The changing environment and regulatory policies in the financial sector have encouraged the use of credit risk models, either in-house or commercial models.
As regulations continue to evolve, there are several challenges with regards to model development and validation that keep model practitioners on their toes, according to a recent article by business analytics company SAS.
- Data Warehousing – The most important part of a credit risk model is the data set behind it, both its granularity and representativeness. In an earlier post, we explained how the phrase “Garbage in, garbage out” holds true for banking credit risk models.
- Scarce default history – If the data set contains only a small sample of defaults, accurate predictions can become difficult, and methods such as under or over sampling must be considered.
- Reject inference – The article mentions several potential methodologies such as parceling, fuzzy augmentation, etc. that can be applied for reject inference.
- Forward-looking indicators – There are several metrics such as Cash to Assets ratio, EBITDA to Assets, Debt Service Coverage Ratio that can help predict probability of default. Knowing which of these or more macroeconomic factors to use, can be a hurdle to model development.
- Accurate models – The accuracy of a credit risk model is incredibly important, both to satisfy requirements with Basel risk assessment and to minimize risk for the institution.
View the methodology behind Sageworks’ Probability of Default Model here.
Read the entire SAS article here.
