Correcting alpha misattribution in portfolio sorts

Abstract

We show that portfolio sorts, as commonly employed in empirical asset pricing applications, are at risk of accidentally misattributing parts of the risk-adjusted return (or "alpha") to the firm characteristic underlying the sort. Such misattribution occurs if the firm characteristic is correlated with an unobservable yet time-persistent factor. We propose a novel, regression-based methodology for analyzing asset returns. Our technique can reproduce the alpha and factor exposure estimates from all variants of sorting assets into (e.g., decile) portfolios. In addition, and contrary to standard portfolio sorts, our approach handles multivariate and continuous firm characteristics and, if firm-specific (fixed) effects are included in the analysis, is robust to alpha misattribution. In our empirical analysis, we indeed find alpha misattribution to be an issue in conventional portfolio sorts as several well-known characteristics-based factors lose their predictive power when we account for firm fixed effects.