Performance measures of LGD models

Bijak, Katarzyna and Thomas, Lyn (2015) Performance measures of LGD models. Credit Scoring and Credit Control XIV, Edinburgh, United Kingdom. 25 - 27 Aug 2015.

Abstract

As far as Probability of Default (PD) assessment is concerned, the model performance (discriminatory power) is typically measured with the Gini coefficient and/or the Kolmogorov-Smirnov (KS) statistic. For Loss Given Default (LGD) models, there are no standard performance measures, though, and a number of different measures are used. In particular, they can be classified as either error measures or non-error measures. The former include Mean Square Error (MSE), Mean Absolute Error (MAE), relative errors etc. Among the latter are coefficient of determination (R-squared) as well as various correlation coefficients between the observed and predicted LGD. Obviously, the lower the error measures and the higher the non-error measures, the better the model performance. Whereas all performance measures have both advantages and disadvantages, some measures cannot be readily recommended for LGD models, even though they have been used for this purpose. It is argued that some measures should only be used for specific types of models (e.g. for linear regression). It is also pointed out that some measures can be used interchangeably to avoid information redundancy. Moreover, both the Area Over the Regression Error Characteristic Curve (AOC) and the Area Under the Receiver Operating Characteristic Curve (AUC) are critically discussed in the LGD context, and an alternative to the latter is proposed.

This record has no associated files available for download.

Contributors

Author: Katarzyna Bijak

Download statistics