Please rotate your device to landscape mode to view the charts.

Machine-Learning the Skill of Mutual Fund Managers

Journal: Journal of Financial Economics

Date: 20231001

Author: Kaniel, Ron; Lin, Zihan; Pelger, Markus; Van Nieuwerburgh, Stijn

Abstract:
We show, using machine learning, that fund characteristics can consistently differentiate high from low-performing mutual funds, before and after fees. The outperformance persists for more than three years. Fund momentum and fund flow are the most important predictors of future risk-adjusted fund performance, while characteristics of the stocks that funds hold are not predictive. Returns of predictive long-short portfolios are higher following a period of high sentiment. Our estimation with neural networks enables us to uncover novel and substantial interaction effects between sentiment and both fund flow and fund momentum.

Link: Google Scholar

Key Findings

Machine Learning Predicts Fund Performance

Fund characteristics consistently differentiate high from low-performing mutual funds, with outperformance persisting for more than three years. Top 10% of funds earn cumulative abnormal returns of 72% while bottom 10% see -119% returns.

Fund Flow & Momentum Are Key Predictors

Fund momentum and fund flow are the most important predictors of future risk-adjusted performance, while characteristics of stocks that funds hold are not predictive. The predictive power is stronger in periods of high investor sentiment.

Long-Lasting Performance Effects

The outperformance persists for 36+ months. Even after 3 years, monthly Sharpe ratio remains at 0.20, compared to 0.30 at 3 months. Performance is robust after fees and for both small and large funds.

Performance by Information Source

  • Fund characteristics combined with sentiment achieve highest Sharpe ratio of 0.25
  • Stock characteristics alone provide no predictive value (negative Sharpe ratio)
  • Adding sentiment improves performance across both fund and stock information

Long-Term Performance Persistence

  • Initial Sharpe ratio of 0.30 remains robust over extended periods
  • Gradual decline to 0.20 over 36 months shows persistent outperformance
  • Performance remains statistically significant even after 3 years

Fund Performance by Sentiment State

  • Returns are highest (0.55%) during high sentiment periods
  • Medium sentiment periods show moderate returns (0.42%)
  • Low sentiment periods have weakest performance (0.23%)

Contribution and Implications

  • First study to show that abnormal returns, obtained as local residuals to a factor model, are statistically better targets for prediction than total returns
  • Demonstrates the importance of non-linear interaction effects between sentiment and fund characteristics that linear models fail to capture
  • Provides new methodology for measuring dependencies on macroeconomic states and interaction effects in machine learning algorithms
  • Results help improve theories of delegation in mutual fund markets and inform investor fund selection strategies

Data Sources

  • Performance by Information Source chart based on Table 3 "Performance of long-short abnormal return portfolios for different information sets"
  • Long-Term Performance Persistence visualization derived from Figure 10 showing results for long-short prediction-weighted portfolios across different holding periods
  • Fund Performance by Sentiment State chart constructed using data from Table A.7 showing long-short abnormal return portfolios in different sentiment terciles