The challenge
A leading bank wanted to estimate the lifetime loss of its mortgage portfolio and track potential loss over time.
Losses could come from two events, loan default and prepayment. Estimating both was essential for future loan origination decisions and for the capital reserves the portfolio would need.
What we did
- Used a competing-hazards risk model to estimate the propensity of default and prepayment over each loan's tenure.
- Included macroeconomic factors such as the home price index, unemployment rate and fed funds rate.
- Included customer variables such as credit score and disposable income.
The result
The client could monitor a $15 billion portfolio in a regulated way and track estimated potential losses of $10 million as economic conditions changed.
The model also tracked actual default losses against forecast losses.
How we would build it today
- The same survival framework, with machine learning challenger models for validation.
- Full model lineage and versioning for model risk management.
- Scenario dashboards, and GenAI drafting of model documentation for validators to review.
This describes our current toolkit for the same problem. It is not part of the original engagement.