The Kolmogorov–Smirnov Statistic, Explained: Measuring Model Power in Credit Risk Modeling

The Kolmogorov-Smirnov (KS) statistic is a performance metric used to measure the discriminatory power of classification models, especially in credit risk modeling for loan approvals. It quantifies the maximum separation between the cumulative distributions of positive (defaulters) and negative (non-defaulters) outcomes when sorted by predicted probability. The calculation involves sorting predictions, computing the cumulative rates for each class, and identifying the largest absolute difference between these rates. This maximum difference is also equivalent to the greatest vertical distance between the True Positive Rate (TPR) and the False Positive Rate (FPR) on an ROC plot. The process is demonstrated with Python code and a logistic regression model applied to a credit dataset.