Work

Design and Evaluation of Machine Learning-Based Investment Strategies in Equity Funds

Year: 2026

Type: article

Abstract: This study examines quantitative investment strategies for Brazilian equity funds, integrating traditional financial performance indicators with machine learning techniques to enhance fund selection. The main objective was to construct and validate predictive models for fund selection. The methodology involved collecting daily data from 2019 to 2025, computing a range of return and risk measures, and trained models to classify 1- and 3-month shifted windows. The 3-month models achieved the strongest predictive accuracy, exceeding 91%, with the Sharpe Ratio emerging as the most influential feature. A 12-month backtest (October/2024–September/2025) showed that ML-constructed portfolios delivered cumulative returns between 14.65% and 91.86%, depending on the selection criterion, substantially outperforming Brazil’s CDI risk-free benchmark (12.70%) and the Ibovespa (11.46%). These findings highlight the practical potential of ML-based fund selection, though successful implementation requires careful risk management and ongoing model validation.

Source: International Journal of Financial Studies

DOI: 10.3390/ijfs14010016

Authors: Danillo Guimarães Cassiano da Silva, Estaner Claro Romão, Fabiano Fernandes Bargos

Institution: Universidade de São Paulo

Language: English

Topic: Stock Market Forecasting Methods

Subfield: Management Science and Operations Research

Field: Decision Sciences

Domain: Social Sciences

Open Access status: gold

Estimated APC paid: $2,172

Funder: Universidade de São Paulo