Robust Dynamic Portfolio Choice Based on Out-Of-Sample Performance

Abstract

We introduce a novel dynamic portfolio choice method, focusing on robust out-of-sample performance rather than on optimal in-sample performance. We therefore devise a strategy that rigorously tackles the problem of estimation error. The method involves defining a discrete set of single-period portfolio allocation policies (candidate portfolio strategies) that accommodate time-varying predictability of individual assets and choosing between them at portfolio revision dates based on bootstrapped out-of-sample portfolio returns. Our approach can handle an extensive menu of important features, in particular, arbitrarily complex types of conditional predictability, a large asset universe, parameter uncertainty, model uncertainty and downside risk aversion. We demonstrate the method's feasibility and usefulness in dynamic investment problems in futures trading, strategic asset allocation and a cross-sectional momentum strategy in equity markets.

Keywords: Out-Of-Sample Robustness, Estimation Error, Risk Management, Downside Risk, Asset Allocation

JEL Classification: G11, C61, C11, C58, C53