Factor Selection in the Cross-section (with Chuanping Sun)

We examine a new approach to selecting factors that have incremental information for explaining the cross-section of asset returns in the factor zoo. Our method can: (1) identify a set of factors that significantly explain the cross-sectional asset returns; (2) establish a hierarchical order to explain the importance of factors; (3) quantify unique contributions of each factor; and (4) address which factors can be subsumed by others and to what extent. In a simulation study with multiple settings for factor structures, we demonstrate that our method outperforms both stepwise regression (i.e., forward selection and backward selection) and LASSO regression. Empirically, we find that selected factors are dense instead of sparse in the stock market, the corporate bond market, and the options market.