Diversifying Diversification Strategies: Model Averaging in Portfolio Optimization

Abstract

The literature on portfolio optimization in the presence of parameter uncertainty has suggested several approaches to mitigate the impact of estimation error on portfolio performance. However, empirical evidence finds no single approach that can achieve a consistently higher risk-adjusted performance than 1/N. In this paper, I propose three averaging rules that synthesize the established approaches in order to mitigate the impact of estimation error on portfolio performance. The evaluation of the proposed averaging rules on empirical and simulated datasets shows that each rule achieves a consistently higher risk-adjusted performance than 1/N, while all individual portfolio strategies considered in the averaging exercise do not. I find that the observed performance gains are economically and statistically significant. The performance gains are attributable to persistent diversification effects between the portfolio strategies under consideration, as well as to empirical characteristics in portfolio returns that are exploited by one of the averaging rules.