American Political Science Review Vol.104,No.3 to control-as completely as possible-for other mea- Sample and Methods sures of culture and familiarity.To that end,we include a number of measures that capture cultural similari- To examine the link between migrant networks and ties between the source and destination countries.The bilateral portfolio investment,we use data from the first-a measure of common legal origin-is more in- International Monetary Fund's (IMF's)Coordinated stitutional than cultural,but it captures the ability of Portfolio Investment Survey(CPIS).The CPIS collects investors from country s to invest in country d with information on the stock of cross-border investments minimal transaction costs because they will already in equities and in short-and long-term bonds broken be familiar with the rules and regulations.We expect down by issuer's country of residence.19 Due to data that country pairs with common legal origins will ex- constraints.we are able to use data on the investment perience higher levels of cross-border investment than portfolio of 56 source (reporting)countries and 154 those pairs with dissimilar legal origins. destination countries.20 The list of source and destina- Our second control for cultural similarity is a mea- tion countries is contained in Appendix C. sure of cultural proximity that is created through the Our data on FDI come from the OECD's Interna- creation of a dummy variable measuring whether the tional Direct Investment.This source is limited in that two countries have a common dominant religion.Com it only provides data for outflows from OECD coun- mon religion proxies for similar beliefs,values,and ex- tries.Therefore,when we look at bilateral FDI,our pectations regarding the existence of social norms and sample is restricted to one of 28 source countries and the internally imposed constraints that are important 158 destination countries. for a business partnership across borders. Our key independent variable-that of migrant The third cultural control is grounded in cultural networks-measures the stock (or total number)of mi- economics and operationalized as a measure of ge- grants from country d residing in country s.These data netic distance between countries.Based on the work come from a World Bank project on South-South mi- of Cavalli-Sforza,Menozzi,and Piazza (1994),schol- gration and remittances.They are based on data from ars have developed measures of genetic distances be- national statistical bureaus (censuses and population tween indigenous populations based on genetic or registers)and secondary sources(the OECD,the Inter- DNA polymorphism.18 This measure of genetic dis- national Labour Organization,and the UN).A 162 x tance has been used to proxy for culture in studies of 162 matrix of the migrant stock in country s from coun- international trade and FDI (Giuliano,Spilimbergo. try d classified according the migrant's country of birth and Tonon 2006:Guiso,Sapienza,and Zingales 2005). is constructed from these national sources (Ratha and economic development(Spolaore and Wacziarg 2008). Shaw 2007).Although some of the underlying data and state formation in Europe (Desmet et al.2007). are from the late 1990s,the majority correspond to Desmet et al.provide evidence that European coun- migrant stock for 2000 or 2001.Consequently,we are tries that are genetically alike have populations that restricted to working with cross-sectional and not time- provide similar answers to World Values Survey ques- series data. tions about cultural,religious,and moral issues. We estimate Equation (1)using ordinary least Finally,we include a more direct measure of cul- squares(OLS)and control for source-and destination tural similarity.Studies in international business find country-specific variables through the use of a double that greater cultural distance between countries is as- set of fixed effects.Inferences based on OLS standard sociated with larger transactions costs,higher uncer- errors may,however.be underestimated.This bias may tainty about business practices,and overall greater un be attributable to two related causes.First,investment ease regarding the prospects for doing business (e.g. by source countries may cluster geographically;conse- Habib and Zurawicki 2002;Kogut and Singh 1988; quently,we may need standard errors that are clustered Siegel,Licht,and Schwartz 2008).Some recent stud- by s.Second,some destination countries,for a multi- ies of international trade find that culturally similar tude of reasons,receive more investment than other countries engage in larger levels of transactions (e.g. Guiso,Sapienza,and Zingales 2005;Siegel,Licht,and Schwartz 2008;White and Tadesse 2008).Following 19 Lane and Milesi-Ferretti (2004)and Eichengreen and Lueng. this lead.we use a measure of cultural difference or naruemitchai(2006)point out some advantages and disadvantages of distance based on questions from the World Values the CPIS data.In designing the survey,the IMF attempted to ensure Survey.Unfortunately,these surveys are only given in comparability across countries;to that end,the surveys are structured to prevent double counting.With that said,the CPIS does not report 95 countries,so their use limits the size of our sample; the domestic holdings of investors,which makes testing theories of consequently,we include these measures as a robust- portfolio allocation and home bias difficult with these data,and it ness check is possible that there is some underreporting.Most significantly,for our purposes,the CPIS does not have data on the foreign holdings of a few large origin countries,including China and Saudi Arabia (although it does have these countries as destinations). 2 As in Rose and Spiegel (2008),we use the average of portfolio investment for 2002,2003,and 2004 because response rates for these 18 The details involved in the derivation of these measures in and of years differ broadly by country.Pooling these years allows us to themselves constitute a paper.The interested reader is directed to almost double the sample size.The correlation between portfolio Spolaore and Wacziarg(2008)for a discussion and application.We investment for 2002 and the average from 2002 to 2004 is 0.91.For are grateful to Spolaore and Wacziarg for generously sharing their the purpose of comparability,we construct the dependent variable data. for FDI in a similar manner. 589
