Asset pricing models with machine-learning method

Abstract

Traditional asset pricing theories and models are facing more and more challenges in empirical study. Machine learning provides a new tool for asset pricing research. Due to the low signal-to-noise ratio and concept drift of financial data, the theoretical constraints of economics are very important for the applicability of machine learning in asset pricing. Firstly, this paper introduces seven multi-factor asset pricing models based on ad hoc sparsity constraints, summarizes the characteristics and shortcomings of traditional asset pricing models. Then, we display the challenges of machine learning facing in empirical application of asset pricing, formulate the targeted economic constraints. Finally, we further discuss the possible future trends of machine learning algorithms in asset pricing.