75165 What Explains the Price of Remittances? An Examination Across 119 Country Corridors ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı Remittances are a substantial source of external �nancing for developing countries that influence many aspects of their development. Though research has shown that remittances are both expensive and price sensitive, little is known about what explains their price. Newly gathered data across 119 country pairs or corridors are used to explore the factors associated with the price of remittances. Corridors with larger numbers of migrants and more competition among providers are found to exhibit lower prices for remittances, when average prices across all types of remittance service providers are considered. Corridors with lower barriers to access banking services and Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 broader regulation of remittance service providers also have lower prices. Remittance prices are higher in richer corridors and in corridors with greater bank participation in the remittance market. Few signi�cant differences emerge when results are compared across banks and, separately, across money transfer operators. However, estimations for Western Union, a leading player in the remittances business, suggest that its prices are less sensitive to competition. JEL classi�cation: F22, F24. In 2008, remittances to developing countries reached $328 billion, more than twice the amount of of�cial aid and over half of foreign direct investment flows (World Bank 2009a). Numerous studies have shown that remittances have a positive and signi�cant impact on economic development along multiple dimensions, including poverty alleviation, education, entrepreneurship, infant Thorsten Beck (corresponding author; T.Beck@uvt.nl) is a professor of economics and CentER fellow and chair of the European Banking Center at Tilburg University. Marı ´nez Perı ´a Soledad Martı ´a (mmartinezperia@worldbank.org) is a senior economist in the Finance and Private Sector Development Research Group of the World Bank. The authors thank Diego Anzoategui and Subika Farazi for excellent research assistance. They received helpful comments from participants at the Second International Conference on Migration and Development and the International Conference on Diaspora for Development, as well as from World Bank colleagues in the Finance and Private Sector Development Research Group and the Payment Systems Unit. The authors are particularly grateful to the journal editor and to three anonymous referees for constructive comments and suggestions. This article’s �ndings, interpretations, and conclusions are entirely those of the authors and do not necessarily represent the views of the organizations with which they are af�liated. THE WORLD BANK ECONOMIC REVIEW, VOL. 25, NO. 1, pp. 105 –131 doi:10.1093/wber/lhr017 Advance Access Publication May 23, 2011 # The Author 2011. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 105 106 THE WORLD BANK ECONOMIC REVIEW mortality, and �nancial development.1 Hence, understanding the market for remittance transactions can be critical for promoting the development process in many low-income countries. Remittance transactions are known to be expensive, with estimates averaging 10 percent of the amount sent (World Bank 2009b). At the same time, these costs are widely dispersed across corridors and range from 2.5 percent to 26 percent. Furthermore, case studies have shown that remittance flows are very sensitive to prices and are likely to increase substantially as prices fall. For example, Gibson, McKenzie, and Rohorua (2006) estimate a 22 percent price elasticity in the New Zealand–Tonga corridor and calculate that lowering the fees to the levels found in the most competitive corridors would raise remit- tances by 28 percent. Using a randomized experiment, Aycinena, Martinez, and Yang (2009) estimate that a $1 lower fee in the United States –El Salvador corridor would boost remittances $25 a month from an average of $290. Because remittances are important for economic development and appear to Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 be sensitive to price, lowering the price has become a priority for policymakers. At the L’Aquila 2009 G-8 Summit, leaders pledged to reduce the price of remittances by half (from 10 to 5 percent) in �ve years.2 Yet, little is known empirically about what explains the price of remittances.3 Are high prices due mainly to factors in the sending or the receiving country? Are high prices related to socioeconomic factors, industry market structure, or government pol- icies and regulations that affect remittance service providers and the mark-ups they are able to charge? Are there signi�cant differences between banks and money transfer operators (MTOs)? Explaining the variation in prices is thus of interest for academics and policymakers alike. Using a new dataset assembled by the World Bank Payment Systems Group on the price of remittances across 119 country pairs or corridors (Remittance Prices Worldwide database (World Bank 2009b), this article explores the 1. For example, see Adams and Page (2003), Adams (2005), IMF (2005), Lopez-Co ´ rdova (2005), Maimbo and Ratha (2005), and Taylor, Mora, Adams and Lopez-Feldman (2005) for studies on the impact of remittances on poverty. Studies such as Cox-Edwards and Ureta (2003), Hanson and Woodruff (2003), Lo ´ rdova (2005), and Yang (2008) �nd that by helping to relax household ´ pez-Co constraints, remittances are associated with improved schooling outcomes for children. Remittances have also been shown to promote entrepreneurship (see Massey and Parrado 1998; Maimbo and Ratha 2005; Yang 2008; Woodruff and Zenteno 2007). Furthermore, a number of studies on infant mortality and birth weight have documented that, at least in the Mexican case, migration and remittances help lower infant mortality and are associated with higher birth weight among children in households that receive remittances (see Kanaiaupuni and Donato 1999; Hildebrandt and McKenzie 2005; Duryea et al. 2005; and Lo ´ rdova 2005). Aggarwal, Demirgu ´ pez-Co ¨c¸ -Kunt, and Martinez Peria (2010) show that remittances can have a positive impact on �nancial development. 2. Paragraph 134, page 49 of the L’Aquila 2009 G8 Summit. http://www.g8italia2009.it/static/ G8_Allegato/G8_Declaration_08_07_09_�nal,0.pdf. 3. Orozco (2006) and Freund and Spatafora (2008) are the exception, but their data is limited to a few countries or a few providers. While Orozco’s work focuses exclusively on Latin America, the Freund and Spatafora study analyzes only the costs of remittances sent from the United States and the United Kingdom exclusively via MoneyGram or Western Union to 66 countries. ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 107 factors associated with remittance prices in 2009.4 It studies corridors that include 13 major remittance sending countries and 60 receiving countries representing approximately 60 percent of total remittances to developing countries. Using data at the corridor level permits bilateral analysis of prices rather than analysis of prices aggregated at the receiving or sending country level only. Furthermore, unlike previous studies focusing on a certain type of remittance service provider (usually the largest international MTOs), the analy- sis here considers the largest providers, whatever the type, in each corridor.5 And by averaging across all types of providers and across each type of provider (banks and MTOs) separately, the factors associated with the price of remit- tances can be compared across different types of institutions. Finally, analyzing the prices charged by Western Union across 98 corridors (80 percent of the sample) alleviates concerns about bias due to differences across �rms and thus sheds light on the factors correlated with the prices charged by a leading remit- tance service provider with worldwide operations. