Definition of Default and Quality of Scoring Functions

A retail bank consumer loan dataset is used to develop logistic regression based scoring functions with different definitions of default from a very broad to a narrow or hard. The performance of the scoring functions is compared with respect to the hard definition of default which indicates real losses suffered by the bank. The results confirm the hypothesis that the scoring functions developed on softer definitions of default perform worse than those developed on harder definitions of default. The conclusion is put into contrast with the observation that the Basel II regulation gives an incentive to use a rather soft definition of default in the rating and scoring process.