Author(s)

ZijunWang

Corresponding Author

ZijunWang

ABSTRACT

To figure out a proper asset pricing factor model among a bunch of candidate factors and anomalies has been always a heated field. This paper applies the Bayesian method proposed by Barillas and Shanken (2018) to investigate the optimal factor models under a more comprehensive collection of candidate factors as well as candidate models. Given a collection of 13 candidate nonmarket factors and asset returns data of U.S. stock market from 1980 to 2018, this paper compares a total of 810 individual models and 64 category models under the same prior specification of a maximum Sharp ratio multiple. This paper also ranks the candidate individual factors and categorical factors based on their posterior probabilities. The results show that the best individual model is the six-factor model {Mkt, UMD, SMB, HMLm, ROE IA} and the optimal categorical models is {Mkt SIZE VALUE PROF INV}. In addition, factors IA and ROE showcased highest predictive power. The results are robust under different prior assumption.

KEYWORDS

Asset pricing model, Bayesian method, candidate factors, categorical models, factor models