Time‐Series Factor Modeling and Selection

Abstract

The article proposes a statistical time-series factor model that incorporates deterministic orthogonal trend polynomials. Such polynomials allow capturing variation in returns without initially identifying a set of robust time-series factors. This modeling approach can serve as a coherent basis for testing and selecting the most relevant factors among a set of possible ones. Additionally, it can help identify whether any factors are missing from a time-series asset pricing model. The use of the proposed model and empirical strategy is illustrated by two empirical applications from the literature, yielding results related to the Fama-French five-factor model and the factor zoo.