A statistical framework for quantifying adaptive behavioural risk for the banking industry

To remain competitive in the credit industry, financial institutions constantly work to improve their credit scoring systems. This research has developed a new approach for providing early warning of default, using only transaction data for the last month and the 'product' information associated with each transaction. Customers with similar spending patterns are grouped together into clusters and risk is modelled at an individual level within each cluster, allowing customised classifiers to be developed for each cluster. The final model aggregates the customised classifiers using neural network technology, providing a tool which can more accurately predict default than previous methods. It is hoped that this tool will allow banks to use customer behaviour to allow early intervention, averting the escalation of financial problems for their clients.