A retail bank wishes to develop a credit risk scoring model to evaluate the credit worthiness of its personal loan borrowers. Accurate prediction of the borrowers’ probability to default on their loans would allow the bank to implement a robust credit approval mechanism apart from pricing its loans in accordance with the associated risks. An effective credit appraisal and pricing mechanism would enable the bank to restrict potentially delinquent customers, thereby freeing up bandwidth and capital to increase business with good customers.
A retail bank wants to predict the likelihood of its personal loan borrowers to default on their loans. The bank has witnessed substantial growth in this segment, and wants to carefully select borrowers for whom it approves the loans. While the bank references credit bureau information for its loan applicants, it wants to derive additional actionable insights from in-house data on its borrowers’ historical repayment behaviour.
Delinquent loans are a major source of losses or profit leakage for a bank, and hence being able to accurately predict the probability of a loan going bad is critical to a bank sustaining its business. Banks that are able to foresee the probability of borrowers defaulting on their loans can devise robust credit approval mechanisms apart from arriving at a proper loan pricing framework.
Being able to foretell good borrowers from the potentially delinquent ones also allows banks to restrict exposure to problematic customers, thereby freeing up bandwidth and capital to execute effective customer-centric strategies that focus on increasing business with good borrowers.
A credit risk scoring model is built for the bank’s borrowers to estimate the borrowers’ credit worthiness. The model is built using data for around 200,000 of the bank’s borrowers. The dataset comprises of close to 100 variables, of which some are demographic data fields, some provide information on the borrowers’ relationship with the bank, a few are related to the commercial & economic terms of the loan taken by the borrowers, and the others provide information on the transaction & repayment track record of the borrower with the bank.
After applying dimensionality reduction techniques to prune the list of predictor variables, the scoring model is built using logistic regression techniques. Subsequent to successful model verification & validation, the credit risk scoring model provides the bank’s management with a borrower credit worthiness appraisal mechanism through which the bank can restrict loan approvals to applicants whose credit scores (a measure of their probability of default) breach a cut-off level.
Using sensitivity analysis techniques and scenario analysis approaches, the bank’s management can also examine the performance of the bank’s personal loan portfolio with respect to variations in the cut-off values of the default probability (or equivalently the credit score).
With the credit worthiness evaluation model in hand, the bank can objectively approve or reject loan applications as well as price its loans in accordance with the risk associated with the respective borrowers. In addition, the bank can focus on increasing its business with good borrowers, through top-up loans, etc.