Kingdom SWE Sweden GRC Greece TUR Turkey HUN Hungary TWN Chinese Taipei IDN Indonesia USA United States IND India ROW Rest of world IRL Ireland WLD World total and test the model and introduce the reliability weights used in the objective function at the first and final stages of the recompilation procedure. 3.1 Data Sources Our objective is to conduct a preliminary test of the model by integrating the individual country Supply and Use Tables, national accounts and international trade statistics. Country SUTs are obtained from WIOD , which covers 27 EU member countries and 13 other major economies in the world from 1995 to 2009 (Table 10.1). We also estimate an SUT for the rest of the world based on official national accounts statistics and the OECD intermediate data sources used to compile the OECD’s Inter-country Input–Output Database: the rest of the world is developed from the input–output/supply and trade in services of 15 countries 13 and trade in goods of all countries where UN Comtrade data are available, with industries aggregated to the 35 sectors used in WIOD, based on ISIC Rev. 3. Therefore, the product and industry classification of our testing data sets are the same as WIOD (see Tables 10.2 and 10.3 for details). We collected and compared various sources for goods and services trade data, including official National Accounts, sourced from the OECD and UNSD, 13 Chile, Iceland, Israel, Norway, Switzerland, Argentina, South Africa, Hong Kong, Malaysia, Philippines, Thailand, Vietnam, Saudi Arabia, Brunei and Cambodia. 268 Trade in Value Added Table 10.2: Product Classification of World Input–Output Database. WIOD CPA Description C1 1 Products of agriculture, hunting and related services C2 2 Products of forestry, logging and related services C3 5 Fish and other fishing products; services incidental of fishing C4 10 Coal and lignite; peat C5 11 Crude petroleum and natural gas; services incidental to oil and gas extraction excluding surveying C6 12 Uranium and thorium ores C7 13 Metal ores C8 14 Other mining and quarrying products C9 15 Food products and beverages C10 16 Tobacco products C11 17 Textiles C12 18 Wearing apparel; furs C13 19 Leather and leather products C14 20 Wood and products of wood and cork (except furniture); articles of straw and plaiting materials C15 21 Pulp, paper and paper products C16 22 Printed matter and recorded media C17 23 Coke, refined petroleum products and nuclear fuels C18 24 Chemicals, chemical products and man-made fibres C19 25 Rubber and plastic products C20 26 Other non-metallic mineral products C21 27 Basic metals C22 28 Fabricated metal products, except machinery and equipment C23 29 Machinery and equipment nec C24 30 Office machinery and computers C25 31 Electrical machinery and apparatus nec C26 32 Radio, television and communication equipment and apparatus C27 33 Medical, precision and optical instruments, watches and clocks C28 34 Motor vehicles, trailers and semi-trailers C29 35 Other transport equipment C30 36 Furniture; other manufactured goods nec C31 37 Secondary raw materials C32 40 Electrical energy, gas, steam and hot water C33 41 Collected and purified water, distribution services of water C34 45 Construction work UNCTAD, IMF’s IFS and BOPS database, Comtrade database, and the OECD database. 14 The same data can often be obtained from several different sources. However, we found there were often significant differences in values between different sources. 15 Because of these differences, it is necessary to analyse the pros and cons of each source to determine which are the most reli- 14 UNSD, United Nations Statistics Division; UNCTAD, United Nations Conference on Trade and Development; IFS, International Financial Statistics; BOPS, Balance of Payments Statistics. 15 There are two major reasons for the difference: valuation (trade valued on an FOB or The Construction of Times Series International Input–Output Database 269 Table 10.2: Continued. WIOD CPA Description C35 50 Trade, maintenance and repair services of motor vehicles and motorcycles; retail sale of automotive fuel C36 51 Wholesale trade and commission trade services, except of motor vehicles and motorcycles C37 52 Retail trade services, except of motor vehicles and motorcycles; repair services of personal and household goods C38 55 Hotel and restaurant services C39 60 Land transport; transport via pipeline services C40 61 Water transport services C41 62 Air transport services C42 63 Supporting and auxiliary transport services; travel agency services C43 64 Post and telecommunication services C44 65 Financial intermediation services, except insurance and pension funding services C45 66 Insurance and pension funding services, except compulsory social security services C46 67 Services auxiliary to financial intermediation C47 70 Real estate services C48 71 Renting services of machinery and equipment without operator and of personal and household goods C49 72 Computer and related services C50 73 Research and development services C51 74 Other business services C52 75 Public administration and defence services; compulsory social security services C53 80 Education services C54 85 Health and social work services C55 90 Sewage and refuse disposal services, sanitation and similar services C56 91 Membership organisation services nec C57 92 Recreational, cultural and sporting services C58 93 Other services C59 95 Private households with employed persons able for our reconciliation model. Ultimately, we chose the National Accounts as the best source for a country’s total gross exports to and imports from the world. For bilateral trade positions we use the OECD’s bilateral merchandise and services trade data (‘Bilateral Trade by Industry and End-Use Category’ and ‘Bilateral Trade in Services by Industry’). Control Totals for Aggregate Trade in Each Country National Accounts data by design often capture estimates of trade that will not be reflected in underlying customs data, since the National Accounts include adjustments to correct for reporting errors, partner country coverage, and CIF basis) and coverage (data missing for some countries, for some sectors and for some years). 270 Trade in Value Added Table 10.3: Industrial classification of World Input–Output Database. WIOT NACE Description 01 AtB Agriculture, hunting, forestry and fishing 02 C Mining and quarrying 03 15t16 Food, beverages and tobacco 04 17t18 Textiles and textile products 05 19 Leather, leather and footwear 06 20 Wood and products of wood and cork 07 21t22 Pulp, paper, paper, printing and publishing 08 23 Coke, refined petroleum and nuclear fuel 09 24 Chemicals and chemical products 10 25 Rubber and plastics 11 26 Other non-metallic mineral 12 27t28 Basic metals and fabricated metal 13 29 Machinery, nec 14 30t33 Electrical and optical equipment 15 34t35 Transport equipment 16 36t37 Manufacturing, nec; recycling 17 E Electricity, gas and water supply 18 F Construction 19 50 Sale, maintenance and repair of motor vehicles and motorcycles; retail sale of fuel 20 51 Wholesale trade and commission trade, except of motor vehicles and motorcycles 21 52 Retail trade, except of motor vehicles and motorcycles; repair of household goods 22 H Hotels and restaurants 23 60 Inland transport 24 61 Water transport 25 62 Air transport 26 63 Other supporting and auxiliary transport activities; activities of travel agencies 27 64 Post and telecommunications 28 J Financial intermediation 29 70 Real estate activities 30 71t74 Renting of m&eq and other business activities 31 L Public admin and defence; compulsory social security 32 M Education 33 N Health and social work 34 O Other community, social and personal services 35 P Private households with employed persons also for unobserved (eg informal) trade. But there are other reasons why dif- ferences across related sources may arise, for example, relating to concepts, including valuation. Table 10.4, for example, shows that UNCTAD, IFS and BOPS world merchandise imports tend to be larger than the National Accounts data we used. This is also a result of valuation differences (UNCTAD and IFS are both in CIF prices; WITS–Comtrade data are also in CIF prices) and defi- nitional differences (IMF’s BOPS data are only for merchandise goods, while Table 10.4: Comparisons of world goods and service trade (various sources as a percentage of National Accounts data). Exports Source Type 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 UNCTAD Goods 97 97 97 98 97 98 98 98 99 99 98 99 98 99 98 WITS Goods 89 92 94 95 95 96 97 97 98 98 97 98 96 97 96 IFS Goods 94 95 98 98 98 98 98 99 99 99 100 100 98 98 97 BOP Goods 84 85 83 85 85 84 84 82 83 83 83 82 82 88 87 BOP2 Goods 90 91 91 110 109 107 108 109 110 110 103 108 108 109 107 UNCTAD Services 101 101 101 101 101 101 101 100 100 100 100 102 104 105 106 BOP Services 85 84 85 88 88 88 88 87 87 87 87 88 91 90 92 Imports Source Type 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 UNCTAD Goods 100 101 101 101 100 101 101 102 102 102 102 102 101 102 101 WITS Goods 92 95 98 98 99 100 100 101 101 101 101 101 100 101 99 IFS Goods 98 99 103 103 102 102 102 102 103 103 103 102 102 101 100 BOP Goods 84 85 84 87 87 87 86 86 86 86 86 86 86 90 89 BOP2 Goods 90 91 91 109 109 109 108 109 110 109 110 110 110 111 110 UNCTAD Services 101 99 99 99 99 99 100 98 98 98 98 99 101 103 104 BOP Services 85 83 82 85 86 85 86 84 83 83 82 82 84 85 86 The Construction of Times Series International Input–Output Database 271 272 Table 10.5: Comparing merchandise trade data for selected countries (various sources as a percentage of National Accounts data). Exports Imports Reporter Source 1995 1997 1999 2001 2003 2005 2007 2009 1995 1997 1999 2001 2003 2005 2007 2009 China UNCTAD 102 100 100 100 100 100 100 100 107 104 104 105 105 105 106 105 WITS 102 100 100 100 100 100 100 100 107 104 104 105 105 105 106 105 IFS 102 100 100 100 100 100 100 100 107 104 104 105 105 105 106 105 BOP 88 45 43 44 45 45 49 49 89 51 55 61 60 58 61 61 BOP2 88 100 100 100 100 100 100 100 89 100 100 100 100 100 100 100 Japan UNCTAD 104 103 104 105 105 105 106 107 113 110 110 111 111 109 109 110 WITS 104 103 104 105 105 105 106 107 113 110 110 111 111 109 109 110 IFS 104 103 104 105 105 105 104 106 113 110 111 111 111 109 108 110 BOP 100 98 99 98 98 98 98 99 100 96 95 96 95 95 95 95 BOP2 100 100 100 100 100 100 100 100 100 99 99 100 99 100 100 100 Germany UNCTAD 100 100 100 100 100 99 99 98 101 102 100 101 100 99 99 98 WITS 100 100 100 100 100 99 99 97 101 102 100 101 100 99 99 97 IFS 100 100 100 100 100 99 98 97 101 102 100 101 100 99 99 98 Trade in Value Added BOP 94 94 93 94 93 93 95 94 92 93 93 93 93 93 95 95 BOP2 99 99 99 99 99 99 100 100 99 100 99 100 100 100 101 102 USA UNCTAD 100 100 99 100 100 100 101 99 102 101 101 101 102 102 102 101 WITS 100 100 99 100 100 100 101 99 102 101 101 101 102 102 102 100 IFS 100 100 100 100 100 100 101 99 102 102 101 101 102 102 102 101 BOP 97 97 97 97 98 97 97 96 98 98 98 98 98 98 98 98 BOP2 98 98 98 98 99 98 98 98 98 98 98 98 98 98 99 98 Source: UN, UNCTAD, WITS–Comtrade, OECD, IMF BOP, and IMF IFS databases. The Construction of Times Series International Input–Output Database 273 105 100 UNCTAD (g) WITS (g) 95 ITS (g) 90 BOP (g) BOP2 (g) 85 UNCTAD (s) BOP (s) 80 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Figure 10.1: Comparing data sources for goods and services: world imports plus exports (various sources as a percentage of National Accounts data). ‘g’ denotes goods; ‘s’ denotes services. BOP2 include merchandise goods plus goods for processing, repair of goods, goods procured in ports by carriers and non-monetary gold). Table 10.5 provides the same comparison for merchandise trade, but look- ing only at the four largest trading countries: China, Japan, Germany and the USA. By focusing on these four major exporters and importers, we can pro- vide a more accurate comparison between the various data sources. By exam- ining these four countries, we can clearly see that the National Accounts data are very close to that of other sources, especially in the case of merchandise exports. For merchandise exports, national accounts data are about 100% for all years for China, Germany and the USA. BOPS data is typically lower but that is expected due to definitional differences with national accounts estimates (see above). Merchandise imports for most sources are clearly larger than the national accounts data, with the exception of the BOP2 database. The data from UNCTAD, WITS, and IFS are on average about 5% larger for China, 1% for Germany, 10% for Japan and 2% for the USA; these differences are a result of the CIF margin. Similar patterns exist for services trade data. For example, world totals found in UNCTAD data on services trade are almost 100% of those of the national accounts based data (see Table). However, national accounts totals are between 9 and 18% larger than those found in the IMF’s BOPS database (Figure 10.1), reflecting the fact that some countries are absent from the BOPS world totals. This difference in totals, however, does not exist in the individ- ual country totals. For example, Table 10.6 shows that services trade data for most years, from most sources, including the BOPS database, are 100% of the national accounts data for both services exports from, and imports to, China and Germany. They are about 30% and 17% larger for Japan’s exports 274 Table 10.6: Comparing services trade data for selected countries (various sources as a percentage of National Accounts data). Exports Imports Reporter Source 1995 1997 1999 2001 2003 2005 2007 2009 1995 1997 1999 2001 2003 2005 2007 2009 China UNCTAD 88 100 100 100 100 100 100 100 89 100 100 100 100 100 100 100 BOPS 88 100 100 100 100 100 100 98 89 100 100 100 100 100 100 100 OECD 100 100 100 100 100 100 100 100∗ 100 100 100 100 100 100 100 100∗ Japan UNCTAD 124 128 130 131 130 131 137 138 111 116 117 117 118 117 120 123 BOPS 124 128 130 131 130 131 137 139 111 116 117 117 118 117 120 123 OECD 124 128 130 131 122 126 134 131∗ 110 111 114 116 115 117 115 116∗ Germany UNCTAD 100 99 95 99 104 105 103 102 100 100 98 98 101 101 101 102 BOPS 100 99 95 100 104 105 106 109 100 100 101 101 101 101 101 102 OECD 100 99 96 100 104 104 101 96∗ 101 102 102 101 101 99 99 101∗ Trade in Value Added USA UNCTAD 95 95 95 94 95 96 96 98 97 97 97 96 98 98 98 101 BOPS 95 95 95 94 95 96 98 98 97 97 97 96 98 98 97 99 OECD 96 96 97 97 97 98 101 100∗ 97 97 97 96 98 99 99 99∗ ∗ Data represent the year 2008. The Construction of Times Series International Input–Output Database 275 and imports, respectably. For the USA, the services trade data are about 5% and 3% larger for US exports and imports, respectively. These differences underscore the difficulty in collecting and estimating accurate trade statis- tics in services and reinforce our position on using National Accounts-based data, where statistics institutes make attempts to deal with inconsistencies or errors within the GDP accounting framework. Selection of Control Total for Aggregate Trade in the World Another benefit of national accounts data a control is that it is fairly balanced. Looking at the share of imports over exports of world totals (see Table 10.7) allows us to compare the global trade balance of the different sources; in a perfectly balanced world this share would equal 100% when both exports and imports are valued in FOB. basis. The data show that on average imports account for 99% of exports (goods, services and total). Imports from UNCTAD, IFS and WITS are predictably larger, by about 2%. This difference reflects the fact that in these databases exports are valued on an FOB basis and imports are valued on a CIF basis. Other Data Sources Each country’s exports to and imports from the world at WIOD product level are obtained directly from WIOD use (for exports at FOB) and supply (for imports at CIF) tables. Initial estimates of CIF margins are also taken from WIOD. We use the GDP by major expenditure components statistics as each coun- try’s macro control variables. The data are downloaded from the ‘National Accounts Official Country Data’ of UN statistics division, and the OECD’s National Accounts database, at current prices, in thousands of US dollars. These provided the source for all countries except Taiwan (Chinese Taipei), which was sourced from the Directorate-General of Budget, Accounting and Statistics (DGBAS) and converted to US dollars. Bilateral and services trade statistics are from OECD sources, but they are only used for source and destination shares after obtaining a globally consis- tent set of exports to and imports from the world at the WIOD product level for each country from our first-stage optimisation procedure. However, both exporter and importer reported data are used as the interval control in our final stage reconciliation when bilateral trade flows are estimated. 3.2 Selection of Reliability Indexes in the Objective Function As pointed out by Wang et al (2010), one of the most desirable analytical and empirical properties of the class of data reconciliation models such as the one we specified by Equations (10.1)–(10.34) is that it uses reliability weights in the objective function to control how much an initial estimate 276 Table 10.7: World trade in total (share of imports over exports by source). Type 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 NA Total 98 98 98 99 99 100 100 99 99 99 99 98 98 98 98 NA Goods 98 98 97 98 99 99 100 99 99 99 99 99 98 99 99 UNCTAD Goods 101 101 101 101 102 103 103 102 102 102 103 101 101 102 101 IFS Goods 102 103 102 102 103 103 104 103 103 103 102 101 102 102 101 BOPS Goods 97 98 99 100 101 103 103 103 103 103 103 103 103 101 101 BOP2 Goods 97 98 98 97 99 101 100 99 99 99 100 100 100 101 100 WITS Goods 101 102 102 101 103 103 103 102 102 103 103 102 103 103 102 NA Services 100 100 100 100 100 100 100 100 100 98 98 96 96 96 97 UNCTAD Services 100 99 98 98 98 98 99 98 98 96 96 94 93 93 94 Trade in Value Added BOPS Services 100 98 97 97 97 96 97 96 96 94 92 90 88 94 94 Source: UN, OECD, UNCTAD, WITS–Comtrade, IMF BOPS and IMF IFS databases. ‘NA’ denotes National Accounts data. The Construction of Times Series International Input–Output Database 277 may be adjusted. From a statistical point of view, the best way to systemati- cally assign reliability weights in the objective function is to obtain estimates of the variance–covariance matrix of the initial estimates, using the inverted variance–covariance matrix as the reliability indicators. The larger the vari- sr sr 2 sr sr sr 2 sr ance, the smaller the associated term (zci − z0ci ) /wzci or (yck − y0 ck ) /wyck contributes to the objective function, and hence the lesser the penalty for the associated variables to move away from their initial value (only the rela- tive, not the absolute size of the variance affects the solution). However, the lack of consistent historical data often makes the estimation of the variance– covariance matrix associated with the initial estimates very difficult to imple- ment. For example, the common practice in SAM balancing exercises is to assign differing degrees of subjective reliabilities to the initial entries of the matrix, following the method proposed by Stone (1942). 16 To date, very few attempts have been made to statistically estimate data reliability such as error variance of the initial estimates from historical data, except Weale (1985), who developed a statistical method that uses time series information on accounting discrepancies to infer data reliability in a System of National Accounts. Theoretically speaking, a similar statistical method can be applied to the historically reported discrepancies of bilateral trade data to derive those variances associated with international trade statistics. In practice, however, the historical data and knowledge of the changes in related country’s trade statistics reporting systems are too demanding and make such a statistical method less attractive for large empirical applications. Therefore, here we use a practical alternative approach to estimate the reliability weights, which is constructed by reporter relative reliability indexes for both exporters and importers. Reporter Reliability Indexes Trade data reported by each country and its partners are often used in the international economic literature to check the quality of trade statistics. An approximate match of mirror statistics suggests that trade data reported via that route are reliable. However, such weights treat the reported trade statis- tics from both reporters equally and do not distinguish which reporter is more reliable. In the case where there is (known) unreliable reporter in the pair, this approach may lead to changes being made to the data reported by the reliable reporter. This is undesirable. To correct this problem, a reporter’s relative reliability index needs to be developed. Such an index should be able to deal with three critical issues. The first issue is related to the difference of reporting countries in their ability to report bilateral commodity trade by end-use categories. Variability 16 Stone 0 0 0 2 proposed to estimate the variance of xij as var(xij ) = (θij xij ) , where θij is a subjective determined reliability rating, expressing the percentage ratio of the standard 0 error to the initial estimates of xij . 