In this part of the book we present the techniques required to identify the body of the return distribution and to describe it via a parametric family of marginals and copulae.

It is important to make an important observation at this point. As we state repeatedly in the book, we hold two related, but logically distinct, sets of beliefs: the first is that information about the ‘next crisis’ is often best gleaned from a forward-looking, causal-model-inspired analysis. The consequences of, say, a possible break-up of the Euro should be examined, we believe, on its own specific terms, not as an instantiation of a generic ‘past crisis’. We also believe that little information of relevance to the crisis at hand is contained in the analysis of past crises. We are less emphatic about this point. If the reader believed that, by culling all of the past outliers out of the return distribution, she would be throwing away useful ‘crisis information’, she could dispense with the culling part of the procedure we recommend, and simply renormalize the joint distribution in such a way to accommodate for the new crisis. So, if the probability of the current crisis not happening obtained via the Bayesian net were, say, 90%, one could renormalize the full past distribution (certainly informative about past crises) to 0.9, and ‘splice’ on top the new tail contributions. This is not the procedure we prefer, but it is a perfectly feasible, and logically defensible, one.