Performance Evaluation with Latent Factors

It is common to evaluate mutual fund (and in general, security) returns by linear factor models. However, performance measures from these models are misleading if there are some omitted factors that explain cross-sectional variation in returns. We propose to use a latent-factor approach, Confounder Adjusted Testing and Estimation (CATE), for performance evaluation. Under reasonable economic assumptions, we show that CATE can consistently separate "alpha" from the return components that are due to common factor exposures, without forcing any particular ex-ante specification of the factors. We demonstrate that CATE outperforms widely used factor models in identifying common variation in mutual fund returns and that CATE alpha positively predicts future fund performance. When ranked by the difference between CATE alpha and CAPM alpha, the most favorable measure used by mutual fund investors, we find that the top decile of funds outperforms the bottom decile by as large as 5% per year. We also find that mutual fund flows become less responsive to returns due to the size, value, and momentum factors over time, yet respond persistently to other factor-related variation