Which Factors with Price-Impact Costs?

Abstract

We show that the squared Sharpe ratio criterion considered by Barillas and Shanken (2017) is no longer appropriate to compare factor models in the presence of price-impact costs. Instead, we propose comparing factor models in terms of their mean-variance utility net of price-impact costs and develop a formal statistical test to compare nested and non-nested factor models. Empirically, we find that price-impact costs change the relative performance of factor models. For instance, while in the absence of costs a seven-factor model considered in DeMiguel, Martin-Utrera, Nogales, and Uppal (2020) is the best low-dimensional model we consider, in the presence of price-impact costs the six-factor model of Fama and French (2018) is better.