Comparing parametric, semi parametric and non-parametric early warning systems for banking crisis: Indian context

This study attempts to develop an early warning system (EWS) for the Indian banking sector for predicting a banking crisis. The early warning system is constructed using signal extraction approach, probit model and artificial neural networks (ANNs). The study has considered and examined the relevance of 15 leading indicators as independent variables that are likely to influence occurrence of a banking crisis, selected based on a comprehensive literature review. Inflation, stock prices, call money rate and ratio of money supply (M3) to foreign exchange reserves are found to be significant variables in predicting the possibility of an approaching banking crisis using probit and Signal Extraction approach. The comparison of predictive power for the three approaches based on quadratic probability score (QPS) and global squared bias (GSB) indicate the superiority of ANNs compared to Signal Extraction approach and probit model in terms of both accuracy and calibration. The uniqueness of this study lies in considering diverse macroeconomic variables and the relevance and reliability of different techniques in anticipating and taking proactive measures to manage the occurrence of banking sector fragility for the Indian economy.