Mutual fund performance: false discoveries, bias, and power

We analyze the performance of mutual funds from a multiple inference perspective. When the number of funds is large, random fluctuations will cause some funds falsely to appear to outperform the rest. To account for such “false discoveries,” a multiple inference approach is necessary. Performance evaluation measures are unlikely to be independent across mutual funds. At the same time, the data are typically not sufficient to estimate the dependence structure of performance measures. In addition, the performance evaluation model can be misspecified. We contribute to the existing literature by applying an empirical Bayes approach that offers a possible way to take these factors into account. We also look into the question of statistical power of the performance evaluation model, which has received little attention in mutual fund studies. We find that the assumption of independence of performance evaluation measures results in significant bias, such as over-estimating the number of outperforming mutual funds. Adjusting for the mutual fund investment objective is helpful, but it still does not result in the discovery of a significant number of successful funds. A detailed analysis reveals a very low power of the study. Even if outperformers are present in the sample, they might not be recognized as such and/or too many years of data might be required to single them out.