Predicting financial cycles with dynamic ensemble selection frameworks using leading, coincident and lagging indicators

This paper develops a model for predicting financial cycles in India, and defines leading, coincident, and lagging indicators to achieve the research objective. The dependent variable is binary, and Synthetic Minority Oversampling Technique (SMOTE) is used for correcting imbalances in the dataset. The study utilizes six distinct Dynamic Ensemble Selection (DES) models, and five different pools of classifiers. Explainable Artificial Intelligence (XAI) is used to identify feature importance. The predictive framework is applied to different time periods with distinct characteristics, and all the DES frameworks yield efficient forecasts. The importance and role of the indicators, however, differ among phases. Our results show, that while during CYCLE phases, exchange rate fluctuations play a significant role in explaining financial cycles, in an UPWARD expansionary phase, expansion in bank credit, capital formation, and realty growth are significant factors. During a DOWNWARD phase and a bearish environment, VIX and oil prices emerge significant.