Validating Credit Default Models with Few or No Defaults

Summary

The document frames a credit risk validation problem under the Basel framework: assess the discriminatory power and calibration of rating classes when predicted probabilities of default are compared with portfolio outcomes. It highlights two difficult cases: portfolios with few borrowers and portfolios with no observed defaults. Such data may arise by chance in a small sample or because the portfolio is genuinely low risk.

The author asks how best practice should handle validation for retail and corporate portfolios, and whether research by Dirk Tasche on low-default settings can be applied beyond model calibration. The document provides no answer, validation procedure, evidence, or recommended statistical test. Its useful contribution is identifying the limits of ordinary outcome-based validation when defaults are scarce, and distinguishing the practical validation question from the related literature on calibrating probability-of-default models. Any method selection would therefore require additional sources and attention to portfolio size and risk profile.

Key ideas

• Credit rating validation evaluates both discriminatory power and calibration against realized portfolio outcomes.
• Few borrowers can make validation difficult even when the model estimates are available.
• A portfolio with no observed defaults may reflect sampling variation or genuinely low underlying risk.
• The document raises whether low-default calibration research can inform validation, but does not resolve the question.
• Retail and corporate portfolios are both included in the stated scope.

Tags

# PD validation in the low/no default setting


# PD validation in the low/no default setting












The topic of this question is the validation as prescribed in the Basel N ($N \ge 2$) framework. The task is given the probability of default $p_k$ for $K$ rating classes at time $t$ and the outcome on a credit portfolio at time $t+1$ to judge the quality (discriminatory power, calibration) of the rating model.

In practice there are cirumstances where the task of validation is difficult: We can face the situation where we have a low number of creditors (due to a special segment or because the bank just entered the market). Furthermore we can have no defaults at all. This is either by pure chance (for example in a small portfolio) or because the portfolio bears low risk.

It appears that Dirk Tasche wrote some papers in this setting and he and his collaborators are most often cited. I somehow can't see the connection to validation. But isn't there more literature?

Some literature deals with calibrating models but what is best practice in validation of no default portfolios? I would be interested in retail and corporate portfolios. I appreciate any remarks. Maybe you can help me find more or apply Tasche's findings to validation.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.