WPS8262 Policy Research Working Paper 8262 The Internet and Chinese Exports in the Pre-Alibaba Era Ana M. Fernandes Aaditya Mattoo Huy Nguyen Marc Schiffbauer Development Research Group Trade and International Integration Team November 2017 Policy Research Working Paper 8262 Abstract This paper uses the dramatic expansion of access to the country such as Alibaba. The paper takes a closer look at Internet in China to analyze the impact of the Internet why, focusing on three questions: what aspects of firm per- on firm performance. The paper combines firm-level pro- formance were affected, what types of firm communication duction data with province-level information on Internet were facilitated, and what dimensions of the new commu- penetration to examine how the rollout of the Internet nication medium were relevant? The paper finds that the across Chinese provinces between 1999 and 2007 influ- Internet not only enhanced trade, but also improved overall enced firms’ export behavior. The econometric strategy firm performance. The results are consistent with improve- enables identifying the impact of the Internet on firm per- ments in communication with buyers and input suppliers. formance in China. The paper shows that the rollout of the The benefits arose not just from better communication, Internet boosted manufacturing exports of firms in China, but from establishing a visible virtual presence, and were even before the rise of major e-commerce platforms in the enhanced by, but not contingent on, access to broadband. This paper is a product of the Trade and International Integration Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at afernandes@worldbank.org, amattoo@worldbank.org, mschiffbauer@worldbank.org, and HNguyen4@imf.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team The Internet and Chinese Exports in the Pre-Alibaba Era* Ana M. Fernandesa Aaditya Mattoob Huy Nguyenc Marc Schiffbauerd JEL Classification codes: F14, O33. Keywords: internet, information and communication technology, export growth, firm-level data, China. a AnaMargarida Fernandes. The World Bank. Email: afernandes@worldbank.org. b AadityaMattoo. The World Bank. Email: amattoo@worldbank.org. c Huy Nguyen. International Monetary Fund Email: hnguyen4@imf.org. d Marc Schiffbauer. The World Bank. Email: mschiffbauer@worldbank.org. * This paper was prepared as a background paper for the World Development Report 2016 Digital Dividends. The authors would like to thank Michael Minges, Rajendra Singh, Tim Kelly, Indhira Santos, Uwe Deichmann, Shawn Tan, Deepak Mishra, and Maggie Chen for helpful discussions, to Michael Ward for sharing data, and to Changjiang Li at CNNIC for help with access to data on ICT in China. Research for this paper has in part been supported by the World Bank’s Multidonor Trust Fund for Trade and Development and the Strategic Research Program on Economic Development. The findings expressed in this paper are those of the authors and do not necessarily represent the views of the World Bank or its member countries. 1. Introduction The scale and speed of the internet roll-out in China after the late 1990s was unprecedented. The total number of internet users in China increased from about 680,000 in 1999 (or 0.7 per hundred people) to 137 million by 2007 (or 16 per hundred people), one of the most rapid increases in internet penetration in the world.1 The number of internet users increased across all of China’s provinces, but growth was stronger in coastal areas in the earlier years and in several inland provinces in the later years. This dramatic expansion of the internet in China is a unique historical event which allows us to analyze the impact of the internet on firm performance. Trade is a major channel through which the internet can affect aggregate economic growth (World Bank, 2016).2 The transactions costs of trade, of which information frictions are an important element, imply that generally only the most productive firms export, especially in developing countries. Access to the internet can reduce information frictions and thus trade costs. The internet helps firms to find information more easily about new buyers in overseas markets and to advertise their products, making it easier for buyers to find them, cutting off the costly middlemen usually needed to establish trade connections.3 The internet also reinforces business and social networks and hence facilitates interaction with potential customers and suppliers (Rauch, 1999; Rauch and Trindade, 2002; Fink et al., 2005). The emergence of major international e-commerce platforms can further reduce trade costs by lowering moral hazard costs between buyers and sellers.4 The internet can thus enable more firms to engage in international trade. In this paper, we combine firm-level manufacturing census data with data on province- level internet penetration in China to examine the effects on firm exporting behavior of the rollout of internet infrastructure across Chinese provinces from 1999 to 2007. Our econometric strategy is based on the hypothesis that increases in internet penetration in a province lead ceteris paribus to larger benefits for firms in industries that rely more intensively on the internet for the business 1 These figures are based on data from the World Bank’s World Development Indicators and from the China Internet Network Information Center, described in Section 3. 2 Other channels, such as increasing labor productivity by raising firms’ utilization of capital and labor, can also be important but they might need more time to materialize. Recent evidence shows that firms must simultaneously invest in complementary skills and management practices to benefit from investments in information and communications technology (ICT). 3 Freund and Weinhold (2004) propose a theoretical model where the internet reduces market-specific costs of trade. 4 To purchase any product, the buyer must find a seller, make a payment before receiving the goods, and trust that the seller will deliver the correct amount and quality on time. The seller has an incentive to deviate from the agreed terms since non-compliance is difficult to enforce internationally. Mutual rating systems embedded in online marketplaces allow the buyer and seller to publicly assess each other’s past performance raising the reputational costs of non-compliance. See World Bank (2016) for a more detailed discussion. 2 opportunities it offers than for firms in industries that rely less intensively on the internet.5 Guided by economic theory, we consider different measures for the internet intensity of industries drawing upon sources based on industry and firm-level data from the United States, China, and Vietnam. A positive effect of internet penetration on firm export performance might be driven by other contemporaneous changes at the province level, such as investments in transport infrastructure. Therefore, our specifications control for province-year fixed effects which capture fixed and time-varying unobservable provincial attributes. Moreover, the specifications control for firm fixed effects to account for the unobserved heterogeneity across firms which could influence their export behavior. Hence, our econometric strategy estimates the within-firm variation in export performance over time due to increases in internet penetration in the firm’s province, which, we believe, helps us to identify the causal impact of the internet on firm performance in China. We address any remaining concerns about the endogeneity of internet penetration by using an instrumental variable strategy based on an indicator of telecommunications development in neighboring provinces. We find that the rollout of the internet boosted manufacturing exports of firms in China. Specifically, growth in the number of internet users per capita in a province raises the probability that a firm participates in export markets, increases firms’ real exports and also their export-output ratios. We take a closer look at the reasons for these findings. First, we examine which aspects of firm performance were affected and find that increased provincial internet penetration did not just enhance firm exports but also overall firm performance. Firms in China are not simply re-orienting their output to export markets but are increasing the scale of their production (measured by real output). The growth in firms’ real output is due to increases in both employment and labor productivity. The effects on firm total factor productivity (TFP) are somewhat less robust, suggesting that labor productivity growth is also driven by capital deepening. Second, we examine what types of firm communication were facilitated and find results consistent with improvements in communication with both buyers and input suppliers. Our estimates show larger effects in more communications-intensive industries that produce differentiated goods as well as in those that rely on a larger share of differentiated intermediate inputs. Third, we examine whether certain aspects of the new communication medium were particularly relevant. We find benefits from establishing 5 This identification strategy is inspired by Rajan and Zingales (1998). 