Text-Based Mutual Fund Peer Groups

Abstract

The proliferation of mutual fund strategies is a longstanding puzzle in the asset management literature. To gain new insight into this topic, we introduce a method for categorizing funds based on the strategy descriptions in their prospectuses. The resulting Strategy Peer Groups (SPGs), constructed using unsupervised machine learning, capture novel information about the funds and are more detailed than existing style categories. Where the prior literature finds that more unique funds experience greater flows, we find instead that investors prefer funds whose portfolio weights and characteristics are closer to the SPG averages. Investors also favor funds with high SPG-adjusted returns, while different investor clienteles—represented by retail, institutional, and retirement share classes—differ in their allocations across peer groups. Our results are consistent with a mutual fund industry that caters to distinct investor clienteles with heterogeneous marginal rates of substitution, rather than investors with a general preference for variety.