Active Machine-Learning-Based Trading and Mutual Fund Performance
Publication Date
3-17-2025
Abstract
We comprehensively examine the utilization of Machine Learning by U.S. equity mutual funds to enhance performance. We propose a novel Active ML-Based Trading (AMLT) measure based on mutual fund holdings that assesses how mutual funds actively align their portfolio decisions with forward-looking ML trading signals, generated from a Deep Neural Network model using extensive quantitative and textual information. A detailed analysis of AI talent of mutual fund employees provides validation to our measure. We document a significant rising trend in AMLT adoption, with top-decile AMLT funds outperforming bottom-decile funds by 2.4% to 3.0% annually on a risk-adjusted basis. The outperformance derives from superior stock selection, lower expenses, and efficient trading cost management, despite the inherently high turnover of ML-based strategies. We identify two key drivers of superior performance. AMLT based on a full-fledged ML model more than doubles the performance based on Linear quasi-ML models or reduced information sets, highlighting ML’s ability to process extensive information and capture complex interactions. Managers’ ability to integrate ML with human expertise contributes to sustained performance over long horizons, across different market conditions, and investment styles.
Document Type
Article
Keywords
Mutual Funds, Mutual Fund Performance, Managerial Skill, Machine Learning, Artificial Intelligence, Active Share
Disciplines
Finance
Source
SMU Cox: Finance (Topic)
Language
English