American Political Science Review Vol. 104, No. 3 to control—as completely as possible—for other measures of culture and familiarity. To that end, we include a number of measures that capture cultural similarities between the source and destination countries. The first—a measure of common legal origin—is more institutional than cultural, but it captures the ability of investors from country s to invest in country d with minimal transaction costs because they will already be familiar with the rules and regulations. We expect that country pairs with common legal origins will experience higher levels of cross-border investment than those pairs with dissimilar legal origins. Our second control for cultural similarity is a measure of cultural proximity that is created through the creation of a dummy variable measuring whether the two countries have a common dominant religion. Common religion proxies for similar beliefs, values, and expectations regarding the existence of social norms and the internally imposed constraints that are important for a business partnership across borders. The third cultural control is grounded in cultural economics and operationalized as a measure of genetic distance between countries. Based on the work of Cavalli-Sforza, Menozzi, and Piazza (1994), scholars have developed measures of genetic distances between indigenous populations based on genetic or DNA polymorphism.18 This measure of genetic distance has been used to proxy for culture in studies of international trade and FDI (Giuliano, Spilimbergo, and Tonon 2006; Guiso, Sapienza, and Zingales 2005), economic development (Spolaore and Wacziarg 2008), and state formation in Europe (Desmet et al. 2007). Desmet et al. provide evidence that European countries that are genetically alike have populations that provide similar answers to World Values Survey questions about cultural, religious, and moral issues. Finally, we include a more direct measure of cultural similarity. Studies in international business find that greater cultural distance between countries is associated with larger transactions costs, higher uncertainty about business practices, and overall greater unease regarding the prospects for doing business (e.g., Habib and Zurawicki 2002; Kogut and Singh 1988; Siegel, Licht, and Schwartz 2008). Some recent studies of international trade find that culturally similar countries engage in larger levels of transactions (e.g., Guiso, Sapienza, and Zingales 2005; Siegel, Licht, and Schwartz 2008; White and Tadesse 2008). Following this lead, we use a measure of cultural difference or distance based on questions from the World Values Survey. Unfortunately, these surveys are only given in 95 countries, so their use limits the size of our sample; consequently, we include these measures as a robustness check. 18 The details involved in the derivation of these measures in and of themselves constitute a paper. The interested reader is directed to Spolaore and Wacziarg (2008) for a discussion and application. We are grateful to Spolaore and Wacziarg for generously sharing their data. Sample and Methods To examine the link between migrant networks and bilateral portfolio investment, we use data from the International Monetary Fund’s (IMF’s) Coordinated Portfolio Investment Survey (CPIS). The CPIS collects information on the stock of cross-border investments in equities and in short- and long-term bonds broken down by issuer’s country of residence.19 Due to data constraints, we are able to use data on the investment portfolio of 56 source (reporting) countries and 154 destination countries.20 The list of source and destination countries is contained in Appendix C. Our data on FDI come from the OECD’s International Direct Investment. This source is limited in that it only provides data for outflows from OECD countries. Therefore, when we look at bilateral FDI, our sample is restricted to one of 28 source countries and 158 destination countries. Our key independent variable—that of migrant networks—measures the stock (or total number) of migrants from country d residing in country s. These data come from a World Bank project on South–South migration and remittances. They are based on data from national statistical bureaus (censuses and population registers) and secondary sources (the OECD, the International Labour Organization, and the UN). A 162 × 162 matrix of the migrant stock in country s from country d classified according the migrant’s country of birth is constructed from these national sources (Ratha and Shaw 2007). Although some of the underlying data are from the late 1990s, the majority correspond to migrant stock for 2000 or 2001. Consequently, we are restricted to working with cross-sectional and not timeseries data. We estimate Equation (1) using ordinary least squares (OLS) and control for source- and destination country–specific variables through the use of a double set of fixed effects. Inferences based on OLS standard errors may, however, be underestimated. This bias may be attributable to two related causes. First, investment by source countries may cluster geographically; consequently, we may need standard errors that are clustered by s. Second, some destination countries, for a multitude of reasons, receive more investment than other 19 Lane and Milesi-Ferretti (2004) and Eichengreen and Luengnaruemitchai (2006) point out some advantages and disadvantages of the CPIS data. In designing the survey, the IMF attempted to ensure comparability across countries; to that end, the surveys are structured to prevent double counting. With that said, the CPIS does not report the domestic holdings of investors, which makes testing theories of portfolio allocation and home bias difficult with these data, and it is possible that there is some underreporting. Most significantly, for our purposes, the CPIS does not have data on the foreign holdings of a few large origin countries, including China and Saudi Arabia (although it does have these countries as destinations). 20 As in Rose and Spiegel (2008), we use the average of portfolio investment for 2002, 2003, and 2004 because response rates for these years differ broadly by country. Pooling these years allows us to almost double the sample size. The correlation between portfolio investment for 2002 and the average from 2002 to 2004 is 0.91. For the purpose of comparability, we construct the dependent variable for FDI in a similar manner. 589
