Are machine learning models more effective than logistic regressions in predicting bank credit risk? An assessment of the Brazilian financial markets

This paper seeks to investigate whether machine learning models are more efficient than logistic regressions to predict credit risk in financial institutions. Through an empirical study that develops the models and applies interpretability techniques to identify the relationships between the variables and their importance, data and economic-financial indicators from Brazilian firms in the wholesale segment are used, combined with the use of supervised machine learning. The results indicate that the model with the best predictor performance is XGBoost, with an accuracy of 0.59 and a ROC curve of 0.97 for out-of-time data. In the interpretability analysis - via sharp value - the results corroborate the importance and economic meaning of the variables. These findings confirm the improvement in the predictive capacity of the models using machine learning techniques and are useful for the financial literature and for financial market agents in general.