Estimation of a Rating Model for Corporate Exposures

This chapter focused on the special difficulties that are encountered when developing internal rating models for corporate exposures. Although the whole process with data collection and processing, model building and validation usually takes quite some time and effort, the job is not yet completed with the implementation of the derived rating model. The predictive power of all statistical models depends heavily on the assumption that the historical relationship between the model’s covariates and the default event will remain unchanged in the future. Given the wide range of possible events such as changes in firms’ accounting policies or structural disruptions in certain industries, this assumption is not guaranteed over longer periods of time. Hence, it is necessary to revalidate and eventually recalibrate the model regularly in order to ensure that its predictive power does not diminish.