I need to know whether a tranche will be long or short correlation.Lets say I don't know the specifics of the attachment, A, and detachment, D. But I do know every thing else like pool decomposition, maturity etc...Then for any tranche on this pool with detachment below K, or attachment above K I can say................Long Credit/Sell Protec.........Short Credit/Long ProtecD<K:........LONG CORRELATION.............SHORT CORRELATIONA>K:........SHORT CORRELATION............LONG CORRELATIONwhere K := "correlation inflection point".The task is to calculate the point of inflection.My method is to calculate the impact, dPV(rho), of bumping correlation, rho, on a series of tranchelets [0%,1%],[1%,2%],...,[99%,100%]The idea being the point at which dPV changes sign which gives me and indication of the point of inflection.I would appreciate suggestions on whether this is a valid approach?Perhaps somebody has had to do something similar?Cheers, Sam.

I thought the concept of being long or short correlation on tranches went out of the window back in '05 along with t-Corr models?

Maybe that is the case. But it is still need to solve this problem. Surely if you are using any gaussian copula type model then the concept of long or short correlation applies?Unless you are saying the market doesn't use these type of models?

Of course the market uses copula models...

In that case katastrofa do you think the approach in my original post is a valid method to find out the point of inflection?Cheers, Sam

The problem with this approach is that you can't hedge on correlation across the capital structure. Plenty of people got taken out by a correlation neutral equity/mezz position when the autos got downgraded in '05. After that people moved on to bCorr copula models and started hedging the entire cap structure rather than a single correlation number.

StructCred,I was actuall talking about bumping base correlation rather than compound correlation.Where I will see a point where dPV of the tranchelet changes sign. Going back to my original post this will represent the point of inflection.Cheers, Sam.

But what is the significance of determiniting the sign if you aren't going to hedge correlation across the term structure?

StructCred,I think the idea here is to try an understand PV movements at a book level.Lets say we don't know the exact decomposition of trades in the book but we do have their deltas, notionals, subordination. The deltas (sensitivity to bumping the underlying single names) imply we know if the trades are long/short protection.Then if we know the point of inflection, and perhaps also say how it has change over time then we can get a better understanding, of which of the trades should be long/short correlation. And therefore a better take on the PV movements of the book.Cheers, Sam

What you're saying implies that correlation sensitivity across capital structure is fungible. This is exactly an idea that cleaned up the market after autos downgrade. People were long correlation by going long equity risk and short correlation by shorting the mezz. They hedged out their deltas and their corr01. When the autos got downgraded, the event was perceived as being very idiosyncratic. The value got pushed into an equity tranche and the corrs on equity and mezz moved in the opposite direction, completely hosing corr hedgers. Trades that were long correlation were losing money (through correlation), trades that were short correlation were also losing money (also through correlation). What makes more sense, is figuring out your risk to bumping each tradable detach point and making sure you're flat that risk (or taking a view on corr for each detachment point). I'd suggest steering clear of assuming parallel shifts in the corr skew.

Hey StructCred - You sound good and as far as studying trades wrt Book Level is concerned the only feasible way is to get roughly the risk sensitivities of the book wrt the tradable parameter be it IR , FX , CDS Spreads etc.But still technically spreaking the question and methodology that Samkhalilian is using just to understand which of the tranchelets are Long/Short Correlation ( though correlation hedging is diffcult these days seeing the volums on FTDs) seems OK to me.I suppose the change in the sign of dPV by only bumping Rho , keeping every parameter constant , seems correct.Am I missing any of the roles of Input parameters in tranche pricing?Cheers

For the gaussian copula the tranchelet that is correlation neutral and the maximal wizou (drho dS), or with maximal correlation position and minimal wizou, are around expected loss when correl is near 0, but go opposite ways when rho goes to 1. For LHP portfolio,K=(1-R)Phi((1-rho)^{+- 1/2} Phi^{-1}(EL)) your method looks reasonnable, note there is no guarantee of a single inflexion point when there is correl skew with values going near 1,so you could see some high mezz going long correl and next one short again.

Hey Sebastian,Indeed for a smaller of bump of the base curve I observe the possiblility for high mezz going long correl and then short again. But using a large bump seems to avoid this. Whether the results are that meaningful though?As a side point how would go about calibrating the tail of the loss distribution in your random recovery model. If you are just pricing up standard index tranches then I guess you don't care what the tail looks, but for bespokes tranches it is a different matter? For example say you want to price up a 60-100 tranche then you don't want to much losses up top, so stripping recovery to 0 maybe to extreme.Cheers, Sam.