Predictive scores: AUC-ROC and BBSS for the three models analyzed.

We also show the metrics for the benchmark model, the majority class. Both the generative model and LightGBM achieve comparable predictive performance in terms of ROC-AUC, outperforming logistic regression. While the generative model exhibits lower calibration performance, as indicated by a reduced BBSS, it uniquely satisfies the key business constraint of spread monotonicity—an essential property for applications such as optimal pricing.