Primo, perfectly agree... know the limits of your modelbut the gaussian copula cannot be accused of everything (although it's quite tempting, check this out).let's at least check the accusations

<t>ok katastrofa, I thought we were talking about conditioning on the value of the factor. But still, even looking at the quantities you mention one has to be very careful, because all the copula construction is based on the terminal distribution at a single time horizon. If you introduce another ti...

<t>@ pine80: may I say that I'm not convinced by the negative forward conditional recovery argument. The factor copula only builds terminal distributions, accordingly the factor Z only makes sense in the context of a terminal distribution. What I mean is that you cannot condition on the same value o...

Anyway, I've just posted this paper on ssrn ssrn.com/abstract=1346414and I would be very interested in having your comments (and in particular Sebastien's comments given that it's an application of his approach). With the hope that one day we'll all be able to think post-crisis!

StringBean, maybe adding stochastic recovery to the gaussian copula is not the solution, but stochastic recovery is part of the solution, otherwise you will always have a supersenior attaching sufficiently high that will be worth zero.

what about the non-homogeneous case?

also, don't forget the super-duper pricing problem! I guess all new modelling proposals must now be able to demonstrate ability to assign non-zero value to tranchelets across the whole capital structure spectrum under reasonable assumptions in terms of market conditions

<t>agree with katastrofa that if anything this model would create a positive correlation between recovery and default prob: names that have a higher prob of defaulting will likely be the first to do so and therefore will be assigned a conditionally higher recovery.anyway.. one comment that I feel li...

ThinkDifferent, apart from having doubts about its feasibility, I really don't see the point of your solution. Did you test this framework of making recoveries a function of the capital structure?Does it produce a good expected loss surface?

how do you calibrate the value of R1, R2 etc? how do you reconcile this with the fact that indices trade with a certain constant level of recovery?

<r>Primo, I still think that the armageddon scenario shouldn't be excluded a-priori. In this respect it's interesting what Morini and Brigo say in section 5 of <URL url="http://www.damianobrigo.it/creditindexoptions.pdf">http://www.damianobrigo.it/creditindexoptions.pdf</URL> precisely about the cor...

for the probability matching method there will also be a difference between stochastic and non-stochastic recovery

I think the only sensible choice for $\tilde{R}$ is zero, otherwise you will never be able to reach 100% loss in the pool, or in other words there will always be a super-super-duper tranche that will be shielded from loss and therefore have zero value. Any other thoughts on this?

well numerically it will make things more stable, although it will increase your cpu time.the good idea in the Amraoui-Hitier is to impose conditional expectation of recovery equal to the recovery that is normally used in the index

<t>what I mean is that stochastic recovery changes the shape of the loss distribution, so if you use deterministic recovery you won't get the same expected tranche loss, even if the value of the deterministic recovery is equal to the expected recovery.so using expected recovery works for the ATM met...