3 a visible virtual presence: the increased availability and use of online services in a province leads to a bigger boost in export outcomes for firms in industries more reliant on an online presence. We also find that these benefits were enhanced by, but not contingent on, access to broadband, approximated by the availability of fiber optic cables in the provinces. An interesting implication of these results is that access to the internet boosted firm-level exports in China even before the emergence of Alibaba as a widely-used, large scale e-commerce platform and before eBay’s expansion in China. The results suggest that the internet significantly reduces trade costs, even in the absence of broadband and major international e-commerce platforms serving the domestic market. This finding is relevant for the many developing countries trying to strike a balance between widening access to basic services and deepening it through the creation of a broadband network and connecting to major e-commerce platforms. The rest of the paper is organized as follows. Section 2 reviews the literature while Section 3 describes the data. Section 4 describes the empirical specification while Section 5 presents the results. Section 6 concludes. 2. Literature Review Early work on the effect of the internet on trade relied on aggregate country-level data (Freund and Weinhold, 2002, 2004; Clarke and Wallsten, 2006). For example, Freund and Weinhold (2004) show a significant effect of the number of web hosts on trade during the period 1995-1999. Their estimates suggest that a 10 percentage-point increase in the growth of the number of web hosts in a country leads to a 0.2 percentage point increase in export growth. Recently Lin (2015) finds similar magnitudes for the impact of internet users on export growth based on data for the period 1990-2006. Other research finds that the relationship depends on destination countries’ income levels: higher internet use in a developing country is related to higher exports to high- income countries but not necessarily to other developing countries (Osnago and Tan, 2015). Recent work in this area has moved from country-level to firm-level analysis. Kneller and Timmis (2016) estimate a strong positive causal impact of broadband internet use by business services firms in the United Kingdom on their propensity to export. To address potential endogeneity, these researchers use spatial differences in broadband availability (linked to the 4 historic telephone network) as an instrument for firms’ use of broadband internet.6 Focusing on China and using firm-level data, Chen and Xu (2015) compare data from Aliexpress and Chinese customs and find that firms exporting through Alibaba are smaller, sell more products per firm, and reach different export markets than firms exporting directly. We contribute to this literature by showing that access to the internet per se can reduce trade barriers not just for business services but also for goods, and boost firms’ goods exports even before large scale, international e- commerce platforms became widely available. Apart from the few studies on the trade impact, there is also a literature on the impact of internet access on productivity at the country and firm levels. This literature is relevant here for methodological reasons and because we will also consider impacts on overall firm performance. The basis for efficiency gains from the internet could include access to information on better management techniques or technology; more efficient marketing and sourcing, management of inventories and other benefits of more coordinated decisions with suppliers and customers; improved human resource practices and communication within the firm; etc. (as argued for instance in the literature on telecommunications reform and manufacturing productivity, e.g. in Arnold et al., 2016). Using cross-sections of firms across developing countries, Clarke et al. (2015) and Paunov and Rollo (2015) show that firm growth and productivity are much higher when internet access is greater and when firms use the internet more intensively. These studies are prone to endogeneity biases as ICT investments and internet-use by firms are not exogenous to their (expected future) productivity.7 Recently, progress has been made in addressing the endogeneity of internet use with respect to firm productivity following different approaches. A few studies use data on sub-national variation in the availability of internet infrastructure as an instrument for internet usage at the firm-level to estimate its impact on firm productivity. Bertschek et al. (2013) and Haller and Lyons (2015) find an insignificant impact of broadband internet on firm productivity in Germany and Ireland, respectively, whereas Grimes et al. (2012) and Akerman et al. (2015) find a positive impact of broadband internet on firm productivity in New Zealand and Norway, respectively. The latter study 6 They draw on the exogenous features of the historic subterraneous fiber-optic and copper-wire telephone network which is the infrastructure through which broadband internet is delivered in the United Kingdom. 7The fact that more productive firms use more ICT suggests there may be barriers to firms using ICT more effectively (Paunov and Rollo, 2015; World Bank, 2016). Two such barriers for firms are workforce skills and complementary organizational capital (Bresnahan et al., 2002; Haltiwanger et al., 2003; Crespi et al., 2007; Bartel et al., 2007; Commander et al., 2011; Iacovone et al., 2015). There is even evidence that investing in ICT without business process reorganization can reduce firms’ productivity growth (Brynjolfsson and Hitt, 2003). 5 argues that limited funding from a public investment program to roll out broadband internet access points in Norway led to a rollout which was exogenous to local productivity growth, allowing for a causal interpretation of their findings. In all these studies, firms already had access to the internet through analog connections before.8 Thus, these studies are likely to capture only incremental improvements due to access to broadband internet. Our paper resembles this earlier work in also exploiting sub-national variation in access to the internet over time, drawing upon the unprecedented scale and speed of internet infrastructure rollout in China in the period 1999-2007. But our primary focus is on the impact of access to the internet per se. We control for time-varying unobserved provincial and firm attributes to isolate the exogenous variation in internet access, and then we also control for fiber optic rollout, as a proxy for access to broadband (for which data are not available for China). We also dig a little deeper than the existing literature into the channels through which access to the internet affects firm performance. Earlier research finds that the impact of communication costs on trade in differentiated products, identified using the Rauch classification of product heterogeneity, is as much as one-third larger than on trade in homogenous products (Fink et al., 2005). The internet is arguably an even more powerful means of reducing information frictions, that is, the costs of search for and communications with buyers and input suppliers. Therefore, we test whether firms that produce communication-intensive differentiated products benefit relatively more from access to the internet. The internet can also encourage firms to use a larger number of differentiated intermediate goods or services by reducing the communication and coordination costs involved in coordinating with a large number of specialized suppliers to produce higher quality products. However, it is also possible that internet-based price comparisons are harder for some differentiated goods and services and these products may, therefore, be less suited for internet-based transactions.9 We test the relative strengths of these competing arguments using the measure of contract intensity proposed by Nunn (2007). 8 For example, in Bertschek et al. (2013) the estimating sample of firms in Germany starts in the early 2000s when 60 percent of firms already had access to broadband internet in Germany. 9 For instance, term life insurances are a standardized product that has seen significant declines from online price comparisons, but more differentiated whole-life insurances are less amenable to online price comparisons and have therefore not seen a comparable decline in prices (Brown and Goolsbee, 2002). Transactions with input providers that require relationship-specific investments and hence contracts may also be less amenable to complete internet-based transactions, although even these transactions may be facilitated by reductions in search costs. 