Testing and Comparing Asset Pricing Factor Models: An Out-of-Sample Perspective

Kan, Wang, and Zheng (2024, JFE) demonstrate that estimation risk drives performance gaps between in-sample (INS) and out-of-sample (OOS) evaluations for factor portfolios. Addressing similar challenges in asset pricing model tests, we introduce the Sample Splitting Alpha (SSA) test, an OOS framework tailored for high-dimensional test assets. Our SSA test utilizes data-splitting techniques to decouple portfolio weight estimation from model testing and evaluates model performance using OOS ridge-regularized portfolios. Under standard regularity conditions, the SSA statistic asymptotically follows a Gaussian distribution, enabling straightforward inference. Furthermore, the SSA test provides a practical method for evaluating asset pricing models, directly tied to the OOS information ratio for efficient model comparison. Simulation results are presented to validate the statistical properties. Finally, our empirical findings indicate that while many models fail to explain the cross section in INS analysis, they are not rejected in OOS tests.