Publication Date



Foreign Portfolio Investment (FPI) has been modelled with the aim of observing patterns, discovering potential inefficiencies and to provide evidence for theories. One particular point of interest is that over the course of many studies, bilateral FPI appears to increase with an increase in the correlation between the GDP growth rates of the sender and the receiver. The Capital Asset Pricing Model (CAPM) in nancial theory predicts the opposite - and this is known as the correlation puzzle. We fit a number of gravity models for bilateral FPI holdings from the CEPII Coordinated Portfolio Investment Survey data using linear mixed models and latent space position models in order to test these theories. We use Maximum Likelihood Heckman Sample Selection estimators to account for potential bias in estimators that can be caused by frequent zeroes in our data. This results in two separate sets of response variables: the presence or absence of FPI between the two countries and the level of FPI between two countries conditional on its presence. Using seven cross-sections of data for the years 2000-2006, we estimate regression coefficients and apply model selection procedures to explain the correlation puzzle by accounting for a number of higher order dependencies such as sender and receiver random effects and then including transivity, clustering and balance. Previous work had ignored these dependencies and had incorrectly assumed independence of residuals. We show that once these factors are captured the correlation fixed effect is occasionally negative, and often not significant, which matches more closely with the predictions of models used in finance, most notably the CAPM. It appears that if there is a presence of FPI between country s and country r , then as the correlation between the economic growth of these countries increases, the level of investment will decrease. However this correlation has no significant impact on the decision of country s to invest in country r.