Rise of the Machines: Factor Investing with Artificial Neural Networks and the Cross–Section of Expected Stock Returns

Abstract

A conventional approach to factor investing entails the use of ordinary least squares (OLS) linear regression between factors (explanatory variables) and stock performance (criterion variable). In this study, we explore the benefits of allowing machines to do more with respect to combining factors, leveraging advancement in artificial intelligence (AI), specifically supervised machine learning. If we are successful in recognizing the benefit of allowing machines to do more, then we believe we are also inching the investment industry toward AI-developed investment strategies. Our findings suggest that market noise, common in the financial markets, during the training period overwhelmed the nonlinear association uncovered in the machine learning process. However, we conclude that the rationality of investor behavior, which constitutes the collective market, predicates the ultimate success of AI and machine learning in factor investing.

TOPICS: Quantitative methods, statistical methods, simulations, big data/machine learning

Key Findings

• • The AI/ML factor model delivered a superior in-sample performance but a mediocre out-sample performance versus a conventional factor model.

• • Market noise during the training period overwhelmed the nonlinear association uncovered in the machine learning process.

• • The rationality of investor behavior predicates the ultimate success of AI and machine learning in factor investing.