Familiarity Breeds Investment August 2010 countries;a phenomena that would call for clustering portfolio investment is positively influenced by reli- on d. gious similarity,but not by a common legal heritage We deal with this potential bias by estimating stan- or by genetic distance.Both latter variables are statis- dard errors that are robust to multiway clustering as tically insignificant.Column 4 includes a more direct developed by Cameron.Gelbach.and Miller (2006). indicator of cultural similarity by including the World Their approach allows for arbitrary correlations be- Values Survey-based measure of cultural difference tween errors that belong to "the same group (along As expected,increasing cultural difference decreases either dimension)"(p.7).As they point out,this esti- cross-border portfolio investment.Adding these mea- mator is applicable in situations when the errors exhibit sures of cultural affinity or institutional familiarity do spatial correlation.Consequently,we report standard not,however,significantly affect the parameter esti- errors that are clustered by both source and desti- mate for migrant stock. nation countries.21 It should be noted that Cameron. It is also possible that patterns of bilateral investment Gelbach,and Miller mention that multiway clustering reflect other economic relationships between countries. increases-by an order of magnitude-the size of stan- Rauch and Trindade (2002)were the first to report a dard errors.In the results reported here,the standard positive relationship between diaspora networks and errors are between 60%and 100%larger than tradi- bilateral trade.If investment follows trade and not mi- tional robust standard errors.Hence,our results are gration,then inclusion of this variable should render very conservative. migrant stock statistically insignificant-or at least de- crease its substantive impact.Consequently,in column EMPIRICAL FINDINGS 5,we include a measure of bilateral trade.Trade has a negative effect on bilateral investment,indicating that these flows are substitutes rather than complements Central Results and its inclusion does not decrease the statistical or Table 1 reports models of dyadic portfolio investment. substantive importance of migrant networks. The specification in column 1 is our benchmark model, Table 2 repeats this exercise,substituting FDI as the where we just control for the variables used in prior dependent variable.Note that due to data limitations, studies of portfolio investment.Consistent with a stan- the FDI models refer to a much smaller number of dard gravity model,portfolio investment is a positive source countries.For the sake of space,we summarize function of country size (as measured by the product rather than walk through the findings from Table 2 of GDPs)and a negative function of distance.Surpris- We find that the gravity specification is a reasonable ingly,common language and common border are sta- benchmark because economic size and distance are tistically insignificant,as is the proxy for diversification statistically significant and consistently signed.The log (correlation of growth rates).Shared policies-a com- of migrant stock has a positive and statistically signif- mon exchange rate peg,a shared dual taxation treaty, icant effect on bilateral FDI that does not go away and membership in a preferential trade agreement- once we use other variables to measure cultural and have a positive and statistically significant effect on institutional familiarity. portfolio investment.We fail to find evidence that bilat- eral telephone traffic-a measure of information costs in previous studies(Portes and Rey 2005)-influences Migrant Networks and cross-border portfolio investment. Heterogeneous Investments In column 2.we add our measure of diaspora networks-the size of the migrant stock from the desti- The findings thus far support the argument that migrant nation residing in the source country.Consistent with networks serve as a conduit for capital flows,and they our hypotheses,we find that migrant networks have a point to the importance of migrant networks in the positive and statistically significant effect on portfolio provision of information.In this section,we test the in- investment.Because both the portfolio investment and formational hypothesis more directly.Following Rauch migrant stock have been transformed into logs,we can and Trindade(2002),we argue that the informational interpret the coefficient as an elasticity.This means role of migrant networks should be more important that increasing the migrant stock from a destination for trade in heterogeneous commodities,where private in a source country by 1%results in 0.2%increase information has greater value.We view FDI opportuni- in portfolio investment.Evaluated at their means,this ties as more heterogeneous than portfolio investment translates to a contribution of $450 per migrant to his opportunities.Not only are there an infinite number or her home country. of FDI opportunities-ranging from joint ownership to The migrant stock,of course,could simply be cap- greenfield investments-they also differ in that their turing cultural affinity or institutional familiarity.In risk of expropriation is greater.Portfolio investment, column 3,we include additional variables to control in contrast,can only be made in assets that are publicly for this possibility.These results are surprising because issued by either governmental or corporate interests entities that provide relatively more information to markets.Because portfolio investment is more liquid 21 We use Cameron,Gelbach,and Miller's (2006)cgmreg ado file. it can more easily be moved from market to market version 3.0,downloaded on August 2,2009,from http://gelbach. and from asset to asset,something that requires rela- eller.arizona.edu/~gelbach/ado/cgmreg.ado. tively less information than FDI.We therefore expect 590
