The scorecard is constructed using our hybrid GAM-Decision Tree model. This process begins with calculating the GAM score based on IMP_CLAGE, IMP_DEROG, and IMP_DELINQ (CLAGE, DEROG and DELINQ after imputation). The GAM score calculation starts at 0 and is adjusted as follows:

Variable Condition GAM Adjustment
IMP_DELINQ IMP_DELINQ = 0 GAM = GAM + 2.57066154046
IMP_DELINQ = 1 GAM = GAM + 2.08231470274
IMP_DELINQ = 2 GAM = GAM + 1.32067935856
IMP_DELINQ = 3 GAM = GAM + 1.12671971708
IMP_DELINQ = 4 GAM = GAM + 1.06024767261
IMP_DELINQ = 5 GAM = GAM - 0.187270502707
IMP_DELINQ = 6 GAM = GAM - 1.16612086452
IMP_DELINQ = 7 GAM = GAM - 1.12016683131
IMP_DELINQ = 8 GAM = GAM - 0.815202415445
IMP_DELINQ = 10 GAM = GAM - 1.29377825181
IMP_DELINQ = 11 GAM = GAM - 1.27805891547
IMP_DEROG IMP_DEROG = 0 GAM = GAM + 2.56838053911
IMP_DEROG = 1 GAM = GAM + 1.88487325753
IMP_DEROG = 2 GAM = GAM + 1.36118958569
IMP_DEROG = 3 GAM = GAM + 0.736978317416
IMP_DEROG = 4 GAM = GAM - 0.552113120056
IMP_DEROG = 5 GAM = GAM + 2.56128811571
IMP_DEROG = 6 GAM = GAM + 1.31328913024
IMP_DEROG = 7 GAM = GAM + 0.607532683746
IMP_DEROG = 8 GAM = GAM - 0.970189363149
IMP_DEROG = 9 GAM = GAM - 0.219770719014
IMP_CLAGE IMP_CLAGE <= 200 GAM = GAM - 0.75
200 < IMP_CLAGE <= 600 GAM = GAM + 0.5
600 < IMP_CLAGE <= 1000 GAM = GAM + 1
IMP_CLAGE > 1000 GAM = GAM - 0.5

After obtaining the GAM score, we classify applicants into risk groups based on their DEBTINC and GAM values.

The final risk classification is as follows:

DEBTINC Condition GAM Condition Risk level Classification
Missing or DEBTINC >= 45 GAM < 1.7 “High risk”
Missing or DEBTINC >= 45 1.7 <= GAM < 5.5 “Medium risk”
Missing or DEBTINC >= 45 GAM >= 5.5 “Low-risk”
DEBTINC < 45 GAM < 1.7 “High risk”
DEBTINC < 45 1.7 <= GAM < 2.5 “Medium risk”
DEBTINC < 45 GAM >= 2.5 “Low-risk”

High Risk: Few applicants are classified as high risk, but all of them defaulted, indicating the scorecard’s accuracy in predicting true positives. Low Risk: Most applicants fall into this category, with only 38 defaults out of 446. This reflects the natural repayment tendency, suggesting the scorecard accurately captures true negatives and aligns with typical default rates, where most borrowers repay their loans. Medium Risk: This group has a moderate number of applicants, with more defaults (66) than non-defaults (39). This mixed outcome suggests cautious lending is needed for medium-risk applicants.