278 Trade in Value Added in reporting quality across countries is highly relevant information for the problem we try to solve in our proposed official approach. As discussed ear- lier, the adjustment process hinges heavily on the relative reliability of each of the reporting countries. An indicator of reporter reliability is a measure of how consistently a country reports its trade in each product relative to all its trading partners. However, judging reliability of a country’s trade based on a single bilateral flow is a poor reference, because a partner can misrepre- sent its trade, thereby potentially discrediting a reliable reporter. Therefore, a good reporter reliability measure should take all reporting countries in the world into account in assessing a country’s reporting reliability. The second issue is what exactly should be captured by the reliability mea- sure. The size of discrepancies could be incorporated into a measure of reli- ability. However, placing emphasis on the magnitude of discrepancies only may over-penalise the reliability of a legitimate reporter. A poor reporter that makes an error for a given trade flow usually makes a similar error with other partners. For example, a reporter that has mistaken the identity of one of its partners has implicitly made a mistake for others. It brings a systemic bias for that reporter. This type of problem should be detected and reflected in the reporter reliability measure without penalising the reliable reporter. The third issue is the capability of the measure to reflect both product- and country-specific reliability information for each country as an exporter and as an importer. Countries typically have specific strengths and weaknesses. For example, one exporting country may have an excellent reporting record on steel used as intermediate goods, but is also highly inconsistent in its reporting practice for trade of organic chemical in final goods. All three issues discussed above are effectively dealt with in the relia- bility index developed by Gehlhar (1996), where reporter reliability indexes were used to make a discrete choice to disregard or accept reported trade flows. The index is calculated as the share of accurately reported transac- tions of a reporter’s total trade for a particular using a threshold level. It assesses reporter reliability from a complete set of global reporting part- ners, captures the reporter’s ability to accurately report without interferences from gross discrepancies in reporting and contains exporter and importer product-specific reliability information. Specifically, the importer-specific and exporter-specific reliability indexes in the objective function (Equations (10.1) and (10.21)) are defined as rs sr MAric sr |Mic − Eic | RIMr ic = sr , where MAr ic = Mic , ALsr ic = rs , s Mic s ∈ALsr 0.20 Mic ic (10.35) XAsic rs |Mic sr − Eic | RIXs ic = sr , where XAs ic = sr Eic , ALsr ic = rs . E r ic Mic s ∈ALsr ic 0.20 (10.36) The Construction of Times Series International Input–Output Database 279 Under such reliability indexes, the size of the discrepancies becomes imma- terial because inaccurate transactions are treated the same regardless of the magnitude of the inaccuracy. The indexes have the flexibility of being imple- mented at the detailed six-digit HS level and can be aggregated to any com- modity group level. We computed such reporter reliability measures for each country and product. Major data are from UN Comtrade with supplements from country sources. 17 Reliability Weights Used in Objective Function After obtaining RIM and RIX for each WIOD product by country, there is an additional issue that needs to be solved before we can empirically compute the reliability weights in the objective function (Equations (10.1) and (10.21)) of the data reconciliation model. There is only one unique number for each trade flow in each route in the global SUTs, which should be a combination of both reporter and partner reported trade statistics based on reporters’ relia- bilities. Therefore, we combine both reporter and partner’s reliability indexes and reported statistics for each trade routine at the WIOD product level to compute the final reporter reliability weights in the objective function. They are assigned by multiplying 1 minus each reporter’s product weighted relia- bility index by their corresponding initial values. For example, the complete set of weights in Equation (10.21) is defined as follows: sr sr wzcit = (1 − RIMr ¯mcit ct )z + (1 − RIXs sr ¯xcit ct )z , (10.37) sr sr wyckt = (1 − RIMr ¯ ckt ct )ym + (1 − RIXs sr ¯ ckt ct )yx , (10.38) sr sr sr sr where z¯mcit ¯xcit ,z ¯ ckt and ym ¯ ckt , yx are the intermediate and final goods trade flows computed based on the share reported by importers and r s exporters, respectively (shares multiple Mct and Ect , the total world trade by products of each country in the balanced individual country SUTs). With such a weighting scheme, we our goal of ensuring that the model has a higher probability of changing unreliable initial data compared reliable data. 4 ADJUSTMENT MADE TO OFFICIAL ACCOUNTS AND STATISTICS BY ENFORCING GLOBAL CONSISTENCY Our model entails enforcing global consistency, which takes place in the first stage. We first establish consistency between country-reported trade in SUTs and official trade statistics in goods and services. The model solves the adjusted country total exports to and imports from the world for each product, and these country/product totals are retained for the second and 17 We are grateful to Dr Mark Gehlhar at the US Department of Interior for helping us to compute the exporter and importer reliability indexes with WIOD product classification from 1995 to 2007. 280 Table 10.8: Reporter reliability indexes, initial inconsistency and mean absolute percentage adjustment of total exports and imports, 1995–2009. Countries RIX %xerr %expadj RIM %merr %impadj Countries RIX %xerr %expadj RIM %merr %impadj AUS 0.504 −0.1 0.4 0.637 0.0 0.9 ITA 0.763 0.7 0.5 0.693 0.7 1.1 AUT 0.598 0.1 0.2 0.665 0.1 1.1 JPN 0.667 −0.4 1.7 0.611 −0.2 1.0 BEL 0.347 0.7 0.7 0.460 0.6 1.2 KOR 0.564 0.0 0.8 0.613 0.0 0.8 BGR 0.623 4.8 4.8 0.439 4.6 5.7 LTU 0.554 3.1 2.3 0.562 2.6 3.7 BRA 0.627 −0.1 1.1 0.605 −0.1 2.7 LUX 0.394 1.1 3.6 0.530 1.0 2.0 CAN 0.862 −0.2 0.4 0.675 −0.2 1.2 LVA 0.496 0.0 0.7 0.600 0.0 2.4 CHN 0.383 14.5 5.8 0.375 12.9 8.8 MEX 0.836 0.0 0.6 0.484 0.0 0.6 CYP 0.300 0.2 3.1 0.494 0.1 2.1 MLT 0.447 −2.4 1.3 0.532 −2.4 1.7 CZE 0.720 7.7 4.0 0.632 7.2 4.7 NLD 0.538 0.6 0.3 0.517 0.6 1.5 DEU 0.739 −0.1 0.5 0.527 −0.2 1.0 POL 0.689 1.2 1.0 0.624 1.1 1.8 DNK 0.572 0.1 0.7 0.629 0.1 1.6 PRT 0.684 1.2 0.5 0.726 1.0 2.7 ESP 0.765 0.9 0.4 0.620 0.9 1.0 ROU 0.644 4.2 1.6 0.497 3.2 4.3 EST 0.523 0.2 0.6 0.440 0.3 1.6 RUS 0.298 0.0 2.0 0.473 0.0 2.3 FIN 0.636 1.0 0.5 0.548 1.1 2.0 SVK 0.694 0.0 0.4 0.492 −0.1 0.7 Trade in Value Added FRA 0.732 0.7 0.3 0.611 0.7 1.2 SVN 0.704 0.9 0.8 0.584 1.1 1.0 GBR 0.567 −1.3 0.9 0.613 −0.5 0.6 SWE 0.623 0.1 1.2 0.682 0.1 1.1 GRC 0.547 −1.1 1.5 0.564 −2.1 3.6 TUR 0.635 0.0 8.4 0.492 −0.2 6.4 HUN 0.639 1.4 1.3 0.584 1.3 1.6 TWN 0.003 −0.2 1.5 0.004 0.3 0.7 IDN 0.506 −0.4 0.6 0.455 0.8 1.3 USA 0.620 0.1 1.9 0.702 0.1 1.6 IND 0.445 0.1 4.8 0.361 −1.5 3.5 ROW 0.000 −64.1 36.1 0.000 −58.1 38.8 IRL 0.489 −0.1 1.2 0.478 0.0 0.7 WLD 9.2 9.4 The Construction of Times Series International Input–Output Database 281 % expadj RIX total log(% expadj) 5 0.7 0.6 4 0.5 3 0.4 2 0.3 0.2 1 0.1 0 0 ROW TWN RUS CYP CHN TWN LUX IND MLT IRL LVA AUS IDN EST NLD GRC LTU KOR GBR AUT Figure 10.2: Reporter reliability and mean absolute percentage adjustment of total exports, 1995–2009. More reliable data get less adjustment. % expadj RIX total log(% expadj) 60 0.7 50 0.6 40 0.5 0.4 30 0.3 20 0.2 10 0.1 0 0 C11 C10 C22 C35 C13 C23 C32 C2 C36 C37 C19 C18 C14 C5 C16 C28 C17 C31 C27 C33 Figure 10.3: Reporter reliability and mean absolute percentage adjustment of world goods by product, 1995–2009. More reliable data get less adjustment. final stages as controls. The data reconciliation procedure produces a set of estimates for both trade and SUT estimates which are different from official statistics, and so it is desirable and important to know how much each set of estimates differs from the officially reported data. However, it is difficult to use a single measure to compare the original and adjusted data, since there are so many dimensions in the model solution sets. It is meaningful to use several measures to gain more insight on the model performance. Generally speaking, it is the proportionate deviation and not the absolute deviation that matters; therefore, we compute the ‘mean absolute percentage adjustment’ (MAPA) with respect to the official data for different and aggregations. Con- sider the following aggregate index measure for country and product group 282 Trade in Value Added total adjustment: T n+1 ¯r r 100 t =1 c =1 |Ect − E0 | MAPAr = T n+1 r ct , (10.39) t =1 c =1 E0ct T g 100 t =1 ¯r r =1 |Ect − r E0 | MAPAc = T g r ct . (10.40) t =1 r =1 E0ct We first focus on results for country total adjustments to illustrate some key characteristics of the adjustment process. Each country’s reliability as an exporter and importer is a key factor that governs the magnitude of adjustment on its total exports and imports (Figure 10.2). The magnitude of adjustment made by the model is relatively small, less than 2% for most countries except a few outliers reflecting the large inconsistencies between National Account total trade data and product level trade data recorded in WIOD national SUTs. We note also that there is a negative correlation between exporters and importers’ reliability and adjustments magnitudes made to cov- ered products (Figure 10.3), although the adjustments are more significant at product level. 18 As expected, both the country and sector patterns of the adjustments reflect their negative relationship with reporter’s reliability, with the exception of a few outliers. This indicates that both country and prod- uct level adjustments are not only impacted by data reliability but also by the initial discrepancies between product level trade data reported in individ- ual country’s SUTs and country totals recorded in the National Accounts. We report in Table 10.8 each country’s reliability indexes, the initial inconsistency between total trade reported in WIOD national SUTs and National Accounts data as well as the mean absolute percentage adjustments. The mean of absolute percentage adjustment for each country’s SUTs from WIOD is summarised in Table 10.9. The extent of adjustment depends not only on the difference between the globally consistent trade data from the first stage of our model and the trade data in the national table, but also on the quality of the individual countries’ statistics and how far their aggregates differ from those in the National Account (GDP by major expenditure com- ponents), which are used as macro controls. Generally speaking, the adjust- ments to sector level value-added product level final-demand related trans- actions are smaller than inputs- and gross outputs-related transactions with exceptions. The reasons for the large magnitude of adjustments to output at 18 The simple correlation coefficient between reporter reliability index with mean abso- lute percentage of adjustment of trade is −0.46. Using RIX and RIM as regressor against MAPA by detailed product level adjustment data, we get the following liner relations: expadj = 0.268 − 0.287RIX and impadj = 0.216 − 0.224RIM. Both coefficient estimates are significant at a 1% level. The Construction of Times Series International Input–Output Database 283 Table 10.9: Mean absolute percentage adjustment of national statistics. Country z-int x -output y -final v -VA Country z-int x -output y -final v -VA AUS 47.9 49.6 0.3 2.6 ITA 30.4 33.8 0.2 0.5 AUT 40.4 39.5 0.3 0.6 JPN 45.7 36.8 0.8 2.2 BEL 37.1 36.7 0.2 0.3 KOR 37.5 52.5 1.4 1.4 BGR 35.7 45.4 0.3 4.5 LTU 39.8 40.4 0.4 0.7 BRA 33.9 29.0 0.4 0.5 LUX 47.6 62.7 1.2 0.8 CAN 39.6 33.9 0.7 0.8 LVA 46.5 42.4 0.3 0.5 CHN 37.2 78.3 1.0 1.4 MEX 48.7 28.8 0.5 1.0 CYP 59.3 32.1 6.3 5.0 MLT 30.6 35.2 0.4 0.8 CZE 35.2 49.6 0.2 4.0 NLD 37.3 37.4 0.3 0.2 DEU 36.3 34.6 0.3 0.4 POL 26.0 34.3 0.2 0.1 DNK 36.5 40.4 0.5 0.4 PRT 49.5 41.4 0.3 2.5 ESP 47.6 40.8 0.3 0.3 ROU 39.0 40.9 0.5 0.2 EST 39.3 52.7 0.4 0.6 RUS 37.5 34.1 0.7 1.0 FIN 38.8 38.9 0.4 0.4 SVK 34.0 42.1 0.2 0.2 FRA 33.3 31.0 0.2 0.3 SVN 42.4 44.8 0.3 0.4 GBR 29.7 26.6 0.2 1.2 SWE 35.1 34.7 0.4 0.2 GRC 37.7 30.2 1.1 0.9 TUR 38.2 34.0 1.0 0.8 HUN 31.8 36.6 0.3 0.9 TWN 39.5 36.7 0.6 1.6 IDN 43.8 31.4 1.1 2.8 USA 35.0 23.0 0.3 0.7 IND 39.5 39.5 0.4 2.4 ROW 120.7 233.4 64.3 178.1 IRL 49.6 50.4 0.5 0.9 WLD 41.0 42.5 3.4 9.8 industry level need further investigation. 19 Computing the adjustment index similar to Equations (10.39) and (10.40) by product groups and final demand categories could help us to identify where the large adjustments come from, providing a means to identify and solve potential problems in the data. If the standard error of national SUT statistics or some sort of reliability index could be developed similar to the index for trade data, the resulting global SUT data could be improved. Finally, we transform the global SUTs in basic prices produced from our data reconciliation model into industry-by-industry ICIO tables using ‘Model D’ discussed in Eurostat (2008, Chapter 11) similar to WIOD. 20 The mean abso- lute percentage difference between the adjusted ICIO tables and WIOD WIOTs is reported in Table 10.10. Generally speaking, the differences in sector level gross outputs are relatively small between WIOD WIOT and the estimated ICIO table by our reconciliation procedure, followed by sector level value added. 19 Ideally, the gross industry or commodity output should be fixed in the reconciliation process, because such data collected by NSI are more reliable than data on intermediate inputs. However, if we fix the gross output recorded in WIOD SUTs, there will be no fea- sible solution for the model, so we have to relax this constraint. The issue is still under investigation. 20 The justification of why ‘Model D’ is chosen is clearly discussed in Section 5 of Timmer et al (2012). 284 Table 10.10: Mean absolute percentage between WIOD industry-by-industry WIOTs and adjusted ICIO tables, 2005. Dom. Imp. Dom. Imp. Gross Value Dom. Imp. Dom. Imp. Gross Value ctr int int’ final final output added ctr int int’ final final output added AUS 62.0 77.1 51.6 317.0 1.0 26.1 ITA 52.2 68.2 47.1 88.9 1.2 23.8 AUT 52.7 63.1 36.7 125.0 1.7 20.0 JPN 56.1 72.1 45.7 103.9 1.4 20.3 BEL 42.7 68.4 34.8 138.2 4.2 19.1 KOR 53.4 77.8 58.5 144.5 2.2 24.1 BGR 52.7 73.9 49.9 243.0 1.8 25.8 LTU 101.7 85.7 52.1 282.5 4.0 27.2 BRA 55.8 68.0 45.0 137.8 1.5 26.8 LUX 88.6 92.7 43.0 355.5 4.8 74.2 CAN 63.8 43.7 48.2 81.9 1.9 13.9 LVA 76.8 85.2 65.6 370.5 3.8 29.2 CHN 41.3 74.3 57.4 91.7 1.8 37.5 MEX 59.7 55.8 31.4 99.6 1.4 9.3 CYP 97.6 102.7 51.1 410.6 8.4 26.3 MLT 76.7 95.9 64.0 500.1 4.9 31.6 CZE 53.8 56.3 55.4 103.4 1.4 29.7 NLD 51.8 65.3 37.0 145.4 6.0 28.5 DEU 51.3 66.8 43.5 99.7 1.3 17.2 POL 34.6 60.3 32.8 97.6 1.3 19.4 DNK 49.6 79.8 43.8 122.4 2.1 18.9 PRT 73.8 68.5 48.4 111.2 1.5 23.8 ESP 59.9 76.3 35.9 62.4 1.3 30.8 ROM 103.9 75.6 70.9 176.7 2.6 41.9 EST 48.3 74.6 69.5 299.2 3.8 15.1 RUS 59.9 75.7 54.3 698.9 1.3 26.9 FIN 51.9 67.9 36.3 182.6 1.3 20.9 SVK 49.0 56.0 40.8 128.1 1.2 25.3 Trade in Value Added FRA 64.0 62.7 49.3 93.3 1.7 14.7 SVN 60.3 60.9 41.6 132.6 1.3 20.6 GBR 82.2 76.9 61.3 149.2 2.0 34.6 SWE 48.4 72.2 42.6 130.8 1.4 16.3 GRC 76.1 93.7 40.5 311.6 2.2 24.6 TUR 75.4 71.8 51.4 70.8 2.1 39.0 HUN 55.4 63.3 43.3 132.8 1.4 21.8 TWN 57.7 64.8 48.9 206.8 2.2 37.8 IDN 81.3 72.5 59.0 248.7 2.6 40.2 USA 53.5 77.1 37.4 126.0 0.8 14.0 IND 54.6 82.2 33.1 123.1 1.3 19.5 ROW 98.9 69.9 68.7 132.3 48.6 41.9 IRL 88.3 86.7 53.4 131.7 1.2 21.4 WLD 59.9 70.3 46.7 122.2 6.8 23.2 The Construction of Times Series International Input–Output Database 285 The difference between domestic transactions is generally less than that of imported transactions, for both intermediate inputs and final demand. The largest difference shows up on imported final demand. 5 DIRECTION OF FUTURE WORK AND CONCLUDING REMARKS This study developed a three-stage mathematical programming model to rec- oncile detailed bilateral goods and services trade statistics with individual country’s Supply and Use Tables to produce a global SUT database. It also documents the major data sources for such a data reconciliation excise and their pro and cons. Tests of the model using WIOD national SUTs and aggre- gate trade statistics from official National Accounts as well as bilateral trade data from OECD produced encouraging preliminary results and shows that the model is feasible and may have great potential in the estimation of an integrated world SUT account. Most importantly, our empirical exercise to test the model using real world data has shown that imposing global consis- tency and eliminating ‘exports to the Moon’ will make no significant changes on NSI’s reported GDP and other major aggregate national account statistics in the balanced global SUT database. However, the model is still in its early stages of development; there are many important issues still to be addressed. We list a few of them as our concluding remarks. 5.1 SUTs with Statistical Discrepancies or Balanced SUTs Both sets of tables may be needed. A globally consistent SUT that keeps major discrepancies may be useful for statistical purposes when evaluating the accu- racy of data recorded in the global SUTs, while a balanced global SUT is nec- essary for analytical purposes, especially for estimating a balanced industry by-industry global IO table that provides the basis for computing trade in value-added estimates. So they are not substitutes but complements. A global SUT with statistical discrepancies could provide initial estimates for a bal- anced analytical world SUT, with the statistical discrepancy information in major accounting identities used to estimate standard errors for each cell in the balanced analytical global SUTs when combined with the adjustment information from the data reconciliation process, as suggested by Lenzen et al (2012). The model developed to produce balanced global SUTs in this chapter can also be used to check the consistency of data from different sources that are needed to construct any global SUTs. 5.2 Re-Exports and Re-Export Mark-Up Theoretically, re-exports can be integrated into the data reconciliation frame- work presented in this chapter without any difficulties. However, we do not include re-exports in our current reconciliation exercise due to the lack of 286 Trade in Value Added reliable total re-exports data at country and product level as controls. We are also not able to estimate exports mark-ups when reconciling individual coun- tries’ SUTs. Further work is needed to identify data sources for re-exports and estimate the mark-up margins for major re-exporting countries in the world in order to treat them as the re-exporting country’s indirect service exports in our future efforts. 5.3 Reliability Weights for National SUT Statistics We did not estimate reliability weights for national and use statistics. Without a properly estimated reliability index, we have to adjust these proportionally during our reconciliation process. Obviously this will impact on the quality of the model solutions. Research efforts will be made to better estimate all initial data reliabilities. 5.4 Structure of International Transportation Sector The use structure of international transportation services in our current rec- onciliation exercise is based on the supply structure estimated from our stage 1 model. Such information is available from detailed trade statistics by transportation modes. We plan to integrate such information into our recon- ciliation procedure and make the international shipping services an integrated part of the global inter-country IO structure in our future efforts. 