A Robust Approach to Optimal Portfolio Choice with Parameter Uncertainty

It is well known that estimated mean-variance portfolios deliver, on average, poor out-of-sample performance. A lesser-known fact that we characterize in this paper is that their out-of-sample performance is also very volatile. Using our analytical characterization of out-of-sample performance volatility, we propose a measure of portfolio robustness defined as the difference between out-of-sample utility mean and a multiple of out-of-sample utility risk. We exploit our measure of portfolio robustness to calibrate shrinkage portfolios and show that they tend to outperform those portfolios that ignore parameter uncertainty or only optimize out-of-sample utility mean.