The Lorenz Curve and associated Gini coefficient are commonly used methods of measuring a model’s efficacy at determining the separation between two groups, in this case, defaulters and survivors.
- The yellow trend represents the performance of RapidRatings’ FHR at issuing ratings in the higher risk level rating categories for companies which default within a time horizon versus issuing ratings in lower risk rating categories for companies which do not default within the time horizon.
- If a model randomly allocated ratings to defaulters and survivors it would plot along the grey trend line, and would have a Gini of 0.
- If a model perfectly allocated ratings to defaulters and survivors it plot along the blue trend line, and have a Gini of 1.
|
The RapidRatings model aspires to match the ‘perfect model’ trend line and the Gini coefficient is a measure of the extent to which this is achieved. |
In constructing the Lorenz Curve, RapidRatings employs a methodology that includes all ratings issued during the time period for the ‘Universe’ sample, and all ratings which were issued during the time period and within 12 months of default are included in the Defaulter sample for the 1-yr Gini (or within 36 months for the 3-yr Gini).
|
A Gini coefficient of 0.82 is very strong and equivalent to an AUC of 0.91. |
Figures 1 and 2 present the one year and three-year Lorenz curves and Gini coefficients.