Hi All,Industry practices typically apply the bump-and-revalue approach to obtain curve sensitivities for instruments. The bumping can be applied to either market quotes or zero rates/forward rates. In the latter case, Jacobian manipulation is required to transformed the risks back to the risks w.r.t. market quotes. If one applies the approach that bumps market quotes, many rounds of curve recalibration is required. It is much more so in a multi-curve framework.My question is, if one can find a way to calculate curve risks for linear IR instruments analytically without any curve recalibration, would that be considered as practically valuable?Thanks,Tw813

Maybe, although that sounds oddly like something Miron and Swannell did about 25 years ago, although admittedly in a single-curve setting.

Yes, that does sound familiar. Miron and Swannell called it the delta vector in their 1990(?) book Pricing and Hedging Swaps.

Hi, I took a look at Miron and Swannell's approach (it is a surprisingly expensive book!). Their approach relies on quite specialized assumptions.1. It requires one to first interpolate input swap rates to obtain all missing swap rates between curve pillars, e.g. by a simple linear interpolation of the input rates.2. It is only directly applicable to curves constructed with a local scheme (a scheme under which a non-pillar discount factor depends only on the two neighboring pillar discount factors). For example, the method cannot be directly used for a cubic spline curve.3. For every scheme one wants to apply the approach, one has to derive a different formula separately. While i do believe it is possible that the approach can be applied to a cubic spline curve if one is willing to works out some heavily messy algebra, this has to be done for every non local scheme one wants to apply. I think the approach is quite intuitive and simple to understand. However, my view is the approach is not general and flexible enough. I believe my approach will be much simpler and it works for all interpolation schemes one wants to put on. I shall put something on SSRN soon. Hopefully can get some feedback from you experienced people.thanks,tw813

Welcome to the real world.

Don't bump the curve inputs -- shift the curve itself. See this thread: http://www.wilmott.com/messageview.cfm? ... SGDBTABLE=

QuoteOriginally posted by: tw813If one applies the approach that bumps market quotes, many rounds of curve recalibration is required. It is much more so in a multi-curve framework.Don't do bump and recompute for the curves inputs (or any other input). Algorithmic Differentiation is there to help you, both in term of performance and precision. This is very generic, applied to linear and non-linear instruments, to curve, volatility surfaces, etc. I have implemented it in our libraries, and would not work in a multi-curve framework without it anymore.

Bump and reval will work wrt underlyings however it will be slow, go with the Algorithmic differentiation approach it is faster for bigger portfoliosIn addition dont worry about bumping zero rates, you should only look at bumping market quotes or forward ratesAnalytic approximation may work, you will have test performance of hedges to see how it pans out