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 The analysis distinguishes three groups of variables that can be associated with cross-corridor variation in the price of remittances. One is the impact of government policies—including exchange rate policies, capital controls, and regulation of remittance service providers—that can influence the price of remittances through their impact on the cost structure of remittance service providers. The second is the role of factors that might affect the ability of remittance service providers to increase their mark-up, such as extent of competition, market structure, and level of education of the migrant population. The third is the role of socioeconomic characteristics in sending and receiving countries that might influence fees through their impact on the cost structure of remittance service providers and on provi- ders’ ability to raise the mark-up. Estimations of the price of remittances across all types of remittance service providers show that prices are consistently lower in corridors with a larger number of migrants and more competition and in corridors with lower access barriers to the banking system and broader regulation of remittance service providers. Remittance prices are higher in richer corridors and in corridors with a higher share of banks among providers. The prices of sending remit- tance using banks or MTOs are associated with similar factors. Western Union prices appear to be less sensitive to competition, perhaps a reflection of the �rm’s market power. This article is related to the literature on the price of banking services. Beck, Demirgu ¸ -Kunt, and Martı ¨c ´nez Perı ´a (2008) document large cross-country 4. The original World Bank database for the period analyzed here contains information on 134 corridors. From that total, 13 corridors (where Russia is the sending country) are missing exchange rate data and 2 other corridors are missing information for some explanatory variables. 5. On average, 8– 10 providers are included for each corridor. In some corridors, primarily those including the United States and Spain as sending countries, the number of providers surveyed exceeds 10. 108 THE WORLD BANK ECONOMIC REVIEW variation in the costs to customers of opening and maintaining bank accounts and in the fees for using automated teller machines and for transferring funds, �nding that �rms report lower �nancing constraints in countries with lower costs of �nancial services. Freund and Spatafora (2008) and Orozco (2006) also present data on remittance prices, but for few countries and providers and not at the corridor level. In a broader sense, this article is also related to the literature on bank inter- est rate spreads (the differences between deposit and lending rates), with higher spreads indicating more expensive banking services. Both institution-speci�c characteristics, such as size and ownership, and market and country character- istics, such as competition and the legal and institutional framework, have been shown to be signi�cant predictors of interest rate spreads (see Demirgu ¸ -Kunt, Laeven and Levine 2004; Laeven and Majnoni 2005; and ¨c Beck 2007 for a general discussion). This article is a �rst exploration of corridor variation in the price of remit- Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 tances and is subject to two important caveats. First, as a pure cross-sectional analysis, it is potentially subject to reverse causation and omitted-variable biases. Hence, only limited, if any, inference can be made on causality.6 Second, the analysis is also limited in scope since it includes data only from formal providers of remittance services. By some estimates, at least a third of remittances are sent through informal channels (Freund and Spatafora 2008). These limitations notwithstanding, the article offers interesting evidence that should stimulate further data collection and analysis. The article is organized as follows. Section I describes the data on the price of remittances. The empirical methodology is in section II and the results are in section III. Section IV summarizes the �ndings and recommends further research. I . D ATA ON THE PRICE OF R E M I T TA N C E S The data on the price of remittances are from the Remittance Prices Worldwide database, a recent survey of remittance service providers conducted by the Payment System Unit of the World Bank in the �rst quarter of 2009 (World Bank 2009b).7 The price of remittances consists of a fee component and an exchange rate spread component. The World Bank dataset covers 14 sending and 72 receiving countries. However, because spread information is missing for remittances from the Russian Federation and data are missing for some explanatory variables, the focus is on 119 corridors, at most, including 6. Most of the variables, however, are likely to be exogenous to remittance prices, including migration flows, distance, and even banking market structure, given the small share of bank pro�ts stemming from remittances. 7. Since then, the data have been updated, and prices are now available through the �rst quarter of 2010. However, because data for most of the correlates used in this analysis are not updated with the same frequency, the panel dimension of the data could not be exploited. ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 109 13 sending countries and 60 receiving countries.8 In most cases, data cover the prices from the main sending location for the corridor in question to the capital city or most populous city in the receiving market. Data were collected by interviewers posing as customers and by contacting individual �rms. Within each corridor, the data were gathered on the same day to control for exchange rate fluctuations and other changes in fee structures. In general, price data were collected for 8–10 major service providers in each cor- ridor, including the main MTOs and banks active in the market.9 Companies surveyed in each segment were selected to cover the maximum remittance market share possible.10 Since the dataset does not provide information on the market shares of each provider, it is not possible to compute weighted averages. Hence, the regression analysis uses both the simple average and the median prices calculated across all providers in a corridor as dependent vari- ables.11 Results are reported using only the simple averages, however, because average and median prices are highly correlated (0.96). Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Prices based on two amounts are available per corridor: the local equiv- alent of $200 and the local equivalent of $500. Since previous studies have found that a typical remittance transaction involves sending close to $200, the analyses are based on the prices associated with this amount.12 Furthermore, the prices of sending $200 and $500 (expressed as a percen- tage of the amount sent) are highly correlated (0.91), so the results do not vary signi�cantly.13 Figure 1 illustrates the variation in average prices across the 119 corridors, calculated across all surveyed remittance service providers in each corridor. The average remittance prices are lowest in the Saudi Arabia–Pakistan corridor (2.5 percent of $200) and highest in the Germany–Croatia corridor (25.8 percent). Averaged across all corridors and providers, the price is 10.2 percent. There is considerable heterogeneity in prices even when the same sending or remittance receiving country is considered. Prices of remittances sent from the United States vary from 3.7 percent to Ecuador to 14.1 percent to Thailand (�gure 2). Remittance prices to India vary from 3.1 percent from Saudi Arabia to 13.3 percent from Germany (�gure 3). This variation underlines the 8. The full data are available at www.remittanceprices.org. Data on exchange rate spreads are also missing for some institutions in Germany, France, and Japan. These institutions are excluded from the calculations of the average remittances costs from those countries. 9. The number of respondents by corridors varies depending on the number of �rms active in the corridor. Some corridors (like the Spain– China corridor) include only two �rms, while others (like the United States– Mexico corridor) go as high as 18. 10. No more information is provided on how �rms were selected. For a discussion of the methodology, see http://remittanceprices.worldbank.org/Methodology/. 11. A priori, it is not clear how having weighted averages instead of simple averages would change the estimations. This problem is interpreted as a potential case of measurement error in the dependent variable, which should not bias the estimates but would affect their ef�ciency. 12. Freund and Spatafora (2008) use the same amount in their study. 13. These results are available on request. 110 THE WORLD BANK ECONOMIC REVIEW F I G U R E 1. Average Price of Remittances Sent Across 119 Migration Corridors Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Source: Remittance Prices Worldwide database. F I G U R E 2. Average Price of Remittances from the United States to 22 Receiving Countries Source: Authors’ analysis based on data from World Bank (2009b). ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 111 F I G U R E 3. Average Price of Remittances to India from Eight Sending Countries Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Source: Authors’ analysis based on data from World Bank (2009b). importance of analyzing the price of remittances at the corridor rather than at the sending or receiving country level. Remittance prices also vary across provider types (table 1). On average, banks charge substantially higher fees (12.4 percent) than do MTOs (8.8 percent). This differential does not control for the fact that banks and MTOs are not active in all corridors and that different banks and different MTOs are active in different corridors. When the analysis focuses on corri- dors where both types of institutions are active, average prices for banks exceed those for MTOs in 43 of the 63 corridors. Furthermore, when prices at the provider level are regressed on a set of corridor dummy variables and a bank dummy variable, bank prices average 3 percentage points higher than MTO fees. Western Union’s prices (10.8 percent) are slightly higher than the average for all MTOs (8.8 percent). Western Union prices also exhibit high cross-corridor variation, ranging from 2.7 percent in the Saudi Arabia –Yemen corridor to 29.9 percent in the United Kingdom–Albania corridor (�gure 4). II. E M PI R I CA L ME T H O D O LO GY To examine the determinants of remittance prices, the average price of sending remittances, Pij, is regressed on a set of sending and receiving country T A B L E 1 . Summary Statistics and Data Sources 112 Number of Description Abbreviation observations Mean Median Date Source All providers 119 10.24 9.47 2009 Remittance Prices Worldwide (World Bank 2009b) Banks’ average prices (% of $200) Banks avg price 70 12.40 11.78 2009 Remittance Prices Worldwide (World Bank 2009b) Money transfer operators’ average prices MTOs avg price 112 8.78 8.07 2009 Remittance Prices Worldwide (World (% of $200) Bank 2009b) Western Union’s average prices (% of $200) WU avg price 98 10.84 10.33 2009 Remittance Prices Worldwide (World Bank 2009b) Log of number of migrants in the corridor Log bil mig 119 11.61 11.88 2006 Ratha and Shaw (2007) Log of GDP per capita in receiving country Log GDP rec 119 7.15 7.40 Average for World Development Indicators (World 2006-07 Bank 2009c) Log of GDP per capita in sending country Log GDP send 119 10.02 10.17 Average for World Development Indicators (World THE WORLD BANK ECONOMIC REVIEW 2006-07 Bank 2009c) Dummy for pegged exchange rate or Peg rec 119 0.33 0.00 2008 Annual Report on Exchange dollarization Arrangements and Restrictions (IMF) Share rural population (%) in receiving Rural pop rec 119 49.48 50.22 Average for World Development Indicators (World country 2006-07 Bank2009c) Share rural population (%) in sending Rural pop send 119 20.56 18.99 Average for World Development Indicators (World country 2006-07 Bank 2009c) Log of distance between sending and Log dist 119 8.22 8.39 - Distances database (CEPII 2010) receiving countries Common language Com Language 119 0.44 0.00 - Distances database (CEPII 2010) Controls on remittances in receiving country Ctrl remit rec 105 0.21 0.00 2007 Annual Report on Exchange Arrangements and Restrictions (IMF 2007) Share of educated (secondary or tertiary Mig educ 88 54.14 53.47 2000 Database on Immigrants and education) migrants Expatriates (OECD 2010) Branches per capita (100,000 people) in Brchs pc rec 89 6.62 6.30 2008 Database on Access to Financial receiving country Services (World Bank 2007) Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Branches per capita (100,000 people) in Brchs pc send 119 33.64 30.86 2008 Database on Access to Financial sending country Services (World Bank 2007) Index of regulations for remittance providers Reg rec 91 2.20 2.00 2008 World Bank Global Payment Systems in receiving country Survey (World Bank 2008) Index of regulations for remittance providers Reg send 119 2.25 2.00 2008 World Bank Global Payment Systems in sending country Survey (World Bank 2008) H-statistic for banking sector in receiving H-Stat rec 111 0.54 0.52 1994-2006 Bankscope database (Bureau van Dijk country 2009) and author calculations H-statistic for banking sector in sending H-Stat send 119 0.52 0.50 1994-2006 Bankscope database (Bureau van Dijk country 2009) and author calculations Number of respondents per corridor Resp per corr 119 7.97 8.00 2009 Remittance Prices Worldwide (World Bank 2009b) Share of banks per corridor (%) % of banks 119 31.35 20.00 2009 Remittance Prices Worldwide (World Bank 2009b) Index importance of banks receiving country Bank imp rec 90 5.32 6.00 2007/8 World Bank Global Payment Systems Survey (World Bank 2008) Index importance of banks sending country Bank imp send 108 5.11 5.50 2007/8 World Bank Global Payment Systems Survey (World Bank 2008) Log of bilateral trade Log Bil Trade 111 21.90 21.87 Average for Direction of Trade Statistics (IMF 2006-07 2009) Savings account annual fee (% of GDP pc) Annual fee rec 85 0.55 0.07 2003 Beck, Demirgu ¸ -Kunt, Martı ¨c ´a ´nez Perı receiving country (2008) Thorsten Beck and Marı Savings account annual fee (% of GDP pc) Annual fee send 72 0.12 0.00 2003 Beck, Demirgu ¸ -Kunt, Martı ¨c ´a ´nez Perı sending country (2008) Minimum amount to open a savings account Min amnt open rec 85 7.36 1.59 2003 Beck, Demirgu ¸ -Kunt, Martı ¨c ´a ´nez Perı (% of GDP pc) receiving country (2008) Minimum amount to open a savings account Min amnt open send 72 0.11 0.00 2003 Beck, Demirgu ¸ -Kunt, Martı ¨c ´a ´nez Perı (% of GDP pc) sending country (2008) ´a Soledad Martı Source: Authors’ analysis based on data referenced in the table. 113 ´nez Perı ´a Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 114 THE WORLD BANK ECONOMIC REVIEW F I G U R E 4. Average Price of Remittances sent through Western Union Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Source: Authors’ analysis based on data from World Bank (2009b). characteristics and on some corridor-speci�c variables captured by the matrix X in equation (1): Pij ¼ a0 þ a1 Sending country factorsi þ a2 Receiving country factors j þ a3 Xij þ uij ð1Þ where Pij is the price of sending $200 from country i to country j expressed as a percentage of $200. All explanatory variables are lagged relative to the price variable. Since this does not completely rule out reverse causation or endogene- ity bias, the results are interpreted as associations rather than as causal impacts. Table 1 provides the summary statistics and data sources for each vari- able, and table 2 reports correlations across the main variables. Government Policies Equation (1) captures an array of factors that might be correlated with remit- tance prices through their association with the costs faced by remittance service providers and the mark-up providers can charge over their marginal costs. First, it controls for different government policies relating to the exchange rate, T A B L E 2 . Correlation Matrix: Pairwise Correlations among Main Variables Min Log Log Log Rural Rural Brchs Brchs Resp Share Log Annual Annual amnt Avg bil GDP GDP Peg pop pop Com pc pc Reg. Reg. H-Stat H-Stat per of bil fee fee open Price mig rec send rec rec send language rec send rec send rec send corr banks trade rec send rec Avg price 1 Log bil mig 2 0.38** 1 Log GDP rec 0.09 0.26** 1 Log GDP send 2 0.14 0.32** 0.18** 1 Peg rec 2 0.08 2 0.10 2 0.12 2 0.14 1 Rural pop rec 0.03 2 0.12 2 0.75** 2 0.19** 0.07 1 Rural pop send 0.36** 2 0.10 0.09 2 0.51** 0.06 2 0.09 1 Com language 2 0.20** 0.15 2 0.11 0.02 2 0.04 0.05 2 0.29** 1 Brchs pc rec 0.05 0.16 0.58** 0.04 0.20 2 0.59** 0.12 2 0.19 1 Brchs pc send 2 0.11 0.12 0.21** 0.09 0.01 2 0.26** 0.22** 2 0.18** 0.19 1 Reg. rec 0.04 0.08 0.18 0.10 0.01 2 0.25** 0.11 2 0.06 2 0.17 0.14 1 Reg. send 2 0.51** 0.03 2 0.14 0.12 0.03 0.06 2 0.66** 0.19** 2 0.15 0.07 2 0.04 1 H-Stat rec 2 0.21** 0.02 0.20** 0.13 2 0.02 2 0.03 2 0.12 2 0.03 0.05 2 0.02 0.00 0.06 1 H-Stat send 2 0.27** 0.35** 0.28** 0.56** 2 0.14 2 0.33** 0.05 2 0.07 0.16 0.55** 0.22** 2 0.11 0.16 1 Resp per corr 2 0.33** 0.35** 0.26** 0.18 0.00 2 0.22** 2 0.20** 0.30** 0.20 0.30** 0.13 0.22** 0.12 0.19** 1 Share of banks 0.55** 2 0.08 0.05 2 0.46** 0.07 0.09 0.60** 2 0.22** 0.17 2 0.26** 0.01 2 0.63** 2 0.16 2 0.46** 2 0.17 1 Log bil Trade 0.02 0.36** 0.35** 0.25** 2 0.25** 2 0.08 0.03 0.00 0.04 2 0.12 0.21 2 0.17 0.12 0.25** 0.18 0.11 1 Annual fee rec 0.29** 2 0.51** 2 0.33** 2 0.25** 0.07 0.28** 0.06 0.26** 2 0.18 2 0.17 2 0.02 2 0.07 2 0.10 2 0.26** 2 0.14 0.10 2 0.27** 1 Thorsten Beck and Marı Annual 0.29** 2 0.42** 2 0.21 2 0.94** 0.11 0.25** 0.65** 0.10 2 0.17 2 0.30** 2 0.12 2 0.50** 2 0.10 2 0.55** 2 0.28** 0.53** 2 0.11 0.40** 1 fee send Min amnt 2 0.01 2 0.23** 2 0.55** 0.01 0.17 0.31** 2 0.12 0.22** 2 0.24** 2 0.16 0.11 0.08 2 0.10 2 0.12 2 0.11 2 0.18 2 0.41** 0.40** 0.07 1 open rec Min amnt 0.39** 2 