Capturing non-linear effects in credit risk modelling
Hi Team, I have few questions on the aspect of credit risk modeling -
1. This questions me on using WOE technique. When we do WOE, if end up not capturing non-linear relationships. Is there any way to capture non-linear effects while still using WOE ?
2. What is the possibility of using ensemble methods like xgboost or neural networks for prediction ?
3. As you mentioned in the content, Scorecard is a representation of PD coefficients. If so,when predicting using non regression techniques like boosting ,How can we create Scorecard for such ?
Regards.
1 answers ( 0 marked as helpful)
Hi Phanindra,
- Provided your intervals are granular enough, WoE would capture a non-linear relationship.
- It is highly likely that such models yield more accurate predictions than a logistic regression model. However, they are not interpretable and, most likely, won't satisfy regulatory requirements in that respect.
- I don't know of any way to create a scorecard from a model like XGBoost, AdaBoost, etc.