Dynamic Portfolio Choice with Predictability and Parameter Uncertainty

Abstract

We study dynamic portfolio choice when expected returns depend on observed and unobserved factors and uncertain parameters, which investors learn over time. The optimal portfolio has myopic and hedging components, but parameter uncertainty complicates its dependence on predictors. We develop a novel approximation that reveals how learning affects hedging. A large predictor induces faster learning and a hedge that reduces portfolio sensitivity to changes in the predictor; ignoring uncertainty overstates predictor influence. Our method extends to multiple assets and predictors, expanding solvable portfolio choice problems. Hedging demands and optimal weights can be computed within an MCMC algorithm for parameter estimation.

Keywords: Dynamic portfolio choice with learning; hedging demands; predictive system; approximate dynamic programming; Markov Chain Monte Carlo.

JEL Classification: G11, C63