0.44** 2 0.22 2 0.93** 0.15 0.27** 0.56** 0.17 2 0.16 2 0.38** 2 0.14 2 0.47** 2 0.13 2 0.72** 2 0.20 0.61** 2 0.10 0.46** 0.96** 0.09 open send **Signi�cant at least at the 5 percent level. ´a Soledad Martı Note: For de�nitions, see table 1. Only the main variables are included in this table. Full results are available on request. Source: Authors’ analysis based on data described in the text. 115 ´nez Perı ´a Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 116 THE WORLD BANK ECONOMIC REVIEW the capital account, and regulation of the remittance market that might influ- ence the costs faced by remittance service providers. It includes a dummy vari- able for receiving countries with pegged exchanged rates (including cases of currency boards, de facto pegged regimes, and no separate legal tender). Lower exchange rate volatility should be associated with lower prices, by lowering the exchange rate costs and uncertainty faced by providers and, thus, the spreads they charge to customers. At the same time, the price of sending remittances is expected to be higher in countries that impose controls on remittance trans- actions, since controls operate like a tax that is likely to be passed onto recipi- ents. Both the dummy variable for pegged exchange rate regimes and the capital controls dummy variable are from the International Monetary Fund (IMF 2007). In 39 corridors (almost 33 percent of the sample), the exchange rate is pegged or the economy is fully dollarized, so there is no exchange rate variability, and in 22 corridors (18 percent of the sample) there are controls on gifts from abroad or remittances. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 The analysis controls for the breadth of regulation of remittance service pro- viders in sending and receiving countries by creating an index of regulation that takes a value of 0 –5 depending on whether providers must be registered, must be licensed, are subject to speci�c safety and ef�ciency requirements, need to comply with anti–money laundering regulations, or need to comply with laws and regulations of general applicability. Data for creating the indexes are from the 2008 Global Payment Systems Survey conducted by the World Bank (2008). While a broader regulatory framework might make the remittance market more transparent and more competitive, greater exposure to regulations can also increase the costs for regulated institutions, so the impact is ambigu- ous a priori.14 Similarly, greater breadth of regulation might reduce the number of service providers, with negative repercussions for competitiveness. The index averages 2.2 among remittance receiving countries and 2.3 among remittance sending countries. Remittance Mark-ups The regressions also include proxies for factors that might be associated with remittance prices because of their effect on the mark-up remittance service pro- viders can charge their customers. The analysis posits that providers can more readily charge a mark-up if there is little competition in the remittance market and if customers are not well informed. Two indirect measures of competition among remittance service providers are used (direct measures are lacking). One is the number of remittance service providers in the database for each corridor. Since the Remittances Prices Worldwide survey tries to cover the most impor- tant providers in a corridor, corridors with more providers are assumed to have 14. Note that the index does not measure the severity of regulations, but only the scope of the regulatory framework. ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 117 more active �rms and, other things equal, to be more competitive.15 The average number of respondents across all corridors is 8, and the number varies from 2 in the Spain–China corridor to 18 in the United States–Mexico corridor. The second measure is of competition among banks in receiving and sending countries. The rationale is that more competitive banking sectors will offer cheaper services, including for remittance transactions. This will create pressure for other providers to lower prices as well. Of course, this implicitly assumes that banks are important players in the remittance business. Following Panzar and Rosse (1982, 1987), the H-statistic is used to measure the degree of com- petition in the banking sector by calculating the sum of the elasticities of banks’ interest revenues to different input prices (see the appendix for a discus- sion of the methodology used to calculate the H-statistic).16 Under perfect competition, an increase in input prices raises both marginal costs and revenues by the same amount and, thus, the H-statistic will equal 1. In a monopoly, an Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 increase in input prices results in a rise in marginal costs, a fall in output, and a decline in revenues, leading to an H-statistic of less than or equal to 0. When H is between 0 and 1, the system operates under monopolistic competition. A negative relationship is expected between the H-statistic in sending and receiv- ing countries and the price of sending remittances. Data for 1994–2006 from Bankscope database (Bureau van Dijk 2009) are used to compute the H-statistic. Among both remittance receiving and sending countries, the H-statistic averages close to 0.53. But the standard deviation is larger for remit- tance sending countries. The signi�cance of the relative importance of banks in the remittance market in explaining cross-corridor variation in remittance prices is also explored using the share of bank respondents among all remittance service pro- viders in the database. To the extent that, as some have argued, banks view remittances as a marginal product and are less likely to offer competitive prices (Ratha and Riedberg 2005), a positive correlation is expected between the share of bank respondents and the average price of remittances. Across the 119 corridors, the share of bank respondents varies from 0 percent in the Italy–Sri Lanka corridor to 100 percent in the South Africa –Swaziland corridor. On average, the share across corridors is 31 percent. Because data were lacking on the share of the remittance market captured by each provider, the percentage of bank respondents described above may not reflect the actual importance of commercial banks. Hence, an alternative measure, obtained from the Global Payment Systems Survey (World Bank 15. Because in most cases, mystery shoppers were used to gather data on the price of remittances, the number of respondents should not be affected by the willingness of certain providers to cooperate. However, it is still possible that in some corridors the number of respondents is small simply because interviewers had dif�culty reaching or locating some providers. 16. Other studies that use this methodology to estimate competition include Bikker and Haaf (2002), Gelos and Roldos (2004), Claessens and Laeven, (2004), and Levy-Yeyati and Micco (2007). 118 THE WORLD BANK ECONOMIC REVIEW 2008), is used to test the sensitivity of the �ndings. The measure indicates the degree to which central banks consider commercial banks to be signi�cant remittance service providers, on a scale from 1 (least relevant) to 6 (most rel- evant).17 The correlation between this variable and the percentage of bank respondents is 0.37 and is signi�cant at the 5 percent level. The �nancial literacy of remittance senders can also affect mark-ups. Since �nancial literacy cannot be captured directly, a measure of the education level of migrants in each corridor is used (migrants with a secondary or tertiary edu- cation as a share of total migrants from the remittance receiving country resid- ing in the remittance sending country). This variable comes from the OECD Database on Immigrants and Expatriates (OECD 2010). This variable is expected to be correlated with �nancial literacy, and to the degree that �nan- cial literacy enables consumers to make better informed choices, prices should be lower. The ratio of secondary and tertiary educated migrants varies from 21 percent for Chinese migrants in Italy to 91 percent for Nigerian migrants in the Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 United States. Because this variable is available for only 88 of the 119 corri- dors, it is not included in the baseline estimations but is shown as an additional variable. Socioeconomic Variables Several socioeconomic variables are included that can influence remittance prices by affecting both costs and mark-ups. One, a proxy for the volume of remittance transactions within corridors, is the number (bilateral stock) of migrants residing in the remittance sending country who are originally from the remittance receiving country. These data are from the World Bank (Ratha and Shaw 2007). Unlike the flow of actual remittances, migrant stock is less likely to be endogenous to the price variable. A negative relationship is conjec- tured between the stock of migrants and the price of remittances; a higher volume of migrants might imply scale economies and, hence, lower costs for providers and more competition among them, resulting in smaller mark-ups.18 The number of migrants is negligible in the South Africa–Zambia corridor and exceeds 10 million people in the United States–Mexico corridor. The average is 379,200 migrants. GDP per capita is also included, as a proxy for economic development and standard of living. This variable comes from the World Bank’s World Development Indicators database (World Bank 2009c). The cost of goods and services will be higher in countries with higher standards of living, so remit- tance prices would also be expected to be higher. Countering that tendency, economic development may be associated with greater ef�ciencies and lower 17. The Global Payment Systems Survey scale uses 1 to indicate most relevant and 6 the least relevant. The scale is inverted here so that higher values indicate that banks are more important. 