Key Issues in Developing Commercial Credit Default Models (WP 2015-2)

Abstract

The stress testing mandated by the Dodd-Frank Wall Street Reform and Consumer Protection Act has reinforced the ongoing development of quantitative models banks use to project expected losses. Banks have faced challenges in developing those models because of limitations on the data available, including incomplete information about obligors and the limited duration of the obligor data. Within the context of such limitations, this paper discusses key issues that can arise while developing commercial credit default models from obligor loan data. Such issues include organizing the data, selecting economic factors for the models, and then validating the models. This paper discusses why it is good practice to organize obligors by industry and rating categories. Organizing obligor data that way, though, requires deciding how granular to make industries and rate categories. The paper illustrates one way of reducing the granularity of risk rating states. Another challenge in data organization discussed here is selecting the temporal frequency. After the data are organized, model development follows, and, in particular, selecting the appropriate economic factors for the models. The paper discusses the current approach some banks use as well as an alternative that is data dependent. Regardless of approach, the models developed need validation. That is because models developed for projecting defaults can fit the data used to develop them very well, but projections by them may not fit new realizations of the data very well. A good approach to model validation is to see how well the models perform "out-of-sample" on data realizations from a time period not used in their development.

Author

Jack Rutner