Predição utilizando a regressão logística com indicadores técnicos como regressores no ETF PIBB11 (período de 2006-2024)

Abstract

This study applies logistic regression to predict the movement of the PIBB11 ETF from January 2022 to December 2024, using training data from 2006 to 2021, and was implemented in Python. The predictions are based on technical indicators that serve as regressors for the model. The selected technical indicators include: simple moving average, exponential moving average, moving average convergence divergence (MACD), Bollinger Bands, stochastic oscillator, and relative strength index. The model was optimized using the GridSearchCV algorithm to automate the search for optimal hyperparameters and evaluated through metrics such as accuracy, precision, recall, and F1-score. Model validation was performed using the TimeSeriesSplit cross-validation technique. The results indicated that models optimized for recall and F1-score achieved the best performance, slightly outperforming the buying and holding strategy in terms of cumulative return over the analyzed period. In contrast, the model optimized for precision and accuracy underperformed compared to the buying and holding strategy. This study contributes to the literature by demonstrating the viability of applying logistic regression with technical indicators, along with hyperparameter optimization, to predict the movement of the PIBB11 ETF.