Model Results

Christian Meyer and Peter Quell

The previous two chapters have focused on credit portfolio models on their own (design, simulation, statistics) and on the data fed into such models on their own (parameter uncertainty, estimation uncertainty). Now, the two will be brought together. Model risk issues in credit risk models will be examined based on actual model results. Finally, various benchmarking exercises will be discussed.

EFFECTS

The risk measure (99.9% VaR) has already been on display, in Chapter 11. However, what can be said about it? Considering the small simulation error, it seems to be fairly easy to compute, but what is its meaning? Does it correspond to intuition, or is intuition a dangerous path? Will it be useful for the computation of economic capital, for example? To answer these questions, we will have a closer look at the loss distribution and then perform extensive sensitivity analysis.

The loss distribution

The VaR is only a single number, and not much information can be extracted from it. Instead, one has to consider the whole loss distribution. From a distance, of course, loss distributions will almost always look like they do in the textbooks: right-skewed (ie, with positive skewness) and