5.5 Conclusions Our data reconciliation exercise has demonstrated that it is feasible to arrive at a balanced global SUT system that preserves the key identities provided by official statistics, or remains very close to them. This is an important improve- ment on other attempts in this field, which often take simple conventions or include balancing items that allocate inconsistencies implicitly or explicitly to a residual, for example, Rest of the World adjustment, or by diverging from official statistics in an uninformed manner (ie without taking into account the relative reliability of the data produced by a given reporting country). However, as noted above, much more can be done to improve the method. Central to this is the identification of sources that create better indicators of reliability throughout the system. Nevertheless, notwithstanding these areas of potential improvement, the model is already an improvement on current procedures and demonstrates that it is a tool to create tables in an efficient manner, for example it will be able to accommodate revisions in underlying data sources even though they may not (yet or never) be included in official SUTs. In addition the tool provides a means to create more timely estimates of SUTs than currently produced by official statistics institutes; thus providing a means to develop more timely estimates of trade in value added. The Construction of Times Series International Input–Output Database 287 The OECD ICIO tables and so the trade in value-added estimates produced in the OECD-WTO initiative currently take national IO tables linked with bilateral trade statistics as their starting point. In coming years, partly because of the increasing availability of national supply–use tables and partly because SUTs are generally more timely than IO tables, the OECD will begin to develop a global SUT that forms the basis of its ICIO database. REFERENCES Canning, P., and Z. Wang (2005). A Flexible Mathematical Programming Model to Esti- mate Interregional Input–Output Accounts. Journal of Regional Sciences 45(3), 539– 563. Eurostat 2008. Eurostat Manual of Supply, Use and input–output Tables, 2008 Edition. http://epp.eurostat.ec.europa.eu/portal/page/portal/product_details/publication ?p_product_code=KS-RA-07-013. Gehlhar, M. (1996). Bilateral Trade Data for Use in GTAP. GTAP Technical Paper 10, Purdue University. Harrigan, J. F. (1990). The Reconciliation of Inconsistent Economic Data: The Informa- tion Gain. Economic System Research 2(1), 17–25. Johnson, R., and G. Noguera (2012). Fragmentation and Trade in Value Added over Four Decades. Working Paper 18186. Lenzen, M., K. Kanemoto, D. Moran and A. Geschke (2012). Mapping the structure of the world economy. Environmental Science and Technology 46(15), 8374–8381. National Research Council. 2006. Analyzing the US Content of Imports and the Foreign Content of Exports. Washington, DC: The National Academies Press. Robinson, S., A. Cattaneo and M. El-Said. (2001). Updating and Estimating a Social Accounting Matrix Using Cross Entropy Methods. Economic Systems Research 13(1), 47–64. Streicher, G., and R. Stehrer (2012). Whither Panama? Constructing a Consistent And Balanced World SUT System Including International Trade And Transport Margins. WIOD Working Paper 13. URL: http://www.wiod.org/publications/papers/ wiod13.pdf. Stone, R. (1942). The Precision of National Income Estimates. The Review of Economic Studies 9(2), 111–125. Stone, R., D. G. Champernowne and J. E. Meade (1942). The Precision of National Income Estimates. Review of Economic Studies 9(2), 110–125. Timmer, M., A. A. Erumban, J. Francois, A. Genty, R. Gouma, B. Los, F. Neuwahl, O. Pindyuk, J. Poeschl, J. M. Rueda-Cantuche, R. Stehrer, G. Streicher, U. Temur- shoev, A. Villanueva and G. J. de Vries (2012). The World Input–Output Database (WIOD): Contents, Sources And Methods. WIOD Background Document. URL http:// www.wiod.org/. United Nations Statistics Division (2011). ‘International Trade Information Systems in 2020’, Global Forum on Trade Statistics, Geneva, 2–4 February 2011. Background Note by UNSD, Eurostat and WTO. van der Ploeg, F. (1984). General Least Squares Methods for Balancing Large Systems and Tables of National Accounts. Review of Public Data Use 12, 17–33. Wang, Z. (2011). A Model to Construct Time Series Databases for Global Trade, Pro- duction, and Consumption Linkages. Paper presented at World Bank Workshop on 288 Trade in Value Added ‘Fragmentation of Global Production and Trade in Value Added: Developing New Measures of Cross Border Trade’, 9–10 June 2011, Washington DC. Wang, Z., M. Gehlhar and S. Yao (2010). A Globally Consistent Framework for Reliability-Based Trade Statistics Reconciliation in the Presence of an Entrepôt. China Economic Review 21(1), 161–189. Wang, Z., M. Tsigas and M. Gehlhar (2012). How a Global Inter-Country Input–Output Table with a Processing Trade Account is Constructed from the GTAP Database. Paper Presented at the 15th Annual Conference on Global Economic Analysis, Geneva, Switzerland. Weale, M. R. (1985). Testing Linear Hypotheses on National Account Data. Review of Economics and Statistics 67, 685–689. 11 Direct Measurement of Global Value Chains: Collecting Product- and Firm-Level Statistics on Value Added and Business Function Outsourcing and Offshoring TIMOTHY J. STURGEON, PETER BØEGH NIELSEN, GREG LINDEN, GARY GEREFFI AND CLAIR BROWN 1 INTRODUCTION: WHY NEW FIRM-LEVEL STATISTICS ON VALUE ADDED AND INTERNATIONAL SOURCING ARE NEEDED NOW International trade and foreign direct investment have long been central fea- tures of the world economy, but their importance has been growing rapidly, especially since the late 1980s. Alongside this quantitative change, a qual- itative shift has also been taking place. Because of advances in information technology, which enable business processes to be segmented and potentially relocated, and the rise of industrial capabilities in less developed countries, which offer more options for relocating them, the production of goods and services has become increasingly fragmented across borders. In other words, it has become more common for value to be added to a product in two or more countries prior to final use in both goods- and services-producing industries. The emergence of global value chains (GVCs) 1 of this sort has led researchers 1 Researchers studying this structural shift in the global economy have generated a very long list of terms to describe it. The international trade literature has stimulated a vast body of research and multiple labels, including a new international division of labour (Frö- bel et al 1980), multistage production (Dixit and Grossman 1982), slicing up the value chain (Krugman 1995), the disintegration of production (Feenstra 1998), fragmentation (Arndt and Kierzkowski 2001), vertical specialisation (Hummels et al 2001; Dean et al 2007), global production sharing (Yeats 2001), offshore outsourcing (Doh 2005) and integrative trade (Maule 2006). The enduring structures that embody these new forms of trade and investment have been referred to as global commodity chains (Gereffi 1994; Bair 2009), global production networks (Borrus et al 2000; Henderson et al 2002), international sup- ply chains (Escaith et al 2010) and global value chains (GVCs), the term we will use here (Humphrey and Schmitz 2002; Kaplinsky 2005; Gereffi et al 2005; Kawakami 2011; Catta- neo et al 2010). 290 Trade in Value Added and the providers of official economic statistics to acknowledge a growing knowledge gap in regard to the flow of intermediate goods and services and the location of value added. Why is this important? It used to be safe to assume that all of an import’s value was added in the exporting country. This gave trade statistics a great deal of analytic value and policy relevance. In this simpler world, indus- trial capabilities could be judged by the quality and technological content of exports, trade rules could be tied to gross levels of trade in specific products or product sets, and exports could be directly related to domestic job creation. ‘Rules of origin’ labelling requirements are also based on the assumption of nationally bounded production, but today it is difficult to know what labels such as ‘made in China’ or ‘made in the USA’ really mean. With GVCs compli- cating the picture, we simply cannot know what share of an imported prod- uct’s or service’s value is added in the country that declares it as an export, and thus, we are less able to judge that country’s level of development from the technological sophistication of its exports, following Lall (2000). Flows of intermediate goods provide hints about the structure of GVCs (see Feenstra 1998; Brülhart 2009; Sturgeon and Memedovic 2010), but because we do not generally know how imported inputs are used in specific products, or how they are combined with domestic inputs and value added, it is not possible to extract concrete information about the geographic distribution and flow of value added from trade statistics alone. What is certain is that using the gross value of trade as a yardstick dis- torts our view of where in the world industrial capabilities lie, creates uncer- tainty about the fairness of trade agreements and even calls into question such fundamental measures as gross domestic product (GDP) and produc- tivity (Houseman 2011). These data and policy gaps have triggered innova- tive efforts to link national input–output (IO) tables into larger international (global and regional) input–output tables (IIOs) that researchers can use to estimate trade in value added, among other things (OECD 2011b). With data of this sort, we can begin to answer the question ‘who wins and who loses from globalisation?’ from the supply side (ie winners and losers in terms of value added, value capture and employment), rather than only the demand side (ie winners and losers in terms of consumer prices versus jobs and wages). Despite the progress that IIO tables represent, the estimation and cross- border harmonisation required to construct them decrease detail and accu- racy. National IO matrices, in countries where they exist, are based on very partial data to begin with, and rely on a range of inferences and (sometimes controversial) assumptions, such as the proportionality of imported inputs across all sectors (Grossman and Rossi-Hansberg 2006; Winkler and Milberg 2009). When national IO data sets are linked across borders, these problems are compounded as industry categories are harmonised at high levels of aggre- gation and additional layers of assumption and inference are added to fill in Direct Measurement of Global Value Chains 291 missing data. Statisticians must ‘cook the books’ to bring IO tables from mul- tiple countries into alignment. Such data gaps are especially acute in services, where product detail is sorely lacking and vast inferences are made to settle national accounts. 2 Almost all of the defining features of services (that is, they are non-tradeable, non-storable, customised and insensitive to price competition) are changing in ways that enable and motivate the formation of GVCs. As a result, task frag- mentation and trade in services are burgeoning, both domestically and inter- nationally, through the twin processes of outsourcing and offshoring. Com- puterisation is allowing a growing range of service tasks to be standardised, codified, modularised and more readily and cheaply transmitted among indi- viduals and organisations that might be at great distance from one another. Clearly, the assumptions behind current data regimes have changed and statistical systems are struggling to catch up. In this chapter, we confront the obvious. It will be exceedingly difficult to fill data gaps without new data. Using existing data in new ways, including generating groupings of traded products that better reflect GVCs, (see, for example, Sturgeon and Memedovic 2010) and linking ‘microdata’ from surveys to administrative sources such as business registers (see, for example, Bernard et al 2005, 2006; Nielsen and Tilewska 2011) can lead to new insights, but they may never be enough. Statis- tical analysis that relies solely on existing data sources will always reflect the limits of the content of surveys and data sources. New data will be needed and, because GVCs are by definition a cross-border phenomenon, interna- tional standardisation will be essential. At the same time, resources for data collection and the political will required to burden private sector respondents with surveys are declining in many countries. Clearly, current priorities will need to be adjusted so new data can be collected without unduly increasing the burden on respondents. 2 Why are the data resources related to services so poor? One reason is that the data are difficult to collect. While companies might track the source of every physical input to man- ufacturing, for warranty or quality control purposes, services expenditures are typically grouped into very coarse categories, such as ‘purchased services’. The absence of tariffs on services, and their non-physical character, mean that when service work moves across bor- ders, no customs forms are filled out and no customs data are generated. Another reason is that service work has historically been thought to consist of non-routine activities that require face-to-face contact between producers and users. Services as different as haircuts and legal advice have traditionally been consumed, in place, as soon as they are produced. The customised and ephemeral nature of many services has led them to be considered ‘non-tradeable’ by economists, or at least very ‘sticky’ in a geographic sense relative to the production of tangible goods. Finally, services have long been viewed as ancillary to manu- facturing, either as direct inputs (eg transportation) or as services provided to people who worked in manufacturing (eg residential construction, retail sales, etc ). As such, services have been viewed as a by-product, not a source, of economic growth. Thus, data collection on services has historically been given a low priority by statistical agencies (Sturgeon et al 2006; Sturgeon and Gereffi 2009), although the need for statistical evidence for policymak- ing has been clearly articulated (Commission of the European Communities, 2003). 292 Trade in Value Added While collecting new data on a globally harmonised basis, for this is what is needed, is a daunting task, we need to begin to test the results of research using IIOs with standardised case studies and proof-of-concept surveys, and, eventually, to replace inferred data with real data in both goods- and services- producing industries. The solution will inevitably include new ‘bottom-up’ business surveys to complement the ‘top-down’ efforts of IIOs. This chapter outlines two such efforts: product-level GVC studies and business function surveys. 2 PRODUCT-LEVEL GVC STUDIES The most direct way to measure the geography of value added is to decompose individual goods and services into their component parts and trace the value added of each stage of production to its source. The procedure yields product- level estimates that identify the largest beneficiaries in terms of value added, value capture (ie profits) and employment. Beneficiaries can be firms, workers, countries or all of the above. Studies in this vein have shown that China’s export values often bear little relation to domestic value added because many exported products contain expensive imported inputs, and the lion’s share of profits tends to be captured upstream from production, in the design and branding activities of the ‘lead firm’ in the value chains, and downstream by distributors, value-added resellers, and retailers. This situation is common when assembly is performed by domestic or foreign-owned contract manufacturers on behalf of multinational brand name or ‘lead’ firms, a pattern of industrial organisation that has been a key driver of economic development in China, elsewhere in developing East Asia, and other places in the world with deep linkages to GVCs, such as Eastern Europe and Mexico (Grunwald and Flamm 1985; Gereffi and Korzeniewicz 1994; Bor- rus et al 2000; Sturgeon and Lester 2004). Because foreign components are commonly specified in designs worked out in the lead firm’s home country, key components and subsystems are often sourced from vendors close to the lead firm, in addition to a palette of well-known component suppliers from countries across the globe. In technologically intensive industries and value chain segments, these supplier and component manufacturing firms tend to be concentrated in OECD or newly industrialised countries, especially Taiwan (Chinese Taipei). To add to the complexity of GVCs, each of these supplier firms might outsource production or have an affiliate in a third country, in a pattern Gereffi (1999) refers to as ‘triangle manufacturing’. Product-level GVC studies are designed to shed light on where value is added and captured in these complex cross-border business networks. The first product-level GVC study, on a specific Barbie doll model, appeared in the Los Angeles Times (Tempest 1996). The Barbie case was then included in a classic paper by trade economist Robert Feenstra (1998) to bolster his Direct Measurement of Global Value Chains 293 Table 11.1: The location of value added and capture for a ‘Tea Party Barbie’ doll, 1996. Production, inputs and contract management Value ($) Materials 0.65 Saudi Arabia: Oil Hong Kong: management, shipping Taiwan (Chinese Taipei): refines oil into ethylene for for plastic pellets for Barbie’s body Japan: nylon hair US: cardboard packaging, paint pigments, moulds Production: China (factory space, labour, electricity) 0.35 Overhead and coordination of production and 1.00 outbound shipping: Hong Kong 1.00 Export value (factory price): 2.00 US: shipping, US ground transportation, wholesale and retail markups 6.99 US: Mattel Inc. (lead firm: design, marketing) 1.00 US retail price: 9.99 Sources: Tempest (1996) from US Commerce Department, Chinese Ministry of Foreign Trade Economic Cooperation, Mattel Inc., Hong Kong Toy Council. argument that the rise of intermediate goods trade was caused, in part, by ‘the disintegration of production in the global economy’ leading to double counting of intermediate goods as they wended their way through interna- tional production networks. The findings of this widely publicised case are summarised in Table 11.1, which shows that only 35 cents (3.5%) of the value of a US$10 ‘Tea Party’ Barbie doll (3.5%) was added in mainland China, where it was assembled, largely of imported materials. The lead firm most commonly used in subsequent product-level GVC research is Apple Inc., the company behind the popular iPod, iPhone and iPad consumer electronics devices, as well as the Macintosh line of personal com- puters (Linden et al 2007, 2009 2011; Hesseldahl 2010). Most recently, the OECD (2011b, p. 40), examining the sources of components for a late-model Apple smartphone (the iPhone 4) that retails for about $600, estimates that only $6.54 (3.4%) of the total factory price of $194.04 was actually added in China, where the product is assembled by the Taiwanese electronics contract manufacturer Foxconn. This is because $187.50 (96.6%) of the factory cost came from imported materials and components, most notably from South Korea, the USA and Germany. Analysis of traded goods from other electronics firms has yielded similar results. For example, a study of a 2005 Hewlett-Packard (HP) notebook com- puter model (model nc6230) found that none of the major components origi- nated in China, where a Chinese Taiwan-based contract manufacturer assem- bled it (Dedrick et al 2010). Yet the full factory price of $856.33 would have counted as part of the gross value of mainland Chinese exports. Ali-Yrkkö et al (2010) obtained similar results in their study of a Nokia mobile-phone handset. 