6 3. Data In our analysis, we use firm-level manufacturing data, province-level internet penetration, and industry-level measures of internet use. 3.1 Firm-Level Manufacturing Data The firm-level data set used in our analysis is the annual survey conducted by China’s National Bureau of Statistics (NBS) covering all industrial firms that are state-owned, and those not state-owned with annual sales exceeding 5 million RMB (about 770,000 USD) for the period 1998-2007.10 Our analysis focuses on manufacturing firms, thus excluding mining and public utilities’ firms. A unique identifier makes it possible to follow firms over time. The data set is an unbalanced panel of firms whose number varies from a low of 147,109 in 1999 to a high of 312,365 in 2007.11 Firms are classified into the 4-digit Chinese Industrial Classification (CSIC) industries, but we work with the more aggregated 3-digit industry classification because that can be concorded with other data we use. The NBS survey collects firm-level information on output, number of workers, expenditures on intermediates, total assets, and exports, which we use to construct our firm outcome measures. A key variable for our analysis is the firm’s location in a given Chinese province. The NBS data provide information on either the actual province of a firm or the address of a firm with a postal code which allows us to identify the province.12 We apply the following cleaning procedures to get to the final estimating sample: we drop from the sample observations with a missing firm identifier or with missing information or negative values for total output, number of workers, total exports, total assets, expenditures on intermediates, province, or industry. The firm-level outcomes used in our analysis are defined as follows. Export participation is an indicator variable taking a value of 1 if firm nominal exports are positive and 0 otherwise. 10 These data have been used extensively in recent research on productivity, export decisions, and many other micro topics, for example in Brandt et al. (2012) and Du et al. (2012). Brandt et al. (2012) show that in comparison to a full census of firms in China the NBS data account for only 20% of industrial firms (the other 80% are smaller than the 5 million RMB cutoff), but they account for more than 70 percent of the industrial workforce, 90 percent of output and, importantly for our analysis, 97.5 percent of exports. 11 The differences in the number of firms across years can be driven by increases or reductions in annual sales relative to the 5 million RMB threshold or by changes in ownership. 12 The NBS data set includes for years 2004-2007 a variable with the Chinese name of the province. For firms present in the data set in these years this province information is used to fill in values for the previous years. For firms not present in the data set in these years, we use information on the zip code of the firm’s address, given that zip codes in China have a structure whereby one part of the code reflects the province. For a tiny share of firms (1,355 firms out of 570,108 or 0.23%) there is a change in province during the sample period. 7 Export intensity is defined as the ratio between firm nominal exports and nominal total output. Real output is defined as firm nominal total output deflated by an industry output deflator taken from Brandt et al. (2012). Real exports are defined as firm nominal exports deflated by the same output deflator. Employment is defined as the firm total number of workers. Labor productivity is defined as firm real value added divided by employment, where real value added is obtained as real output minus real intermediates, defined as nominal expenditures on intermediates divided by an input deflator taken from Brandt et al. (2012). A total factor productivity (TFP) Törnqvist index is obtained following Caves et al. (1982) with firm real output as the output measure and employment, real intermediates, and capital as inputs. Capital is obtained as firm net fixed assets deflated by a regional fixed investment price index from the NBS, following Du et al. (2014). For the TFP index calculation: the elasticity of labor is obtained as the ratio of nominal wages to nominal output; the intermediates elasticity is obtained as the ratio of nominal intermediates to nominal output; and constant returns to scale are assumed so the elasticity of capital is one minus the elasticities of labor and intermediates; all three are averaged at the 4-digit industry level. Two remarks should be made about the NBS data. First, the NBS data do not include information on Chinese firms’ use of ICT, such as the internet or computers. This is not necessarily a handicap for our analysis since firms’ decisions to use the internet are correlated with their production and export decisions as well as their productivity, and this endogeneity would bias the estimated effects of the internet on firm outcomes. We thus rely on a subnational data set on internet rollout, described in Section 3.2, to measure firms’ access to the internet depending on their location. Second, the NBS data include information on firm exports to foreign countries, but unfortunately not on domestic sales to other provinces which may also be affected by increased access to the internet. Table 1 presents some descriptive statistics on the estimating sample, including information on the number of firms per year (Panel A), and the means and standard deviations of our firm-level outcome variables (Panel B). As a check of the quality of our firm TFP measure, we calculated average annual TFP growth for Chinese manufacturing firms over the 1998-2007 period and found that to be about 2.89 percent, which is extremely close to the magnitude of 2.85 percent provided by Brandt et al. (2012) in their paper focused on TFP estimation. Figures 1 and 2 show the number of exporting firms per province and total exports per province (obtained as the sum of exports across all firms) over time. Exporting firms and exports as a share of the population 8 were initially highly concentrated in the coastal areas also increased in several inland provinces in later years. 3.2 Provincial Internet Penetration Data Data on the internet rollout across Chinese provinces over time is central to the analysis. Our main measure of internet penetration is the number of internet users per province-year, obtained from various editions of the annual reports on the internet produced by the China Internet Network Information Center (CNNIC) for years 1999-2007. To account for the heterogeneous sizes of Chinese provinces and the private good nature of internet use, our final measure is scaled by province population; that is, the number of internet users per 10,000 people, designated in what follows as number of internet users per capita. An alternative measure of internet penetration is the number of .CN domain names, also obtained from the CNNIC for years 1999-2007. A .CN domain name represents a personal computer used to access the Internet, a server computer hosting a website, the website itself, or any other service communicated via the Internet. The fact that our measure covers top level .CN domain names, rather than just domain names of industrial, commercial, and financial enterprises, implies that it represents a broader measure of internet penetration or activity (including that of academic institutions, government agencies, non-profit organizations, and others) and is thus less vulnerable to endogeneity problems with respect to manufacturing firms’ outcomes. The number of .CN domain names can be interpreted as the total internet content produced in a province by businesses, government, and households, and hence goes beyond the number of internet users per capita by also capturing the depth of the internet penetration. Since the number of .CN domains have both private and public good aspects, it is not clear whether our final measure should be the number of .CN domains scaled by the population of the province or the absolute number of .CN domains. We will, therefore, use both the number of .CN domains per 10,000 people, designated in what follows by number of .CN domains per capita as, and the number of .CN domains. We will consider, as an additional measure, the length of the fiber optic line (in kilometers) per province-year obtained from the NBS for years 1999, 2002, 2003, and 2005-2007. At the beginning of the period, fiber optic lines in China were primarily in the backbone (city-to-city) network. The final user had access to the network through digital subscriber lines (DSL) technology but had to be within a relatively short distance of the fiber optic line to get the full 9 benefits. Towards the end of the period, almost every Chinese county had broadband internet connections though it is not clear how many final users had access to broadband.13 Since the optical fiber line is a public good, our final measure seeks to capture access to it by multiplying the length of the optic fiber line by the population density in the province, and is designated in what follows by number of people per fiber optic line kilometer. The optical fiber line is arguably a relatively exogenous measure of internet penetration, but since it is available for a smaller number of sample years and since by itself it does not constitute a perfect measure of internet penetration, we use it as a complementary variable to address a specific question. For an instrumental variable specification, we use data from the Ministry of Information Industry of China on the telecommunication industry’s mobile revenues and the volume of minutes used for each Chinese province in 1998-2007 to calculate telecommunication prices. These prices are obtained as the ratio between mobile revenues and the volume of minutes used, deflated by a consumer price index taken from the from the Chinese Statistical Yearbook.14 For each province in a particular