Integrating Factor Models

Abstract

This paper proposes a comprehensive framework to address uncertainty about the correct factor model. Asset pricing inferences draw on a composite model that integrates over candidate models using posterior probabilities as weights. While the integrated model is weighted against deviations from the CAPM, evidence shows that (i) considerable time-varying mispricing exists, (ii) unconditional models record near-zero probabilities, and (iii) the post-earnings announcement drift, quality-minus-junk, and intermediary capital factors are incremental to the market. Moreover, equities appear considerably riskier in the presence of uncertainty about the correct model and underlying parameters. Out-of-sample, the integrated model delivers stable and outperforming efficient portfolios.