Dear all,The Market conditions of these last days are such that the Gaussian copula model isn't able to reflect the Credit market. More precisely, we can't find a base correlation between 0 and 1 e.g. for the iTraxx S7V1 7Y 0-22% tranche.Have anyone an idea of the "best" copula which would be able to reflect stress markets? Or maybe another model "better" in this case than copula approach for pricing bespoke CDOs tranches?It is an open question and I am very interesting in reading your practical experience about this question.Thanks by advance,Amb

There is so much going on inside a CDO that I don't think we are anywhere near a good model yet. As soon as a model becomes popular you will find all banks' prices converging, spreads narrowing, profits falling, there is less margin for error and then...Kapow! Something 'unexpected' happens because of the hidden complexity of the product and lots of people lose lots of money. (It's the same story over and over again with complex products, it's just that CDOs are more complex than most because of the interaction between all the underlyings.)So either you accept blow ups now and then, or you accept wide spreads to cope with the unexpected. Some experiments with CDO prices here.P

Dear Paul,Thank you very much for your answer. I totally agree with you and your post on your blog is interesting.Unfortunately it isn't really helpful... 8o)Amb

I know, sorry! Just some preliminary stuff at the moment!P

Wim Schoutens just gave a presentation on recent developments in credit risk, including Levy processes. If you are a CQF alum, you can access the lecture in the CQF Extra Classes. If not, you can check out his website. More specifically, his Levy Base Correlation paper is very interesting.As an alternative solution, you could try using a stochastic recovery rate instead of a deterministic one or adjust your recovery assumption. This may be a quick fix solution.

Hi Structurer,I read this article few months ago. Have you ever implemented Levy model? If yes, have you used it in stress markets (e.g. on this friday February 29th)?Levy base correlation values sound to be smaller than Gaussian and then it maybe solves my problem. Nevertheless, I would be glad to hear different points of view. In my mind, if I met difficulties friday for pricing bespoke CDOs based on iTraxx Main S7, it is because of a conjunction between illiquidity on the iTraxx tranches and stress CDS markets and then I'm afraid that no model can reflect this "unexpected" situation...Amb

QuoteOriginally posted by: AmbHave anyone an idea of the "best" copula which would be able to reflect stress markets? Or maybe another model "better" in this case than copula approach for pricing bespoke CDOs tranches?Sure. Currently I am not authorised to disclose details (ithis stuff is all about money) but may give some suggestions - copulas are total crap: conterintuitive, nummerically complex and inefficient, not flexible. Try to solve the problem from the very beginning, without using any known tools, common assumptions and sophisticated mathematics. The solution will be surprisingly simple and powerful.

QuoteSure. Currently I am not authorised to disclose details (ithis stuff is all about money) but may give some suggestions - copulas are total crap: conterintuitive, nummerically complex and inefficient, not flexible. Try to solve the problem from the very beginning, without using any known tools, common assumptions and sophisticated mathematics. The solution will be surprisingly simple and powerful.Are you able to calculate some single name or index delta with you solution ?

QuoteOriginally posted by: AmbHi Structurer,Levy base correlation values sound to be smaller than Gaussian and then it maybe solves my problem. Nevertheless, I would be glad to hear different points of view. AmbIn my experience the opposite is true. Look at Joao Garcia and Serge Goossens (2007) "Levy Base Correlation Explained " on www.defaultrisk.com. They show that Levy Base Correlations are actually higher. Since the problem with Gaussian Base Correlations are, that they can't be high enough to fit CDX senior tranches, Levy Copulas probably won't help that much.

Random recovery rate following e.g. Beta distribution could solve your problem. Also, consider fat tailed copulas like Student-or Double-t.

Dear all,For Gamal: my curiosity is woken up by your post. Back to the beginning, interesting, I will think about it...For Keanu, I have just read quickly the article Lévy Base Correlation Explained. My first assumption on Lévy BC sounds to be confirmed: the Lévy BC are smaller than Gaussian BC for senior tranches. Have you remarked somethings different in the last days (if you use this model for bespoke CDO pricing)? For Torquant and Structurer, have you any good references for stochastic recovery rates models?Amb

QuoteOriginally posted by: AmbDear all,For Keanu, I have just read quickly the article Lévy Base Correlation Explained. My first assumption on Lévy BC sounds to be confirmed: the Lévy BC are smaller than Gaussian BC for senior tranches. Have you remarked somethings different in the last days (if you use this model for bespoke CDO pricing)? AmbI have implemented some of the Levy Copulas, and found that they are not able to calibrate senior tranches in Base Correlation framework either (in the current market enviroment). What I also dislike about them, that the Base correlation is not stable in maturity (see Figure 4 in article).

Thanks Keanu for your practionner's point of view.Amb

QuoteOriginally posted by: AmbFor Torquant and Structurer, have you any good references for stochastic recovery rates models?AmbI don't know any particularly good reference on this. The quick-and-dirty model extension is conditioning on the recovery rate risk factor (one could suffice, but it depends on the underlying basket) and using numerical integration afterwards (note that you would need to rebootstrap your hazard curves).

QuoteOriginally posted by: KeanuAre you able to calculate some single name or index delta with you solution ?Of course, you may do it with any model. I may calculate even more: bucket deltas and decompose market deltas into risk factors.