Comparing Factor Models with Price-Impact Costs

Abstract

The Sharpe-ratio criterion proposed by Barillas and Shanken (2017) to compare factor models is insufficient in the presence of price-impact costs because the efficient frontier spanned by the factors is nonlinear. Instead, we propose a statistical test to compare factor models in terms of mean-variance utility net of price-impact costs. Empirically, model performance depends not only on the turnover required to rebalance the factor portfolios, but also on the liquidity of the stocks traded and the absolute risk-aversion parameter. The q-factor model, the six-factor Fama-French model, and a high-dimensional model outperform for high, medium and low absolute risk aversion, respectively.

Keywords: trading costs, mean-variance utility, statistical test.

JEL Classification: G11, G12