year, we compute the average of mobile prices across all its neighboring provinces and use this as the instrument for provincial internet penetration, adapting the approach proposed by Zhang and Ward (2012). The rationale for this instrument is that telecommunication prices reflect the state of development of telecommunications services in neighboring provinces, and indirectly in the province of interest, but do not directly affect firm performance in that province.15 To place the growth of internet penetration in context, Chinese authorities began to introduce reforms of telecommunications policies in the early 1990s, but the country’s entry into the World Trade Organization (WTO) in 2001 accelerated the reform process and stimulated the development of the telecommunications sector (Yu et al. 2004 and Zheng and Ward, 2011). The expansion of the telecommunications network was initially limited to major cities along the eastern coast of China due to a lack of capital. Later, additional capital was raised in the stock market to expand the telecommunications networks in other provinces in the central and western regions. China’s WTO commitments also reflect a pattern of gradual liberalization of foreign entry, staggered across regions in fixed voice and data (2004-2007), mobile (2001-2006) and value added 13 Importantly, while reports discuss the number of broadband internet users in China in aggregate, information on the number of broadband internet users by province-year in our sample period is not published by CNNIC or any other source. Thus, we are unable to use the number of broadband internet users as a measure of internet access. 14 The telecommunication price variables were provided to us by Michael Ward. 15 Data for Hainan, a Chinese province consisting of various islands, are not considered in this exercise since no neighboring provinces can be identified. 10 services (2001-2003). Typically, in the first stage foreign service suppliers were first permitted to establish joint venture enterprises with stringent limits on foreign ownership (less than 25-30 percent) to provide services in and between the cities of Shanghai, Guangzhou and Beijing. In the second stage, the ownership restrictions were relaxed to a limited extent (about 35 percent) and the geographic scope of activities was extended to services in and between Chengdu, Chongqing, Dalian, Fuzhou, Hangzhou, Nanjing, Ningbo, Qingdao, Shenyang, Shenzhen, Xiamen, Xi'an, Taiyuan and Wuhan. In the final stage, foreign investment limits were further relaxed (to 49 or 50 percent) and geographic restrictions eliminated. The evidence above suggests that during our sample period the Chinese government managed the spread of telecommunications services, and hence the internet, to a certain extent. While the rollout of the internet across Chinese provinces was not strictly exogenous, it is highly unlikely that the internet network expansion was driven by the performance of any particular firm in China. Therefore, we will argue that our identification strategy, and the comprehensive controls for constant and time-varying unobservable provincial and firm attributes in our econometric specifications, mitigate the concerns of endogeneity between provincial internet penetration and firm performance in China. Figures 3 through 5 show the internet rollout across Chinese provinces over the sample period according to each of our measures. The number of internet users per capita increased substantially from a low base in all Chinese provinces between 1999 and 2007. The increase was universal across provinces but was stronger in coastal areas in the earlier years and in several inland provinces in later years. The number of people per fiber optic line kilometer also increased across Chinese provinces between 1999 and 2007, though it was already high to start with in the Henan province in 1999. A comparison of Figures 3 to 5 on internet penetration and Figures 1 and 2 on the number of exporting firms and exports in each province suggests a positive correlation. But there are many confounding factors and possible reverse causality; therefore, we follow a specific empirical identification strategy, which is described in Section 4 and based on the following additional data. 3.3 Measures of Intensity of Internet Use 11 The last element of information we use in our analysis is a measure of the intensity with which each industry uses the internet. The internet is likely to be more important for some industries than others due to differences in their products, production processes and transactions processes. We consider three measures, each with its own limitations, but each providing some information on the intensity with which industries use the internet. Our first measure is the share of post and telecommunications expenses in total intermediate expenses at the 2-digit industry- level for China. This measure is designated as China share of post and telecommunications, and based on an input-output table for China in 1998, obtained from the World Input-Output Database. Our second measure is the share of information and communications technology (ICT) capital services in total capital (the sum of ICT and non-ICT capital) services at the 2-digit industry level for the United States in 1998. This measure is designated as US share of ICT capital services, and based on U.S. data in the KLEMS database. This measure of sectoral ICT intensity follows most previous studies on ICT and productivity, starting with Stiroh (2002).16 By focusing on the U.S. this measure can be viewed as providing information on the technological characteristics of the industry, and its use alleviates concerns that may arise with the China share of post and telecommunications being endogenous to firm performance in China. A third measure, which we will only use in conjunction with the number of .CN domains as a measure of internet penetration, is the share of firms with a website at the 3-digit industry level in Vietnam in 2007. This measure is designated as Vietnam share of firms with websites, and based on the 2007 Vietnam Enterprise Census.17 Vietnam’s production structure and processes exhibit similarities to those in China, since Vietnam followed a similar growth path based on exports and foreign investment. Therefore, using data for Vietnam is an appropriate choice and again empirically preferable to the use of data on firms with websites for China (if it were available) due to endogeneity concerns. Our use of measures based on data for countries other than China implies an assumption that the internet intensity of industries in those countries is a good proxy for the potential degree of reliance on the internet of firms in the same industries in China. For illustrative purposes, Appendix Table 1 shows the actual measures of intensity of internet use at the 2-digit level. Industries with generally higher reliance on the internet are 16 In contrast to Stiroh (2002) we do not exclude IT-producing industries from our analysis―they naturally exhibit some of the highest shares of ICT capital services. 17 The Vietnam Enterprise Census is collected by the General Statistics Office in Hanoi and covers about 300,000 firms each year classified according to the 4-digit of the Vietnamese Industrial Classification which is aggregated and concorded to the Chinese firm-level data at the 3-digit level based on the verbal descriptions, as the two classifications are quite similar drawing on the ISIC. 12 transport equipment, electrical and optical equipment, and machinery not elsewhere classified, while those with generally lower reliance on the internet are food, beverages, and tobacco, pulp, paper, printing, and publishing, and coke, refined petroleum, and nuclear fuel. The simple and rank correlations among measures of intensity of internet use at the 2-digit or 3-digit levels are shown in Appendix Table 2 and are generally positive, in some cases significant. 4. Empirical Specification To estimate the effects of internet penetration on the export performance of Chinese manufacturing firms we consider empirical specifications of the form: = + _ ∗ _ + + + (1) where j stands for a firm, p for a province, s for a sector and t for a year, is an outcome measure, and is an independent and identically distributed error (i.i.d). The regressor of interest is the interaction between a measure of internet penetration at the province-year level _ with a measure of the intensity of internet use at the industry level _ . The interaction term reflects the idea that Chinese firms in industries that rely more intensively on the internet should benefit more from increases in internet penetration in their province than firms in industries that rely less intensively on the internet (ceteris paribus). This assumption is inspired by Rajan and Zingales (1998) and implies that there are no restrictions (other than cost) preventing access to the internet by firms in certain provinces.18 Eq. (1) controls for province-year fixed effects, which capture any fixed and time-varying unobserved provincial attributes.19 Eq. (1) also includes firm fixed effects, which control for unobserved differences across firms that influence their export performance.20 Eq. (1) thus tests whether the effect of provincial internet penetration is stronger for exports of firms in industries that are more reliant on the internet while controlling for fixed and time-varying unobserved provincial and firm attributes. The inclusion of province-year fixed effects in all our empirical specifications controls for other improvements contemporaneous with internet penetration, such as investments in provincial 18 Our specification is a generalized difference-in-differences specification as in Rajan and Zingales (1998) who study the effects of financial development on industry growth. In their analysis, the dependent variable is industry growth, the main coefficient of interest is the interaction between a country-year measure of financial development and an industry measure of dependence on external finance, and there are controls for industry fixed effects and country fixed effects. 