From Stock Return Predictability To Mutual Fund Performance: A Machine-learning Approach

Abstract

I use machine learning methods to identify stock characteristics important for predicting both stock returns and mutual fund performance. My customized machine learning models can successfully predict both stock returns and fund performance, and a nonlinear model delivers better performance. While existing studies find that technical signals play a dominant role in return prediction, I show that after accounting for high correlations among variables, fundamental stock characteristics are more influential than originally documented. Across funds, performance varies widely with fund exposure to anomalies; however, the aggregate performance of funds does not benefit from fund trading on market anomalies. I also find that the negative relation between stock selection and market timing – a long-standing intrigue in the fund literature – is not due to fund pursuit of anomaly-based investment strategies.

Keywords: Mutual Funds, Performance Evaluation, Anomalies, Factor Zoo, Machine Learning

JEL Classification: G11, G12, C53, C55, C58