Paulo Maio's webpage

Disciplining Factor Models: The Role of Conditional Risk

Abstract
I propose a new method to evaluate and compare factor models - the capacity of each model in pricing its factors both unconditionally (“cross-sectional" risk prices) and conditionally (“time-series" risk prices). In a well specified model, the two sets of (covariance) risk price estimates should not deviate substantially. I show that most empirical workhorse models fail the cross-sectional-time-series consistency, both qualitatively (discrepancy in statistical significance) and quantitatively (large discrepancy in the magnitudes). The Fama - French 5-factor model is the main outlier to such a negative pattern. My analysis suggests that several prominent factors (including momentum) are not credible risk factors.