19 The province-year fixed effects also account for the actual level of provincial internet penetration as part of the interaction regression model. 20 The firm fixed effects are also a way to account for industry attributes (since each firm is in a single industry), including the industry-level intensity of internet use as part of the interaction regression model. 13 transport or logistics infrastructure which facilitate exports as well as for the presence of special economic zones and distance to ports. These fixed effects address a concern that could be raised by Figures 3-5, that is picking up the effect of not internet penetration but of other province- level time-varying developments. The interaction of _ with industry internet intensity in Eq. (1) further ensures that we identify the impact of provincial internet penetration only through firms whose businesses rely more intensively on the internet as opposed to other infrastructure services. A potential reverse causality concern could still arise with Eq. (1): industries with higher internet intensity could be seeing faster export growth in provinces where internet penetration is growing faster for reasons unrelated to improvements in internet availability. It is difficult to construct a compelling alternative economic argument, other than the one suggested in this paper, for this specific correlation. It would require, for example, the existence of unobserved export promotion industrial policies, exclusively targeting industries intensive in the use of the internet in the provinces where internet penetration is also growing faster. One way of addressing this concern, to estimate a variant of Eq. (1) including industry-province time trends, proved to be computationally impossible for the Chinese firm-level data set. As an alternative, we estimated a variant of Eq. (1) that accounts for time variation at the province-industry level by controlling for the following additional variables: province-industry growth in output and province-industry growth in exports.21 We address any remaining concerns about the endogeneity of internet penetration by employing an instrumental variable strategy. We use as an instrument for internet penetration in a province, an indicator of the state of telecommunications services in neighboring provinces: the average price of mobile telecommunication services.22 The dependent variables in our specifications vary at the firm-year level but our regressor of interest varies at a more aggregated province-industry-year level, therefore, inference will be based on Huber-White standard errors robust to heteroskedasticity clustered at the province- industry-year level, following Moulton (1990).23 To sum up, the main coefficient of interest, , is estimated based on the within-firm variation in performance over time due to increases in internet penetration in the firm’s province for firms that operate in industries using the internet 21 Additionally, a selection problem could arise with Eq. (1) if a disproportionally large number of exporting firms in industries that use the internet more intensively relocated (for incumbent firms) or entered (for new firms) into provinces with higher internet penetration during our sample period, which our data show was not the case. 22 We acknowledge that mobile prices reflect also the level of competition in the telecommunication market. 23 Results are robust to the clustering of standard errors at less stringent levels of disaggregation. 14 more intensively, which, we argue, provides a causal impact of the internet on firm performance in China. 5. Results We begin by presenting evidence of the effect of internet penetration measured by the number of internet users per capita on firm export performance in China. The estimates in columns (1)-(2) of Table 2 show that higher provincial internet penetration increases the probability of export participation significantly more for firms in industries that rely more heavily on the internet. Focusing only on the firms that export, columns (3)-(4) show that the share of exports in firms’ total output, i.e., their export intensity, is significantly higher for firms in industries that use the internet heavily in provinces with higher internet penetration. Finally, columns (5)-(6) show that in provinces with higher internet penetration, real exports per firm are also significantly higher for firms in industries relying more on the internet. For the three firm export outcomes in Table 2, the effects are strong regardless of whether we rely on the China share of post and telecommunications or the US share of ICT capital services as the measure of industry intensity of internet use. The estimated coefficients in Table 2 suggest that a one standard deviation change in the log of the provincial number of internet users per capita is associated with a 0.4 percent increase in the firm likelihood of exporting, a 0.005 increase in the firm export-output ratio, and a 0.08 increase in firm log real exports, for firms in an industry with the average intensity of internet use, as measured by the China share of post and telecommunications.24 The results in Table 2 are robust to the inclusion of province-industry growth in real output and in real exports, as shown in Appendix Table A.3. Interestingly, while province-year growth in real exports has a significant positive correlation with firm export performance, province-year growth in real output is either uncorrelated or negatively correlated with firm export performance. As noted above, a potential reverse causality concern could still arise in our empirical specification if industries with higher internet intensity were seeing faster export growth in provinces where internet penetration is growing faster, for reasons unrelated to improvements in internet availability. We address this concern by employing an instrumental variable strategy. We 24The magnitudes are obtained as the product of three values: the coefficient in columns (1), (3) or (5), respectively, the standard deviation of the logarithm of the provincial number of internet users per capita (1.4 for the sample used in column (1) and 1.25 for the sample use in columns (3) and (5)), and the China share of post and telecommunications for the average industry (0.006). 15 use as an instrument for internet penetration in a province, an indicator of the state of telecommunications services in neighboring provinces: the price of mobile telecommunication services. Table 3 presents the IV estimates which are qualitatively similar to those in Table 2. The F-statistic shows a strong first-stage while the reported p-value for the Kleibergen-Paap under- identification test suggests that the proposed instrument is valid. Next, we delve into the reasons why firm export performance is positively affected by provincial internet penetration, focusing on three broad questions: (1) What other aspects of firm performance were affected? (2) What types of firm relationships were changed? and (3) What dimensions of the new communication medium were most relevant? First, we explore which and how different aspects of firm performance in the domestic market were affected. Table 4 presents the results from estimating Eq. (1) for firm real output, productivity, and employment. Columns (1)-(2) show a significant increase in firm real output as provincial internet penetration grows, regardless of the measure of industry reliance on the internet used. Hence, firms in China are not simply re-orienting their existing output levels to export markets with the advent of the internet but rather they are increasing their scale of production. The growth in real output originates, at least partly, in higher firm efficiency as columns (3)-(6) show positive and generally significant effects of internet penetration on firm labor and total factor productivity. The weaker effects on total factor productivity suggest that growth in firm labor productivity is also accompanied by firm capital deepening. We also investigate whether the increase in firm productivity in face of growing provincial internet penetration was achieved at the expense of firm employment. Columns (7)-(8) show that the effects of internet penetration on firm employment are positive (though not significant when industry internet use is proxied for by the US share of ICT capital services). This evidence suggests that growing internet penetration enabled