Less is More: Asset Selection with Estimation Risk

摘要: Diversification across a large number of assets results in poor portfolio performance in the presence of estimation risk. We propose the reduction of the number of assets in order to resolve this problem. We develop a generic framework that determines the optimal set of assets for a mean-variance investor, ie, the set that maximises the out-of-sample performance of her portfolio strategy. We apply this framework to three popular strategies, namely the mean-variance and minimum variance portfolios as well as the 1/N rule. In a comprehensive empirical analysis, we demonstrate that asset selection drastically improves the performance of the portfolios considered. The resulting optimal portfolios are generally small, including less than 10 assets for practical sample sizes. Our findings allow fresh interpretations of three empirical puzzles related to underdiversification, home-bias and asset allocation.