294 Trade in Value Added China $30, 4% US $334, 39% Japan $286, 33% Rest of World $171, 20% Korea $35, 4% Figure 11.1: Geography of value added in a Hewlett-Packard notebook computer. Source: based on Dedrick et al (2010, Table A-3). The factory cost of the product in 2005 was $856. The amounts shown for each country, except China, are the total cost of inputs from firms headquartered in that country. No inputs came from Chi- nese companies, so the $30 assigned to China is an estimate of value added that was subtracted from the cost of inputs from ‘Rest of World’. Clearly export value is a highly misleading measure of China’s benefit from export trade. A more meaningful measure of the benefit to China’s economy would be calculated in value-added terms. A simple approximation of value added is the sum of operating profit, direct labour wages, and depreciation. Going back to the study of the HP notebook computer by Dedrick et al (2010), because there were no Chinese firms among the major suppliers, China earned no profit (and thus booked no depreciation related to this product). That leaves direct labour as a source of value added. The cost of assembly and test, which took place in China and is mostly wages, came to $23.76, some of which would be retained as profit by the Taiwanese assembly company. Some of the smaller inputs may have received final processing in China, but this typically amounts to a very small percentage of value added, no more than a few dollars in this case. On this basis, Dedrick et al estimate that China’s value added to this product at $30. In this example, then, assigning China the full factory price of $856.33 overstates its value added by more than 2,800%! This is because $826.33 (96.5%) of the factory cost went to imported materials Direct Measurement of Global Value Chains 295 and components, mainly from firms based in South Korea, the USA and Japan (see Figure 11.1). Judging from prior research on similar GVCs (Sturgeon 2003), it is very likely that most if not all high-value components were specified by HP’s design group in the USA, and purchased by the company’s contract manufacturer under terms that HP negotiated directly with its main component suppliers. This underscores the powerful role played by HP—the ‘lead firm’ in the GVC— even though the company may have taken no physical ownership of work- in-process inventory. HP’s role is as a buyer of manufacturing and logistics services, a conceiver and marketer of the product and an orchestrator of the GVC. While this role allows HP to extract the lion’s share of profit from the ultimate sale of the computer, it is mostly or even entirely invisible in trade statistics. This creates a difficult methodological problem. To fill in this gap Linden et al (2009, 2011) estimated value added and employment in upstream activities, such as research and development (R&D) and marketing, from the ratio of the target product’s sales in total firm revenues. One outcome of this exercise was an estimate that the share of US-based employees in the total iPod-related wages (from R&D to retail) paid worldwide in 2006 was 70%, considerably higher than the estimated share of US-based companies in the global distribution of gross profit from the iPod hardware alone. Product-level GVC studies typically look only one value-chain level upstream from final assembly. However, a sub-system company may produce or purchase high value sub-assemblies and components in a third country (eg Singapore and Malaysia are common locations for the production of head assemblies for hard drives). Estimates of the actual geography of value added must be made, and these require a great deal of industry knowledge. In IO analysis, industry knowledge is not required because both direct and indirect value added for any imported or domestic intermediate inputs are taken into account as a standard part of the estimates. However, as discussed below, GVC analysis can potentially separate the geographical assignment of the two chief elements of value added (wages and profits), whereas IO analysis cannot. The focus of the product-level GVC research cited in this section is on highly popular consumer electronics products such as those from Apple, Hewlett- Packard and Nokia. 3 This is no accident, since the research mainly relies on data from private consulting firm ‘teardown reports’ itemising and nam- ing the suppliers of the high-value components used in each product. These reports are based on physical disassembly and examination of component parts. Because such reports are available for only the most high profile items, product-level GVC study methods have been difficult to generalise. Moreover, the electronics products that teardown reports analyse typically contain hun- dreds of clearly identifiable components with relatively transparent world 3 An exception is a set of five case studies from the shoe industry conducted by the Swedish National Board of trade (2007). 296 Trade in Value Added For the finished product... 1. Make, model/SKU and average selling price of the product. 2. Value when it leaves the factory (‘factory price’). 3. The percentage of factory costs accounted for by ‘materials’, ‘labour’ and ‘other (specify)’. 4. List of top material inputs (however many it takes to For each of the inputs... account for 75–80% of factory costs), typically listed in the BOM. 1. Short description. 5. The cost of assembly (converting inputs into final products) 2. Name of as they were in a specific time period (eg late 2010) when the manufacturere/supplier. product was being made. 3. Country where 6. Approximate number of units manufactured in the manufactured. specified period. 4. Average cost (price) 7. Share of shipments within the specified period to each of input to company type of recipient (eg direct to consumer, OEM customer, assembling the product distributor, value-added resellers, retailers). in the specified 8. Share of shipment in 2010 by country or regional location time period. (eg USA, Japan, China, other Asia, Europe, other North America). Figure 11.2: Basic data needed for product-level GVC studies. prices. The most valuable components tend to bear the names of their manu- facturers, and can thus be traced to their country of manufacture. Studies of automobiles, which have many model-specific parts without published prices, or apparel products made from fabrics that might have been produced by a number of suppliers in multiple locations, are more difficult to decompose and value after the fact. Asking firms for the data directly is possible, but most firms tend to be unwilling to share this sort of strategically sensitive information with researchers, even with assurances of confidentiality. Despite the difficulties of extending the method to different industries, product-level GVC studies continue to proliferate. Although it has not yet been used in published work, we are aware of several active research projects that are using the product-level GVC approach to study a variety of industries, including wind turbines and other mechanical products, small cc motorcycles and women’s apparel. For consistency and comparability, a standardised, or least mutually compatible, approach is needed. In the interest of moving in this direction, we specify a set of research requirements for product-level GVC studies below. The best-case approach we lay out here assumes full coopera- tion or mandatory compliance by participating firms. While such compliance may be difficult or even impossible to come by, our goal is to set a high ini- Direct Measurement of Global Value Chains 297 tial standard that can be adjusted in the face of pragmatic considerations. Ideally, factory prices and costs would be directly from manufacturing com- panies, at the point of production, or from some other corporate office where data itemising the bill of materials (BOM) for specific products is held. A BOM typically designates the part number (or other designation) and cost of each input. The basic data needed to collect information on value added at the product level are presented in Figure 11.2. First, the product needs to be identified, either by its make and model or by its stock keeping unit (SKU) number. Then, the factory price of the prod- uct is collected, along with internal costs for labour, materials and other costs (mostly overhead) directly related to production. Then, a list of the most valu- able materials and other inputs, perhaps derived from the BOM, is collected. The next step is to estimate the profit margins and/or employment asso- ciated with the final product and with each of the key inputs. If the analysis extends to the retail end of the value chain, then data about the structure and geography of sales channels (items 7 and 8 in Figure 11.2) should also be analysed and the average selling price at retail estimated. As this brief description shows, the data requirements for a product-specific analysis are considerable. Again, the data are often hard to obtain because of their com- mercial sensitivity and the results are difficult to generalise because they only represent a single product. An approach that avoids targeting a single product or company is the use of average breakdowns of component values for a generic product type (eg notebook PC, 2 MW wind turbine). Sometimes data of this sort can be obtained through industry associations willing to cooperate with researchers by requesting data from their membership. These average values can be com- bined with qualitative value chain analysis (see Gereffi and Fernandez-Stark 2011) to identify the industry’s key lead firms and main suppliers. With this information it is possible to construct industry- or subsector-level estimates of the geography of value capture. Again, although it has not yet been used in published work, we are aware of active research using this approach. As we mentioned earlier, product-level GVC studies can complement stud- ies using official statistics. For example, Koopman et al (2008) combine stan- dard IO tables with information that separates processing and normal trade, all from official sources in China. This study estimates that about half of the gross value of total Chinese exports is derived from imported inputs, rising to 80% for technology-intensive sectors such as electronics. For export process- ing production as a whole, primarily consisting of products branded by non- Chinese firms, foreign value added was estimated to be 82% in 2006 (Koopman et al 2008, p. 19). These findings suggest that the product-level cases of iPods, iPhones, iPads and similar consumer electronics goods produced in China for export, may not be that extreme. 298 Trade in Value Added Again, the product-level approach makes it conceivable to go further and separate out the labour and profit components of value added. 4 Consider the example of a Japanese-branded hard disk drive assembled in China from imported parts before it is included in a notebook PC such as the Hewlett- Packard model nc6230 notebook computer discussed above. According to information from an executive in the hard drive industry, the value added attributable to hard drive assembly wages is about 7% ($4.76) of the $68 whole- sale price of the drive, and the value added corresponding to the Japanese firm’s gross profit is about 20% ($13.60). If all of the value added of the hard drive (ie 27% of the wholesale price, or $18.36) is assigned to China (assuming the drive was assembled there), then local value added is overestimated by nearly 300%. If, on the other hand, all of the value is assigned to Japan, then Japanese value added is only overstated by 35% and Chinese value added is underestimated by a relatively small amount. Since pragmatic considerations may limit the number of value-chain levels in which these types of detail can be collected, it is clearly better to err on the side of assigning value to the country where the sub-system company is headquartered in industries where labour accounts for a much smaller share of value added than does profit. International IO studies, however, would do the opposite, assigning all the value added to the location where the work is performed. Product-level GVC studies are demanding in terms of industry knowledge, but they are the only method to enable separate treatment of the labour and profit components of value added. They require knowledge of the headquar- ter locations of participating firms (for profit accounting) and their factory locations (for labour accounting) and must have a means to estimate the split between them. International IOs, by default, assign all the value added to the factory location. Despite the challenges, product-level studies are worth per- forming from time to time as a check on the robustness of measures of the distribution of value from world trade that are derived from official statistics. Product-level GVC studies are important not only because they suggest that the local value in manufactured goods exports can be vastly overstated, but also because exports may overstate the exporting country’s technological attainments. Goods manufactured in developing countries are often leading edge in terms of markets and technology. Hence, the technological sophistica- tion and competitive stature of an exporter’s industrial base can be exagger- ated when exports are used as a measure of industrial capability. Not only are most technology-intensive parts produced in industrialised countries, but so too are the ‘knowledge work’ and the intangible assets involved in system-level design, product strategy, marketing, brand management and supply chain orchestration. 4 Value added is the difference between the selling price and the cost of acquired inputs. In practice, however, this is equal to some measure of profit plus wages plus some account- ing values such as depreciation. Direct Measurement of Global Value Chains 299 This is important not only for the value that these activities create, but also because they are the key elements in competitive performance, innova- tion and new industry creation: the bedrock of economic development. Even the cutting-edge production equipment and logistics systems used for the manufacture of products such as notebook computers and smart phones are not ‘native’ to mainland China or other less developed countries in East Asia, but implanted there by firms based in Taiwan (Chinese Taipei), South Korea and OECD countries (Steinfeld 2004). This has important policy implica- tions. While product-level GVC studies suggest that the competitive ‘threat’ to advanced economies posed by indigenous Chinese capabilities may be vastly overstated, not only in the popular press but in policy circles, massive exports do reflect large-scale employment, even if they are based on non-indigenous innovations and market success. The result could be an increasing disjunc- ture between innovation and employment that will lead, if not to wholesale economic decline, at least into uncharted waters. 3 BUSINESS FUNCTION SURVEYS There is a pervasive dynamic working against the usefulness of current busi- ness statistics. On the one hand, production is becoming increasingly bundled with services. On the other hand, it has become easier to fragment the value chain geographically. Thus, value added cannot be fully determined by tally- ing up the physical inputs to products listed as outputs. A range of largely intangible ‘support’ functions (eg R&D, sales, marketing, IT systems) also add value and, like production, these support functions are available from suppli- ers and service providers outside the firm and in a variety of locations around the world. Thus, GVCs are expanding the arena of sourcing and competition beyond main products to the vertical business function that can be offered (horizon- tally, to diverse customers) as more or less generic goods and services within and across industries. Firms not only outsource the assembly of goods, and source tangible inputs in GVCs (as captured by product-level GVC studies), but they increasingly outsource and sometimes even offshore intangible ser- vices and support functions as well. These include IT services, back-office work such as payroll and accounting, call centres for sales or customer sup- port, and even engineering and elements of R&D (Dossani and Kenney 2003; Gereffi and Fernandez-Stark 2010). We argue that these trends require a new statistical unit of analysis to sup- plement the main activity/industry of the firm—ie the business function— and new surveys to capture how they are sourced and to quantify their value. Business function surveys are ideal for collecting data on the location of value added for three reasons. First, because they consist of intangible services, the value added by support functions has proven very difficult to capture, classify 300 Trade in Value Added and quantify. Second, the parsimony of business function lists (see Box 11.1) reduces respondent burden, while still generating data that can be compared and aggregated across firms, countries and industries. In fact, the business function approach does away with any hard distinction between goods- and services-producing firms. The primary output of a firm may be a good or a service, but the array of support functions that may or may not be done by the firm are roughly the same. Third, experience with ground-breaking sur- veys (Brown 2008) suggests that data quality tends to be high because busi- ness functions are in keeping with the way many managers think about and account for their operations. Box 11.1. Seven business functions used in the european survey on interna- tional sourcing. 5 In the European International Sourcing survey, seven busi- ness functions (plus a residual ‘other’ category) were identified using the Euro- pean Central Product by Activity (CPA) classification. 1. Core/primary business functions: production of final goods or services intended for the market or third parties carried out by the enterprise and yielding income. The core business function usually represents the primary activity of the enterprise. It may also include other (secondary) activities if the enterprise considers these to comprise part of its core functions. 2. Support business functions: support business functions (ancillary activi- ties) are carried out in order to permit or facilitate production of goods or services intended for sale. The outputs of the support business func- tions are not themselves intended to be directly for sale. The support business functions in the survey are divided into the following. (a) Distribution and logistics: this support function consists of trans- portation activities, warehousing and order processing functions. In figures and tables, ‘distribution’ is used as an abbreviation for this function. (b) Marketing, sales and after-sales services including help desks and call centres: this support function consists of market research, advertising, direct marketing services (telemarketing), exhibitions, fairs and other marketing or sales services. It also includes call- centre services and after-sales services, such as help desks and other customer support services. In figures and tables ‘marketing, sales’ is used as an abbreviation for this function. (c) Information and communications technology (ICT) services: this support function includes IT services and telecommunications. IT services consist of hardware and software consultancy, customised software data processing and database services, maintenance and Direct Measurement of Global Value Chains 301 repair, web-hosting, other computer related and information ser- vices. Packaged software and hardware are excluded. In figures and tables ‘ICT services’ is used as an abbreviation for this function. (d) Administrative and management functions: this support function includes legal services, accounting, bookkeeping and auditing, business management and consultancy, HR management (eg train- ing and education, staff recruitment, provision of temporary per- sonnel, payroll management, health and medical services), corpo- rate financial and insurance services. Procurement functions are included as well. In figures and tables ‘Administration’ is used as an abbreviation for this function. (e) Engineering and related technical services: this support function includes engineering and related technical consultancy, technical testing, analysis and certification. Design services are included as well. In figures and tables ‘Engineering’ is used as an abbreviation for this function. (f) Research & Development: this support function includes intramu- ral research and experimental development. In figures and tables ‘R&D’ is used as an abbreviation for this function. Not only is the business function classification useful for tracing the organ- isational and geographic location of value added, but also as a high-level stand-in for occupational categories, since jobs can also be tallied according to their general function within the organisation. Since the business func- tion approach aggregates product and services into a limited number of well- defined categories, it has proven feasible for large-scale surveys. Two of these implementations are described in some detail in the latter sections of the chapter. 3.1 Business Function Lists We are only just beginning to develop standard methods for collecting eco- nomic data according to business functions. In this section we provide some examples from recent and current surveys. Firms or their main operations units 6 typically have a main output, be it a good or service. In a statistical context, the function that produces this out- put typically determines the firm’s industry classification using standardised activity/industrial codes such as its ISIC, NACE or NAICS classification. Instead of counting all output and employment under this main output classification, 6 Large firms may have several distinct operational units with distinct outputs. These are variously called divisions, lines of business or business segments. For such firms it is sometimes best to collect data at this level. 