Resumo

The fund industry has become a vital component of the global financial market, appealing to both individual and institutional investors due to its liquidity, diversification, and low-cost professional management services. When investing in a fund, investors aim to outperform benchmarks and create value. However, predicting high-performing funds is challenging due to various influencing factors. Recent studies use of machine learning to predict fund performance, capturing nonlinear relationships and multiple variables, demonstrating superior predictive power over traditional methods.

This study investigates the effectiveness of machine learning algorithms in forecasting the performance of equity investment funds in the Brazilian market.

Mutual funds, popular due to their liquidity, diversification, and professional management services. Research on fund performance falls into three primary areas: past performance persistence, fund characteristics, and manager traits. Recent studies employ machine learning to predict fund performance, revealing the effectiveness of algorithms in capturing nonlinear relationships and enhancing predictive accuracy (DeMiguel et al., 2023).

In our approach, we employ machine learning models to generate monthly forecasts for fund performance. To determine which funds underperform and which outperform, we use the abnormal return of Carhart's four-factor model as our dependent variable. To evaluate the outcomes, we construct both long-only and long-short portfolios by deciles.

The results reveal that ensemble models outperform traditional linear models, highlighting the predictive capabilities of these advanced methods. The study indicates that fund characteristics are comparatively less significant than return and risk-based metrics.

Our results confirms the effectiveness of machine learning in predicting mutual fund performance, with Ensemble Models outperforming others. Our findings emphasize the importance of return-based metrics, especially idiosyncratic volatility, over fund characteristics.

DeMiguel, V., Gil-Bazo, J., Nogales, F. J., & Santos, A. A. (2023). Machine learning and fund characteristics help to select mutual funds with positive alpha. Journal of Financial Economics, 150(3), 103737. Kaniel, R., Lin, Z., Pelger, M., & Van Nieuwerburgh, S. (2023). Machine-learning the skill of mutual fund managers. Journal of Financial Economics, 150(1), 94–138.