Did You Know how gini coefficients are used to gauge credit-score accuracy?
When comparing the accuracy of one credit scoring model with that of another, as in this month’s description about VantageScore 4.0 outperforming VantageScore 3.0, data scientists often cite a metric known as gini, or the gini coefficient. What is gini, and how does it apply to credit scores?
As noted in an article at the Entrepreneurial Finance Lab, the math involved in calculating a gini coefficient is pretty complicated, but the concept behind it is fairly straightforward.
Economists devised gini to measure inequality of quantities, such as wealth, income and debt, across populations. Gini values range from zero to 100, with zero indicating an absolute lack of inequality, or, in other words, total equality. For example, in a study of a population’s income, a gini of zero would indicate that every member of the population earned exactly the same amount. A gini of 100, on the other hand, indicates complete inequality; in the income study, it would mean no two people in the population had identical incomes.
When applied to credit scoring, the gini coefficient compares the distribution across the credit score range of consumers who defaulted on their loans (ie., those who went 90 days or more past due) to that of consumers who did not default on their loans. In this case, a gini value of 0 would indicate that defaulting consumers are equally distributed across the entire credit score range—in other words, that the credit score failed to assign lower credit scores to more of the consumers who defaulted, as a more predictive score would do. A coefficient value of 100 would indicate that all of the defaulting consumers were successfully assigned the lowest possible scores. Gini coefficients of 45 or above are deemed to be indicators of strong credit-scoring accuracy.