A BAYESIAN APPROACH TO THE ESTIMATION OF THE CAPITAL ASSET PRICING MODEL

Abstract

In this study, we apply the Bayes estimator associated with Cauchy-type g-prior in the estimation of Capital Asset Pricing Model. We simulate a simple linear regression model with error terms generated from mixture of normal distributions and mixture of normal and Cauchy distributions respectively, to represent the 'flat-tail' distribution of security and portfolio returns. The simulation results show that the Bayes estimator is more efficient than Least Squares estimator in terms of mean square error. The Bayes estimator associated with Cauchy-type g-prior is subsequently applied to U.S. monthly and annual data and found empirically to be better than Least Squares estimator in terms of relative efficiency of one step ahead forecast mean square error.