firms to increase their exports based on scaling up their output by hiring more employees but also by enhancing the productivity of the workforce. Export growth was thus not purely a substitution of exports for domestic output but reflected an overall expansion of firms fueled by efficiency gains presumably for the reasons discussed in Section 2. Of course, the fact that Table 2 shows an increase in the export share of output suggests that internet penetration had a stronger impact on foreign sales than domestic sales. This finding is plausible because the internet can be expected to deliver greater marginal informational benefits on foreign than domestic markets. Estimates (not 16 reported here) following the same IV strategy as in Table 3 show qualitatively similar findings for the impacts of internet penetration on firm real output, labor productivity, and employment. Second, we explore the impact on firms’ relationships with buyers and input suppliers. Some of the conceptual mechanisms for the gains in firm export performance from internet penetration discussed in Section 2 relate to the internet being a means to reduce information frictions, and hence costs of search and communications with potential and actual buyers and input suppliers. Since transactions in industries that manufacture differentiated products are more communications-intensive than transactions for industries that manufacture homogeneous products, there is likely to be greater potential in the former for the internet to reduce communications costs faced by firms and hence boost exports. We define a dummy variable identifying the 3-digit industries with differentiated products, based on the classification proposed by Rauch (1999), designated in what follows as Rauch differentiated products.25 We re-estimate Eq. (1) interacting the Rauch differentiated products measure with the provincial number of internet users per capita. The estimated effects on firm export outcomes are shown in columns (1), (3), and (5) of Table 5. The probability of participating in export markets, the ratio of exports to total output, and real exports all increase significantly more for firms in the more communications- intensive industries - the Rauch differentiated products industries – as provincial internet penetration increases. To explore heterogeneity of effects on the input side, we identify a measure that captures the degree to which highly specialized and customized inputs are used in an industry’s production process. As we noted in Section 2, the internet can help reduce the costs of communicating and coordinating with a large number of specialized suppliers; but the internet may also be less suitable for conducting transactions for these inputs because price comparisons are harder to make or because sophisticated contracts are needed to address relationship-specific investments. We use the measure of contract intensity proposed by Nunn (2007), which is given by the share of specialized and customized intermediate inputs used in the production of the final good for each 3-digit industry based on a U.S. input-output table, designated in what follows as Nunn contract intensity.26 We re-estimate Eq. (1) interacting the Nunn contract intensity measure with the 25 The Rauch classification is provided at the 4-digit SITC level which is concorded to the Chinese firm-level data at the 3-digit level based on verbal descriptions and on an SITC to ISIC concordance. We focus on Rauch’s “liberal” definition of industries with differentiated products. 26 Nunn uses data on the fraction of each input that is not sold on an organized exchange nor reference-priced according to Rauch (1999) to construct the share of intermediate inputs that require specialized and customized business relationships. Nunn’s contract 17 provincial number of internet users per capita. The estimated effects on firm export outcomes are shown in columns (2), (4), and (6) of Table 5. We find clear evidence of a significant increase in firm export participation, export intensity and real exports in the specialized input-intensive industries as a province’s internet penetration grows, suggesting that the trade facilitation role of the internet outweighs other offsetting considerations. All in all, the evidence in Table 5 shows that the gains from increased internet penetration for firm export performance have arisen from improvements in communications with clients and suppliers. Third, we examine what dimensions of the new communication medium were most relevant. Specifically, we test if industries in which firms typically rely more on websites see a greater improvement in export performance when internet penetration increases in a province. We use as measures of provincial internet penetration, the number of .CN domains per capita as well as the total number of .CN domains, which reflect the internet content that firms and households generate in that province. In particular, the number of .CN domains increases with the number of online services such as websites in the province. We pair these measures with the share of firms with websites in specific industries in Vietnam as the interaction term in re-estimating Eq. (1). The results shown in Table 6 show a significant increase in the probability of participating in export markets, the ratio of exports to total output, and real exports for firms in industries relying more heavily on websites in provinces where the numbers of .CN domains per capita or of .CN domains in absolute terms grew faster. Hence, we conclude that the gains from increased internet penetration for firm export performance in China have arisen also from using online services, including by establishing a visible virtual presence.27 We also examine whether it is access to broadband internet that improves firm export outcomes or access to the internet per se. Our sample period covers the beginning of the rapid expansion of broadband internet penetration in China but data on the provincial number of broadband internet users before 2007 are not available. However, we can approximate expansion in the access to broadband internet by using changes in the kilometers of optical fiber line per province, which is one of the key means through which broadband internet is delivered. We estimate a variant of Eq. (1) where a second interaction term is included between the number of intensity measure (based on Rauch’s liberal classification) is available at the 3-digit ISIC revision 2 level which is concorded to the Chinese firm-level data at the 3-digit level based on verbal descriptions. 27 Estimates (not reported here) interacting the numbers of .CN domains or numbers of .CN domains per capita with either the China share of post and telecommunications or the US share of ICT capital services, are qualitatively similar to those in Table 6. 18 people per fiber optic line kilometer and each of the industry intensity of internet use measures used in Table 2, in addition to the interaction term using the number of internet users per capita. The results are shown in Table 7.28 The estimates show that firms’ export participation, export intensity, and real exports are significantly higher in provinces with more people per fiber optic line kilometer. Interestingly, the number of internet users per capita is still significant after controlling for the effect of more fiber optic line kilometers per capita. The results thus suggest that while broadband internet connection availability enhances firm export performance, the positive and significant pure effect of internet penetration is maintained. 6. Conclusion We take advantage of the scale and speed of the internet rollout in China to analyze how the surge in internet penetration, including in less developed inland provinces, improved firms’ export performance. We combine firm-level production data with provincial information on internet penetration in China between 1999 and 2007. We focus on the impact on firm exports since trade is a major channel through which the internet can raise aggregate economic growth. Our econometric strategy allows us to identify a causal impact of the internet on firm performance. We find that growth in internet penetration has raised Chinese firms’ manufacturing exports. We take a closer look at why, and reach three broad conclusions. We find, first, that the internet did not just enhance trade but improved overall firm performance; higher exports go along with an increase in firms’ real output as provincial internet penetration increases. Thus, firms in China are not simply re-orienting their output to export markets but also increasing the scale of their production. The growth in firms’ real output is due to an increase in productivity as well as in employment. Second, we show that the gains from increased internet penetration on firm export performance are likely to have arisen from a reduction in communications costs with buyers and input suppliers. Third, the benefits from increased internet penetration on firm export performance were also linked to the establishment of a visible virtual presence and were enhanced by but not contingent on access to broadband. 