Abstract
We present a methodology for constructing robust credit default estimates using Bayesian mixture models. Robust models explicitly take parameter uncertainty into account by allowing the modeller to formally express his degree of confidence in the model he is using and thus generate new model parameters that more accurately reflect his views on the soundness of his model. In the context of credit risk modelling, robust models are beneficial to practitioners because they provide a more structured way to err on the side of caution when estimating default probabilities. We conclude by briefly comparing the model presented with so-called credit risk models of incomplete information.
Acknowledgements
The author would like to thank Josh Davis and Niels Pederson for helpful feedback on this article. This article contains the current opinions of the author and not necessarily those of PIMCO (Pacific Investment Management Company LLC). These opinions are subject to change without notice.