Machine learning in factor investing: can ai predict excess returns?

Abstract

Factor investing has revolutionized asset allocation, yet its reliance on traditional, often linear, models may leave alpha on the table. This paper explores the potential of machine learning (ML) to enhance factor investing by predicting excess returns. We review the limitations of traditional factor models and delve into the theoretical underpinnings of ML for excess return prediction. We then analyse empirical evidence, contrasting the performance of ML models with conventional approaches. We find that ML, particularly non-linear models like neural networks, can outperform linear models in capturing complex relationships and generating alpha. However, challenges remain, including data overfitting, interpretability, and dynamic market adaptation. We conclude by outlining promising avenues for future research and discussing the ethical implications of AI-driven factor investing.