Remarks on Statistical Measures for Assessing Quality of Scoring Models

Abstract

Granting a credit product has always been at the heart of banking. Simultaneously, banks are obligated to assess the borrower’s credit risk. Apart from creditworthiness, to grant a credit product, banks are using credit scoring more and more often. Scoring models, which are an essential part of credit scoring, are being developed in order to select those clients who will repay their debt. For lenders, high effectiveness of selection based on the scoring model is the primary attribute, so it is crucial to gauge its statistical quality. Several textbooks regarding assessing statistical quality of scoring models are available, there is however no full consistency between names and definitions of particular measures. In this article, the most common statistical measures for assessing quality of scoring models, such as the pseudo Gini index, Kolmogorov‑Smirnov statistic, and concentration curve are reviewed and their statistical characteristics are discussed. Furthermore, the author proposes the application of the well‑known distribution similarity index as a measure of discriminatory power of scoring models. The author also attempts to standardise names and formulas for particular measures in order to finally contrast them in a comparative analysis of credit scoring models.