Risk prices and model selection: bad news about sparse estimators and an uniformly valid inference theory

Abstract

Lots of risk factors have been published in Finance papers in the last 20 years. Under a large menu, it’s hard to manually construct factor models with data-driven discipline and, more importantly, it’s difficult to assess the contribution of each newly proposed factor. We present some new literature on the usage of Machine Learning techniques to tackle this problem and discuss how to perform uniformly valid statistical inference on linear factor models for the stochastic discount factor. We provide further simulation evidence in favor of [Belloni and Chernozhukov, 2014] and discuss the method in [Feng et al., 2019] in detail.