Aplicação de aprendizado de máquina à teoria moderna do portfólio: uma avaliação no mercado acionário brasileiro no biênio 2018-2019

Abstract

This study evaluated the application of the Random Forest machine learning algorithm, integrated with Modern Portfolio Theory, in the selection and optimization of financial portfolios in the Brazilian stock market during the 2018-2019 biennium. The model was employed to identify assets with a higher probability of positive returns, incorporating these results into an optimizer based on the mean-variance criterion and comparing the approach to the Capital Asset Pricing Model. The results indicate that the Random Forest classifier outperforms the Capital Asset Pricing Model in terms of absolute and risk-adjusted returns, as measured by metrics such as the Sharpe Ratio, Treynor Ratio, and Jensen’s Alpha. The algorithm exhibited superior performance in portfolios with a limited number of assets, while, as diversification increased, the strategies tended to converge in terms of performance. These findings highlight the potential of machine learning as a robust tool for optimizing the risk-return trade-off in the Brazilian financial market