28 The sample sizes in Table 7 differ from those in Table 2 due to the availability of fewer years of data for the number of people per fiber optic line. 19 It is interesting that access to internet boosted exports in China even before the emergence of Alibaba as a widely-used, large-scale e-commerce platform (and before eBay’s market expansion in China). These findings suggest that the internet significantly reduces trade costs even in the absence of broadband and major international e-commerce platforms serving the domestic market. Recognition of these benefits of access to more basic internet services could inform policy in developing countries that are trying to strike a balance between widening access to basic services and deepening it through the creation of a broadband network. 20 References Akerman, A., Gaarder, I., and M. Mogstad (2016). “The Skill Complementarity of Broadband Internet,” Quarterly Journal of Economics 130 (4): 1781-1824. Arnold, J., Javorcik, B., Lipscomb, M., and A. Mattoo (2016). “Services Reform and Manufacturing Performance: Evidence from India,” Economic Journal 126 (590): 1-39. Bartel, A., Ichniowski, C. and K. Shaw (2007). “How Does Information Technology Really Affect Productivity? Plant-Level Comparisons of Product Innovation, Process Improvement and Worker Skills,” Quarterly Journal of Economics 122 (4): 1721-1758. Bertschek, I., Cerquerab, D., and G. Kleinc (2013). “More Bits – More Bucks? Measuring the Impact of Broadband Internet on Firm Performance,” Information Economics and Policy 25 (3): 190–203. Blum, B. and A. Goldfarb (2006). “Does the Internet Defy the Law of Gravity?,” Journal of International Economics 70 (2): 384–405. Brandt, L., Van Biesebroeck, J. and Y. Zhang (2012). “Creative Accounting or Creative Destruction? Firm-level Productivity Growth in Chinese Manufacturing,” Journal of Development Economics 97 (2): 339-351. Brown J. and A. Goolsbee (2002), “Does the Internet Make Markets More Competitive? Evidence from the Life Insurance Industry,” Journal of Political Economy 110 (3): 481-507. Bresnahan, T., Brynjolfsson, E. and L. Hitt. 2002. “Information Technology, Workplace Organization, and the Demand for Skilled Labor: Firm-Level Evidence,” Quarterly Journal of Economics 117 (1): 339-76. Brynjolfsson, E. and L. Hitt. (2003). “Computing Productivity: Firm-level Evidence,” Review of Economics and Statistics 85 (4): 793–808. Caves, D., Christensen, L., and E. Diewert (1982). “The Economic Theory of Index Numbers and the Measurement of Input, Output, and Productivity,” Econometrica 50 (6): 1393-1414. Chen, M. and M. Xu (2015). “Online International Trade in China,” Background Paper for the World Development Report 2016. World Bank, Washington, DC. Clarke, G., Qiang, C. and L. Xu (2015). “The Internet as a General-Purpose Technology: Firm- Level Evidence from around the World,” World Bank Policy Research Working Paper 7192. Clarke, G. and Scott J. Wallsten (2006). “Has the Internet Increased Trade? Developed and Developing Country Evidence,” Economic Inquiry 44 (3): 465–84. 21 Commander, S., Harrison, R. and N. Menezes-Filho (2011). “ICT and Productivity in Developing Countries: New Firm-Level Evidence from Brazil and India,” Review of Economics and Statistics 93 (2): 528-541. Crespi, G., Criscuolo, C., and J. Haskel (2007). “Information Technology, Organizational Change and Productivity Growth: Evidence from UK Firms,” CEP Discussion Paper no. 783. Du, L., Harrison, A., and G. Jefferson (2014). “FDI Spillovers and Industrial Policy: The Role of Tariffs and Tax Holidays,” World Development 64 (C): 366-383. Fink, C., Mattoo, A., and I. Neagu (2005). “Assessing the Impact of Communication Costs on International Trade,” Journal of International Economics 67 (2): 428-445. Freund, C. and D. Weinhold (2002). “The Internet and International Trade in Services,” American Economic Review 92 (2): 236-240. Freund, C. and D. Weinhold (2004). “The Effect of the Internet on International Trade,” Journal of International Economics 62 (1): 171–89. Grimes, A., Ren, C. and P. Stevens (2012). “The Need for Speed: Impacts of Internet Connectivity on Firm Productivity,” Journal of Productivity Analysis 37 (2): 187-201. Haller, S. and S. Lyons (2015). “Broadband Adoption and Firm Productivity: Evidence from Irish Manufacturing Firms,” Telecommunications Policy 39 (1): 1-13. Haltiwanger, J., Jarmin, R., and T. Schank (2003). “Productivity, Investment in ICT and Market Experimentation: Micro Evidence from Germany and the US.” Technical report, US Census, Economic Studies. Iacovone, L., M. Pereira-Lopez and M. Schiffbauer (2015): “ICT Use, Competitive Pressures and Firm Performance in Mexico”, World Bank Economic Review, Papers and Proceedings special issue of the World Bank 2015 ABCDE Conference. Kneller, R., and J. Timmis (2016). “ICT and Exporting: The Effects of Broadband on the Extensive Margin of Business Service Exports,” Review of International Economics 24 (4): 757-796. Kshetri, N. and Nikhilesh and Dholakia (2007): “Drivers of the Broadband Industry in China and India: What Can We Learn?” PTC proceedings, unpan1.un.org Lin, F. (2015). “Estimating the Effect of the Internet on International Trade,” The Journal of International Trade & Economic Development 24 (3): 409-428. Moulton, B. (1990). “An Illustration of a Pitfall in Estimating the Effects of Aggregate Variables on Micro Units,” Review of Economics and Statistics 72 (2): 334-338. Nunn, N. (2007). “Relationship-Specificity, Incomplete Contracts, and the Pattern of Trade,” Quarterly Journal of Economics 122 (2): 569-600. 22 Osnago, A. and S. Tan (2015). “The Effects of the Internet on Trade Flows and Patterns,” Background Paper for the World Development Report 2016. World Bank, Washington, DC. Paunov, C. and V. Rollo (2016). “Has the Internet Fostered Inclusive Innovation in the Developing World?,” World Development 78 (C): 587-609. Rajan, R. and L. Zingales (1998). “Financial Dependence and Growth,” American Economic Review 88 (3): 559-586. Rauch, J. (1999). “Networks versus Markets in International Trade,” Journal of International Economics 48 (1): 7-35. Rauch, J. and V. Trindade (2002). “Ethnic Chinese Networks in International Trade,” Review of Economics and Statistics 84 (1): 116-30. Stiroh, K. (2002). “Information Technology and the U.S. Productivity Revival: What Do the Industry Data Say?” American Economic Review 92 (5): 1559-1576. Ward, M. and S. Zheng (2012). “Mobile and Fixed Substitution for Telephone Service in China,” Telecommunications Policy 36 (4): 301-310. World Bank (2016). World Development Report: Digital Dividends. The World Bank, Washington D.C. Yu, L., Berg, S. and Q. Guo (2004). “Market Performance of Chinese Telecommunications: New Regulatory Policies,” Telecommunications Policy 28 (9): 715-732. Zhang, S. and M. Ward (2011). “The Effects of Market Liberalization and Privatization on Chinese Telecommunications,” China Economic Review 22 (2): 210-220. 23 Figure 1. Number of Manufacturing Exporting Firms per ‘000 Inhabitants across Chinese Provinces and Over Time 24 Figure 2. Total Manufacturing Exports per Million of Inhabitants across Chinese Provinces and Over Time 25 Figure 3. Number of Internet Users per Capita across Chinese Provinces and Over Time 26 Figure 4. Number of .CN Domains per Capita across Chinese Provinces and Over Time 27 Figure 5. Number of People per Fiber Optic Line Kilometer across Chinese Provinces and Over Time 28 Table 1. Summary Statistics on Manufacturing Firm-Level Data Panel A. Number of Firms per Year 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Number of Firms 149,693 147,109 148,267 156,804 166,859 181,064 256,595 251,043 278,726 312,365 Panel B. Summary Statistics on Firm-Level Outcome Variables across 1998-2007 Number of Standard Average Minimum Maximum Observations Deviation Dummy for Export Participation 2,048,525 0.269 0.443 0 1 Export Intensity (If Export Participation = 1) 521,606 0.582 0.371 0.000 1 Real Exports (If Export Participation = 1) 550,390 9.398 1.716 -0.173 19.039 Log Real Output 2,010,232 9.900 1.408 -1.065 18.883 Log Employment 2,027,205 4.740 1.172 -2.303 12.145 Log Labor Productivity 1,997,993 5.146 1.184 -8.137 15.535 Log Total Factor Productivity 1,770,596 0.185 0.399 -2.793 9.261 Table 2. Impact of Internet Penetration (Internet Users) on Firm Export Performance Dependent Variable: Firm-Level Dummy for Export Firm-Level Export Intensity Firm-Level Real Exports (If Participation (If Export Participation = 1) Export Participation = 1) (1) (2) (3) (4) (5) (6) Province Number of Internet Users per Capita X Industry China share of Post and Telecommunications 0.410*** 0.661*** 9.722*** (0.084) (0.121) (0.939) Province Number of Internet Users per Capita X Industry US Share of ICT Capital Services 0.008*** 0.019*** 0.304*** (0.003) (0.004) (0.034) Firm Fixed Effects Yes Yes Yes Yes Yes Yes Province-Year Fixed Effects Yes Yes Yes Yes Yes Yes Number of Observations 1,762,394 1,762,394 437,407 437,407 464,661 464,661 R-squared 0.812 0.812 0.850 0.850 0.821 0.821 Notes: Robust standard errors in parentheses, clustered at the province-industry-year level. *** indicates significance at the 1% confidence level. 