The paper develops a model with ¡§lumpy¡¨ setup costs, which govern the flow of bilateral foreign direct investment (FDI). Every country is potentially both a source for FDI flows to several host countries, and a host for FDI flows from several source countries. But technologically-advanced countries have a comparative advantage in setting up foreign subsidiaries. Thus, the model generates two-way, rich-rich and rich-poor, FDI flows. We employ a sample of 24 OECD countries, over the period 1981-1998. We observe many pairs of countries with no FDI flows between them. Zero reported flows could indicate either true zeros stemming from marginal productivity conditions, measurment errors, or true zeroes that are due to fixed costs (which dominate marginal productivity conditions). Previous empirical literature on the determinants of FDI flows imposes a no-fixed cost assumption on the estimation procedure (Tobit). In contrast, by employing the Heckman selection procedure, we show that the Tobit restriction is not consistent with the data, and yields biased estimates. Controlling for the selection into source-host pairs of countries, and for time and country fixed effects, we find: (1) FDI flows respond positively to advances in host country level of education relative to the source country level of education, whereas the source-country level of education is a predictor of the formation of source-host country pairs; (2) FDI flows respond positively to improvements in host country financial risk ratings relative to the source country ratings; (3) existence of rich-poor pairs hinge on surpassing an education-income threshold, whereas rich-rich FDI flow volumes depend on education and income levels.
