Predictive Analytics for Financial Crisis Detection Using Economic Indicators
DOI:
https://doi.org/10.71086/IAJE/V13I3/IAJE1319Keywords:
Financial Crisis, Early Warning System, Machine Learning, Economic Indicators, Gradient Boosting, SHAP.Abstract
Financial crises are not an exception, and the impact is felt in output, jobs, and public finances. This paper is an attempt to build a machine learning model for early warning of financial crises, based on macroeconomic and financial indicators. The framework integrates country-level data, establishes a binary classification variable using lagged classifiers, addresses the imbalance in class sizes, and compares four models: signal-extraction logit, Random Forest, Neural Network, and Gradient Boosting. The area under the ROC curve, recall, and precision are used to evaluate models. Gradient Boosting is the best performing, with an AUC of 0.92, recall of 0.83, and precision of 0.71. Analysis using SHAP values highlights that the credit-to-GDP gap, asset price volatility, and interest rate spreads are the most important variables to trigger a crisis alert. The findings have implications for using explainable machine learning as an auxiliary early warning tool to traditional EWS systems.