29 Table 3. Impact of Internet Penetration (Internet Users) on Firm Export Performance – IV Estimation Dependent Variable: Firm-Level Dummy for Export Firm-Level Export Intensity (If Export Firm-Level Real Exports (If Export Participation Participation = 1) Participation = 1) (1) (2) (3) (4) (5) (6) Province Number of Internet Users per Capita X Industry China share of Post and Telecommunications 1.346*** 3.065*** 53.065*** (0.307) (0.689) (7.601) Province Number of Internet Users per Capita X Industry US Share of ICT Capital Services 0.060*** 0.089*** 1.861*** (0.018) (0.023) (0.293) Firm Fixed Effects Yes Yes Yes Yes Yes Yes Province-Year Fixed Effects Yes Yes Yes Yes Yes Yes Number of Observations 1,758,329 1,758,329 437,056 437,056 464,246 464,246 R-squared 0.812 0.812 0.850 0.850 0.816 0.813 F-statistic from first-stage 193 64 50 37 47 34 Kleibergen-Paap underidentification test (p-value) 0.000 0.000 0.000 0.000 0.000 0.000 Notes: Robust standard errors in parentheses, clustered at the province-industry-year level. *** indicates significance at the 1% confidence level. The table reports two-stage least squares estimates of the second-stage given by Eq. (1). Province internet use is instrumented by an average of mobile telecommunication prices in the province’s neighboring provinces. Hainan province is excluded from the estimating sample. Table 4. Impact of Internet Penetration (Internet Users) on Firm Performance Dependent Variable: Firm-Level Real Firm-Level Labor Firm-Level Total Factor Firm-Level Output Productivity Productivity Employment (1) (2) (3) (4) (5) (6) (7) (8) Province Number of Internet Users per Capita X Industry China share of Post and Telecommunications 5.674*** 4.513*** 0.099 1.433*** (0.469) (0.462) (0.504) (0.264) Province Number of Internet Users per Capita X Industry US Share of ICT Capital Services 0.112*** 0.102*** 0.053*** 0.017* (0.017) (0.016) (0.015) (0.010) Firm Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Province-Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Number of Observations 1,731,278 1,731,278 1,729,848 1,729,848 1,525,083 1,525,083 1,752,904 1,752,904 R-squared 0.874 0.874 0.821 0.820 0.712 0.713 0.888 0.888 Notes: Robust standard errors in parentheses, clustered at the province-industry-year level. *** and * indicate significance at the 1% and 10% confidence levels, respectively. 30 Table 5. Impact of Internet Penetration (Internet Users) on Firm Export Performance – Differentiated and Contract-Intensive Industries Dependent Variable: Firm-Level Dummy for Export Firm-Level Export Intensity Firm-Level Real Exports (If Participation (If Export Participation = 1) Export Participation = 1) (1) (3) (5) (7) (9) (11) Province Number of Internet Users per Capita X Industry Rauch Differentiated Products 0.003*** 0.003*** 0.019*** (0.000) (0.001) (0.004) Province Number of Internet Users per Capita X Industry Nunn Contract Intensity 0.010*** 0.020*** 0.111*** (0.002) (0.004) (0.025) Firm Fixed Effects Yes Yes Yes Yes Yes Yes Province-Year Fixed Effects Yes Yes Yes Yes Yes Yes Number of Observations 1,723,388 1,660,611 426,546 421,746 452,956 447,985 R-squared 0.812 0.812 0.851 0.850 0.819 0.818 Notes: Robust standard errors in parentheses, clustered at the province-industry-year level. *** indicates significance at the 1% confidence level. Table 6. Regressions of Firm Export Outcomes on Number of .CN Domains Dependent Variable: Firm-Level Dummy for Export Firm-Level Export Intensity Firm-Level Real Exports (If Participation (If Export Participation = 1) Export Participation = 1) (1) (2) (3) (4) (5) (6) Province Number of .CN Domains per Capita X Industry Vietnam Share of Firms with Websites 0.003** 0.005** 0.092*** (0.001) (0.002) (0.014) Province Number of .CN Domains X Industry Vietnam Share of Firms with Websites 0.041*** 0.035*** 0.684*** (0.005) (0.006) (0.048) Firm Fixed Effects Yes Yes Yes Yes Yes Yes Province-Year Fixed Effects Yes Yes Yes Yes Yes Yes Number of Observations 1,583,690 1,583,690 400,714 400,714 425,691 425,691 R-squared 0.817 0.817 0.852 0.852 0.827 0.828 Notes: Robust standard errors in parentheses, clustered at the province-industry-year level. *** and ** indicate significance at 1% and 5% confidence levels, respectively. 31 Table 7. Regressions of Firm Export Outcomes adding People per Fiber Optic Line Dependent Variable: Firm-Level Dummy for Export Firm-Level Export Intensity Firm-Level Real Exports (If Participation (If Export Participation = 1) Export Participation = 1) (1) (2) (3) (4) (5) (6) Province Number of Internet Users per Capita X Industry China share of Post and Telecommunications 0.552*** 0.917*** 14.266*** (0.126) (0.186) (1.466) Province Number of People per Fiber Optic Line Kilometer X Industry China share of Post and Telecommunications 0.109** 0.149** 2.563*** (0.043) (0.056) (0.355) Province Number of Internet Users per Capita X Industry US Share of ICT Capital Services 0.015*** 0.029*** 0.479*** (0.005) (0.006) (0.046) Province Number of People per Fiber Optic Line Kilometer X Industry US Share of ICT Capital Services 0.004*** 0.004** 0.095*** (0.001) (0.002) (0.010) Firm Fixed Effects Yes Yes Yes Yes Yes Yes Province-Year Fixed Effects Yes Yes Yes Yes Yes Yes Number of Observations 1,175,202 1,175,202 292,135 292,135 310,922 310,922 R-squared 0.835 0.835 0.868 0.868 0.844 0.844 Notes: Robust standard errors in parentheses, clustered at the province-industry-year level. *** and ** indicate significance at 1% and 5% confidence levels, respectively. 32 Appendix Appendix Table A.1. Measures of Intensity of Internet Use for 2-Digit Industries China Share of US Share of ICT Vietnam Share Rauch Nunn Contract Post and Capital of Firms with Differentiated Intensity Telecommunic. Services Websites Products Average across 3-Digit Industries 15t16 Food, Beverages, and Tobacco 0.002 0.112 0.127 0.3 0.765 17t19 Textiles, Wearing Apparel, Leather, and Related Products 0.005 0.100 0.106 0.6 0.885 20 Wood and Paper Products 0.004 0.083 0.119 1 0.871 21t22 Printing and Publishing 0.003 0.189 0.075 0.446 0.965 23 Coke and Refined Petroleum Products 0.004 0.155 0.294 0.414 0.769 24 Chemicals and Chemical Products 0.005 0.201 0.172 0.555 0.869 25 Rubber and Plastics Products 0.005 0.074 0.154 0.204 0.967 26 Other Non-Metallic Mineral Products 0.005 0.124 0.135 1 0.962 27t28 Basic Metals and Fabricated Metal Products 0.011 0.139 0.111 0.578 0.847 29 Machinery and Equipment N.E.C. 0.011 0.337 0.157 1 0.971 30t33 Electrical and Optical Equipment 0.006 0.302 0.208 1 0.958 34t35 Transport Equipment 0.009 0.312 0.289 1 0.986 36t37 Other Manufacturing 0.008 0.231 0.135 1 0.946 Notes: The 2-digit industries shown are those in the KLEMS database that are manually concorded to the Chinese firm-level data N.E.C. stands for not Elsewhere Classified. The measures in the last three columns are originally available at the 3-digit level and here their averages for each 2-digit industry are shown. 33 Appendix Table A.2. Correlation across Measures of Intensity of Internet Use Panel A. For 2-Digit Industries China Share of US Share of ICT Vietnam Share Rauch Nunn Contract Post and Capital of Firms with Differentiated Intensity Telecommunic. Services Websites Products Simple Correlation China Share of Post and Telecommunic. 1 US Share of ICT Capital Services 0.597** 1 Vietnam Share of Firms with Websites 0.178 0.441 1 Rauch Differentiated Products 0.500* 0.563** 0.136 1 Nunn Contract Intensity 0.397 0.481* -0.065 0.505* 1 Spearman Rank Correlation China Share of Post and Telecommunic. 1 US Share of ICT Capital Services 0.539* 1 Vietnam Share of Firms with Websites 0.302 0.517* 1 Rauch Differentiated Products 0.491* 0.483* 0.119 1 Nunn Contract Intensity 0.462 0.379 0.154 0.399 1 Panel B. For 3-Digit Industries Vietnam Share Rauch Nunn Contract of Firms with Differentiated Intensity Websites Products Simple Correlation Vietnam Share of Firms with Websites 1 Rauch Differentiated Products 0.115 1 Nunn Contract Intensity 0.058 0.620*** 1 Spearman Rank Correlation Vietnam Share of Firms with Websites 1 Rauch Differentiated Products 0.119 1 Nunn Contract Intensity 0.179** 0.522*** 1 Notes: ***, **, and * indicate significance at the 1%, 5%, and 1% confidence levels, respectively. 34 Appendix Table A.3. Impact of Internet Penetration (Internet Users) on Firm Export Performance, Adding Measures of Province-Industry-Year Growth Dependent Variable: Firm-Level Dummy for Export Firm-Level Export Intensity Firm-Level Real Exports (If Participation (If Export Participation = 1) Export Participation = 1) (1) (2) (3) (4) (5) (6) Province Number of Internet Users per Capita X Industry China share of Post and Telecommunications 0.423*** 0.659*** 9.537*** (0.085) (0.120) (0.936) Province Number of Internet Users per Capita X Industry US Share of ICT Capital Services 0.009*** 0.018*** 0.298*** (0.003) (0.004) (0.034) Province-Industry Growth in Real Output -0.005 -0.004 -0.034*** -0.032*** -0.166*** -0.131*** (0.004) (0.004) (0.007) (0.007) (0.044) (0.043) Province-Industry Growth in Real Exports 0.004*** 0.004*** 0.010*** 0.010*** 0.122*** 0.124*** (0.001) (0.001) (0.002) (0.002) (0.016) (0.016) Firm Fixed Effects Yes Yes Yes Yes Yes Yes Province-Year Fixed Effects Yes Yes Yes Yes Yes Yes Number of Observations 1,750,021 1,750,021 437,291 437,291 464,541 464,541 R-squared 0.812 0.812 0.850 0.850 0.821 0.821 Notes: Robust standard errors in parentheses, clustered at the province-industry-year level. *** indicates significance at the 1% confidence level. 35