Credit risk modeling
The discipline of quantifying the likelihood of borrower default and expected loss, using statistical and machine learning techniques on historical and current data. Outputs feed pricing, provisioning, and capital decisions.
Credit risk models estimate probability of default (PD), loss given default (LGD), and exposure at default (EAD), often segmented by product, customer type, and lifecycle stage. Inputs range from bureau scores and financial ratios to cash-flow features and behavioral signals. Regulated models face strict validation, monitoring, and explainability requirements. The current frontier is combining classical statistical models with ML on richer feature sets (transactions, alternative data) while keeping models interpretable enough for supervisory scrutiny.