Talked about this last week

http://www.noelwatson.com/blog/PermaLink,guid,d0edfd9e-c0ab-4276-a0b5-d204829bba2a.aspx

I was worried that it would be similar to his previous lecture, but was pleasantly surprised to discover that it had been updated to include recent events.

Probably the most interesting part of the evening was the analysis of the CPDOs. Taking ABN's SURF as an example, the original version of this paid 200bps over LIBOR and was rated AAA by both S&P and Moodys. Wim's argument was that using the Gaussian copula approach does not accurately model fat tail events (the credit markets are shock driven and do not follow a normal distribution), whereas the Levy model gets a lot closer. Using the Levy process would have given a rating of A.

Also discussed were CDO index tranches (ITRAXX Europe and CDX) and how the Gaussian copula gives a base correlation smile (high correlations have to be entered for more senior tranches to get accurate spreads). Using the Levy process gets a much flatter base correlation.

The next few years will be interesting for credit risk modelling. Could it be that we will find a better way to model correlation than using one variable to model the interaction between 125 companies? Will people finally start to move away from using Gaussian models to those that model the tails more accurately? Is it a good time to launch a CPDO now that spreads are historically high?

Some Fitch analysis on the first generation CPDOs