We helped an Icelandic digital financing company to develop a statistical underwriting scorecard supporting the ultimate goal of building a stable and healthy portfolio of repetitive clients and keeping reasonable default rate.
We reviewed existing risk policies of the company and proposed several adjustments in order to improve overall performance of the process and decrease associated costs. The main adjustments were associated with:
- Modification of existing KO rules
- Inclusion of a statistical scorecard
- Introduction of score-based loan limits
The proposed process optimization focused on decrease if risk costs as well as costs reduction. The final processed consisted of the following steps:
After the new process design, we developed 4 different statistical scorecards for new and existing customers covering two different types of products. The scorecards were were trained, tested and validated for stability. The main method used for the scorecards development was logistic regression and following further advanced statistical methods were tested as challangers:
The two most promising models (logistic regressin and Random Forest) for each scorecard were tested in pilot mode as challangers to the existing risk policies.
The newly designed risk policies and statistical scorecards significantly decreased probablitiy of default and increased total profitability of the products by 21%.
Increase in profitability of the loan products in the project scope.