Aplicação de técnicas de machine learning na optimização de retornos financeiros no Mercado de Forex

Abstract

This work investigates the application of machine learning techniques in optimizing financial returns within the Forex market, combining rigorous statistical analysis with an economic evaluation of forecasts. The study was conducted using data extracted from MetaTrader 5, covering the period from September 9, 2010, to January 31, 2025. Supervised predictive models were considered, including L1-penalized logistic regression, decision trees, and arti- ficial neural networks, evaluated using walk-forward temporal validation with an expanding window and block bootstrap resampling, ensuring the robustness and stability of out-of- sample results (30% of the dataset), with hyperparameters optimized via grid search with time series k-fold cross-validation. The results indicate that penalized logistic regression showed greater consistency and predictive stability, statistically outperforming the decision tree and exhibiting performance comparable to the artificial neural network, with out-of- sample accuracies around 60%. Temporal stability analysis revealed that the progressive expansion of the training sample can improve algorithmic accuracy. From an economic pers- pective, forecasts were incorporated into trading strategies with explicit risk management rules, including the use of Value at Risk (VaR) as a stop-loss mechanism. This integration allowed for the conversion of moderate statistical advantages into significant improvements in the risk-return ratio, reflected in increased Sharpe and Sortino ratios and reduced draw- downs. The findings suggest that the practical utility of machine learning models depends not exclusively on algorithmic complexity, but on the combination of statistical parsimony, rigorous temporal validation, and effective risk control. In summary, the study contributes to the empirical literature by demonstrating that machine learning techniques can support decision-making in the Forex market when applied with methodological rigor and cautious interpretation, even in environments characterized by high uncertainty and instability