Tactical asset allocation and estimation risk

Abstract

In investment practice, expected returns are assumed to be time-varying. Instrumental variables like dividend yields or term spreads are employed to predict expected returns. However, there is a substantial amount of estimation risk (or, parameter uncertainty) attached to these predictive relationships. In this paper, we investigate several approaches to incorporate estimation risk in predictive regressions. We compare them to each other in an out-of-sample study of sector rotation strategies and find that incorporating estimation risk improves the risk-adjusted performance of dynamic and active asset allocation strategies and reduces turnover. The most promising strategies are based on Bayesian statistics.