Portfolio return prediction and risk price heterogeneity

A model of portfolio return dynamics is considered in which the price of risk is permitted to be heterogeneous. In doing this, a novel method is proposed that delivers improved out-of-sample forecasts of portfolio returns. The main innovation is the use of a set of predictors that account for variation in risk prices across (segmented) markets. These predictors are the conditional covariances between the returns to the components of the portfolio under consideration and commonly used state variables (that is, Fama–French factor returns). The results indicate that the proposed method dominates competing methods (including those that assume homogeneous risk prices) when applied to domestic and international data—a finding that is robust to the sample period, performance measure, and the state variables used. The use of clustered conditional covariances leads to further improvements in out-of-sample performance.