302 Trade in Value Added as business censuses typically do, business function surveys supplement the primary output function with a standardised, generic list of support func- tions (see Box 11.1). In other words, firm-level data (eg occupational employ- ment, wage levels paid, internal, external and international sourcing costs) is collected for specific functions rather than for the firm as a whole. In the business function frameworks developed so far, the main productive func- tion of the firm has been designated variously as ‘production’ (Porter 1985), the ‘core function’ (Nielsen 2008), ‘operations’ (Brown 2008) and the ‘primary’ business function (Brown and Sturgeon, forthcoming). Even if the terminology used differs, the approach is similar in the sense that it distinguishes between the primary business function and a generic list of functions that ‘support’ it. Conceptually, Michael Porter pioneered the business function approach. In his 1985 book, Competitive Advantage, he identified a list of nine generic business functions: R&D; design; production; marketing and sales; distribu- tion; customer service; firm infrastructure; human resources; and technology development. To our knowledge, the earliest use of a business function list to collect eco- nomic data was for the EMERGENCE Project (Huws and Dahlman 2004), funded by the European Commission. This research used a list of seven business func- tions tailored to collect information about the outsourcing of information- technology-related functions, such as software development and data pro- cessing. Such industry-specific bias in business function lists can simplify data collection and focus research on specific questions (such as IT outsourcing), but the results cannot be easily compared with or aggregated with other data, and they increase the risk of creating non-exhaustive lists. When business function lists are non-exhaustive, they leave some functions unexamined and block a comprehensive firm-level view of employment or value added. Again, while non-exhaustive business function lists are useful for examining specific business practices and firm-level characteristics, they are not well suited for general use as a parsimonious alternative for, or supplement to, industry and occupational classifications. An exhaustive list similar to Porter’s was devel- oped for the European Union (EU) Survey on International Sourcing (Nielsen 2008) and adopted by Statistics Canada for the 2009 Survey of Innovation and Business Strategy (SIBS) 7 (again, see Box 11.1). 8 Business function data can be used to inform a wide variety of research and policy questions. For example, they can be used to characterise patterns of business function bundling in respondent firms (ie organisational design 7 See http://www.ic.gc.ca/eic/site/eas-aes.nsf/eng/h_ra02092.html. 8 In contrast, the EMERGENCE project list (Huws and Dahlman 2004) and a more recent list developed by the Offshoring Research Network for the purpose of detecting R&D off- shoring (Lewin et al 2009) did not include a category for the firm’s main operational func- tion, but instead used a list of commonly outsourced functions (product development, IT services, back-office functions, call centres, etc ). Again, non-exhaustive lists of this sort cannot provide a full picture of firm organisation or sourcing patterns. Direct Measurement of Global Value Chains 303 as indicated by employment or costs/revenues by function), to collect data on wages by function as a high-level stand-in for detailed data on occupa- tional employment and, critically for the purposes of this volume, to exam- ine firm-level patterns of domestic and international sourcing (value added). Potentially, business function lists might supplement, or even partially sub- stitute for, the long lists of industry-specific product trailers that underlie IO tables in settings with severe resource constraints. The main strength of the business function approach is its potential to identify and measure support activities and other intangible assets within the firm (R&D or customer service capabilities) in a way that is easily comparable across sectors and countries. 3.2 Using Business Function Surveys to Collect Data on External and International Sourcing: The Eurostat International Sourcing Survey Sam: fix cls file! This section provides some illustrations of business function data from the 2007 Eurostat International Sourcing Survey (Nielsen et al 2008). The results show how business function surveys can provide insights into a complex and hard-to-research topics such as international sourcing. The survey was an economy-wide ad-hoc survey carried out by 12 European countries in 2007, covering the so-called non-financial business economy. The survey asked about sourcing decisions made by European firms in the period 2001–6. The focus of the survey was on larger enterprises, as multinational groups of enterprises were considered to be the key players and drivers for international sourcing. A bottom threshold of 100 or more employees was used, although statistical offices in several countries decided to lower the threshold to enterprises with 50 or more employees. This section uses the information from 4–12 European countries, based on data availability. The survey did not ask respondents to quantify the value of their external and international sourcing, only to indicate if they had made such choices or not. (However, subsequent business function surveys have quantified the value of sourcing by business function, as we will see in the following section.) For the 12 European countries listed in Figure 11.3 the 2007 Eurostat Inter- national Sourcing Survey found that 16% of the enterprises with 100 or more employees had sourced one or more business function abroad. More than twice as many enterprises in Ireland and the United Kingdom did so (38% and 35%, respectively). The two small and open Nordic economies, Denmark (25%) and Finland (22%), were also significantly above the average. Germany (13%) was just below the average. Figure 11.3 shows the frequency of international sourcing for R&D and engineering functions. The business function most frequently outsourced internationally was the core (primary) function. Interestingly, the core business function is the only function sourced more frequently internationally than domestically. This was especially true for manufacturing firms in high wage countries such as Den- mark. More surprisingly, R&D was as frequently sourced internationally as it was domestically. 304 Trade in Value Added Engineering functions R&D functions Czech Republic Norway Italy Slovenia Finland Netherlands Ireland Denmark Portugal 0 5 10 15 20 25 30 35 40 45 Share of enterprises (%) Figure 11.3: R&D and engineering functions sourced internationally by enterprises in selected European countries, 2001–6. Source: Eurostat report data, http://epp.eurostat.ec.europa.eu/statistics_explained/ index.php/Global_value_chains_-_international_sourcing_to_China_and_India. In the four Northern European countries listed in Table 11.2, the study found that 30–40% of the firms surveyed made decisions to source support functions internationally. Manufacturing enterprises sourced a variety of sup- port functions internationally, but engineering, distribution and ICT func- tions were the most common. Compared to manufacturing enterprises, ser- vice enterprises were more likely to keep their core function in-house while sourcing support functions internationally, as shown in Table 11.3. For ser- vices enterprises, the functions most commonly sourced internationally are ICT and administration. 3.3 Using Business Function Surveys to Shed Light on the Relationship Between International Sourcing and Employment International sourcing has mainly been perceived as a driver of lower-skilled job loss, especially in labour-intensive manufacturing activities, such as prod- uct assembly. Indeed, as we have just shown, the 2007 Eurostat International Sourcing Survey found that manufacturing enterprises were more likely to be engaged in international than other enterprises. Why are some jobs vulnerable Direct Measurement of Global Value Chains 305 Table 11.2: Business functions sourced internationally by manufacturing enterprises in selected European countries, 2001–6: share of enterprises carrying out international sourcing (%). Denmark Finland Netherlands Norway Core/primary function 70 71 73 60 Distribution 20 21 17 13 Marketing and sales 12 23 15 13 ICT services 17 21 25 12 Administration 9 14 19 11 Engineering 22 11 7 17 R&D 14 10 15 7 Other functions 5 2 2 20 Source: Nielsen (2008). Enterprises have 50 or more employees, except for the Netherlands, covering 100 or more employees. Table 11.3: Business functions sourced internationally by services enterprises in selected countries, 2001–6: share of enterprises carrying out international sourcing (%) Denmark Finland Netherlands Norway Core/primary function 28 39 42 16 Distribution 28 18 27 7 Marketing and sales 24 28 10 27 ICT services 41 33 27 42 Administration 30 30 25 37 Engineering 17 9 4 11 R&D 17 21 11 7 Other functions 6 10 3 12 Source: Nielsen (2008). Enterprises have 50 or more employees, except for the Netherlands, covering 100 or more employees. to international sourcing while others are less so? Economists have developed a variety of measures based on occupational or job characteristics to deter- mine the ‘offshorability’ of jobs (Kletzer 2009; Blinder and Krueger 2009). In one example of this approach, survey respondents were directly asked about the difficulty of having their work performed by someone in a remote location (Blinder and Krueger 2009). Based on the worker’s description of his or her job tasks, the researchers decided how ‘offshorable’ each job was by using professional coders to rank the ‘offshorability’ of each occupation. Another example identified a list of US occupations (at the three-digit level) that are ‘potentially affected by offshoring’ based on ‘offshorability attributes’ of occu- pations, including the use of information and communication technologies, the use of highly codifiable knowledge and the degree of face-to-face contact (van Welsum and Reif 2009). The most sophisticated attempt to classify jobs according to their vulner- 306 Trade in Value Added ability to trade is the movability index (‘M Index’) developed by Jensen and Kletzer (2006). The M Index uses the detailed job descriptions in the Occu- pational Information Network (O∗ NET) database 9 that describe the degree of face-to-face customer contact, use of codifiable information and appearance of Internet-enabled work processes to characterise work in specific occupa- tions. They assign a value to each six-digit occupational code based on an examination of the O∗ NET job description and researchers’ characterization of how movable the occupation is. The M Index is based upon eleven job char- acteristics divided into two categories: information content (eg getting, pro- cessing, analysing information; Internet enabled) and job process (eg face-to- face contact; performing or working directly with the public; routine nature of work in making decisions and solving problems). A similar concept is behind the literature on ‘trade in tasks’, which also uses O∗ NET descriptions to con- sider which work tasks are vulnerable to relocation (see, for example, Gross- man and Rossi-Hansberg 2012). However, there is a fundamental conceptual flaw in using individual tasks and jobs as a unit of analysis in determining how easy it is to fragment and relocate work in the context of geographically extensive, yet operationally integrated production networks. Qualitative field research on how companies set up GVCs (see, for example, Dossani and Kenney 2003; Berger et al 2005) suggests that the processes of outsourcing and offshoring are rarely domi- nated by the shift of individual jobs to distant locations or outside suppliers. Although it is certainly possible, 10 this is even less likely with individual tasks. More common is the outsourcing (and possible offshoring) of larger groups of employees working on a coherent body of activities, such as manufacturing, accounts payable or after-sales service. In other words, it is more likely that business functions will be outsourced, rather than individual jobs and tasks. The character (tacitness versus codifiability) of the tasks, jobs and occupa- tions may be far less important than the character of the linkages between domestic and foreign operations, ie if instructions and requirements can be easily and clearly transmitted to the remote work site, as well at the ease of transferring the output to the following stage in the value chain. The busi- ness function may require the exchange of a great deal of tacit information, but as long as those exchanges occur within the work group and the inbound and outbound information flow can be codified and transported efficiently, the function can be readily outsourced and offshored, all other factors being equal (eg there has to be enough competence in the supply base to take on the function, following Gereffi et al 2005). 9 The O∗ NET, formerly the Dictionary of Occupational Titles (DOT), is the US Bureau of Labor Statistics’ primary source for occupational information. See https://onet.rti.org/. 10 For example, incoming calls for customer service are sometimes routed to various call centres in different locations, depending on the customer’s question or value to the company (Askin et al 2007). Direct Measurement of Global Value Chains 307 125 120 115 Manufacturing firms internationally 110 sourcing support functions only 105 All firms internationally sourcing support functions only 100 All firms internationally 95 sourcing core functions 90 85 Manufacturing firms internationally sourcing core functions 80 2000 2001 2002 2003 2004 2005 2006 2007 Figure 11.4: Employment trends by type of function sourced internationally, Denmark, 2000–7. Source: Nielsen and Tilewska (2011). Based on median values of full-time equivalent number of employees. Index 2000 = 100. To be fair, not all of the literature on trade in tasks falls into the trap of equating job characteristics with ‘offshorability’. A study by Lanz et al (2011) estimates the task content of goods and services by combining information on 41 tasks from the O∗ NET database with information on employment by occupation and industry for large sets of occupations. This finds the tasks that can be digitised and offshored are often complementary to tasks that cannot. What is the evidence regarding employment from business function sur- veys? The 2007 Eurostat International Sourcing Survey found that 20–25% of all surveyed manufacturing enterprises sourced internationally, compared with about 10% of all enterprises in the other sectors of the economy. How- ever, concerns about job loss in Europe due to international sourcing could go beyond the issue of manufacturing job loss to knowledge-intensive job loss as well. The survey shows that around 10–15% of the enterprises that did source internationally in the period 2001–6 sourced R&D functions, as shown by Figure 11.3. Analysis of firm-level employment patterns in Denmark in the period 2000– 7, using an exercise linking data at enterprise level from the 2007 Euro- stat International Sourcing Survey to the Danish structural business statis- tics register, found differences between enterprises sourcing only their core function internationally, and those enterprises sourcing only support func- tions internationally (see Figure 11.4). This exercise shows that enterprises sourcing their core function internationally had a considerable decline in 308 Trade in Value Added their employment— down to an index of 93 in 2007—compared with the enterprises only sourcing support functions internationally, which increased employment to an index 108. Enterprises with no international sourcing at all increased employment even faster, to an index of 125. When manufacturing enterprises were analysed separately, this pattern was even more pronounced. Manufacturing enterprises internationally sourcing only core activities lost the most employees, down to an index of 86 in 2007. 3.4 Quantifying Value Added with Business Function Surveys: The 2011 National Organizations Survey Both economic theory and research based on extensive field interviews sug- gest that managers often experiment with a variety of ‘make’ or ‘buy’ choices and on- or offshore sourcing (Bradach and Eccles 1989; Berger et al 2005). Quantifying internal and external sourcing costs is important because firms can, and often do, combine internal and external sourcing of specific business functions. For example, the primary business function (eg component man- ufacturing or assembly) may be outsourced, but only when internal capacity is fully utilised. Or firms might combine internal and external sourcing for strategic reasons, such as pitting in-house operations against external sources for competition in the realms of cost, quality or responsiveness (Bradach and Eccles 1989). Combinations of internal and external sourcing might show a transitional phase of outsourcing, bringing work back in-house (sometimes referred to as insourcing), or building up new in-house functions, and quan- titative data collected over time can capture these trends. The same can be said of location. Managers can decide to locate business functions in proximate or distant locations, in high or low cost locations, near customers, suppliers, specialised labour markets, and so on, and sometimes they combine these approaches and motives. Figure 11.4 captures the four choices managers have in regard to combining the organisational and geo- graphic location of work: 1. domestic in-house (‘domestic insourced’ in EU terminology); 2. offshore in-house or foreign affiliate (‘international insourced’ in EU ter- minology); 3. domestic outsourced; and 4. offshore outsourced (‘international outsourced’ in EU terminology). The central question in GVC research, then, is not which of these four choices managers make, but how they combine them. Quantitative employment, wage and sourcing information by business func- tion was recently collected in the USA by the 2011 National Organizations Direct Measurement of Global Value Chains 309 Table 11.4: Organisation and offshoring: four possibilities. Location Organisation Domestic International Internal: function within EU terminology: domestic EU terminology: the enterprise or insourced international insourced enterprise group US terminology: domestic US terminology: offshore in-house in-house Function performed within Function performed within the enterprise or the enterprise or enterprise group within enterprise group outside the compiling country the compiling country (by affiliated enterprises) External: function outside EU terminology: domestic EU terminology: the enterprise or outsourced international outsourced enterprise group US terminology: domestic US terminology: offshore outsourced outsourced Function performed Production outside the outside the enterprise or enterprise or group and enterprise group by outside the compiling non-affiliated enterprises country (by non-affiliated and within the compiling enterprise, eg suppliers, country service providers, contractors) Source: Based on Nielsen (2008). Survey (NOS), funded by the National Science Foundation. 11 The purpose of the study is to generate direct comparison of domestic employment charac- teristics with outsourcing and offshoring practices. The 2011 NOS was admin- istered online and by telephone to a representative sample of US businesses, plus a sample of the largest US companies. The survey includes two randomly sampled frames: 900 organisations representative of total US employment linked to the General Social Survey (GSS), and a large firm sample of 975 busi- ness segments drawn from the largest companies in the USA (drawn from the 2009 list of ‘Fortune 1000’ firms), 12 referred to hereafter as the F1K. For these large firms, business segments (also known as divisions or lines of busi- ness) are used rather than the firm in its entirely because these sub-units are typically managed with some independence and sometimes make products with very different characteristics than other segments of the same company (eg financial products versus manufactured goods). This two-tier sampling incorporated firms/segments of all sizes and also provided a larger sample 11 See the US Office of Science and Technology Policy website: http://www.scienceof sciencepolicy.net/award/national-survey-organizations-study-globalization-innovation -and-employment. 12 In addition, the F1K sample was oversampled for firms with high levels of R&D spend- ing because of keen interest in the topic of R&D outsourcing and offshoring. 310 Trade in Value Added Figure 11.5: Data collection grid for outsourcing and offshoring by business function. Source: National Organizations Survey. of firms (the F1K) likely to be globally engaged. After eliminating duplicates and foreign-owned enterprises, the overall response rate was 30% and was comparable across firms by size. In the 2011 NOS, questions about business functions were apparently easily understood and answered by senior executives at large and small firms, non- profits and public organisations. 