wilmott.com

Dear All,I am looking for help with the following:I wrote a simple function that discounts all the flows from a given bond using just a bond's payment schedule and 6 parameters of zero coupon curve (nelson siegel form - with a humb). Now I usually "fit" the function's results of series of bonds with empirical values from the market by changing the parameters of NS zero curve to minimize the differences between the model and market. It also takes 20-30 seconds to produce best fit answers. I know that using SOLVER for such a calculation is a very primitive and inaccurate solution. Does anybody have any expirience with solving that problem in a clever way using matrices or some smart add-in ?Thank you in advanceMD-------------------------Marcin Dec

QuoteOriginally posted by: mdecDear All,I am looking for help with the following:I wrote a simple function that discounts all the flows from a given bond using just a bond's payment schedule and 6 parameters of zero coupon curve (nelson siegel form - with a humb). Now I usually "fit" the function's results of series of bonds with empirical values from the market by changing the parameters of NS zero curve to minimize the differences between the model and market. It also takes 20-30 seconds to produce best fit answers. I know that using SOLVER for such a calculation is a very primitive and inaccurate solution. Does anybody have any expirience with solving that problem in a clever way using matrices or some smart add-in ?Thank you in advanceMD-------------------------Marcin DecI don't understand...why don't you bootstrap to get the zero curve.

I agree with Nonius, why not just bootstrap? What is your objective in building the zero curve – relative value analysis or actual pricing and mark to market? If for pricing, be careful using fitted models because while they result in smooth curves they are generally not arbitrage free (that is, they do not identically reproduce the set of input bond prices). Unless you are valuing derivatives whose payoffs are a function of short-maturity forward rates, why would you care if the curve is not smooth? If you are using the curve for rich/cheap analysis then a smooth curve is more of a visual aid then anything else. Bootstrapping is pretty robust and arbitrage free.

Dear Nonius and DavidJN,The purpose for that calculation is neither pricing nor relative value analysis. I want to know the modelled NS curve for term structure analysis, market expectations regarding short term interest rates in the future. Of course I do bootstraping for the pricing and RVA. But the problem is that it is hard to get smoothed results in Polish, Hungarian and Czech bonds. Only few sectors of the curve are very liquid, so the curve lacks updated prices in some sectors. Sometimes very short forward (1week- 4 weeks) is negative. Therefore I tend to model zero curve using the belowmentioned methodology (despite I am aware that it will produce just theorethical picutre of the market). Anybody there uses excel to fit the NS curve?MD

QuoteOriginally posted by: mdecDear Nonius and DavidJN,The purpose for that calculation is neither pricing nor relative value analysis. I want to know the modelled NS curve for term structure analysis, market expectations regarding short term interest rates in the future. Of course I do bootstraping for the pricing and RVA. But the problem is that it is hard to get smoothed results in Polish, Hungarian and Czech bonds. Only few sectors of the curve are very liquid, so the curve lacks updated prices in some sectors. Sometimes very short forward (1week- 4 weeks) is negative. Therefore I tend to model zero curve using the belowmentioned methodology (despite I am aware that it will produce just theorethical picutre of the market). Anybody there uses excel to fit the NS curve?MDhmmm...I see the problem now...sort of like a similar problem in stripping credit curves for illiquid names...so, lemme see...one day you've got 2, 5, and 10 mat bond prices. you interpolate then bootstrap.then, the next day, shit, you only have a 2 and 10 mat bond price...so if you interpolate and bootstrap, the interpolated 5 year adds fictitious rate volatility, if you are trying to analyse curve dynamics...I'm I getting the drift here?

Hi mdec!There is a thread Nelson-Siegel to model the yield curve?. But I'm affraid that every attached worksheet uses SOLVER - but anyway, take a look Best Regards,MaTT