American Political Science Review Vol.104,No.3 TABLE 1.Determinants of Cross-border Portfolio Investment (1) (2) (3) (4) (5) Log(migrant stock from din s) 0.204* 0.189* 0.252* 0.209* (0.0584) (0.0608) (0.0660) (0.0596) Log(product of GDPs) 0.181* 0.178* 0.172* 0.0902* 0.157* (0.0511) (0.0481) (0.0501) (0.0449) (0.0476) Log(e) -1.421* -1.209* -1.191* -0.672* -1.337* (0.210) (0.225) (0.233) (0.282) (0.227) Common border -0.365 -0.832* -0.906* -0.802* -0.922* (0.407 (0.438) (0.438) (0.483) (0.427 Official common language 0.359 0.0387 -0.0903 0.0526 0.0265 (0.267 (0.275) (0.288) (0.327 (0.268) Correlation of growth rates 0.0926 0.0777 0.0564 -0.170 0.0458 (0.204) (0.199) (0.197) (0.274) (0.193) Common exchange rate peg 1.375 1.191* 1.147* 1.131* 1.169* (0.364) (0.347) (0.341) (0.475) (0.343) Dual taxation treaty 1.334* 1.264* 1.253* 0.795* 1.228* (0.347) (0.343) (0.341) (0.302) (0.338) Preferential trade agreement 0.806* 0.807* 0.810* 0.838* 0.810*4 (0.456) (0.423) (0.418) (0.404) (0.410) Log(bilateral telephone volume) 0.109 0.0935 0.0875 0.110 0.0822 (0.0736) (0.0702) (0.0712) (0.123) (0.0722) Common legal heritage 0.212 0.180 0.218 (0.216) (0.229) (0.211) Common dominant religion 0.550* 0.784* 0.569* (0.277) (0.272) (0.276) Genetic distance -0.000110 0.0000764 (0.000484) 0.0000707 (0.000510) (0.000493) Cultural difference -0.560* (0.196) Log(bilateral trade) -0.137* (0.0551) Constant -133.2* -134.6* -129.9* -61.02 -111.9* (44.76) (41.99) (43.82) (39.70) (41.48) Observations 4.980 4,980 4.980 2,971 4,980 Adjusted R2 0.783 0.786 0.787 0.800 0.788 GDP,gross domestic product. Robust standard errors in parentheses. Dependent variable:log(portfolio investment from origin to destination). Robust standard errors clustered by both origin and destination country. All models include both origin and destination dummy variables. p<.10:*p<.05. that migrant networks should be substantively more Table 3 contains the results of this analysis.It is im- important for FDI than for portfolio investment. portant to point out that we require complete obser- We examine this hypothesis in Table 3,where we vations for both portfolio and FDI.The overlapping estimate a seemingly unrelated regression of the de- sample results in investment from 24 source countries terminants of both portfolio investment and bilateral (all OECD)into 152 destination countries.We find that trade in commodities.This approach allows for testing larger migrant networks increase both portfolio invest- cross-equation restrictions-the null being that the ef- ment and FDI,but that the effect on FDI is substan- fect of migrant stock on portfolio investment is equal tively larger,and statistically different at the 95%con- to its effect on FDI.Rather than reporting standard fidence level,than the effect on portfolio investment. errors,we provide 95%confidence intervals.22 This is consistent with our expectations:as portfolio in- vestment represents a more homogenous opportunity 22 Hypothesis testing using seemingly unrelated regression assumes set,private information provided by migrant networks that the errors from both equations are asymptotically normal.In the becomes increasingly valuable for investors evaluating case of trade,the residuals are not due to a large number of zeros. more heterogeneous options.23 We therefore calculate standard errors and associated confidence in- tervals of Huber/White standard errors.Unfortunately,the approach of Cameron,Gelbach,and Miller(2006)is not directly applicable to 23 We note an important caveat to this result:if we divide the sam. seemingly unrelated regression. ple,as we do later,so that we are examining investments from rich 591American Political Science Review Vol. 