Specification Test of International Asset Pricing Models

Abstract

The benchmark model for domestic asset pricing is the Sharpe-Lintner CAPM, which claims that the asset returns are determined by their covariances with the return on the (domestic) market portfolio. But since the late 1970's, the major developed countries have gradually rescinded capital controls and opened the domestic market to international investors. As the global risk factors become more influential, is the local benchmark the right benchmark for asset pricing? This study tries to answer this question by evaluating and comparing several domestic and international asset pricing models. The set of models includes the Sharpe-Lintner CAPM, the International single-beta CAPM (with and without exchange risk), the Fama-French three-factor model (1993), and the International Fama-French multi-factor model (1998). These models are tested on several common sets of assets, which are comprised of size and B/M portfolios both within and across individual countries. (When the investors are constrained to the local market or a fund manager wants to construct a country fund, it is useful to know how local factors and global factors affect asset returns in the single national market. But if the investors have access to foreign markets or the fund manager wants to diversify globally, it is more relevant to price portfolios from different countries.) I use the Hansen-Jagannathan (1997) distance measure (hereafter HJ-distance) to test and compare various models. In addition, this article explores the relation between the Fama-French empirical factors and the business cycle variables, and how exchange risks affect individual portfolio returns. The asset pricing models are mainly required to price the B/M spread in returns, because the size effect is absent from all countries. The asset pricing test results show that the models with constant risk prices fail to price the cross-sectional returns. By allowing the risk prices to fluctuate with business cycle variables, the local Fama-French factor models and the conditional international models are able to pass the HJ-distance test within US and UK. Since the business cycle variables are highly correlated across countries, it suggests the importance of common world risks in local asset pricing. The market integration hypothesis (with the uniform risk prices across countries) is not rejected in cross-country asset pricing, when risk prices are scaled by global industrial production. The conditioning business cycle variables greatly improve the performance of each model, and they are always significantly priced (together with the global market risk when cross-country). Both the value premiums as in Fama and French (1993, 1998) and the exchange risks are priced in unconditional models. However, these factors either are never priced or lose their significance in the presence of industrial production.