Second-order accuracy metrics for scoring models and their practical use

The paper proposes new second-order accuracy metrics for scoring/rating models, which show the target preference of model - it is better to diagnose "good" objects or "bad" ones a constant generally accepted predictive power determined by first-order metric Gini index. Two proposed, they have both an integral representation and numerical one. two types, first based on binary events evaluate model, second default probability calibration given model. Comparison results calculating allows you validate settings reveals its distortions. article provides examples ratings several rating agencies, as well well-known approach van der Burg's ROC curves.