Data Mining Corrections and Mutual Fund Performance

The authors extend Markowitz and Xu's data mining correction test to allow quantitative models' return to have different market betas and apply them to estimate and test the statistical significance of variedness of mutual fund managers' skills. The authors also calculate the probability of making type I and type II errors when a registered investment advisor uses past returns to recommend buying or selling a mutual fund. In the data mining correction test setting, the authors derive the approximate formula for the information coefficient of using past returns to forecast the future holding returns and compare it with the empirical observed information coefficient.