18. The presence of more migrants might encourage entry of a larger number of remittance service providers, leading to more contestability and lower mark-ups. ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 119 costs for �nancial intermediation (Harrison, Sussman, and Zeira 1999) and thus lower remittance prices. In the sample, GDP per capita for receiving countries varies from $148 in Malawi to close to $14,000 in the Republic of Korea. Among remittance sending countries, GDP per capita varies from $3,640 in South Africa to $40,200 in Japan (all prices in U.S. dollars).19 The geographic distribution of the population in sending and receiving countries might also be an important driver of the price of remittances, as a more sparsely distributed population might be harder to reach, thus raising transaction costs for providers. A more sparsely distributed population might also increase the pricing power of providers, as geographic access is more dif�- cult for senders and recipients of remittances. The share of rural population in both sending and receiving countries is used to proxy for the disparity in geo- graphic distribution.20 These data come from the World Bank’s World Development Indicators (World Bank 2009c). Among receiving countries, the rural population varies from 13 percent of the total in Lebanon to 87 percent Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 in Uganda and averages 48 percent. By contrast, among sending countries, the rural population varies from 0 for Singapore to 40 percent for South Africa and averages 21 percent. Bank Variables To measure access to �nancial services more directly, some estimations also control for the number of bank branches per capita in sending and receiving countries.21 This variable is expected to have a negative association with the price of sending remittances, as higher branch penetration will reduce trans- action costs and increase scale. The ratio of branches per capita averages about 6 per 100,000 inhabitants in receiving countries and close to 34 per 100,000 in sending countries. Measures of the costs of accessing formal banking services in both sending and receiving countries (the minimum amount to open a savings account and the annual fee to maintain an account) are also included (Beck, Demirgu ¸ -Kunt, and Martı ¨c ´a 2008). Easier and cheaper access is conjec- ´nez Perı tured to increase the options for both senders and recipients of remittances and thus to boost competition. The minimum balance to open a savings account averages 7.36 percent of GDP per capita in receiving countries and 0.11 19. Regressions were also run that controlled separately for the level of �nancial development using the ratio of liquid liabilities to GDP. The results are very similar to those including GDP per capita. Since these variables are highly correlated—(0.2) among receiving countries and (0.4) among sending countries—these estimations are not reported, and GDP per capita is included instead as a broader measure of development. 20. The share of rural population is a better proxy for the effect of service delivery than population density, which is an average within a country and does not take into account which share of the population actually lives in more remote areas. The population density variable yielded similar results. 21. These data are from Beck, Demirgu ¸ -Kunt, and Martı ¨c ´nez Perı´a (2007) and can be accessed at http://go.worldbank.org/EZDOBVQT20. Because these data are not available for all corridors, this variable is not included in all estimations. 120 THE WORLD BANK ECONOMIC REVIEW percent in sending countries; fees average 0.55 percent of GDP per capita in receiving countries and 0.12 percent in sending countries. Corridor-speci�c Variables Finally, several corridor-speci�c variables are included that might influence the extent and ease of remittance transactions and, therefore, their costs. These are the distance between sending and receiving countries (from capital city to capital city) and a dummy variable for a common language (takes a value of one if both countries have at least one common language spoken by more than 9 percent of the population). These data come from the French Research Center in International Economics (CEPII) Distances database (CEPII 2010). Smaller geographic and linguistic distances might also foster the emergence of informal remittance service providers, adding competitive pressure to the formal remittance market. Some estimations also include the log of bilateral Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 exports and imports, a measure of bilateral trade. These data come from the IMF Direction of Trade Statistics (IMF 2009). Correlations Between Variables in Our Dataset The average prices of remittances are signi�cantly lower in corridors with a higher number of migrants, smaller share of rural population, and a common language (see table 2). Prices are also lower in corridors where competition is higher and bank participation in the remittance industry is lower. Finally, prices are lower in corridors where sending countries have a broader regulatory framework for remittance service operators and where minimum balances to open a savings account and annual fees to maintain them are lower. Some explanatory variables are highly correlated with others. For instance, GDP per capita in receiving and sending countries is signi�cantly correlated with competition among providers, rural population share, branch penetration, cost of using banking services, and extent of bilateral trade. II I. EM P I R I CA L R ES ULT S Table 3, column 3.1 reports the baseline estimation considering average remit- tance prices charged across all providers with variables for which data are available for all 119 corridors. Though information on the number of respon- dents and the percentage of banks among respondents is also available across all corridors, these variables are not included in the baseline estimations since, as discussed earlier, they might not adequately capture the degree of compe- tition and the importance of banks in the remittance market. The baseline regression shows that, across all providers in 119 corridors, remittance prices are signi�cantly associated with the number of migrants in the corridor, the level of income, and the share of rural population in receiving and sending countries. Corridors with a higher number of migrants exhibit T A B L E 3 . Regressions for the Average Prices of Sending $200 in Remittances for all Remittance Service Providers Variable (3.1) (3.2) (3.3) (3.4) (3.5) (3.6) (3.7) (3.8) (3.9) (3.10) (3.11) (3.12) Log number of migrants 2 1.09 2 0.98 2 1.12 2 1.10 2 1.05 2 1.02 2 1.47 2 0.44 2 0.35 2 1.07 2 0.87 2 1.06 ( 2 4.96)*** ( 2 4.40)*** ( 2 6.20)*** ( 2 4.37)*** ( 2 4.44)*** ( 2 4.40)*** ( 2 10.54)*** ( 2 1.25) ( 2 2.26)** ( 2 4.41)*** ( 2 2.50)** ( 2 4.33)*** Log GDP per capita 1.91 2.11 2.76 1.31 3.57 1.96 3.20 3.18 3.23 2.25 2.64 2.13 receiving (2.40)** (2.66)*** (3.74)*** (1.88)* (3.98)*** (2.28)** (3.60)*** (2.30)** (2.50)** (2.69)*** (3.17)*** (2.51)** Log GDP per capita 2.04 1.95 3.90 2.73 2.38 1.03 2.33 6.43 10.64 1.94 2.36 2.19 sending (2.21)** (2.11)** (3.96)*** (3.32)*** (2.82)*** (0.91) (2.92)*** (2.56)** (7.19)*** (2.02)** (1.12) (2.03)** Pegged or dollarized 2 1.16 2 1.02 2 0.77 2 1.20 2 0.22 2 0.53 2 1.40 1.06 0.98 2 1.37 2 0.07 2 1.16 ( 2 1.38) ( 2 