Hi, I have a question regarding the calculation of swap DV01s under the post-crisis multicurve framework. Let say a swap uses Curve A for projection and Curve B for discounting, but Curve A is built from Curve C (eg Curve C may be an OIS curve). When we calculate the DV01 of this swap, do we simply bump Curves A and B and with the bumped CUrve A built from original Curve C?Thanks for your help on this, much appreciated.

QuoteOriginally posted by: quanteric[br]I have a question regarding the calculation of swap DV01s under the post-crisis multicurve framework. Let say a swap uses Curve A for projection and Curve B for discounting, but Curve A is built from Curve C (eg Curve C may be an OIS curve). When we calculate the DV01 of this swap, do we simply bump Curves A and B and with the bumped CUrve A built from original Curve C?The multi-curve framework existed before the crisis, just that is was not used so widely. But I guess this is not the point of your question.It is not clear for me what is the set-up your are describing. If C is OIS and you build your projection curve A zero-coupon with C as discounting curve, there is only one way to use A: projection with A and discounting with C. Using curve B as discounting is incoherent (except if B=C). Can you give an example of A, B and C?Then you have to clarify what you mean exactly by DV01. Different people have different definition of DV01/PV01.

Hi, sorry for be being vague. I am trying to match Bloomberg.I have a tenor basis swap GBP 5y starting 30/01/2012 3m vs 6m.So I have the standard GBP curve (Curve 22) for discounting and projecting the 6m leg and for the 3m leg, project using the 3m curve (Curve 222) and discounting using the standard curve. I am able to match Bloomberg if i switch off OIS DC Stripping. By DV01, they mean (Vd-Vu)/20 where Vd is the value with the curves bumped up by 10 bps and Vd bumped down by 10bps. Now If I turn on OIS DC Stripping, both Curve 22 and 222 will be bult using the discount factor from the GBP OIS curve (Curve 141). I have tried bumping and not bumping Curve 141 and still could not quite match Bloomberg exactly, though am quite close......

QuoteOriginally posted by: quanteric I am trying to match Bloomberg.I can not guarantee that this is the best thing to do. I don't know how Bloomberg is doing it and usually it is not easy to obtain from them very detailed information on there modeling methods. Their curve constructions don't have the reputation to be the cleanest in the world.QuoteBy DV01, they mean (Vd-Vu)/20 where Vd is the value with the curves bumped up by 10 bps and Vd bumped down by 10bps. I don't know why it is done with a finite difference so large bump. There are more stable methods to obtain differentiation that the "bump and recompute". In particular well implemented Algorithmic Differentiation is in general more stable and faster. In the special simple case you mentioned this is probably not very important, but I'm still surprise by this approach.QuoteI am able to match Bloomberg if i switch off OIS DC Stripping. Now If I turn on OIS DC Stripping, both Curve 22 and 222 will be bult using the discount factor from the GBP OIS curve (Curve 141). I have tried bumping and not bumping Curve 141 and still could not quite match Bloomberg exactly, though am quite close......Again I can not comment on the Bloomberg case as I'm not using Bloomberg curves. From my experience a lot of systems that claim to handle multi-curves (including known names) do it relatively superficially. There is often limitation on how you can construct curves (usually even if it is a multi-curve framework, you are restricted to construct the curves one by one) and the delta/PV01/curve sensitivity does not include all the relevant movements. One problem often encountered is that the sensitivity to the discounting market rates (OIS) comes directly in the discounting curve construction and indirectly as the discounting curve is used to construct the forward curve. This indirect contribution is missing in some places.

Depends on your calculation methodology. I would personally bump Curve C, as there is exposure to Curve C rates. By bumping curve C, you will effectively change forward rates projected from curve A. You can then bump curve A separately to compute DV01 on curve A.In this case, we have the following:dNPV(A, B, C) = dNPV/dA + dNPV/dB + dNPV/dA * dA/dCwhere the notation is hopefully obvious.This will get a bit more involved in the gamma calculation.However an alternative methodology where you do not bump curve C is not completely meaningless either. In that case, you estimate the discounting risk of your trade on the projected flows provided that these flows do not change. You are missing out on the last term above, with unpleasant consequences such as not being able to explain day-to-day P&L variation accurately, but well, it's up to you.

Are you referring SWPM when you say Bloomberg?QuoteI am trying to match Bloomberg.I don't think Bloomberg use OIS discounting when deriving estimation curve like AUD 3M, from what I see in "ICVS", but I might be wrong.