The Factor Model Failure Puzzle

Abstract

After six decades of research work, we still do not have a factor model that can price the cross section of U.S. equities. We empirically show that the most recent, cutting-edge asset pricing models developed to date using economic theory and machine learning cannot explain the average returns of the implied mean-variance efficient portfolios of other models. We create a theoretical model that suggests we may need hundreds of years of additional data to develop a factor model that can price the cross section.

Keywords: asset pricing, machine learning, factor models, stochastic discount factors, mean-variance efficiency, random forests, neural networks, prediction, Sharpe ratio optimization

JEL Classification: G10, G12