1.23) ( 2 1.03) ( 2 1.64) ( 2 0.20) ( 2 0.49) ( 2 1.46) (0.57) (0.60) ( 2 1.45) ( 2 0.07) ( 2 1.21) Share rural population 0.09 0.09 0.10 0.05 0.13 0.09 0.12 0.14 0.14 0.12 0.11 0.10 receiving (2.91)*** (2.75)*** (3.23)*** (1.83)* (4.01)*** (2.64)*** (2.59)** (2.16)** (2.63)** (3.34)*** (2.83)*** (2.85)*** Share rural population 0.22 0.21 0.33 0.08 0.23 2 0.01 0.25 0.16 0.16 0.22 0.08 0.22 sending (4.43)*** (4.27)*** (7.19)*** (1.71)* (4.67)*** ( 2 0.11) (4.81)*** (1.47) (2.31)** (4.21)*** (1.00) (3.83)*** Log distance 2 0.36 2 0.29 0.23 2 0.02 0.47 0.19 0.58 2 0.10 0.75 2 0.20 2 0.12 2 0.36 ( 2 0.70) ( 2 0.57) (0.55) ( 2 0.05) (0.87) (0.34) (1.00) ( 2 0.08) (0.90) ( 2 0.39) ( 2 0.18) ( 2 0.65) Common language 0.06 0.49 0.23 0.46 1.31 2 0.41 0.15 1.35 0.78 2 0.36 2 1.32 2 0.29 (0.08) (0.66) (0.34) (0.69) (1.43) ( 2 0.44) (0.19) (0.75) (0.61) ( 2 0.44) ( 2 1.46) ( 2 0.34) Number respondents 2 0.30 per corridor ( 2 2.82)*** H 2 statistic receiving 2 5.15 ( 2 2.65)*** Thorsten Beck and Marı H 2 statistic sending 2 16.12 ( 2 5.08)*** Share of banks per 0.09 corridor (6.70)*** Index banks importance 0.70 receiving ´a Soledad Martı (1.65) Index banks importance 1.88 sending (4.40)*** ´nez Perı ´a Index of regulation 0.24 receiving (Continued ) 121 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 122 TABLE 3. Continued Variable (3.1) (3.2) (3.3) (3.4) (3.5) (3.6) (3.7) (3.8) (3.9) (3.10) (3.11) (3.12) (0.74) Index of regulation 2 2.78 sending ( 2 2.93)*** Bank branches per 0.12 capita receiving (0.62) Bank branches per 2 0.05 capita sending ( 2 3.20)*** Savings accounts fee 0.49 receiving (1.33) Savings accounts fee 16.28 sending (1.76)* THE WORLD BANK ECONOMIC REVIEW Min. amount to open 2 0.02 account receiving ( 2 0.42) Min. amount to open 27.97 account sending (7.24)*** Controls on remittances 0.06 (0.06) Share of educated 0.02 migrants (0.73) Log bilateral trade 2 0.06 ( 2 0.25) Observations 119 119 111 119 84 91 89 53 53 105 88 111 R-squared 0.36 0.38 0.56 0.53 0.54 0.45 0.52 0.38 0.62 0.39 0.25 0.36 Diff. max-min prices 0.26 0.25 0.27 0.35 0.41 0.33 0.28 0.36 0.25 0.23 0.23 0.28 predicted over actual *Signi�cant at the 10 percent level; **signi�cant at the 5 percent level; ***signi�cant at the 1 percent level. Note: Number in parentheses are robust t- statistics. Constant is included but not shown. Source: Authors’ analysis based on data described in the text. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 123 lower prices, while those with higher incomes per capita and a larger percen- tage of rural population face higher prices. These results are consistent across all estimations reported in table 3. As expected, greater competition among providers (measured by number of respondents or the H-statistic for the banking sector) is associated with lower remittance prices (table 3, columns 3.2 and 3.3). Corridors where banks play a larger role in the remittance market exhibit higher prices (columns 3.4 and 3.5). Corridors with broader regulation of remittance service providers in the sending country have lower prices, while the regulatory breadth in the receiving country does not seem to matter (column 3.6). Greater access to and lower costs of banking services are associated with lower prices of remittances (columns 3.7–3.9). In particular, corridors with more bank branches per capita in the sending country face lower prices, while corridors with higher minimum amounts to open accounts and higher annual fees have higher remittance prices. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 The results discussed so far are economically as well as statistically signi�- cant. For example, an increase in the number of migrants from the corridor at the 25th percentile (United Kingdom–China with 56,774) to the corridor at the 75th percentile (Spain–Colombia with 384,621) is associated with a 2 per- centage point drop in average prices. An increase in competition (as measured by the H-statistic) in the sending country from the 25th percentile to the 75th implies a 4.4 percentage point reduction in remittance prices, while an increase in the receiving country is associated with a 1.4 percentage point reduction. A similar change in the number of remittance service respondents (from 6, the 25th percentile, to 10, the 75th percentile) is associated with a 1.2 percentage point drop in prices, while an increase in the scope of remittance regulation in the sending country implies a reduction of 2.8 percentage points. A comparable increase in the number of branches per capita in the sending country is associ- ated with a 1.6 percentage point decline in prices. Even stronger, an increase in the percentage of banks among survey respondents from the 25th (0 percent) to the 75th percentile (50 percent) implies an increase in prices of more than 4 percentage points. Note that the average price across corridors associated with these changes is close to 10 percent, so the effects are considerable. In contrast, no robust association is found between remittance prices and measures of exchange rate stability or the presence of capital controls on remit- tances (columns 3.1 and 3.10). Similarly, the distance between sending and receiving countries, the extent of bilateral trade, and whether countries share a common language are not correlated with remittance prices (columns 3.1 and 3.12).22 Finally, the share of educated migrants does not have a signi�cant effect (column 3.11). 22. If common language is replaced with a dummy variable for whether the receiving and sending countries have colonial ties, the main results do not change and the dummy variable for colonial ties tends to be positive and signi�cant. These results are available on request. 124 THE WORLD BANK ECONOMIC REVIEW Using alternative indicators for several variables, such as the Parson and others (2007) data on bilateral migration and a Barro and Lee (2001) measure of educational attainment, yields similar �ndings.23 Also, running the regressions for median instead of average prices does not change the results sig- ni�cantly; neither does using prices based on sending $500 instead of $200. The results are not reported here but are available on request. Overall, the estimations have good predictive power. The R-squared for the baseline regression (table 3, column 3.1) is 0.36 and varies from 0.25 (column 3.11) to 0.56 (column 3.3), depending on the additional controls included. Similarly, the estimations are reasonably good at predicting the difference between extreme observations (the difference between the corridors with the maximum and minimum prices). Depending on the estimation, the share of the actual difference between the maximum and minimum prices that is predicted by the estimations varies from 0.23 (column 3.11) to 0.41 (column 3.5). Finally, partial plots of remittance prices against the variables found to be Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 consistently signi�cant (log of migrants, log of GDP per capita in sending country, share of rural population in sending country, number of respondents in the corridor, H-statistic for bank competition in the sending country, share of bank respondents, index of importance of banks among remittance service providers, and index of regulation of remittance service providers) show that these variables do a good job of predicting prices and that the correlations are not driven by outliers (�gure 5). The log of migrants appears to be an excep- tion, with large outliers for the South Africa–Zambia and South Africa – Angola corridors (top left corner of �gure 5). However, when these two out- liers are removed, the log of migrants remains signi�cant at the 1 percent level and the other results in the baseline estimations do not change signi�cantly. Next are the factors that influence remittance prices across service provider types. Table 4 shows separate estimations for average prices among banks (columns 4.1–4.4), MTOs (columns 4.5–4.8), and Western Union (4.9–4.12). To save space, only some of the speci�cations shown to be signi�cant in the regressions for all providers (see table 3) are reported here; others are available on request. In columns 4.1–4.4, the dependent variable is the average price across all bank respondents in a corridor. Since there are corridors where banks do not play a signi�cant role in the remittance market (and so were not included in the database), the sample size is smaller than that in table 3. Most of the results discussed so far hold when the sample is restricted to banks. In particular, a larger number of migrants, lower levels of per capita income in the receiving country, and a smaller share of rural population are still 23. The correlation between the World Bank bilateral migration data and the Parson and others (2007) data is 0.66, and results do not change when the Parson and others data are used. These results are available on request. Barro and Lee’s (2001) average years of schooling of the population over 25 for the receiving country was used. Results remain unchanged. The results using the data on the education of migrants are presented here, since those data more directly relate to the population that conducts remittance transactions. ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 125 F I G U R E 5. Partial Plots of Selected Regressors against Remittance Prices Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Source: Authors’ analysis based on data described in the text. 126 T A B L E 4 . Regressions for the Average Prices Charged by Banks, Money Transfer Operators, and Western Union on $200 in Remittances Banks Money transfer operators Western Union Variable (4.1) (4.2) (4.3) (4.4) (4.5) (4.6) (4.7) (4.8) (4.9) (4.10) (4.11) (4.12) Log number of 2 1.07 2 1.57 2 1.07 2 1.04 2 1.34 2 1.20 2 1.24 2 0.71 2 2.00 2 1.86 2 2.07 2 2.03 migrants THE WORLD BANK ECONOMIC REVIEW ( 2 2.64)** ( 2 3.97)*** ( 2 2.15)** ( 2 2.50)** ( 2 7.85)*** ( 2 6.42)*** ( 2 5.40)*** ( 2 2.23)** ( 2 7.44)*** ( 2 6.97)*** ( 2 7.32)*** ( 2 4.10)*** Log GDP per capita 3.06 6.28 5.92 15.79 1.02 1.59 0.69 1.63 1.62 1.95 1.33 2.06 receiving (1.79)* (3.42)*** (3.63)*** (2.14)** (1.82)* (2.95)*** (0.87) (2.15)** (2.19)** (2.56)** (1.42) (1.97)* Log GDP per capita 4.81 3.28 0.26 14.63 1.41 1.67 1.01 3.63 2.31 3.37 2.23 3.84 sending (2.88)*** (1.46) (0.14) (2.44)** (2.42)** (2.58)** (0.93) (2.70)** (2.93)*** (2.59)** (1.68)* (0.49) Pegged or dollarized 2 1.22 0.51 1.15 3.97 2 0.91 2 0.84 2 0.70 0.18 2 2.10 2 2.08 2 2.00 2 1.16 ( 2 0.72) (0.23) (0.45) (0.85) ( 2 1.58) ( 2 1.64) ( 2 0.88) (0.19) ( 2 2.70)*** ( 2 2.56)** ( 2 1.87)* ( 2 0.72) Share rural population 0.07 0.15 0.18 0.49 0.04 0.07 0.04 0.09 0.04 0.05 0.05 0.07 receiving (1.18) (2.15)** (3.02)*** (2.04)* (1.67)* (2.86)*** (1.47) (2.62)** (1.13) (1.45) (1.23) (1.53) Share rural population 0.02 0.33 2 0.23 2 0.14 0.06 0.12 0.04 0.04 0.06 0.15 0.05 0.05 sending (0.23) (3.77)*** ( 2 1.33) ( 2 0.87) (1.44) (2.56)** (0.56) (0.54) (0.90) (1.55) (0.43) (0.31) Log distance 0.14 1.14 0.99 2.69 2 0.26 2 0.19 2 0.27 0.14 2 0.28 2 0.28 2 0.48 2 0.34 (0.18) (1.13) (1.20) (1.14) ( 2 0.77) ( 2 0.61) ( 2 0.60) (0.25) ( 2 0.54) ( 2 0.52) ( 2 0.74) ( 2 0.32) Common language 0.73 2 0.87 2 1.71 2 6.93 0.39 0.13 2 0.23 1.01 2 0.20 2 0.35 2 1.14 2.62 (0.49) ( 2 0.57) ( 2 0.93) ( 2 2.41)** (0.58) (0.20) ( 2 0.27) (0.74) ( 2 0.22) ( 2 0.38) ( 2 1.07) (1.02) Share of banks per 0.19 0.03 0.02 corridor (5.71)*** (2.48)** (0.89) Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 H 2 statistic receiving 2 7.27 2 4.55 2 2.61 ( 2 1.47) ( 2 3.46)*** ( 2 1.20) H 2 statistic sending 2 14.34 2 5.13 2 7.99 ( 2 1.80)* ( 2 2.02)** ( 2 1.60) Index of regulation 0.54 2 0.03 0.48 receiving (0.82) ( 2 0.10) (1.18) Index of regulation 2 6.39 2 0.63 2 0.25 sending ( 2 3.70)*** ( 2 0.72) ( 2 0.21) Min. amount to open 0.54 2 0.02 2 0.04 account receiving (0.99) ( 2 0.48) ( 2 0.70) Min. amount to open 50.75 52.99 2 19.66 account sending (3.85)*** (1.02) ( 2 0.27) Observations 70 66 58 26 112 106 86 50 98 92 75 38 R-squared 0.54 0.47 0.56 0.63 0.36 0.42 0.33 0.35 0.44 0.46 0.50 0.44 Diff. max-min prices 0.45 0.38 0.61 0.72 0.28 0.18 0.37 0.19 0.66 0.68 0.70 0.69 predicted over actual *Signi�cant at the 10 percent level; **signi�cant at the 5 percent level; ***signi�cant at the 1 percent level. Note: Number in parentheses are robust t-statistics. Constant is included but not shown. Source: Authors’ analysis based on data described in the text. Thorsten Beck and Marı 127 ´a Soledad Martı ´nez Perı ´a Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 128 THE WORLD BANK ECONOMIC REVIEW associated with lower prices, as is broader regulation of remittance service pro- viders in the sending country. As before, a higher share of banks among respondents and higher minimum balances to open accounts are positively cor- related with prices. The measures of competition are no longer signi�cant at the 5 percent level, a result likely due to the lower number of observations.24 Most of the earlier �ndings are con�rmed when the sample is restricted to MTOs (columns 4.5–4.8 of table 4). A larger number of migrants and greater competition in the banking system are associated with lower prices, while higher levels of income and bank participation are associated with higher prices. A larger share of rural population is associated with higher remittance prices among MTOs, but regulation of remittance service providers and costs of opening bank accounts are not signi�cantly associated with remittance prices among MTOs. Columns 4.9–4.12 of table 3 show results for the prices charged by Western Union, one of the world’s largest MTOs, active in 98 corridors of the sample. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Focusing on one �nancial institution permits controlling for any bias arising from differences in institutions across corridors (composition bias), even within the group of banks and MTOs. The price data for Western Union con�rm that a larger number of migrants and lower GDP per capita in the receiving and sending countries are associated with lower prices. In addition, exchange rate stability (as a result of pegged rates or dollarization) is also correlated with lower prices. Contrary to previous estimations, however, none of the competition-related indicators enter signi�cantly, which could be due to Western Union’s dominant position in the remittance business across most cor- ridors.25 Similarly, the share of rural population is generally not signi�cantly associated with remittance prices across corridors for Western Union. I V. C O N C L U S I O N S This article on 119 migration corridors �nds that remittance prices are associ- ated with a number of factors. First, the number of migrants is negatively and signi�cantly associated with the price of remittances across different samples and providers. This seems to suggest an important volume effect that works through scale economies and lower costs for providers or through higher com- petition in a larger market leading to a lower mark-up. Second, remittance prices are higher in corridors with higher income per capita, which could reflect higher prices of nontradable goods, such as services, in general. Third, competition and market structure matter, except in the case of Western Union. Corridors with a larger number of providers and countries with more 24. This is established by rerunning the regression for the average fee across all providers for the same sample as used in table 4. 25. This could be due to the fact that Western Union might have been operating longer in some corridors than other �rms. Also, Western Union might have better network coverage than other providers in some countries. ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 129 competitive banking sectors exhibit lower prices, although prices are higher in corridors with a higher share of banks among providers. Fourth, banking sector outreach, as measured by branch penetration and cost barriers, is associ- ated with lower remittance prices. Finally, a broader regulatory framework for remittance service providers in the sending country is associated with lower remittance prices, especially among banks. Several factors were not found to be consistently correlated with remittance prices, In particular, exchange rate stability, capital controls, and �nancial lit- eracy. However, this might be due to the use of imperfect variables to capture these policies. While this article offers some interesting �ndings on an important topic, it is only a �rst exploration into what drives remittance prices. Future research should be able to exploit panel variation to get deeper into the issues, while also addressing some of the limitations of the data and analysis. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 REFERENCES Adams, Richard H. 2005. “Remittances and Poverty and Investment in Guatemala.� In International Migration, Remittances and the Brain Drain, ed. M. Schiff, and C. Ozden. Washington DC: World Bank. Adams, Richard H., and John Page. 2003. “International Migration, Remittances and Poverty in Developing Countries.� Policy Research Working Paper 3179. World Bank, Washington DC. ¨c Aggarwal, Reena, Asli Demirgu ´a Soledad Martı ¸ -Kunt, and Marı ´a. Forthcoming. “Do Workers’ ´nez Perı Remittances Promote Financial Development?