American Political Science Review Vol.104,No.3 TABLE 3.Comparing Portfolio and Foreign Direct Investment Portfolio Investment FDI Log(migrant stock from din s) 0.225* 0.319 [0.146.0.304] [0.227,0.412] Log(product of GDPs) 0.138* 0.180* [0.0904,0.185) [0.117,0.243] Log(distance) -0.708* -0.967* [-1.015,-0.401) [-1.357,-0.577 Common border -0.475 0.363 [-1.191,0.240] [-0.541.1.266] Official common language 0.0154 1.406* [-0.477,0.5081 [0.684,2.129] Correlation of growth rates -0.423* 0.197 [-0.761,-0.0851] [-0.154,0.548] Common exchange rate peg 0.929* 0.296 「0.482,1.377J [-0.419,1.010] Dual taxation treaty 0.820* 0.656* [0.377.1.262] [0.150,1.162] Preferential trade agreement 0.262 0.706* [-0.392,0.916] [0.0103,1.401] Common legal heritage 0.333* 0.351* [0.0255,0.641] [-0.0189,0.721] Common dominant religion 0.448* 0.337 [0.109.0.786] [-0.0891,0.7631 Genetic distance -0.000301 -0.0000469 [-0.00101,0.000406] [-0.000834,0.0007401 Log(bilateral telephone volume) 0.233* -0.0803 [0.0620,0.404] [-0.269,0.109] Constant -104.0 -148.1* 【-145.8,-62.24 [-204.0,-92.30j Observations 2,337 Adjusted R2 FDI,foreign direct investment;GDP,gross domestic product. Seemingly unrelated regression estimated via maximum likelihood estimate (MLE);confidence intervals based on robust covariance matrix. 95%robust confidence intervals in brackets. Column 1 dependent variable:log(portfolio investment from origin to destination). Column 2 dependent variable:log(FDI). All models include both origin and destination dummy variables p<.10;*p<.05. in the FIRE sector.In columns 4,5,and 6,we repeat across different subsamples.25 This is especially im- this exercise using FDI as the dependent variable and portant as pointed out by Blonigen and Wang(2005), reach the opposite conclusion because the tertiary ed- who experience problems when pooling FDI data ucation variable is statistically insignificant,whereas across developed and developing countries.26 Table 5 employment in the FIRE sector is not.Although not breaks up the sample and examines flows of portfo- statistically consistent across all models.these results lio investment and FDI from rich countries into rich do point to the importance of specific information when it comes to portfolio investment-something associated with training in the FIRE sectors.The findings also support our inclination regarding different information 25 In an earlier version of this article,the models were estimated without fixed effects,and the findings reported were robust to the requirements of heterogeneous versus homogenous inclusion of a large number of source and destination country con- investments:heterogeneous investment opportunities trol variables.Those results were also robust when estimated via across a large number of sectors-like those associated instrumental variables using the source country's immigration policy with FDI-do not necessarily privilege those with spe- as the instrument.Unfortunately,it is not possible to repeat that cific knowledge of banking or financial assets. here because immigration policy is country specific and would be completely collinear with the set of source country fixed effects. Other plausible instruments for immigration policy(e.g.,common frameworks for labor mobility within the structure of the General Robustness of the Central Results Agreement on Trade in Services)would not meet the excludability criteria. We check the robustness of our results by estimat- ing the effect of the migrant stock on investment 26 Iam grateful to an anonymousreviewer for bringing this reference to my attention. 593American Political Science Review Vol. 104, No. 3 TABLE 3. Comparing Portfolio and Foreign Direct Investment Portfolio Investment FDI Log(migrant stock from d in s) 0.225∗∗ 0.319∗∗ [0.146, 0.304] [0.227, 0.412] Log(product of GDPs) 0.138∗∗ 0.180∗∗ [0.0904, 0.185] [0.117, 0.243] Log(distance) −0.708∗∗ −0.967∗∗ [−1.015, −0.401] [−1.357, −0.577] Common border −0.475 0.363 [−1.191, 0.240] [−0.541, 1.266] Official common language 0.0154 1.406∗∗ [−0.477, 0.508] [0.684, 2.129] Correlation of growth rates −0.423∗∗ 0.197 [−0.761, −0.0851] [−0.154, 0.548] Common exchange rate peg 0.929∗∗ 0.296 [0.482, 1.377] [−0.419, 1.010] Dual taxation treaty 0.820∗∗ 0.656∗∗ [0.377, 1.262] [0.150, 1.162] Preferential trade agreement 0.262 0.706∗∗ [−0.392, 0.916] [0.0103, 1.401] Common legal heritage 0.333∗∗ 0.351∗ [0.0255, 0.641] [−0.0189, 0.721] Common dominant religion 0.448∗∗ 0.337 [0.109, 0.786] [−0.0891, 0.763] Genetic distance −0.000301 −0.0000469 [−0.00101, 0.000406] [−0.000834, 0.000740] Log(bilateral telephone volume) 0.233∗∗ −0.0803 [0.0620, 0.404] [−0.269, 0.109] Constant −104.0∗∗ −148.1∗∗ [−145.8, −62.24] [−204.0, −92.30] Observations 2,337 Adjusted R2 FDI, foreign direct investment; GDP, gross domestic product. Seemingly unrelated regression estimated via maximum likelihood estimate (MLE); confidence intervals based on robust covariance matrix. 95% robust confidence intervals in brackets. Column 1 dependent variable: log(portfolio investment from origin to destination). Column 2 dependent variable: log(FDI). All models include both origin and destination dummy variables. ∗ p < .10; ∗∗ p < .05. in the FIRE sector. In columns 4, 5, and 6, we repeat this exercise using FDI as the dependent variable and reach the opposite conclusion because the tertiary ed￾ucation variable is statistically insignificant, whereas employment in the FIRE sector is not. Although not statistically consistent across all models, these results do point to the importance of specific information when it comes to portfolio investment—something associated with training in the FIRE sectors. The findings also support our inclination regarding different information requirements of heterogeneous versus homogenous investments: heterogeneous investment opportunities across a large number of sectors—like those associated with FDI—do not necessarily privilege those with spe￾cific knowledge of banking or financial assets. Robustness of the Central Results We check the robustness of our results by estimat￾ing the effect of the migrant stock on investment across different subsamples.25 This is especially im￾portant as pointed out by Blonigen and Wang (2005), who experience problems when pooling FDI data across developed and developing countries.26 Table 5 breaks up the sample and examines flows of portfo￾lio investment and FDI from rich countries into rich 25 In an earlier version of this article, the models were estimated without fixed effects, and the findings reported were robust to the inclusion of a large number of source and destination country con￾trol variables. Those results were also robust when estimated via instrumental variables using the source country’s immigration policy as the instrument. Unfortunately, it is not possible to repeat that here because immigration policy is country specific and would be completely collinear with the set of source country fixed effects. Other plausible instruments for immigration policy (e.g., common frameworks for labor mobility within the structure of the General Agreement on Trade in Services) would not meet the excludability criteria. 26 I am grateful to an anonymous reviewer for bringing this reference to my attention. 593