Tests of International CAPM with Time-Varying Covariances

Abstract

We perform maximum likelihood estimation of a model of international asset pricing based on CAPM. We test the restrictions imposed by CAPM against a more general asset pricing model. The "betas" in our CAPM vary over time from two sources -- the supplies of the assets (government obligations of France, Germany, Italy, Japan, the U.K. and the U.S.) change over time, and so do the conditional covariances of returns on these assets. We let the covariances change over time as a function of macroeconomic data. We also estimate the model when the covariances follow a multivariate ARCH process. When the covariance of forecast errors are time-varying, we can identify a modified CAFM model with measurement error -- which we also estimate. We find that the model in which the CAPM restrictions are imposed (which involve cross-equation constraints between coefficients and the variances of the residuals) perform much better when variances are not constant over time. Nonetheless, the CAPM model is rejected in favor of the less restricted model of asset pricing.