Model card
- Purpose
- Give lenders an early, segment-level view of where arrears are likely to build, before loans reach default.
- Inputs
- Loan-level repayment and arrears history
- Product, vintage and term
- Collateral and guarantee flags
- Selected macroeconomic indicators
- Method
- Multi-state transition model across delinquency buckets, with transition probabilities adjusted for borrower, product and macroeconomic covariates.
- Outputs
- Transition matrices by segment
- Projected arrears by bucket and horizon
- Early-warning flags for segments moving faster than expected
- Validation
- Back-testing against partner portfolio data is planned for the pilot stage. Current version tested on synthetic data only.
- Limitations
- Needs roughly two years of loan-level history to calibrate well.
- Small segments are pooled, which reduces granularity.
- Not designed as a regulatory capital model.
Version history
- v0.3September 2026Covariate layer added.
- v0.2June 2026Prototype on synthetic portfolio data.
- v0.1April 2026Specification and data requirements.