Thanks - btw: is there any difference between the 5y and the 10y base curve?Regards,SurferD

We'll be covering this in the next research note. Suffice to say that the ability to compare 5 vs 10 year is _another_ big advantage of using Base Correlations.RegardsLee

that's fantastic! Can you post it when it will be available?Thanks,SurferD

Am I right that, according to the Base Correlation idea, the upside is the ability to match implied correlation across tranches, but the associated downside is the need to use a simplied tranche pricing model (Homogeneous Large pool model).In particular, I notice the remark "the model cannot differentiate between a single name widening by 10,000bp and100 names widening by 100bp""We believe because of the infrequent occurrence of big single name moves that thislimitation is more than made up for by its simplicity and transparency."Isnt the whole point of tranches to capture single name blowouts? Isnt this a severe kind of approximation?

Nope. There are two separate sections in our framework. The use of the simple Large Pool Model and Base Correlations separately.It is perfectly possible to use Base Correlations with any other model.The use of the Large Pool Model is driven by practicalities of the market, and a desire to standardise.It is a fact of life that it is not possible to input 100 CDS spreads in an accurate and timely fashion to quote correlations, if they could even be agreed upon - no exchanges here.RegardsLee

Ah yes I see now- apologies for the misunderstanding. Basically you have different implied correlations for the lower and upper attachment points, right. Cool idea. Why didnt i think of it?!By the way, is it really that hard to deal with 100 underlying spreads? you just copy them over from a spreadsheet, press shift-F9, and computation shouldnt take more than a minute! Or is the point that its not practical for a dealer to show all his underlying spreads corresponding to each quote, say in his bloomberg messages?

You should really use curves for each obligor in the portfolio as equity tranches are sensitive to individual issuer curve shape (e.g. idiosyncratic factors rather than systemic factors) & not just assume flat cost of protection across all maturities.Lee - - I have locked into this "base correelation" thing and I have a question. Do you calibrate to a separate bid and offer or just assume mid is half way between bid and offer?Unless of your you have trade information for all tranches at ropughly a similar time...RegardsMark

Some excellent questions here - I'm glad I joined!Bid/offers are very interesting. What we have done, for now is to use the (tranche and traded index) mid spread, and calculate the mid Base Correlations from that.There are some possible extensions. Calculate Bid Correlations from Bid spreads (not quite an arbitrage scenario, but equivalent value), and ask correlations from ask spreads.Alternatively, it would be possible to build an arbitrage position where long risk 0-3 + long risk 3-6 + short risk 0-6 must be arbitrage free. The problem with this is that the bid/offers tend to get very wide very quickly. An improvement may be to come from both ends (0-x and x-100) at the same time.One interesting point is that the Base Correlation methodology fails if the total losses in the tranches so far is greater than those in the portfolio. (To me this is like Black Scholes failing to calculate an implied vol if option price is less than intrinsic value). This tends to happen at current bid/offer spreads for the most senior tranches if you use offers only...And yes, to emphasise mrowell's points, you do need to use a full curve ideally, which aren't published, but even if you took only 100 single name 5 year points (and ignored different recovery rates for now) you would still struggle.The names in these tranches indices are the most liquid and most traded in the market, but the bid offers are still around 5bp on each name. (We often disagree with other dealers in our calculation of the "basis to theoretical" in the unleveraged indices, which is due to different dealers not agreeing on the single name spreads at an instant.)Moreover, single name sensitivities only matter most when there is a single name blow up, and at that time price discovery is very difficult. We have a good history of this as Parmalat moved to its Credit Event last year - typical bid offer spreads were 5-10 points (CDS was traded upfront only). But as I said before, there is no right way to do this. I hope we've come up with a usable and useful methodology, but we are very happy to keep the discussion going, and adopt whatever becomes the consensus.RegardsLee

I have one more question: do you expect the base curve of the EU indices and US indices to look the same? If not what makes them look different, higher/lower diversification? Higher/lower spreads?Thanks again for the help,SurferD

Hi LeeAnother question from me also - I hope you were prepared for this bombardment of questions!According to your paper, base correlations are backed out from expected losses of tranches, stating that they are traded. In other words, upfront prices are quoted rather than running coupons.is this really the case? Looking at Trac-X, I can see prices for running coupons. These would have to be converted to upfront prices for your methodology, using some effective duration, which would need some assumption about underlying correlation. As I understand it, if you deal with running coupons, with separate fixed and floating legs, you dont have enough information to bootstrap out the values of the legs for the different tranches. Is this right? Thanks

No, you have enough information from the running spreads to get the expected losses. For an individual tranche it is not a function of correlation at all.In other words, correlation gives you the expected loss distribution, that is the expected loss per tranche. And then those expected losses give you the spreads. So it's pretty easy to go backwards.I would imagine this will become a lot easier to understand once I can send the model out.RegardsLee

On the different markets, you can certainly compare Base Correlations across regions if you do an appropriate transformation - and the interesting conclusion is that current correlations in Japan, Europe and US are pretty much exactly the same.This is coming in the next research note.RegardsLee

hi leehmm maybe im being obtuse about this.. but.. isnt the time dimension missing from your argument?as i understand it, for a running coupon, you dont need one loss distribution, but a whole set, ie. one for each payment date.If you have say a 5 year tranche, then its running coupon is determined by 5 expected tranche losses, one for each year, say, E(L_i). The fixed leg is s \sum (N- E(L_i)), where N is the tranche notional, and s is the running coupon (lets put interest and recovery rates to zero). The floating leg is E(L_5) (expected tranche loss after 5 years).Putting PV to zero, you haves = E(L_5)/\sum(N - E(L_i))so from a single tranche spread s, its not possible to back out all 5 values of the expected tranche loss, E(L_i), for each year i.On other hand , you could if you have a whole curve of running coupons for the tranche against maturity.Is this right?Thanks

I just realised it does work if you put the interest rate to zero, since \sum E(L_i) = E(L_5). Does your model rely on interest rates being zero then?

Thanks a lot for your quick answer. Just to understand: what kind of transformation do you propose in order to be able to compare say the EU base curve with the US base curve? And: are the current actual base curves for EU, US and JP the same or are they the same only after you transform them?Thanks again for your help,SurferD