• Abstract: Data snooping is a major concern in empirical asset pricing. By exploiting the “blessings of dimensionality” we develop a new framework to rigorously perform multiple hypothesis testing in linear asset pricing models, while limiting the occurrence of false positive results typically associated with data-snooping. We first develop alpha test statistics that are asymptotically valid, weakly dependent in the cross-section, and robust to the possibility of omitted factors. We then combine them in a multiple-testing procedure that ensures that the rate of false discoveries is ex-ante bounded below a pre-specified 5% level. We also show that this method can detect all positive alphas with reasonable strengths. Our procedure is designed for high-dimensional settings and works even when the number of tests is large relative to the sample size, as in many finance applications. We illustrate the empirical relevance of our methodology in the context of hedge fund performance (alpha) evaluation. We find that our procedure is able to select - among more than 3,000 available funds - a subset of funds that displays superior in-sample and out-of-sample performance compared to the funds selected by standard methods.

• Paper: pdf file

• Appendix: pdf file