Re: [Quantlib-users] Nelson Siegel Calibration - Freeze exponential decay?

[Luigi B. <lui...@gm...>]

You can choose the calibration method (see
https://github.com/lballabio/QuantLib/blob/v1.36/ql/termstructures/yield/nonlinearfittingmethods.hpp#L99)
but if none is given, it uses the Simplex method.

Luigi


On Fri, Dec 6, 2024 at 3:35 PM philippe hatstadt <pha...@ma...> wrote:

> Thanks Luigi. Do you know what calibration method is used, as I find that
> it has a hard time finding a good global calibration. Furrhermore, the
> solution is unstable to minor changes in inputs if I don’t pass an initial
> guess.
>
> Regards
>
> Philippe Hatstadt
> +1-203-252-0408
>
>
> On Dec 6, 2024, at 7:08 AM, Luigi Ballabio <lui...@gm...>
> wrote:
>
> 
> Hello Philippe,
>     no way to fix a single parameter, I'm afraid.  If you have a set of
> calibrated parameters and want to use them without performing the
> calibration, you can pass maxEvaluations=0 to achieve that but you'll still
> have to pass at least one helper in order to establish a max date.
> Starting from next release, there will be another constructor available
> that takes no helpers and an explicit max date and does the same.
>
> Hope this helps,
>     Luigi
>
>
> On Mon, Dec 2, 2024 at 4:15 PM Philippe Hatstadt via QuantLib-users <
> qua...@li...> wrote:
>
>> Is there a way to fix one or more parameters when calibrating NS from the
>> python API?
>> Specifically, I would like to fix the exponential decay lambda parameter.
>>
>> After building instrument helpers, I use this call:
>> ql.FittedBondDiscountCurve(*params, ql.NelsonSiegelFitting(),, accuracy,
>> numIter, guess, simplex_lambda)
>>
>> I looked at the c++ code, and it seems that there is a array parameter
>> that represents weights, but I didn't see anything related to freezing one
>> parameter.
>>
>> Also, I was looking for a way to just build a NS curve by passing all
>> parameters, and not perform any calibration, but I didn't see one. It's
>> easy to implement, but wondering as the solution might be related to my
>> first question?
>>
>>
>>
>> Best regards Philippe Hatstadt 203-252-0408
>> https://www.linkedin.com/in/philippe-hatstadt/
>>
>>
>> _______________________________________________
>> QuantLib-users mailing list
>> Qua...@li...
>> https://lists.sourceforge.net/lists/listinfo/quantlib-users
>>
>