Credit risk
Production Credit Risk Models
Credit scoring models for origination, first loans, and renewals, monitored in production with discrimination and stability metrics.
3credit stages covered: origination, first loans, renewals
The problem
A fintech lender needs to decide who receives credit, and on what terms, at every stage of the customer relationship: the first application, the first loans, and each renewal. Each stage has different information and different risk.
My role
I develop, implement, and monitor the credit risk models for origination, first loans, and renewals at Solventa.
Technical approach
- Models trained with XGBoost, Random Forest, and Neural Networks.
- Feature development and behavioral analysis to segment risk profiles and identify warning signals and atypical patterns, with early work on early warning and fraud detection models.
- Variable definition and validation together with Data Engineering, ensuring consistency between training and production.
Monitoring
Models in production are monitored with discrimination and stability indicators (AUC, KS, PSI), with periodic backtesting and recalibration.
Conceptual representation. Data, variables, and internal implementation details are intentionally omitted.