The Risk of Out-of-Sample Portfolio Performance

Abstract

We show theoretically and empirically that estimated portfolios bear substantial out-of-sample utility risk in high-dimensional settings or when these portfolios exploit in-sample near-arbitrage opportunities. We use our novel analytical characterization of out-of-sample utility risk to propose a robustness measure that balances out-of-sample utility mean and volatility. We demonstrate that while individual portfolios do not offer maximal robust performance, portfolio combinations achieve the optimal tradeoff between out-of-sample utility mean and volatility and are more resilient to estimation errors. Our analysis of out-of-sample performance risk has implications for constructing and evaluating quantitative investment strategies and models of the stochastic discount factor.

Keywords: parameter uncertainty, mean-variance portfolio, shrinkage

JEL Classification: G11, G12