13 Senior executives were able to quantify 13 ‘Costs’ are defined as follows. For a manufacturing business the costs of goods sold (COGS) are materials, labour and factory overhead. For a retail business the COGS is what the company pays to buy the goods that it sells to its customers. For a service business, it is the cost of the persons or machines directly applying the service, typically called ‘cost of sales’ by accountants. For a consulting company, for example, the cost of sales would be the compensation paid to the consultants plus costs of research, photocopying Direct Measurement of Global Value Chains 311 Table 11.5: Average share of employment (in percent) by business function and organ- isation type, December 2011 (US-owned firms’ US operations). For-profit Non- Public All F1K non-F1K profit sector cases A Primary business function 49.1 61.3 66.8 68.3 60.1 B Management, admin and back office 9.6 9.6 14.5 11.4 10.6 C Sales and marketing 11.9 7.3 2.7 1.3 6.6 D Customer and after-sales service 8.2 6.5 4.4 2.8 5.8 E Transportation, logistics, and dist. 6.6 5.2 2.7 4.7 5.2 F R&D of products, services, or tech. 7.7 4.4 2.1 2.3 4.6 G Facilities maintenance and repair 2.4 2.9 4.2 5.2 3.5 H IT systems 4.0 2.4 2.4 3.5 3.1 Average size (US employment) 15,022 1,616 2,333 4,217 6,272 Number of cases (n) 99 109 39 85 332 Source: 2011 National Organizations Survey, preliminary, 17 March 2012. the number of jobs, wage ranges and sourcing locations by business function according to their ‘best estimate’. For example, in the 336 completed surveys, only 4.5% (15) respondents indicated ‘don’t know’ to the question about the percentage of total US employment in their organisation according to busi- ness function. Of these, 12 were able to supply information about ranges of employment for each function (eg 1–10%, 11–30%), leaving only 3 respondents unable to answer the question. The survey also asked for sourcing as a per- centage of costs, either the cost of goods sold or the cost of services sold, known as ‘cost of sales’ (see Figure 11.5). This question was also well received by respondents, again according to their ‘best estimate’. We present some of the study’s preliminary findings here. First, Table 11.5 lists the percentage of costs for eight business functions in four types of US organisations where the survey was administered: 1. F1K business segments; 2. for-profit companies (not included in the F1K); 3. non-profit firms and organisations such as religious organisations and hospitals; and 4. public sector organisations, such as local, state, and federal government agencies. Taken together, samples 2-4 comprise a nationally representative sample of organisations, based on employment. There are some clear differences in employment allocation (on average) across the four organisational types. Comparing F1K firms with other for- profit firms, we see in Table 11.5 that, on average, F1K firms have fewer and production of reports and presentations. For a public organisation, costs are typically defined in its operating budget. 312 Trade in Value Added 100 80 60 % 40 20 In-house domestic 0 Outsource domestic g affiliate oreign eign outsource G H Figure 11.6: Location of business functions as a percentage of costs of goods or services sold (all cases, n = 306). Source: National Organizations Survey, preliminary, 17 March 2012. Categories on the horizontal axis refer to those defined in Table 11.5. 20 16 12 % 8 4 0 Outsource domestic g affiliate oreign reign outsource Figure 11.7: Location of outsourced/offshored business functions as a percentage of costs of goods or services sold: F1K cases, n = 86. Source: National Organizations Survey, preliminary, 17 March 2012. Categories on the horizontal axis refer to those defined in Table 11.5. employees working in their primary business function and more working in R&D and sales and marketing. Figure 11.6 shows the breakdown in costs for each of the eight business functions for the four possible combinations of organisational and geographic location discussed above and shown in Table 11.5 and Figure 11.5. A striking finding of the study is the low levels of international sourcing, on average, across all business functions, with the highest found in sales and marketing (7% of the function’s costs from international sourcing) and customer services and after-sales service (6% of the function’s costs from international sourc- ing). In the USA, firms and other organisations tend to source most business functions in-house. Functions with the highest domestic outsourcing, on aver- age, are facilities maintenance (13.5% of the function’s costs), IT systems (12% Direct Measurement of Global Value Chains 313 20 16 12 % 8 4 0 Outsource domestic oreign affiliate reign outsource Figure 11.8: Location of outsourced/offshored business functions as a percentage of costs of goods or services sold: private sector non-F1K cases, n = 104. Source: National Organizations Survey, preliminary, 17 March 2012. Categories on the horizontal axis refer to those defined in Table 11.5. of the function’s costs), and transportation and logistics services (9% of the function’s costs). On average, all firms in the sample spent only 3% of their primary function’s costs on domestic outsourcing and 5% of their primary function’s costs on international sourcing. Global engagement among US firms appears to be roughly comparable to, if slightly more common than among European firms. Recall that the 2007 Euro- stat International Sourcing Survey found that 20–25% of all surveyed manu- facturing enterprises sourced internationally, compared with about 10% of all enterprises in the other sectors of the economy. The preliminary analysis of NOS data has not yet broken out manufacturing firms for separate analysis, but of the 191 for-profit firms in the NOS study that answered the question, 24% outsourced at least some of their primary function domestically, while 30% sourced some portion of their primary function abroad (26% from foreign affiliates and 15% from offshore suppliers; 11% did both). While more analysis needs to be done to make direct comparisons between the surveys (the 2007 Eurostat International Sourcing Survey did not include firms with fewer than 100 employees, or 50 employees in some countries and covers an earlier time period, 2003–6 as opposed to calendar year 2010), the findings appear to be roughly consistent. The picture from the USA changes when only the largest firms in the NOS study are considered. When F1K business segments are broken out and com- pared to the rest of the for-profit cases as in Figures 11.7 and 11.8, F1K cases show a much higher level of international sourcing, especially though foreign affiliates, as expected. Interestingly, non-F1K for-profit companies engaged in average higher levels of domestic outsourcing than F1K companies for three functions: transportation, facilities maintenance and IT services. Finally, we present preliminary finding from the 58 NOS cases that were engaged in international sourcing (through affiliates, independent suppliers or both) and answered a question about the type of offshore location used: 314 Trade in Value Added 80 60 % 40 20 0 The same or higher g y lower ightly ch lower Figure 11.9: Percentage of international costs by type of location (operating costs in relation to the USA) and business function, 2010, organisations engaged in international sourcing (n = 58). Source: National Organizations Survey, preliminary, 17 March 2012. Categories on the horizontal axis refer to those defined in Table 11.5. those with costs equal to or greater than the USA, slightly lower than the USA or much lower than the USA. The results, presented in Figure 11.9, show that the lion’s share of international sourcing is to locations with costs that are equal to or higher than the USA. This suggests that the main motivation for international sourcing is to access skilled labour and advanced county product markets rather than low costs and emerging markets. It may also reflect the long-standing investments sourcing and other business relation- ships held by firms in the USA, especially with Canada and Western Europe. Next in importance are countries with costs much lower than the USA. Inter- national sourcing in countries with costs slightly lower than the USA is quite low, which might help explain the low level of integration of middle-income countries (eg in Latin America versus East Asia) in GVCs, contributing to the ‘middle-income trap’ experience of some developing countries (Giuliani et al 2005; Rodrik 2007). These preliminary findings indicate that, despite the concerns voiced in aca- demic literature and in media coverage about economic globalisation, GVCs and the outsourcing and offshoring of service work, these practices are in fact far from pervasive among US organisations. While GVCs are real and growing, they might be said to be in their infancy. Identification of trends will only come with follow-up surveys using the same framework. 4 CONCLUSIONS Scalable, comparable data are sorely needed in order to build accurate meso- level portraits of the location of value added and international sourcing pat- terns. On the one hand, macro-statistics and the IIOs that seek to combine them into larger cross-border matrices are too aggregated to provide reli- Direct Measurement of Global Value Chains 315 able, detailed industry-level estimates, and they are difficult to extend into the developing world, where input–output data is less developed or entirely missing. On the other hand, it is not feasible to collect product-level GVC data in large-scale surveys with the purpose of producing aggregated data at industry or country levels, mainly because it places too high a burden on respondents and data agencies, a problem exacerbated by the strategically sensitive nature of the data. Business function surveys can help fill this void. The importance of developing international standards in connection with new business surveys cannot be overstated. Global integration is first and foremost a cross-border phenomenon, and understanding it fully will require the collection of compatible, if not identical, data. A coordinated, sustained and iterative effort is needed. The inclusion of developing countries in these efforts is essential. At the same time, current data-collection programmes need to be evaluated on a constant basis in order to make negative priorities (eg reduce the number of collected variables, change the frequency of or abandon surveys) in order to make room for new surveys on emerging issues without increasing the overall respondent burden. Currently, official business statistics are under consider- able pressure, partly to achieve reductions in respondent burden, and partly because of budget constraints. Even under these conditions, it is important to identify new emerging topics of vital importance for understanding the cur- rent structure and dynamics of economic development for which no official statistical evidence is available. Such evidence can partly be established by methods that create no additional burden on enterprises, such as the link- ing of micro data and the construction of IIOs, but new surveys designed with minimal respondent burden in mind, such as business function surveys, must also be systematically deployed. Ideally, a global data-collection effort can come to rely on automated reporting systems that reduce the burden on organisations while increasing accuracy. While these goals will take time and be difficult to achieve, a concerted and well-coordinated effort is needed now. REFERENCES Aksin, Z., M. Armony and V. Mehrotra (2007). The Modern Call Center: A Multi- Disciplinary Perspective on Operations Management Research. Production and Oper- ations Management 6, 665–688. Ali-Yrkkö, J., P. Rouvinen, T. Seppälä and P. Ylä-Anttila (2010). Who Captures Value in Global Supply Chains? Case Nokia N95 Smart Phone. ETLA Discussion Papers 1240, February. Arndt, S., and H. Kierzkowski (eds) (2001). Fragmentation: New Production Patterns in the World Economy. Oxford University Press. Bair, J. (2009). Global Commodity Chains: Genealogy and Review. In Frontiers of Com- modity Chain Research (ed. J. Bair). Stanford, CA: Stanford University Press. Berger, S., and the MIT Industrial Performance Center (2005). How We Compete. New York: Doubleday. 316 Trade in Value Added Bernard, A. B., J. B. Jensen, and P. K. Schott (2005). ‘Importers, Exporters and Multi- nationals: A Portrait of Firms in the US that Trade Goods’, NBER Working Paper 11404. Bernard, A. B., J. B. Jensen, and P. K. Schott (2006). Survival of the Best Fit: Exposure to Low-Wage Countries and the (Uneven) Growth of US Manufacturing Plants. Journal of International Economics 68, 219–237. Blinder, A., and A. Krueger (2009). Alternative Measures of Offshorability: A Survey Approach. National Bureau of Economic Research, Working Paper 15287. Borrus, M., D. Ernst, and S. Haggard (eds) (2000). International Production Networks in Asia. London and New York: Routledge. Bradach, J., and R. Eccles (1989). Price, Authority, and Rust: From Deal Types to Plural Forms, Annual Review of Sociology 15, 97–118. Brown, C., and T. Sturgeon (forthcoming). The Outsourcing and Offshoring Practices of United States Organizations’. IRLE Working Paper, University of California and Berkeley. Brown, S. (2008). Business Processes and Business Functions: A New Way of Looking at Employment. Monthly Labor Review December, 51–70. Brülhart, M. (2009). An Account of Global Intra-industry Trade, 1962–2006. World Economy 32(3), 401–459. Cattaneo, O., G. Gereffi, and C. Staritz (2010). Global Value Chains in a Postcrisis World: A Development Perspective. Washington, DC: The World Bank. Commission of the European Communities (2003). The Competitiveness of Business- Related Services and Their Contribution to the Performance of European Enter- prises, COM(2003) 747 final. Dean, J., K. C. Fung, and Z. Wang (2007). Measuring the Vertical Specialization in Chinese Trade. US International Trade Commission, Office of Economics Working Paper 2007-01-A. Dedrick, J., K. Kraemer, and G. Linden (2010). Who Profits from Innovation in Global Value Chains? A Study of the iPod and Notebook PCs. Industrial and Corporate Change 19(1), 81–116. Dixit, A., and G. Grossman (1982). Trade and Protection with Multistage Production. Review of Economic Studies 49(4), 583–594. Doh, J. (2005). Offshore Outsourcing: Implications for International Business and Strategic Management Theory and Practice. Journal of Management Studies 42(3), 695–704. Dossani, R., and M. Kenney (2003). Lift and Shift: Moving the Back Office to India. Information Technologies and International Development 1(2), 21–37. Escaith, H., N. Lindenburg, and S. Miroudot (2010). International Supply Chains and Trade Elasticity in Times of Global Crisis. Report. Economic Research and Statistics Division, World Trade Organization, January 30. Feenstra, R. (1998). Integration of Trade and Disintegration of Production in the Global Economy. Journal of Economic Perspectives 12(4), 31–50. Fröbel, F., J. Heinrichs, and O. Kreye (1980). The New International Division of Labor. Cambridge University Press. Gereffi, G. (1994). The Organization of Buyer-Driven Global Commodity Chains: How US Retailers Shape Overseas Production Networks. In Commodity Chains and Global Capitalism (ed. G. Gereffi and M. Korzeniewicz), pp. 95–122. Westport, CT: Praeger. Gereffi, G. (1999). International Trade and Industrial Upgrading in the Apparel Com- modity Chain. Journal of International Economics 48(1), 37–70. Gereffi, G., and M. Korzeniewicz (eds) (1994). Commodity Chains and Global Capital- ism. Westport, CT: Praeger. Direct Measurement of Global Value Chains 317 Gereffi, G., J. Humphrey, and T. Sturgeon (2005). The Governance of Global Value Chains. Review of International Political Economy 12(1), 78–104. Gereffi, G., and K. Fernandez-Stark (2010). The Offshore Services Value Chain: Devel- oping Countries and the Crisis. World Bank Policy Research Working Paper 5262 (April). Gereffi, G., and K. Fernandez-Stark (2011). Global Value Chain Analysis: A Primer. Durham, NC: Center on Globalization, Governance and Competitiveness (CGGC), Duke University. Giuliani, E., C. Pietrobelli, and R. Rabellotti (2005). Upgrading in Global Value Chains: Lessons from Latin American Clusters. World Development 33(4), 549–573. Grossman, G., and E. Rossi-Hansberg (2006). The Rise of Offshoring: It’s Not Wine for Cloth Anymore. The New Economic Geography: Effects and Policy Implications, pp. 59–102. Jackson Hole, WY: Federal Reserve Bank of Kansas City. Grossman, G. and E. Rossi-Hansberg (2008). Trading Tasks: A Simple Theory of Off- shoring. American Economic Review 98(5), 1978–1997. Grossman, G. and E. Rossi-Hansberg (2012). Task Trade between Similar Countries. Econometrica 80(2), 593–629. Grunwald, J., and K. Flamm (1985). The Global Factory. Washington, DC: Brookings Institution. Henderson, J., P. Dicken, M. Hess, N. Coe, and H. Yeung (2002). Global Production Net- works and the Analysis of Economic Development. Review of International Political Economy 9(3), 436–464. Hesseldahl, A. (2010). The iPad: More than the Sum of Its Parts; $270 More, Actually. Bloomberg Business Week, February 22, p. 24. Houseman, S. (2011). Offshoring and Import Price Measurement. Survey of Current Business February, 7–11. Hummels, D., J. Ishii, and K.-M. Yi (2001). The Nature and Growth of Vertical Special- ization in World Trade. Journal of International Economics 54(1), 75–96. Humphrey, J., and H. Schmitz (2002). How Does Insertion in Global Value Chains Affect Upgrading in Industrial Clusters? Regional Studies 36(9), 1017–1027. Huws, U., and S. Dahlmann (2004). Outsourcing of ICT and Related Services in the EU: A Status Report. European Foundation for the Improvement of Living and Working Conditions. Luxembourg: Office for Official Publications of the Euro- pean Communities. URL: http://www.eurofound.europa.eu/emcc/publications/ 2004/ef04137en.pdf. Jensen, J. B., and L. Kletzer (2006). Tradable Services: Understanding the Scope and Impact of Services Offshoring’. In Brookings Trade Forum 2005, Offshoring White- Collar Work (ed. S. Collins and L. Brainard). Washington, DC: Brookings Institution Press. Kaplinsky, R. (2005). Globalization, Poverty and Inequality. Cambridge, UK: Polity Press. Kawakami, M. (2011). Inter-Firm Dynamics of Notebook PC Value Chains and the Rise of Taiwanese Original Design Manufacturing Firms. In The Dynamics of Local Learn- ing in Global Value Chains: Experiences from East Asia (ed. M. Kawakami and T. J. Sturgeon), Chapter 2. London: Palgrave Macmillan. Kletzer, L. (2009). Understanding the Domestic Labor Market Impact of Offshore Ser- vices Outsourcing: Measurement Issues. In Measurement Issues Arising from the Growth of Globalization (ed. W. E. Upjohn). Washington, DC: Institute for Employ- ment Research and the National Academy of Public Administration. 318 Trade in Value Added Koopman, R., Z. Wang, and S.