Constructing and Using Double-Adjusted Alphas to Analyze Mutual Fund Performance

We propose a new approach for estimating mutual fund performance that simultaneously controls for both factor exposure and firm characteristics. This double-adjusted alpha is motivated by the recent findings that traditional Fama-French style factor models do not fully adjust returns for the anomalies related to the factors. We formulate a hierarchical Bayesian model which separates the part of the traditional alpha that can be related to firm and asset characteristics from the true alpha. Our Bayesian approach is straightforward, has theoretical advantages over the traditional two-pass estimation and leads to higher precision. Our double-adjusted alphas produce a different ranking of mutual funds than the traditional alphas. We show that as a consequence, the double-adjusted alphas lead to stronger evidence of persistence of mutual fund performance. On the other hand, we find that the link between selectivity and alpha is driven by the effect of characteristics. Finally, we show that fund flows are mostly driven by the true skill part of the return and hardly by the effect of characteristics. We conclude that good measurement of the true outperformance of mutual funds is crucial for understanding skill.