Modelling default probabilities in a credit risk portfolio

Although latent variable models are a well-known tool for evaluating a portfolio credit risk, concerns are raised regarding tractability of a subsequent analysis/simulation. The project attempts to address such concerns by incorporating a special class of Bernoulli mixture models. Then, efforts will be made to compare efficiency of these models with the commonly used latent variable counterparts and the benchmark model suggested by the regulator. In addition, the intern will produce analogous comparisons of the Bernoulli mixture model with the portfolio credit risk model that is currently employed by Sun Life Financial. At the end, the undertaken work may recommend the ways of further improving the internal quantitative techniques of Sun Life Financial that would allow more accurate and reliable assessment of a portfolio credit risk.

Evghenii Furman
Faculty Supervisor: 
Hongmei Zhu