Can Machine Learning Help to Select Portfolios of Mutual Funds?

Identifying outperforming mutual funds ex-ante is a notoriously difficult task. We use machine learning to exploit fund characteristics and construct portfolios of equity funds that earn positive and significant out-of-sample alpha net of all costs. In contrast, alphas of portfolios selected with OLS are indistinguishable from zero. We show that the performance of machine-learning methods is the joint outcome of exploiting multiple fund characteristics and allowing for flexibility in the relation between characteristics and performance. Our results hold also for portfolios of only retail funds, for various measures of fund performance, for different methodological choices, and across different market conditions.