For this thesis, we applied the Cross Entropy method on a credit risk model for the ING wholesale lending portfolio and some synthetically created realistic portfolios. The Cross Entropy method is found to be able to find appropriate Importance Sampling parameters within a relative modest resource budget. With the new parameters, the standard deviation of the estimate that the losses will exceed the available buffer can be decreased with more than 95%. A similar reduction with regular Monte Carlo would require the number of scenarios to increase four hundred times. Alternative methods provide similar reductions, but these use numerical methods that are more complex to implement and require more resources to calculate.Further tests show that the method is robust to the parameters used in the Cross Entropy method (within reasonable limits), it is not influenced significantly by the constitution of the portfolio and that none of the problems occur that the scientific literature warns about (in particular the “degeneracy of the likelihood ratio”).
展开▼
