An Early Warning System Based on Fuzzy CMAC

Many statistical models such as the Cox´s model have been applied to the study of bank failure. However, these classical models have not attempted to identify the possible traits of financial distress that eventually leads to bank failure. It is difficult to explicitly specify what constitutes a financial distress and the intrinsic relationship between financial distress and a failed bank. This paper attempts to apply a fuzzy system named FCMAC-TVR to bank failure analysis. The FCMAC-TVR network is a generic network, in which the numerical operation is carried out by neural network, but the readable rules are generated by TVR inference scheme. The trained FCMAC-TVR operates as a bank failure classification and prediction system and the formulated fuzzy rule base shed lights on the inherent contributions of the selected financial covariates to bank failure. Experiments have demonstrated that the FCMAC-TVR network consistently outperforms the Cox´s model in classifying failed and survived banks using a set of US banking data.