Out-of-Sample Alphas Post-Publication

Abstract

Anomaly strategies generate positive and significant CAPM alphas post-publication. Existing explanations include non-market risks, trading costs, and investment frictions. This paper introduces a complementary channel: when a new anomaly strategy is published, investors face uncertainty in identifying the optimal weight to allocate to the anomaly in order to achieve a positive alpha post-publication, making the strategy less appealing. Empirically, we find that the average post-publication alpha of anomaly strategies is close to zero when optimal weights are estimated out-of-sample using pre-publication data. This finding is robust across specifications, including those using empirical Bayesian shrinkage and machine learning to estimate weights. Conceptually, this suggests investors have little incentive to add a new anomaly strategy to their portfolios. While investors can generate positive out-of-sample alphas by combining multiple anomaly strategies via shrinkage methods, we show the demand from such investors is insufficient to eliminate alphas in equilibrium.

Keywords: Anomalies, CAPM Alphas, Out-of-Sample Alphas

JEL Classification: G10, G11, G12, G14