Statlog - Processing more variables to estimate the risk of loan default
Banking
Processing more variables to estimate the risk of loan default
Background
- For banks, default probability is the estimate of the likelihood borrowers will be unable or unwilling to pay down their loans.
- Loan origination and account management strategies depend on default probability estimates.
- Banks also rely on these estimates to meet regulatory requirements and measure regulatory capital, economic capital, and loan loss allowances.
Our approach
- We developed tools to automatically conduct feature engineering and variable selection using thousands of variables available from the credit bureau and the client’s internal systems.
- We estimated logistics and survival models.
- We provided models that were deployed into production and are quarterly backtested to validate maintenance of their discriminatory and predictive power.
Our results
We created reusable tools that automated feature engineering and variable selection to speed up model development.