New Factor Models and the APT
57 Pages Posted: 12 Dec 2016 Last revised: 19 Jul 2019
Date Written: July 18, 2019
We examine the consistency of the existing multifactor models in the literature with the Arbitrage Pricing Theory (APT) framework. We follow the APT-related literature and estimate the common factor structure from a cross-section containing 420 equity portfolios (associated with 42 major CAPM anomalies) by employing the asymptotic principal components (APC) method. Our model contains six statistical factors from the first nine estimated APCs associated with this cross-section. The model successfully prices the 42 anomalies, with cross-sectional explanatory ratios around or above 50%. Critically, our benchmark statistical model clearly dominates most of the current multifactor models, which implies that most of these models are not good empirical proxies for an APT-based model. The sole exception to this pattern is the five-factor model of Hou, Mo, Xue, and Zhang (2019), which is equivalent (in statistical terms) to our benchmark model. Our results represent a major challenge to most current multifactor models.
Keywords: asset pricing; linear multifactor models; APT; equity risk factors; stock market anomalies; cross-section of stock returns; principal components
JEL Classification: G10; G12
Suggested Citation: Suggested Citation