Semi-analytic CDO Pricing Model

trancheitup · July 20th, 2007, 3:37 pm

I have a full MC CDO pricing framework in place, but for risk analysis I'm looking to speed up the pricing process with a semi-analytical model implementation. Any opinions/references on the best place to start? Thanks!

stigko · July 20th, 2007, 9:44 pm

A paper from Kalemanove et.al, or the paper written by grad student named Annelis Luescher will serve a quick introduction to the semi-analytical gaussian case. If your port. is large and sufficiently homogeneous, I am sure that this can be done fast. Pricing more than one tranche require more than one lossdist of course, but with just one lossdist (i.e when pricing equity)took less then 5 seconds on my terribly slow computer, implemented in even slower software. (R/S-plus)

meteor · July 20th, 2007, 10:19 pm

you can check :gibson "understanding the risk of CDO" (or something like that) where he prices the tranches with the gaussian copula in a closed form (easy to understand)

sadsack2 · July 21st, 2007, 12:21 pm

I think Hull & White's explanation ("Pricing CDOs without Simulation" (?)) is the most straightforward and can be coded easily in VBA or C.

SK980 · August 9th, 2007, 9:33 pm

3/4th of the way to the end of the paper, there is a section on what I think you are seeking.

SK980 · August 9th, 2007, 9:39 pm

Does anyone know of the impact that assumptions in these models, in particular, a flat recovery rate assumption, could have on the cdo market? Ex: if recovery rates are assumed to be stochastic, how much impact can it have on current ratings?

trancheitup · August 10th, 2007, 11:29 am

Thanks for the reference. I actually implemented the model per Gibson, but the itraxx paper looks interesting. I'll give it a solid read. Regarding your question, I think for pricing, the standard is fixed recovery (I can't say I've ever seen anything different in practice). However, for ratings/risk management random recoveries are the norm. Obviously it depends on several factors, but in my experience it's not uncommon to see a cdo rating drop a notch or more when moving from say a 40% fixed recovery to a random recovery drawn from a beta distribution with 40% mean and 35% std (in-line with Moody's CDOROM assumptions).