On Wed, 2009-03-25 at 10:31 +0000, [hidden email] wrote:
> Revision: 16065 > http://quantlib.svn.sourceforge.net/quantlib/?rev=16065&view=rev > Author: nando > Date: 2009-03-25 10:31:08 +0000 (Wed, 25 Mar 2009) > > Log Message: > ----------- > added flat extrapolation Care to discuss? Luigi -- The shortest way to do many things is to do only one thing at once. -- Samuel Smiles ------------------------------------------------------------------------------ Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are powering Web 2.0 with engaging, cross-platform capabilities. Quickly and easily build your RIAs with Flex Builder, the Eclipse(TM)based development software that enables intelligent coding and step-through debugging. Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com _______________________________________________ QuantLib-dev mailing list [hidden email] https://lists.sourceforge.net/lists/listinfo/quantlib-dev |
On Wed, Mar 25, 2009 at 11:38 AM, Luigi Ballabio
<[hidden email]> wrote: > On Wed, 2009-03-25 at 10:31 +0000, [hidden email] wrote: >> Revision: 16065 >> >> Log Message: >> ----------- >> added flat extrapolation >> >> Modified Paths: >> -------------- >> trunk/QuantLib/ql/termstructures/yield/forwardcurve.hpp > > Care to discuss? glad to oblige. The InterpolatedForwardCurve used the interpolant to extrapolate, which easily leads to implausible rates, unless one used piecewise-constant interpolation. Even "basic" linear interpolation could lead to negative rates or implausible high rates depending on its slope in the final segment Flat forward rate extrapolation keeps forward rate continuous, at the cost of an angular point for non piecewise-constant interpolations. This is a non-issue for linear interpolation since in this case the curve already has an angular point at every knot. When it comes to higher order interpolation it is still a bearable cost compared to implausible extrapolation values. In any case this cost is negligible in the overall smoothness evaluation of the curve, as the forward rate curve is integrated in order to obtain zero rate (and then discount factor) Of course extrapolation is always a dangerous exercise, but if one needs it then flat extrapolation of instantaneous forward rates is usually the preferred approach, with no real feasible alternatives I know of. This is also the approach adopted in the InterpolatedHazardRateCurve. Any contrarian out there? ciao -- Nando ------------------------------------------------------------------------------ Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are powering Web 2.0 with engaging, cross-platform capabilities. Quickly and easily build your RIAs with Flex Builder, the Eclipse(TM)based development software that enables intelligent coding and step-through debugging. Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com _______________________________________________ QuantLib-dev mailing list [hidden email] https://lists.sourceforge.net/lists/listinfo/quantlib-dev |
On Wed, 2009-03-25 at 12:23 +0100, Ferdinando Ametrano wrote:
> glad to oblige. Thanks. I agreed with the change, but I though it would have been a good thing to have its rationale explained---which you did nicely. Luigi -- I'd never join any club that would have the likes of me as a member. -- Groucho Marx ------------------------------------------------------------------------------ Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are powering Web 2.0 with engaging, cross-platform capabilities. Quickly and easily build your RIAs with Flex Builder, the Eclipse(TM)based development software that enables intelligent coding and step-through debugging. Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com _______________________________________________ QuantLib-dev mailing list [hidden email] https://lists.sourceforge.net/lists/listinfo/quantlib-dev |
On Wed, Mar 25, 2009 at 12:33 PM, Luigi Ballabio
<[hidden email]> wrote: > I agreed with the change, but I though it would have been a > good thing to have its rationale explained which now prompts me to tackle a possible next step. While I thought that flat fwd extrapolation was uncontroversial when modelling fwd rates (InterpolatedForwardCurve), it might become slightly controversial when applied to discount (InterpolatedDiscountCurve) and zero (InterpolatedZeroCurve). In the current situation for InterpolatedDiscountCurve we are extrapolating discounts, which is not sensible and might lead to negative and/or increasing discounts Any flat discount extrapolation would simply make no sense at all as it would imply null fwd rates. As for InterpolatedZeroCurve we are now extrapolating zero, which again is not sensible and might lead to negative rates. In this case flat zero extrapolation might look tempting, but it is actually equivalent to extrapolate flat fwd rates equal to their average value over the interpolation interval: a quite bizzarre assumption which also introduce a jump in fwd rates. So I propose to extrapolate flat fwd rates and this could be quite easily implemented in the InterpolatedXXXCurve classes. Any opinion? ciao -- Nando ------------------------------------------------------------------------------ Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are powering Web 2.0 with engaging, cross-platform capabilities. Quickly and easily build your RIAs with Flex Builder, the Eclipse(TM)based development software that enables intelligent coding and step-through debugging. Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com _______________________________________________ QuantLib-dev mailing list [hidden email] https://lists.sourceforge.net/lists/listinfo/quantlib-dev |
On Wed, Mar 25, 2009 at 1:36 PM, Ferdinando Ametrano <[hidden email]> wrote:
> So I propose to extrapolate flat fwd rates and this could be quite > easily implemented in the InterpolatedXXXCurve classes. I just did it in Rev16082 http://quantlib.svn.sourceforge.net/quantlib/?rev=16082&view=rev ciao -- Nando ------------------------------------------------------------------------------ _______________________________________________ QuantLib-dev mailing list [hidden email] https://lists.sourceforge.net/lists/listinfo/quantlib-dev |
Free forum by Nabble | Edit this page |