Nonlinear Factor Attribution

Abstract

Factor attribution based on linear regression often fails to satisfactorily explain the performance of systematic investment strategies. A volatile or persistent residual suggests nonlinear interactions between factor returns and portfolio construction. We propose a nonparametric adjustment to attribute the impact thereof to better reconcile realized performance with the investment process. The approach classifies stocks based on their squared standardized factor exposures and identifies which segments are most responsible for the unexplained portfolio return. The resulting nonlinear attribution is robust and testable for statistical significance.

Keywords: Factor Investing, Performance Attribution, Nonlinear Factor Model

JEL Classification: G12, G15, C13, C51, C52