Diversifying Estimation Errors: An Efficient Averaging Rule for Portfolio Optimization

Abstract

We propose an averaging rule that combines established minimum-variance strategies to minimize the expected out-of-sample variance. Our rule overcomes the problem of selecting the "best" strategy ex-ante and diversifies remaining estimation errors of the single strategies included in the averaging. Extensive simulations show that the contributions of estimation errors to the out-of-sample variances are uncorrelated between the considered strategies. This implies that averaging over multiple strategies offers sizable diversification benefits. Our rule leverages these benefits and compares favorably to eleven strategies in terms of out-of-sample variance on both simulated and empirical data sets. The Sharpe ratio is across all data sets at least 25% higher than for the 1/N portfolio.