Exploring predictive prowess of ensemble machine learning models in banking stocks: A technical, fundamental, and macroeconomic analysis

The work scrutinises the predictive prowess of ensemble machine learning models, namely Random Forest, Gradient Boosting, and XGBoost, in the domain of stock prediction by training models at two different stages. In stage 1, we restrict our evaluation to 18 technical indicators alongside holding period returns of 20 Indian Banks between January 2013 and March 2022, for short-term (20, 40, and 60 days), medium-term (180 days), and long-term durations (240 days); in stage 2, we further develop the study by including an additional combination of 6 firm-specific fundamental factors and 6 macroeconomic variables along with 18 technical indicators designed using price, volume, and momentum. During stage 1, we observe a modest range of performance, that is, between 62% and 78% across metrics, namely, accuracy score, F1 score, precision values, recall score, and specificity numbers. However, with the inclusion of fundamental and macroeconomic factors in stage 2, we observe a significant improvement in performance metrics to the tune of 90% across models for different holding periods. Particularly, for XGBoost, the reported accuracy range lies between 96% and 98%. Results indicate that while technical indicators are extremely important for short-term returns, the feature importance of fundamental and macroeconomic variables is highlighted during medium-term and long-term, respectively.