| Abstract: | Many large-N cross-national studies claim to show that political institutions and phenomena determine where foreign direct investment (FDI) flows. In this article, I argue that these studies tend to overemphasize statistical significance and often neglect to assess the explanatory or predictive power of their theories. To illustrate the problem, I estimate variations of a statistical model published in an influential article on “Political Risk, Institutions, and FDI.” I find that none of the political variables that the authors consider accounts for much of the variation in aggregate FDI inflows. To ensure that this underwhelming result is not driven by misspecification or measurement error, I leverage a large firm-level data set on the investment location decisions of thousands of multinational firms. Using nonparametric machine-learning techniques and out-of-sample tests, I show that gravity variables can help us develop very accurate expectations about firm behavior but that none of the 31 “political determinants” of FDI that I consider can do much to improve our expectations. These findings have important implications because they suggest that governments retain some room to move in the face of economic globalization. |
