Evaluating asset pricing models with non-traded factors using the method of maximum-correlated portfolios

Abstract

This paper examines the performance of several important asset pricing models with non-traded factors. We propose to test the asset pricing models using the method of Maximum-Correlated (MC) Portfolios. This method is particularly useful when evaluating models with non-traded factors, where the models are potentially mis-specified and factors are possibly noisy. The Q-statistics and Sharpe ratios, derived from MC portfolio method, are used as the goodness-of-fit measures. Smaller Q-stats and higher Sharpe ratios indicate better model performance. We find that Campbell (1996) and Jagannathan and Wang (1996) models are among the best models to price the test assets. These results differ significantly from the existing methods which may be biased by noisy factors.