Create Rating Models to assess borrower creditworthiness and calculate capital charges
Client : US Commercial Bank
Objective
To develop a rating model that allows a US commercial bank with a large corporate portfolio to assess the creditworthiness of borrowers/applicants and calculate credit risk capital charges.
CRISIL's Solution
- Develop a borrower model based on a hybrid approach employing logistic regression (LR) that directly predicts probability of default (PD) based on borrower characteristics
- Borrower data screened and variables shortlisted using suitable approaches such as binning, weight of evidence and information value
- Using stepwise logistic regression, established an econometric relationship between PD and risk attributes
- Using expert judgement-based model overlay, incorporated qualitative information and early warning signals on existing borrowers to calculate adjusted PD for each borrower, enhancing model effectiveness
- Tested model performance and stability on different test datasets using classification table and ROC curve
Client Impact
- Client used borrower rating model to strengthen its credit approval and pricing mechanism by enhancing its assessments of borrower creditworthiness
- The model helped lower losses from default by a significant extent