International Portfolio Diversification and Returns Correlation: Theory, Empirics, and Methodology for Panel Country-Pair Data

My dissertation attempts to explain one of the diversification puzzles in international macro-finance, and develops specification tests for the fixed effects econometric models that are widely used in international economics. The first two chapters shed light on an understudied aspect of the 'international portfolio diversification puzzle': not only do investors diversify too little abroad, but when they invest abroad, they prefer a country with less diversification benefit. Recent empirical evidence has suggested that countries that have higher stock return correlation also have higher bilateral financial asset holdings (Portes and Rey 2005, Aviat and Coeurdacier 2007, and Lane and Milesi-Ferretti 2008).;The first chapter of my dissertation argues that understanding this puzzling empirical finding requires a multi-county perspective theoretically. I begin by constructing an N-country DSGE model with heterogeneous stock return correlations. The N-country model shows that the effect of stock return correlation on bilateral asset holdings depends upon the stock return correlation with the other countries. It also shows that the overall level of equity home bias depends on the heterogeneous stock return correlations among all countries.;In the second chapter, I tested the prediction on portfolio choice with a large data set on international equity holdings. The empirical result controlling for the multilateral stock return correlations with other countries overturns the result of preceding literature, and confirms that a higher stock return correlation lowers bilateral equity asset holdings as theory predicts.;In the third chapter, I evaluated econometric methodologies that I used in the first two chapters. The country-pair fixed effects (pair FE) model is one of the commonly used models in cross-country panel research as it yields consistent parameter estimates by capturing unobserved country-pair specific heterogeneity that may be correlated with the error term. However, the pair FE model has drawbacks. Not only is there a loss of efficiency due to many dummy variables but also coefficients of time-invariant variables within a country-pair are not identified. As an alternative to the pair FE model, I estimate a country two-way fixed effects (two-way FE) model that controls for an individual country's heterogeneity within a pair. If the two-way FE model gives a consistent result as compared to the pair FE model, then the two-way FE model is able to provide estimates of time-invariant variables within a pair and to improve efficiency. The Monte-Carlo exercises compare the pair FE and the two-way FE estimators, and show which specification is appropriate. To test whether the two-way FE model sufficiently captures unobserved heterogeneity, I propose robust Hausman (1978) specification tests that can be applied even if neither estimator is fully efficient.