Background
We were engaged by an engagement with a major bank with SME lending expertise approximately one year after they had engaged Moody’s to provide risk scoring. Opportunity to compare performance in a challenging, real-world environment.
Challenge
- Moody’s delivered risk assessment of each customer
- Moody’s defined required customer and financial information
- Performing a back cast, we were not able to collect additional information or our regular information requests
- Financial information available to us was limited to only a few summary ratios
Solution
We used our proprietary modeling solution to build a risk rating system. Highly detailed analysis of internal data consistency led to a confidence metric on each element of the application. Principal component analysis and inter-correlation built minimal spanning trees, extracting the most important data and highlighting the areas of weakness.
Additionally, we matched origination time risk ratings from Moody’s to the resulting account and repayment history.
Results
- We produced a unified model across all products within the SME & Business Banking silo
- Our models deployed and reduced risk by 80%, while maintaining a greater than 90% acceptance rate
- Moody’s risk ratings explained 1.2% of risk (R-squared: 0.0121) while ours explained 70.2% in our deployed model (R-squared: 0.7015), a 58x improvement
- On a back cast: a) Our risk rating derived 97% of the economic value from the portfolio b) Moody’s was statistically indistinguishable from randomness