Familiarity Breeds Investment August 2010 countries; a phenomena that would call for clustering on d. We deal with this potential bias by estimating standard errors that are robust to multiway clustering as developed by Cameron, Gelbach, and Miller (2006). Their approach allows for arbitrary correlations between errors that belong to “the same group (along either dimension)” (p. 7). As they point out, this estimator is applicable in situations when the errors exhibit spatial correlation. Consequently, we report standard errors that are clustered by both source and destination countries.21 It should be noted that Cameron, Gelbach, and Miller mention that multiway clustering increases—by an order of magnitude—the size of standard errors. In the results reported here, the standard errors are between 60% and 100% larger than traditional robust standard errors. Hence, our results are very conservative. EMPIRICAL FINDINGS Central Results Table 1 reports models of dyadic portfolio investment. The specification in column 1 is our benchmark model, where we just control for the variables used in prior studies of portfolio investment. Consistent with a standard gravity model, portfolio investment is a positive function of country size (as measured by the product of GDPs) and a negative function of distance. Surprisingly, common language and common border are statistically insignificant, as is the proxy for diversification (correlation of growth rates). Shared policies—a common exchange rate peg, a shared dual taxation treaty, and membership in a preferential trade agreement— have a positive and statistically significant effect on portfolio investment.We fail to find evidence that bilateral telephone traffic—a measure of information costs in previous studies (Portes and Rey 2005)—influences cross-border portfolio investment. In column 2, we add our measure of diaspora networks—the size of the migrant stock from the destination residing in the source country. Consistent with our hypotheses, we find that migrant networks have a positive and statistically significant effect on portfolio investment. Because both the portfolio investment and migrant stock have been transformed into logs, we can interpret the coefficient as an elasticity. This means that increasing the migrant stock from a destination in a source country by 1% results in 0.2% increase in portfolio investment. Evaluated at their means, this translates to a contribution of $450 per migrant to his or her home country. The migrant stock, of course, could simply be capturing cultural affinity or institutional familiarity. In column 3, we include additional variables to control for this possibility. These results are surprising because 21 We use Cameron, Gelbach, and Miller’s (2006) cgmreg.ado file, version 3.0, downloaded on August 2, 2009, from http://gelbach. eller.arizona.edu/∼gelbach/ado/cgmreg.ado. portfolio investment is positively influenced by religious similarity, but not by a common legal heritage or by genetic distance. Both latter variables are statistically insignificant. Column 4 includes a more direct indicator of cultural similarity by including the World Values Survey–based measure of cultural difference. As expected, increasing cultural difference decreases cross-border portfolio investment. Adding these measures of cultural affinity or institutional familiarity do not, however, significantly affect the parameter estimate for migrant stock. It is also possible that patterns of bilateral investment reflect other economic relationships between countries. Rauch and Trindade (2002) were the first to report a positive relationship between diaspora networks and bilateral trade. If investment follows trade and not migration, then inclusion of this variable should render migrant stock statistically insignificant—or at least decrease its substantive impact. Consequently, in column 5, we include a measure of bilateral trade. Trade has a negative effect on bilateral investment, indicating that these flows are substitutes rather than complements, and its inclusion does not decrease the statistical or substantive importance of migrant networks. Table 2 repeats this exercise, substituting FDI as the dependent variable. Note that due to data limitations, the FDI models refer to a much smaller number of source countries. For the sake of space, we summarize rather than walk through the findings from Table 2. We find that the gravity specification is a reasonable benchmark because economic size and distance are statistically significant and consistently signed. The log of migrant stock has a positive and statistically significant effect on bilateral FDI that does not go away once we use other variables to measure cultural and institutional familiarity. Migrant Networks and Heterogeneous Investments The findings thus far support the argument that migrant networks serve as a conduit for capital flows, and they point to the importance of migrant networks in the provision of information. In this section, we test the informational hypothesis more directly. Following Rauch and Trindade (2002), we argue that the informational role of migrant networks should be more important for trade in heterogeneous commodities, where private information has greater value. We view FDI opportunities as more heterogeneous than portfolio investment opportunities. Not only are there an infinite number of FDI opportunities—ranging from joint ownership to greenfield investments—they also differ in that their risk of expropriation is greater. Portfolio investment, in contrast, can only be made in assets that are publicly issued by either governmental or corporate interests, entities that provide relatively more information to markets. Because portfolio investment is more liquid, it can more easily be moved from market to market and from asset to asset, something that requires relatively less information than FDI. We therefore expect 590