Reappraising Empirical Validity of Asset-Pricing Models with Consideration of Statistical Power

Abstract

This paper re-evaluates the empirical validity of popular asset-pricing models, explicitly considering the statistical power of the employed test. We argue that past studies conduct the test with the power of 1 or extremely close to 1, leading to an acute imbalance between Type I and II error probabilities. As a result, at a conventional level such as 0.05, the test rejects the empirical validity of the asset-pricing models even when the null hypothesis is violated by an economically negligible margin. We re-evaluate the validity of the models by adjusting the level of significance; and by conducting the equal-probability test with moderate sample sizes. Based on these alternatives, we find evidence that the popular asset-pricing models have been adequately capturing the systematic variation of portfolio returns.