Thank you Yossarian;I want to make sure that I got this base correlation approach correct. I read the the document "A model for base correlation" that explains how the calculations were carried out by spread sheet.For each tranche, by the spread quoted in the market , a cumulative default probability is calculated for each tranche= PD(horizon)Then by using gaussian factor setting and spread for m(market factor proxy), an expected loss is calculated as a fuction of rho (base correlation), and this expected tranche loss gives expected survival, which is then translated into a "survival rate" for a given tranche, again as a function of rho. Then by equating this "survival rate" to PD(horizon) which is defautl probability, we get the base correlation.My question is, for the first base tranche 0-3 it is easy and we will get a base correlation. Then to get the base correlation for the second base tranche 0-6, we have spread for 3-6, the rho for 0-3, and expected loss for 0-3. We need to calculate expected loss for 3-6 and sum with 0-3 loss to get, expected loss for 0-6. I am confused about carrying out the steps following this, after getting rho1.Could someone explain how we process from this point? Do I get PD and use it as a survival rate to get expected loss for 3-6 (backward). And then sum with the previous expected loss to get loss for 0-6. Then how do I get rho?I would appreciate if someone guide through the calculations like "using these values you get this and substituting that gives you this etc."Thank you guys.

Hi to all!just a quick question. I have historical data for Itraxx (mid spread). is there anyone that knows where I could find the data for the tranches? I am looking for the spreads. Bloomy or reuters' ric should be perfect.tks a lot!

Hi all,I heard some people are quoting nowadays in base-correlation and I was wondering: does anyone use the ATM Correlation concept described in the JPM paper 'A Relative Value Framework for Credit Correlation'? In particular for 'calibrating' a custom portfolio into an index is this the market standard method?Regards,SurferD

Some dealers reject the idea, but increasingly it seems more banks are using it regularly for bespoke STCDOs. The skew observed from CDX seems to be preferred over iTRAXX skew as the "market". However it is not clear that mapping the loss from the correlation skew is done on CDO^2. Curious to find that out.

QuoteOriginally posted by: LuckymanHi to all!just a quick question. I have historical data for Itraxx (mid spread). is there anyone that knows where I could find the data for the tranches? I am looking for the spreads. Bloomy or reuters' ric should be perfect.tks a lot!Not sure whether you can get historical iTraxx tranche quotes from Bloomberg. But in case you're also interested in dj cdx.na.ig, you can easily get the historical MS indications from Bloomberg. Their index names are: MSCX03 Index, MSCX37, etc. for the standard tranches.

QuoteOriginally posted by: HelenSome dealers reject the idea, but increasingly it seems more banks are using it regularly for bespoke STCDOs. The skew observed from CDX seems to be preferred over iTRAXX skew as the "market". However it is not clear that mapping the loss from the correlation skew is done on CDO^2. Curious to find that out.You cannot use base correlation to value CDO^2. You need a model, which can explain the smile. A number of houses use variations of the random factor loading model.

CreditFlux have tranche prices (and a tranche implied correlations for what it's worth)http://www.creditflux.com/sz/data/impcorr.htm

Thanks. Too bad that I don't have the subscription for CreditFlux. Any chance that you can post the file?Complexity: multi-factor (random or not) correlation models are all well intended but seriously lack the tractability thus hard to trade. What will become the "norm" for CDO or CDO^2 - correlation skew ( and to go through the trouble of calibration) or the "real model" with factors that everyone agrees ?

Helen,you don't need multiple factors to match the correlation skew using a random factor loading model. Calibration is not too difficult using constraint optimization algorithms. The parameters should be relatively stable, once calibrated. Alternatively, Variance Gamma type models ala Joshi & Stacy may be a solution.

Thanks a lot for the preso, looks really interesting! Which one of the random factor loading models do you like most? Regards,SurferD

Is there more than one random factor model?I've only heard of the Andersen and Sidenius modelAndersen & SideniusCould you give references for the other ones?

complexity: in the Joshi&Stacey pdf you posted, the authors state that the Gaussian copula model is not time homogeneous.does that basically mean that if you calibrate it for 5 years it is unlikely to produce sensible 7 year CDO tranche prices (with the sme model parameters)?and that "stochastic time" or "intensity Gamma" models in general are more likely to "work" over a wider maturity horizon?

erstwhile: you're right. calibrating a copula model to one maturity will not give you the right results for another. the problem is that the forward loss distribution cannot be modelled correctly with a gaussian copula model.

fodao: this is the only one I know as well - sorry for the confusion. I was more thinking in terms of specification. Regards, SurferD

I can't find the Joshi & Stacey paper you mentioned. Could you please post it again? Sounds very interesting.