104, No. 3 TABLE 1. Determinants of Cross-border Portfolio Investment (1) (2) (3) (4) (5) Log(migrant stock from d in s) 0.204∗∗ 0.189∗∗ 0.252∗∗ 0.209∗∗ (0.0584) (0.0608) (0.0660) (0.0596) Log(product of GDPs) 0.181∗∗ 0.178∗∗ 0.172∗∗ 0.0902∗∗ 0.157∗∗ (0.0511) (0.0481) (0.0501) (0.0449) (0.0476) Log(e) −1.421∗∗ −1.209∗∗ −1.191∗∗ −0.672∗∗ −1.337∗∗ (0.210) (0.225) (0.233) (0.282) (0.227) Common border −0.365 −0.832∗ −0.906∗∗ −0.802∗ −0.922∗∗ (0.407) (0.438) (0.438) (0.483) (0.427) Official common language 0.359 0.0387 −0.0903 0.0526 0.0265 (0.267) (0.275) (0.288) (0.327) (0.268) Correlation of growth rates 0.0926 0.0777 0.0564 −0.170 0.0458 (0.204) (0.199) (0.197) (0.274) (0.193) Common exchange rate peg 1.375∗∗ 1.191∗∗ 1.147∗∗ 1.131∗∗ 1.169∗∗ (0.364) (0.347) (0.341) (0.475) (0.343) Dual taxation treaty 1.334∗∗ 1.264∗∗ 1.253∗∗ 0.795∗∗ 1.228∗∗ (0.347) (0.343) (0.341) (0.302) (0.338) Preferential trade agreement 0.806∗ 0.807∗ 0.810∗ 0.838∗∗ 0.810∗∗ (0.456) (0.423) (0.418) (0.404) (0.410) Log(bilateral telephone volume) 0.109 0.0935 0.0875 0.110 0.0822 (0.0736) (0.0702) (0.0712) (0.123) (0.0722) Common legal heritage 0.212 0.180 0.218 (0.216) (0.229) (0.211) Common dominant religion 0.550∗∗ 0.784∗∗ 0.569∗∗ (0.277) (0.272) (0.276) Genetic distance — −0.000110 — 0.0000764 (0.000484) 0.0000707 (0.000510) (0.000493) Cultural difference −0.560∗∗ (0.196) Log(bilateral trade) −0.137∗∗ (0.0551) Constant −133.2∗∗ −134.6∗∗ −129.9∗∗ −61.02 −111.9∗∗ (44.76) (41.99) (43.82) (39.70) (41.48) Observations 4,980 4,980 4,980 2,971 4,980 Adjusted R2 0.783 0.786 0.787 0.800 0.788 GDP, gross domestic product. Robust standard errors in parentheses. Dependent variable: log(portfolio investment from origin to destination). Robust standard errors clustered by both origin and destination country. All models include both origin and destination dummy variables. ∗p < .10; ∗∗p < .05. that migrant networks should be substantively more important for FDI than for portfolio investment. We examine this hypothesis in Table 3, where we estimate a seemingly unrelated regression of the de￾terminants of both portfolio investment and bilateral trade in commodities. This approach allows for testing cross-equation restrictions—the null being that the ef￾fect of migrant stock on portfolio investment is equal to its effect on FDI. Rather than reporting standard errors, we provide 95% confidence intervals.22 22 Hypothesis testing using seemingly unrelated regression assumes that the errors from both equations are asymptotically normal. In the case of trade, the residuals are not due to a large number of zeros. We therefore calculate standard errors and associated confidence in￾tervals of Huber/White standard errors. Unfortunately, the approach of Cameron, Gelbach, and Miller (2006) is not directly applicable to seemingly unrelated regression. Table 3 contains the results of this analysis. It is im￾portant to point out that we require complete obser￾vations for both portfolio and FDI. The overlapping sample results in investment from 24 source countries (all OECD) into 152 destination countries. We find that larger migrant networks increase both portfolio invest￾ment and FDI, but that the effect on FDI is substan￾tively larger, and statistically different at the 95% con- fidence level, than the effect on portfolio investment. This is consistent with our expectations: as portfolio in￾vestment represents a more homogenous opportunity set, private information provided by migrant networks becomes increasingly valuable for investors evaluating more heterogeneous options.23 23 We note an important caveat to this result: if we divide the sam￾ple, as we do later, so that we are examining investments from rich 591