Strategic Asset Allocation for Long-Term Investors: Parameter Uncertainty and Prior Information

Abstract

We study the effect of parameter uncertainty on the long-run risk of three alternative asset classes: equity, nominal bonds and short-term T-Bills. We estimate the long-run risk as the annualized predictive variance of returns at different horizons implied by a vector autoregression using alternative Bayesian priors. Under an uninformative prior we conclude that not only equity becomes more risky relative to estimates that are conditional on known parameter values. The long-run risk of long and short-term bonds increases proportionally with the same factor. Correlations among returns appear robust against parameter uncertainty. Alternative informative priors imply large differences in expected returns, which lead to different optimal portfolios. To limit the effect of a single prior we derive a robust portfolio rule that associates a portfolio with the worst prior for that portfolio. The optimal robust portfolio appears well-diversified and stable with respect to the investment horizon.

Keywords: strategic asset allocation, Bayesian vector autoregression, parameter uncertainty, robust portfolio choice

JEL Classification: C32, G11, C11