� Journal of Development Economics. Aycinena, Diego, Claudia Martinez, and Dean Yang. 2009. “The Impact of Remittance Fees on Remittance Flows: Evidence from a Field Experiment among Salvadoran Migrants.� Department of Economics, University of Michigan, Ann Arbor. Barro, Robert J., and Jong-Wha Lee. 2001. “International Data on Educational Attainment: Updates and Implications.� Oxford Economic Papers 53 (3): 541–63. Beck, Thorsten. 2007. “Ef�ciency in Financial Intermediation: Theory and Empirical Measurement.� In Bernd Balkenhol, ed., Micro�nance and Public Policy: Outreach, Performance and Ef�ciency, New York, Palgrave, MacMillan. ¨c Beck, Thorsten, Asli Demirgu ´a Soledad Martı ¸ -Kunt, and Marı ´a. 2007. “Reaching Out: Access ´nez Perı to and Use of Banking Services across Countries.� Journal of Financial Economics 85: 234 –66. ———. 2008. “Banking Services for Everyone? Barriers to Bank Access and Use around the World�. World Bank Economic Review 22 (3): 397– 430. Bikker, Jacob A., and Katharina Haaf. 2002. “Competition, Concentration, and Their Relationship: An Empirical Analysis of the Banking Industry.� Journal of Banking and Finance 26 (11): 2191–214. Bureau van Dijk. 2009. Bankscope Database. www.bvdinfo.com/ProductsContainer/Bankscope.aspx. CEPII (French Research Center in International Economics). 2010. Distances database. CEPII, Paris. http://www.cepii.fr/anglaisgraph/bdd/bdd.htm. Claessens, Stijn, and Luc Laeven. 2004. “What Drives Bank Competition? Some International Evidence.� Journal of Money, Credit, and Banking 36 (3): 562– 83. Cox-Edwards, Alejandra, and Manuelita Ureta. 2003. “International Migration, Remittances, and Schooling: Evidence from El Salvador.� Journal of Development Economics 72 (2): 429– 61. ¨c Demirgu ¸ -Kunt, Asli, Luc Laeven, and Ross Levine. 2004. “Regulations, Market Structure, Institutions, and the Cost of Financial Intermediation.� Journal of Money, Credit, and Banking 36 (3): 593–622. 130 THE WORLD BANK ECONOMIC REVIEW Duryea, Suzanne, Ernesto Lo ´ rdova, and Alexandra Olmedo. 2005. “Migrant Remittances and ´ pez Co Infant Mortality: Evidence from Mexico.� Inter-American Development Bank, Washington, DC. Freund, Caroline L., and Nikola Spatafora. 2008. “Remittances, Transaction Costs, Determinants, and Informality.� Journal of Development Economics 86 (2): 356–66. ´ n, and Jorge Roldo Gelos, R. Gasto ´ s. 2004. “Consolidation and Market Structure in Emerging Banking Systems.� Emerging Markets Review 5 (1): 39–59. Gibson, John, David J. McKenzie, and Halahingano Rohorua. 2006. “How Cost Elastic Are Remittances? Evidence from Tongan Migrants in New Zealand.� Paci�c Economic Bulletin 21 (1): 112 –28. Hanson, Gordon H., and Christopher Woodruff. 2003. “Emigration and Educational Attainment in Mexico.� School of International Relations and Paci�c Studies, University of California, San Diego. Mimeo. Harrison, Paul, Oren Sussman, and Joseph Zeira. 1999. “Finance and Growth: Theory and New Evidence.� Finance and Economics Discussion Series, FEDS Working Paper 99-35. Federal Reserve Board, Washington, DC. Hildebrandt, Nicole, and David McKenzie. 2005. “The Effects of Migration on Child Health in Mexico.� Economia 6 (1): 257– 89. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 IMF (International Monetary Fund). 2005. World Economic Outlook 2005. Washington, DC: IMF. ———. 2007. Annual Report on Exchange Arrangements and Restrictions. Washington, DC: IMF. ———. 2009. Direction of Trade database. DOT Online. IMF, Washington, DC. http://www2. imfstatistics.org/DOT/help/DOThelp.htm. Kanaiaupuni, Shawn Malia, and Katharine M. Donato. 1999. “Migradollars and Mortality: The Effects of Migration on Infant Survival in Mexico.� Demography 36 (3): 339 –53. Laeven, Luc, and Giovanni Majnoni. 2005. “Does Judicial Ef�ciency Lower the Cost of Credit?� Journal of Banking and Finance 29 (7): 1791– 812. Levy-Yeyati, Eduardo, and Alejandro Micco. 2007. Concentration and Foreign Penetration in Latin American Banking Sectors: Impact on Competition and Risk. Journal of Banking and Finance 31 (6): 1633–47. ´ pez Co Lo ´ rdova, Ernesto. 2005. “Globalization, Migration, and Development: The Role of Mexican Migrant Remittances.� Economia 6: 217–49. Maimbo, Samuel Munzele, and Dilip Ratha. 2005. Remittances: Development Impact and Future Prospects. Washington, D.C.: World Bank. Massey, D., and E. Parrado. 1998. “International Migration and Business Formation in Mexico.� Social Science Quarterly 79: 1 –20. OECD (Organisation for Economic Co-operation and Development). 2010. Database on Immigrants and Expatriates. OECD, Paris. http://stats.oecd.org/Index.aspx?lang=en#. Orozco, Manuel. 2006. “International Flows of Remittances: Cost, Competition and Financial Access in Latin America and the Caribbean—Toward an Industry Scorecard.� Inter-American Dialogue, Washington, DC. Panzar, John C., and James N. Rosse. 1982. “Structure, Conduct, and Comparative Statistics.� Economic Discussion Paper 248. Bell Laboratories, Murray Hill, NJ. ———. 1987. “Testing for ‘Monopoly’ Equilibrium.� The Journal of Industrial Economics 35 (4): 443 –56. Parsons, Christopher, Ronald Skeldon, Terrie Lousie Walmsley, and L. Alan Winters. 2007. “Quantifying International Migration: A Database of Bilateral Migrant Stocks.� Policy Research Working Paper 4165. World Bank, Washington, DC. Ratha, Dilip, and Jan Riedberg. 2005. “On Reducing Remittance Costs.� Development Prospects Group. World Bank, Washington, DC. Ratha, Dilip, and William Shaw. 2007. South– South Migration and Remittances. World Bank Working Paper 102. Washington, DC: World Bank. http://go.worldbank.org/U4RXL56V20. ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 131 Taylor, J. Edward, Jorge Mora, Richard Adams, and Alejandro Lopez-Feldman. 2005. “Remittances, Inequality, and Poverty: Evidence from Rural Mexico.� Working Paper 05-003. Department of Agricultural and Resource Economics, University of California, Davis. Woodruff, Christopher, and Rene Zenteno. 2007. “Migration Networks and Microenterprises in Mexico.� Journal of Development Economics 82 (2): 509– 28. World Bank. 2007. Data on Access to Financial Services. World Bank, Washington, DC. http://econ. worldbank.org/WBSITE/EXTERNAL/EXTDEC/EXTRESEARCH/0,,contentMDK:21546633 page PK:64214825 piPK:64214943 theSitePK:469382,00.html ———. 2008. Payment Systems Worldwide, a Snapshot: Outcomes of the Global Payment Systems Survey 2008. Washington, DC: World Bank. http://siteresources.worldbank.org/ INTPAYMENTREMMITTANCE/Resources/Global_Survey_Book.pdf ———. 2009a. Migration and Remittances Website. http://go.worldbank.org/SSW3DDNLQ0. ———. 2009b. Remittance Prices Worldwide database. World Bank, Washington, DC. http:// remittanceprices.worldbank.org. ———. 2009c. World Development Indicators database. World Bank, Washington, DC. http://data. worldbank.org/data-catalog/world-development-indicators. Yang, Dean. 2008. “International Migration, Remittances, and Household Investment: Evidence from Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Philippine Migrants’ Exchange Rate Shocks.� Economic Journal 118 (528): 591– 630. A P P E N D I X : O B TA I N I N G T H E PA N Z A R AND ROSSE (1987) H- STATIS TIC Based on the Panzar and Rosse (1987) methodology and following the empiri- cal strategy pursued by Classes and Laeven (2004), the H-statistic is obtained by estimating equation (A1): LnðPit Þ ¼ ai þ b1 lnðW1;it Þ þ b2 lnðW2;it Þ þ b3 lnðW3;it Þ þ g lnðZ;it Þ ðA1Þ þ dD þ eit where P is the ratio of gross interest revenues to total assets ( proxy for banks’ output price); W1 is the ratio of interest expenses to total deposits and money market funding ( proxy for input price of deposits); W2 is the ratio of personnel expenses to total assets ( proxy for input price of labor); W3 is the ratio of other operating and administrative expenses to total assets ( proxy for input price of equipment/�xed capital); Z is a matrix of controls including the ratio of equity to total assets, the ratio of net loans to total assets, and the logarithm of assets; D is a matrix of year dummies; ai denotes bank-level �xed effects; i denotes banks; and t denotes years. Annual balance sheet and income state- ments from Bureau van Dijk’s Bankscope database (Bureau van Dijk 2009) were used to calculate the H-statistic for each sending and receiving country banking sector during 1994–2006. The H-statistic equals b1 þ b2 þ b3, the sum of the input price elasticities of total revenues. Conceptually, the statistic measures the responsiveness of bank revenues to input prices. An H-statistic less than or equal to 0 is a sign of a monopoly, H equal to 1 indicates perfect competition, and H between 0 and 1 indicates monopolistic competition.