Credit Card Behaviour Scorecard Model
For a Large Bank
CLIENT & PROBLEM STATEMENT
- The Client is a leading Universal bank in Asia.
- It has strong customer base in retail, corporate and NR space.
- The Client wanted to create a behaviour scorecard model in order to predict whether the existing customer who is holding credit card is going to default.
APPROACH
- Internal behaviour data i.e. payment, transaction, delinquency, transaction, month on books etc. of customers was used to develop the model.
- Roll rate analysis was done in order to get the good and bad definition. Vintage analysis was done in order to get the performance widow.
- Two out of time samples were taken to check the performance of the model.
- Binning of variable was done to get the weight of evidence(WOE) and information value(IV).
- Finally, logistic regression was fitted to predict the probability of default.
- Model performance was checked using KS , AUROC , GINI , PSI , Accuracy , F1 score , Precision , Recall.
- Points to double the odds methodology was used to arrive at the scores.
SOLUTION & OUTPUT
- Three scorecards were created for three types of cards.
- The segmentation of the cards was done on the basis of days past due (DPD).
