Bringing Discipline to the Factor Zoo: Evidence from China

Abstract

Determining the true contribution of each factor recently found in the literature poses a challenge due to the multidimensionality of control factors and omitted variable bias. To address these issues, this thesis applies advanced model selection approach in machine learning, namely double-selection LASSO in the context of Chinese A-shares market. Results presented in this paper confirms that, although most of these new factors are redundant relative to existing factors, some factors have significant statistical explanatory power compared to the hundreds of factors proposed in the past. Further, we compared the results of alternative methods to identify whether the newly proposed factors that were significant in the main analysis are robust.