-J. Wei (2008). How Much of Chinese Exports is Really Made in China? Assessing Domestic Value-Added When Processing Trade Is Perva- sive. NBER Working Paper 14109. Krugman, P. (1991). Increasing Returns and Economic Geography. Journal of Political Economy 99, 483–499. Lall, S. (2000). The Technological Structure and Performance of Developing Country Manufactured Exports, 1985–98. Oxford Development Studies 28(3), 337–369. Lanz, R., S. Miroudot, and H. Nordås (2011). Trade in Tasks. OECD Trade Policy Working Paper 117. Lewin, A., S. Massini, and C. Peeters (2009). Why Are Companies Offshoring Innovation? The Emerging Global Race for Talent. Journal of International Business Studies 40(6), 901–925. Linden, G., K. Kraemer, and J. Dedrick (2007). Who Captures Value in a Global Innova- tion System? The Case of Apple’s iPod. Personal Computing Industry Center Work- ing Paper 06-01-2007, University of California Irvine. Linden, G., K. Kraemer, and J. Dedrick (2009). Who Captures Value in a Global Innova- tion System? The Case of Apple’s iPod. Communications of the ACM 52(3), 140–144. Linden, G., K. Kraemer, and J. Dedrick (2011). Innovation and Job Creation in a Global Economy: The Case of Apple’s iPod. Journal of International Commerce and Eco- nomics May, pp. 223–240. Maule, C. (2006). Integrative Trade: Issues for Trade Analysis, Statistics and Policy’, Paper prepared for the CTPL Conference on Integrative Trade between Canada and the United States: Policy Implications. Nielsen, P. B. (ed.) (2008). International Sourcing: Moving Business Functions Abroad. Statistics Denmark. URL: http://www.dst.dk/publ/InterSourcing. Nielsen, B., and Z. Tilewska (2011). Data Linking—Creating New Evidence by Utilis- ing Existing Statistical Registers. Case: International Sourcing. In Proceedings of the 58th World Statistics Congress of the International Statistical Institute (ISI), Dublin, pp. 3247–3265. OECD (2011a). Global Value Chains: Preliminary Evidence and Policy Issues, Direc- torate for Science, Technology and Industry. Paper presented to the Committee on Industry, Innovation and Entrepreneurship, CIIE meeting, 31 March–1 April 2011. OECD (2011b). Developments in Measuring Trade In Value Added (and Employment), Statistics Directorate. Paper presented to the Working Party on International Trade in Goods and Trade in Services Statistics, 7–9 November 2011. Porter, M. (1985). Competitive Advantage. New York: Free Press. Rodrik, D. (2007). One Economics, Many Recipes. Globalization, Institutions and Eco- nomic Growth. Princeton University Press. Steinfeld, E. 2004. Chinese Enterprise Development and the Challenge of Global Inte- gration. In Global Production Networking and Technological Change in East Asia (ed. S. Yusuf, A. Altaf and K. Nabeshima). Oxford University Press. Sturgeon, T. (2002). Modular Production Networks. A New American Model of Indus- trial Organization. Industrial and Corporate Change 11(3), 451–496. Sturgeon, T. (2003). Exploring the Risks of Value Chain Modularity: Electronics Out- sourcing During the Industry Cycle of 2001–2002. MIT IPC Working Paper 03-002, May. Sturgeon, T., and G. Gereffi (2009). Measuring Success in the Global Economy: Inter- national Trade, Industrial Upgrading, and Business Function Outsourcing in Global Value Chains. Transnational Corporations 18(2), 1–35. Sturgeon, T., and R. Lester (2004). The New Global Supply-base: New Challenges for Local Suppliers in East Asia. In Global Production Networking and Technological Direct Measurement of Global Value Chains 319 Change in East Asia (ed. Y. Shahid, M. A. Altaf and K. Nabeshima), Paper 2. Oxford University Press. Sturgeon, T., and O. Memedovic (2010). Mapping Global Value Chains: Intermedi- ate Goods Trade and Structural Change in the World Economy. UNIDO Working Paper 05/2012. Vienna: United National Industrial Development Organization. Sturgeon, T., F. Levy, C. Brown, J. B. Jensen, and D. Weil (2006). Why We Can’t Measure the Economic Effects of Services Offshoring: The Data Gaps and How to Fill Them. MIT Industrial Performance Center Working Paper 06-006. Swedish National Board of Trade (2007). Add Value to the European Economy: How Anti-Dumping Can Damage the Supply of Globalized European Companies. Five Case Studies From The Shoe Industry. Kommerskollegium, Stockholm, Sweden. Tempest, R. (1996). Barbie and the World Economy. Los Angeles Times, September 22. van Welsum, D., and X. Reif (2009). We Can Work It Out: The Globalization of ICT- Enabled Services. NBER Working Paper 12799. Winkler, D., and W. Milberg (2009). Errors from the ‘Proportionality Assumption’ in the Measurement of Offshoring: Application to German Labor Demand. SCEPA Working Paper 2009-12. Schwartz Center for Economic Policy Analysis and Department of Economics New School for Social Research, New York. Yeats, A. (2001). Just How Big Is Global Production Sharing? In Fragmentation: New Production Patterns in the World Economy (ed. S. Arndt and H. Kierzkowski). Oxford University Press. 12 Integrating Value-Added Trade Statistics into the System of National Accounts A: Perspectives from the World Trade Organization Andreas Maurer The meeting’s objectives were to identify the state-of-the-art, existing data gaps and the direction of future work for developing ‘new measures of cross- border trade’. I wish to highlight the aspect of data gaps and what official statistics can do. The focus here is on statistics and measures that are developed and used for defining and monitoring trade policy and economic development. Are current statistics useful to gauge trade policy and economic development? The issue is therefore whether current statistics are useful to gauge trade policy and economic development. This is a question taken up in WTO’s Public Forum in September 2011, where the objective was geared towards trade policy. The Forum was due to debate how the measure of trade flows in value-added terms affects the way we analyse international economics and conduct trade policy. However, before we discuss how ‘measuring’, or better ‘estimating’, trade in value added impacts on the economic analysis; we have to see how we integrate this ‘animal’ in official statistics, such as the System of National Accounts, balance of payments and customs-based merchandise trade statistics. The System of National Accounts is a consistent and integrative account- ing system. The central question is ‘who does what by means of what for what purpose with whom in exchange for what with what changes in stocks?’ (see SNA 1993, Paragraph 2.12). The best-known macroeconomic aggregate is gross domestic product (GDP), which equals the sum of gross value added of all resident producer units. The balance of payments is concerned with transactions between residents and non-residents whereby transactions between the two groups involve a change in economic ownership of the product, be it a good or a service. Customs-based merchandise trade statistics use the principle of physical crossing for recording international transactions. The recording not only com- prises the value, quantity and other data elements; it also assigns to each 322 Trade in Value Added transaction a country of origin and country of (last known) destination. For the country of origin, rules of origin are used to identify the respective countries. It is here where trade in value added comes into play and where misunder- standings in the interpretation of statistics in the community of users start. In an era of international supply chains and production, where a single prod- uct can be made in many countries, the recording of trade flows based on the concept of country of origin may not reflect the way global business is done today or where the resulting income flows will be registered. The questions therefore are whether official statistics produce the right data and whether we (can) draw the right signals out of current statistics. 1 DO WE HAVE THE RIGHT STATISTICS? International supply chains and trade finance have often been quoted as con- tributing factors to the steep fall in trade flows in late 2008 and 2009. As pointed out elsewhere in this volume, do official statistics tell us • ‘who produces for whom?’ if we analyse the real economy, or • ‘who finances whom?’ if we consider the monetary economy? We face new business models, changes in transportation and communication and mass consumer demand, especially in the West. The latter are met by rising manufacture capacities in Eastern Asia, which makes Asia a global hub of manufacturing (which makes up more than 80% of its exports). The centre of economic gravity seems to have shifted to Asia. The way in which businesses are run has changed greatly over the last 20 years. We need to have information on the interconnectivity of national economies through linking firm activity (production) with export activity (trade). This will help policymakers base their decisions on economically meaningful data. As a result, governments will better understand that rais- ing trade barriers hurts domestic companies, which are dependent on the availability of competitive inputs for their competitiveness. It will also alle- viate some of the misunderstandings resulting from inflated bilateral trade imbalances based on the gross calculation of trade figures rather than on a value-added basis. 2 INTER-AGENCY COOPERATION FROM A WTO PERSPECTIVE Following an exchange of letters between Director-General Pascal Lamy and Secretary-General Angel Gurría in February 2009, the WTO and OECD have worked together, hosting seminars and conferences involving other key play- ers such as the Japanese Institute of Development Economies (IDE-JETRO), the US International Trade Commission (USITC) and the World Input–Output Integrating VA Trade Statistics into the SNA 323 Database (WIOD) Project. In parallel, through Eurostat and the UN Statistical Commission, there the WTO had a fruitful dialogue with the community of official statisticians, aimed at strengthening international cooperation. Both the OECD and WTO feel that further benefits could be gained by devel- oping more formal mechanisms of cooperation that would help to (i) mainstream existing research and results on trade in value added into the wider policymaker community, (ii) provide an institutional forum for existing initiatives, be they rooted in national, non-official or academic organisations, and (iii) motivate and accelerate further developments in this work domain. Results should focus on clarifying the concept of trade in value added to enable gross trade flows to be decomposed into domestic value-added compo- nents and import components through exploring a common understanding on the definitions, methodologies and challenges, based on state-of-the-art methodology. Further work could include reference to common institutional objectives, such as the development of a joint-branded database on value-added trade flows. Improving the coverage (eg to African countries) of such a database will be a priority for both organisations. The WTO has launched the ‘Made in the World Initiative’ (MIWI), as an effort to bridge the gap between researchers and trade policymakers and to develop a network of interested researchers and industries. As both OECD and WTO have built strong links with other key players in Asia, Europe and the USA, the two institutions are now in a position to coordi- nate efforts towards the estimation of trade flows in value-added terms based on official trade statistics and national accounts. Both organisations have been cooperating very closely with IDE-JETRO. Another key project in this area is the World Input–Output Database (WIOD), which was financed by the EU and aims at producing time-series of inter- country IO tables benchmarked on national accounts for 40 countries. The project produced important results in terms of both data and methodologies in March 2012. The long-term sustainability and mainstreaming of the project after this date needs to be addressed. Another stakeholder, the USITC, has also been developing a methodology for measuring trade in value added. The OECD is part of the WIOD consortium and has long been in the busi- ness of producing and maintaining an IO database. In cooperation with other stakeholders, OECD, with the support of WTO, could build on the WIOD expe- rience to coordinate the efforts and expertise of a large network of experts and institutions, such as IDE-JETRO, to deliver long-term benefits beyond the life-time of WIOD. The OECD and WTO would also promote a closer dialogue between researcher and official statisticians. As countries move to the new 2008 SNA and Balance of Payments Manual recommendations (International Monetary 324 Trade in Value Added Fund 2009, henceforth BPM6), as well as new industrial classification sys- tems such as ISIC Rev. 4 and NACE Rev. 2, it will be important to tap into the expertise of national accounts, input–output, business and trade statisticians. Promoting such a dialogue would involve a close cooperation with important stakeholders such as Eurostat, the UN Statistical Commission and all the rel- evant international agencies. Such a dialogue has been included in a 2020 Vision of an International Trade Information System as agreed upon by four organisations in the Global Forum on Trade Statistics, held in February 2011 in Geneva. 3 TRADE IN VALUE ADDED AMONG INTERNATIONAL ORGANISATIONS Each organisation could be contributing to this joint OECD/WTO project. MIWI is a hub for discussion and exchange of information of importance to this project (by holding events, etc ). For example, in February 2011, WTO hosted a UNSD/Eurostat-organised Global Forum on Trade Statistics. One of the major outcomes was the fostering of trade and business statistics linkages by integrating trade and business registers to explore the database for analysing business processes. Others included the lack of detail in trade in services statistics or information on bilateral trade flows. These subjects are not only important for trade in value added but also for export promotion, and involve not new surveys but the use of administrative sources. 4 DO WE DRAW THE RIGHT SIGNALS OUT OF CURRENT STATISTICS? DO WE BRING UNDERSTANDING TO THE MEASUREMENT? As previously mentioned, the global division of labour has emerged through an intense inter-industry trade in intermediate goods and services, benefitting from the efficient allocation of tasks across the globe for producing and trad- ing. There is a responsibility for statisticians to explain their statistics in this context. For example, if we look at trade flows, conclusions may be drawn in respect of bilateral imbalances on exchange rate policy, but analysis of inter- national fragmentation of production through estimating imported inputs in domestically produced exports may view these imbalances differently. 5 HOW SHOULD OFFICIAL STATISTICS REACT? As the 2008 SNA, MSITS 2010 and IMTS 2010 1 are put into place, interna- tional organisations have to jointly assist implementation of these concepts 1 Manual on Statistics of International Trade in Services 2010 and International Merchan- dise Trade Statistics 2010. Integrating VA Trade Statistics into the SNA 325 and definitions to develop data sets that are more apt for analysing globali- sation. However, in fostering implementation, it has to be ensured that there is no information loss between the different versions of the respective statis- tical frameworks. Keeping track of the flow of intermediate goods exchanged within global value chains when there is no change of ownership is one of the implementation issues to be addressed. As all statistical frameworks have undergone revision, no new additional concepts and definitions can be defined. However, instead of devising new concepts, existing statistics on the external sector can be used in a more sys- temic way. In this vein, a new statistical tool in form of a satellite account could be developed to complement national accounts. This tool would bring together a country’s foreign activities with respect to trade—goods, services, intellectual property, capital (foreign direct investment) and income flows, labour (movement of workers)—in one integrated presentation, similar to tourism satellite accounts. B: Perspectives from the United Nations Ronald Jansen Global production has become increasingly fragmented and different stages of production are now regularly performed in different countries. As inputs cross-borders multiple times, traditional statistics on trade values—measured in gross terms—do not reflect economic reality in respect of the value added in any particular country. This is the opening of the workshop programme and the main theme of the workshop. Similarly, global production and trade in value added were among the main themes of the Global Forum on Trade Statistics, which was organised in Febru- ary 2011 by the United Nations Statistics Division (UNSD) together with Euro- stat, WTO and UNCTAD. The forum received high-level attention from policy- makers and was attended by almost 200 trade statisticians from all around the globe. Pascal Lamy stressed in his presentation the importance of relevant trade statistics in a globalised world, stating that all trade negotiations, in the end, deal with numbers. A number of issues have been raised explicitly or implicitly in the discussion paper on ‘Tracing Value Added in International Trade’ (Mattoo et al 2011). I wish to clarify these issues from the perspective of official trade statistics and will highlight the ongoing efforts to improve trade statistics, also to the benefit of the research on trade in value added. This chapter should be read 326 Trade in Value Added in connection with the discussion paper and the documents of the Global Forum. 1 1 ISSUE 1: COLLECTION OF VALUE-ADDED TRADE DATA First of all, I want to state that detailed trade statistics by product and part- ner countries in terms of gross values will remain necessary input for many analytical purposes, including IO research. It is not desirable to collect trade statistics in other than gross values. Aside from the fact that such statistics are necessary for agriculture, energy, environment and transportation statistics, quality assurance frameworks of trade statistics are for a large part based on a consistent relation between the value and the quantity of the traded goods. This will hold true whether data is collected via enterprise surveys or through customs documents. Additional information will need to be collected if we want to decompose the gross values into domestic and foreign content, or further refinements. I shall mention some of those additional elements below. The objective of our workshop discussions is to find ways to publish trade data in value-added terms, but such an objective is not equivalent to collecting trade data in value-added terms. 2 ISSUE 2: CUSTOMS RECORDS OR ENTERPRISE SURVEYS? We need both sources of data. The most important source of trade data remains the customs data. In fact, trade statisticians should advocate more forcefully the keeping of detailed customs information on importation and exportation documents. The trade community (traders and enterprises) puts pressure on the government to facilitate customs procedures, and has been successful in some ways. We should realise that enterprise surveys can be nowhere near as detailed or as timely as customs records. Enterprise surveys will cover necessarily fewer goods, give less detail on trading partners and will be obtained less frequently. The greatest value of enterprise surveys will be as an addition to customs records. These surveys could then focus on specific questions, such as how much of the manufacturing processes of an enterprise are done under contract on behalf of foreign enterprises. 3 ISSUE 3: LINKING TRADE AND BUSINESS STATISTICS The main topic of our discussion is the fragmentation of the global production processes. The implication is that we want to know more about the strategies 1 See http://unstats.un.org/unsd/trade/s_geneva2011/outcome.htm. Integrating VA Trade Statistics into the SNA 327 of businesses that operate globally in their production. To reiterate a point often made, trade is not done between countries, but between businesses. Session 4 of the Global Forum on Trade Statistics was devoted in full to the issue of global production and outsourcing of business functions 2 with pre- sentations by, among others, Timothy Sturgeon (on ‘Measuring Global Value Chains’) and by Peter Boegh Nielsen (on international sourcing of business functions). These research projects investigate directly the global business strategies and need of the statistical community for further development of classifications on intermediate products and on business functions. Another related outcome of the Global Forum on Trade Statistics 3 was to better link trade and business statistics by • developing a common basis across all relevant national institutions to identify enterprises active in international trade, including multina- tional enterprises and their foreign affiliates, • developing and maintaining a statistical trade information system at micro-level around the enterprise register, including multinational enterprises and their foreign affiliates, and • establishing this statistical information system—under observance of relevant confidential rules—by making optimal use of and connecting existing data sources, such as custom-based merchandise trade statis- tics, trade and business registers, economic census data, existing enter- prise surveys, other administrative records and possibly data sources for employment, environment or energy. 4 ISSUE 4: CROSS-BORDER TRADE AND THE CHANGE OF OWNERSHIP PRINCIPLE The main area of contention between trade statisticians and national accoun- tants has been not valuation but the issue of ‘change of ownership’. According to SNA, an international transaction in goods takes place only if there has been a change of ownership between a resident and a non-resident. When a good crosses the border, it does not necessarily mean that there has been a change in ownership. International Merchandise Trade Statistics (IMTS) cover goods which add to or subtract from the stock of material resources of a country by entering (imports) or leaving (exports) its economic territory. This basis differs from the change of ownership between residents and non-residents required for balance of payments 4 and national accounts. This controversy 2 See http://unstats.un.org/unsd/trade/s_geneva2011/outcome.htm. 3 See the United Nations Statistics Division, http://unstats.un.org/unsd/trade/s_geneva 2011/Global_Forum_on_Trade_Statistics-detailed_vision_statement-15Mar2011.pdf. 4 See http://www.imf.org/external/pubs/ft/bop/2007/pdf/chap10.pdf. 328 Trade in Value Added is the backdrop to the discussion on the international sourcing of produc- tion processes, better known as the issue of ‘goods for processing abroad’ or ‘processing trade’ (Mattoo et al 2011) or ‘manufacturing services on physical inputs owned by others’ (International Monetary Fund 2009). In the context of Global production and GVCs, this issue is probably the most important one. 5 ISSUE 5: INTERNATIONAL SOURCING OF PRODUCTION PROCESSES International trade has been at the centre of many recent discussions on glob- alisation, be it through the offshoring of the production process, operations of multinationals, foreign direct investments or trade negotiations. Produc- tion processes of garments, motor vehicles, televisions or computers are now often spread across several countries not only to reduce labour and capital costs but also, for instance, to benefit from investment incentives offered by the host countries. Even though treatment of goods for processing in the stat- istical sense is by no means a new discussion, it gained a lot of recent attention because of its increasing economic importance, especially for economies like China and Mexico. My proposal for measuring trade statistics in relation to international sourc- ing of production processes is as follows. 