Investment styles and the multiple testing of cross-sectional stock return predictability

Abstract

The scheme of simultaneously testing many profitable strategies may conceal the hazard of data-snooping bias. However, certain portfolio returns are also more likely to exhibit codependency because of their same investment styles. Aiming at the phenomena of stock return anomalies, we consider two multiple testing approaches: one ignores the classification of portfolios and the other utilizes such information. The results based on grouped multiple testing suggest that the implied adjusted critical values for t-statistics may vary across investment styles, and several statistically significant portfolios may be unidentified under the pooled setup.