Ensembles of Portfolio Rules

Abstract

We propose a framework for combining portfolio rules while mitigating the impact of estimation error. Our main goal is to integrate heterogeneous rules that previously proposed combination methods cannot accommodate, enabling researchers and investors to leverage established and ongoing advances in portfolio choice. The proposed framework relies on the (pseudo) out-of-sample returns of the considered rules, thus avoiding estimation of the PRs’ return moments. The optimal combination is determined by an ensemble approach that maximizes the utility generated jointly by the candidate rules while allowing for learning about the PRs’ relative performance. Based on out-of-sample evaluations of over forty years, we document substantial utility gains for our approach compared to both individual rules and previously proposed combination strategies.

Keywords: Portfolio choice; Combination of estimators; Ensemble learning; Estimation risk

JEL Classification: G11, C10