1. Link detailed merchandise trade statistics to the business register. This matching process may not be perfect, but is essential in deriving results. 2. Conduct a survey among exporting enterprises of the manufacturing industries and determine the percentage of processing done under con- tract by enterprise and industry. 3. Link the enterprise survey to the merchandise trade statistics via the business register, and determine the volume and kind of imported and exported goods that are associated with ‘processing under contract’. The end result will be trade statistics broken down by product, industry and partner country, with a separate breakdown of processing under contract. Balance of payment compilers could then use this information to adjust the trade in services and trade in goods statistics. Such survey could be validated and complemented by an economic census, which is ideally done at five-year intervals. (For instance, Malaysia conducted an economic census in 2011 and included questions on processing under contract.) This approach produces official statistics on intermediate goods processing by industry and product. Note that change of ownership always needs checking, since even within multinationals it is possible that inputs in the production process are actually acquired by the foreign affiliate. The Bank of Thailand conducted a survey which showed that the top three electronics manufacturers in Thailand buy the inputs into their production from their Integrating VA Trade Statistics into the SNA 329 mother companies. This raises the issue of transfer pricing, which I shall not discuss here. The international recommendations for IMTS were revised in 2010 (IMTS 2010) and contain new recommendations for a number of additional data elements useful in the analysis of the globalisation issues, namely 1. additional valuation of FOB for imports, 2. country of consignment for imports and exports, which will facilitate tracing the routes the goods take, 3. indication of customs procedures for inward and outward processing, and 4. mode of transport. IMTS 2010 also recommends linking trade to business statistics; this recom- mendation has been emphasised in recent months. 6 ISSUE 6: UNSD AND IO TABLES Paul Cheung, Director of the United Nations Statistics Division, spoke at the 19th International Input–Output Conference in Alexandria, VA on the relation between official statistics and IO analysis (Cheung 2011). One of the points he made is that if a country does not have the source data or the resource capacity and expertise to provide value added by industry, gross domestic product by expenditures in current and constant prices, and gross national income, then the country will not be in a position to produce a fully articulate IO table. In this regard, it will be useful to update the UN Input–Output Handbook to reflect all relevant changes introduced with the 2008 SNA, keeping in view that the handbook should be a practical compilation guide for countries at varying levels of statistical development. In conclusion, I am advocating two parallel and mutual supportive devel- opments: on one hand, to improve official trade statistics by linking them to business statistics, and on the other hand to improve the compilation of IO tables with support of, for instance, an updated handbook on input–output tables. C: Perspectives from the Organisation of Economic Cooperation and Development Nadim Ahmad 330 Trade in Value Added The OECD is strongly in favour of an internationally coordinated approach to the development of value-added trade estimates and supports the idea that this could best be achieved with an inter-Secretariat approach that brings together a number of international agencies which are able to tap into their existing institutional networks of official statistics. The ultimate goal is of course a global IO table, which will in practice require agreement on the opti- mal level of industry or product detail across the different agencies respon- sible for the collection and harmonisation of national IO tables. We feel that it is perhaps premature and ambitious to encourage official statistics offices to produce national input–output (IO) or supply–use (SU) tables that differ- entiate imports by IO industries and final demand on the basis of the source industry and country which the import came from, but certainly we feel that statistics institutes could be encouraged to provide more detailed information on imports made by IO industries; this would significantly improve the qual- ity of value-added trade estimates, as for many countries these are created using a simplistic proportionality assumption. Many developed economies could be encouraged to do this by tapping into firm-level data, in particular firm-level data that links business and trade registers. Developing countries should be encouraged to develop similar capacities, such that IO tables are able to reflect industry or product classifications in as homogeneous a way as is possible. Particular attention in this regard should be made to classifica- tions that are able to differentiate between ‘ownership’, ie foreign or domes- tic, and import–export intensities. In this context we should also retain some scope for differentiating between ‘processors’ and conventional producers, noting in particular the changes in the 2008 SNA: improvements and indeed potential data sources have been identified in the deliberations of a Eurostat- led task force looking at goods for processing. 1 The value-added trade indi- cators we produce should be as detailed and useful as possible. In that sense the objective should be to produce estimates that reflect the whole economy and industries, broken down by factors of value added, ie labour and capital or operating surplus. Many major economies already produce annual SU tables but not all. Clearly, the construction of internationally recognised estimates of value-added trade could serve as an important catalyst to motivate the development of more timely annual SU tables. In this regard it is useful to note that the OECD is currently engaged in a project with the Chinese National Bureau of Statis- tics to produce SU tables. The OECD, along with its partner agencies in the Inter-Secretariat Working Group on the National Accounts, will continue to encourage countries, including developing economies to implement the SNA, which recommends and included annual SU tables. Certainly more work is needed on the allocation of trade to industries espe- cially services. In the meantime, however, it will remain necessary for some 1 See http://search.oecd.org/officialdocuments/publicdisplaydocumentpdf/?cote=STD /TBS/WPTGS%282012%2910&docLanguage=En. Integrating VA Trade Statistics into the SNA 331 estimation, even if crude, to be done, although clearly there will be merit in providing ranges for the estimates that reflect changes in the assumptions used to allocate imports. This will also help in determining the level of detail at which value-added trade estimates could and should be presented, noting that the level of detail in IO tables is probably likely to be greater than the level of detail published. A great deal of work remains to be done. But the importance of having some measure of value-added trade now means that we cannot wait for improve- ments in the underlying data to come first. Indeed, it is hoped that by demon- strating a credible need for these improvements by the production of value- added estimates we will be able to accelerate matters and motivate official statistics offices to allocate resources to improving the underlying data. D: Perspectives from the US Bureau of Economic Analysis Robert E. Yuskavage One of the key features of globalisation has been the international fragmen- tation of goods production as firms take a global approach to reducing costs and expanding markets. This fragmentation has manifested itself in the devel- opment of global value chains characterized by rapid growth in the trade of intermediate products across borders. For some products, inputs cross bor- ders multiple times before a final product is completed. Conventional foreign trade statistics that are based on these cross-border gross flows assign the full value of imports and exports to countries of origin and destination. Because these conventional trade measures have major limitations for assessing inter-country linkages and bilateral trade balances, the Bureau of Economic Analysis (BEA) supports further research designed to develop accu- rate value-added trade statistics that could ultimately be included as sup- plementary measures in the System of National Accounts (SNA). However, BEA strongly believes that the conventional gross flows should remain as the featured measures of cross-border trade because of their important role in calculating the net exports component of gross domestic product and in pro- viding high-frequency bilateral trade balances that are very timely and highly detailed. Gross flows attribute the full value of imports entirely to the country where the final product is produced regardless of how much value was added by other countries upstream in the supply chain. Although economists have long minimised the importance of bilateral balances, these balances continue to receive considerable attention among policymakers and play important roles 332 Trade in Value Added in discussions about trade policy and exchange rate management. However, large merchandise trade surpluses or deficits can be misleading for policy purposes when the domestic factor content of imports is high or the foreign factor content of exports is high. A recent Wall Street Journal article (Batson 2010) about the global value chain for the Apple iPhone underscores this point. The iPhone is imported from China and included in the US import statistics at its wholesale value even though that value consists largely of intermediate goods and services produced in other countries, including the USA. The cost of assembly in China contributes only a small portion to the wholesale cost. Many of the papers presented at this workshop propose alternative approaches to mea- suring bilateral trade flows that focus on the value added to the final prod- uct by each country in the value chain. In the iPhone example, the overall US trade deficit would remain the same but the deficit with China would be considerably smaller and the deficits with other countries that supply parts, components, and services would be larger. 1 THE BUREAU’S APPROACH For its economic accounts, the BEA closely follows international guidelines that are designed to increase the comparability of economic statistics across countries. The current approach followed by the USA and other countries is based on guidelines issued by the International Monetary Fund (IMF) in vari- ous editions of the Balance of Payments Manual. In 2009, the IMF released the sixth edition of the Balance of Payments and International Investment Posi- tion Manual (BPM6). This update, the first since 1993, was coordinated with an update in 2008 of the System of National Accounts (2008 SNA) in order to maximise the overall consistency between these two key sets of interna- tional guidelines for economic accounts. Several of the provisions in the most recent updates of these new international statistical standards were designed to at least partly address concerns about the impact of global production and global value chains on economic statistics. These provisions are described briefly later in this section. It is important to recognise that the value-added approach has no impact on a country’s overall trade balance and therefore no impact on gross domes- tic product (GDP) calculated as the sum of final expenditures. In effect, it reallocates a country’s overall trade balance among its trading partners and expands the set of trading partners to include other countries in the global value chain. For GDP, net exports are calculated as gross exports minus gross imports. In the value-added approach, measures of gross exports and gross imports would each be smaller, but net exports overall would be the same. Integrating VA Trade Statistics into the SNA 333 2 THE VALUE-ADDED APPROACH Value-added-based trade measures have been proposed as an alternative that better reveals the primary resources provided by countries to produce final products. Arguments for this approach have been made in the past in the con- text of measuring the factor content of international trade and identifying the export content of imports and the import content of exports. However, direct measurement of value-added trade is extremely difficult, if not impossible. A 2006 National Academy of Sciences (NAS) study concluded that meaningful value-added trade measures could not be developed primarily because the information was not available in business record-keeping systems. 1 In general, US business firms do not maintain information in their account- ing systems that would allow them to readily identify whether their material inputs are from domestic or foreign sources. Firms typically obtain their mate- rial inputs from wholesale suppliers and distributors and are not necessarily concerned about the country of origin for these materials. In addition, for for- eign source materials, the country of origin may change frequently, depending on relative prices and other market factors. However, information developed by firms for supply chain management could prove helpful in this endeavour. 3 INPUT–OUTPUT METHODS As a result, perhaps the most promising approach to developing comprehen- sive and consistent value-added trade measures that go beyond case studies of individual high-profile products involves the use of world IO tables. Even the NAS study acknowledged that the IO approach might prove viable for this effort but raised serious concerns about data quality and the assumptions required to obtain results. At that time, the largest reservation concerned the lack of a consistent time series of supply and use tables that could be linked across trading partner countries. These linkages are critical to deriving mea- sures that take into account not only the countries of origin and destination for traded goods throughout the value chain but also the production technol- ogy that is employed in the countries that provide inputs both directly and indirectly. Previous chapters in this volume provide a strong testament to the impres- sive advances that have taken place in the last few years in the development of consistent and comparable linked regional supply and use tables. As a result of these advances, the reliability of value-added trade measures based on the IO approach has increased significantly, for both the measures for individ- ual countries and the related bilateral trade balances between trading partner countries. Of course, significant further work is required before a consistent 1 See National Research Council and National Academy of Sciences (2006). 334 Trade in Value Added set of reliable measures can be developed on a time-series basis for a wide range of countries on a timely basis. One area in particular that requires further work is the development of the import use tables that play a key role in generating the estimates of bilateral supply and use of imported intermediate inputs. For the reasons described above, direct measures of the use of imported intermediate inputs by indus- try are not available, and existing import use tables rely heavily on the import comparability assumption. BEA has conducted research evaluating the reli- ability of this assumption and has found that while it works well for some industries at aggregate levels it is not as accurate for the more detailed indus- tries that are critical for understanding the use of imported inputs for prod- ucts involved in cross-border trade. Other research based on Census Bureau micro data holds promise for future improvements in import use tables. 4 NEW INTERNATIONAL GUIDELINES Two provisions that were introduced in SNA 2008 and BPM6 that are directly related to the impact of global manufacturing are a new treatment of goods sent abroad for processing without a change in ownership (goods for process- ing) and the purchase and subsequent resale of goods abroad without sub- stantial transformation and without the goods entering or exiting the coun- try (merchanting). Under the treatment of goods for processing, no change in ownership is imputed, the goods are excluded from merchandise trade gross flows for both exports and imports and the value of the service provided by the contract manufacturer is recorded as trade in services. This treatment has no impact on the overall trade balance, but it shifts the composition of the balance between goods and services. If intermediate materials were acquired from other countries, in principle those goods would be counted as imports from those countries rather than from the country of final assembly. The new treatment of goods for processing in particular has the potential to signifi- cantly reduce the distortions associated with the traditional measures based on gross trade flows. Some global manufacturing activities may also qualify for the new merchanting treatment, but this treatment would not necessarily address the issue of bilateral balances. 5 CONCLUSION BEA encourages further research to develop IO based value-added measures of foreign trade in order to more clearly articulate the nature of bilateral trade flows and balances. However, BEA does not believe that value-added measures should supplant conventional gross flow trade statistics as the featured mea- sures of cross-border trade and for calculating GDP. The existing bilateral Integrating VA Trade Statistics into the SNA 335 gross trade flow statistics are timely, long-standing, useful for a wide range of statistical and analytical purposes, well reported by countries around the world and consistent across countries. BEA will support the development of value-added measures by continuing to improve the accuracy and timeliness of its input–output accounts for the USA. However, for IO-based measures to be useful for calculating bilateral trade balances, a coordinated approach across countries would be required. Practical problems would arise if each country were responsible for compiling its own value-added trade statis- tics. Issues of frequency, timeliness and consistency would also need to be addressed. Finally, it is important to point out that the new international standards introduced in SNA 2008 and BPM6 related to goods for processing in princi- ple at least partly address some of the concerns raised in this volume. How- ever, national statistical agencies face major challenges in implementing these new standards because of limited source data and resources. For statistical agencies, implementing the new standards should be a higher priority than developing new analytical measures. Researchers should work closely with the statistical agencies to help implement these important new standards while continuing to advance the state of the art for world IO tables. REFERENCES Batson, A. (2010). Not Really ‘Made in China’, Wall Street Journal, 10 December. Cheung, P. (2011). Input–Output Analysis in Contemporary Official Statistics. Paper presented to 19th International Input–Output Conference, Alexandria, VA, 13– 17 June 2011. URL: http://www.iioa.org/Conference/19th/keynote.html. International Monetary Fund (2009). Balance of Payments Manual (6th edition). URL: http://www.imf.org/external/pubs/ft/bop/2007/pdf/bpm6.pdf. Mattoo, A., Z. Wang and S.-J. Wei (2011). Tracing Value-Added in International Trade: An Overview of Issues and a Proposal. Discussion Paper presented at World Bank Trade Workshop. URL: http://go.worldbank.org/R156ABXQQ0. National Research Council and National Academy of Sciences (2006). Analyzing the US Content of Imports and the Foreign Content of Exports. Washington, DC: National Academies Press. SNA (1993). System of National Accounts. United Nations Statistics Division. URL: http://unstats.un.org/unsd/nationalaccount/sna1993.asp.