Prediction Value of Camel: Evidence from Support Vector Machine

This study aims to analyze the predictive strength of the CAMEL framework with respect to the stability of Rural Banks by evaluating two methods of analysis: Logistic Regression (Logit) and Support Vector Machine (SVM). SVM, unlike conventional methodologies, has benefits in dealing with patterns that are intricate, non-linear, and multifaceted, often resulting in better prediction outcomes. The analysis is based on a panel dataset consisting of 1,461 Rural Banks for 56 consecutive quarters from September 2010 to March 2024. This provides 81,816 bank-quarter observations. The Incremental Predictive Value (IPV) based on Classification Performance (CP) and Receiver Operating Characters (ROC) offers assessment benchmarks for the outperforming method. Results show that SVM augments the predictive capability of CAMEL, attaining an IPV of 33.94% and for out-of-sample CP 75.72% for in-sample CP. Sensitivity assessment from an array of checks confirms these results including cuts in thresholds, varies forecasting windows, ownerships, bank sizes, regions, and locations. Through this research, we have contributed to the literature by (a) advocate the claim of diluting the predictive power of CAMEL serves as an early warning system; and (b) on the machine learning front, shining light on SVM’s scope for use in stability prediction and highlighting the need to change outlook on SVM use in financing stable projections.