Re: [Quantlib-users] Svensson fitting

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

Hello Frederick,
    unfortunately the optimizer can impose constraints on the parameters,
not the fitted quotes.  I would do as you already did, that is, modify
FittingCost::values() to add a penalty when the quotes are outside the
range.  I'm not sure why that would give you a worse curve, but optimizers
are hard to reason about.  If I had to guess, it might be that the
unconstrained optimization goes through the forbidden range before going to
the minimum, and by adding a penalty you're blocking that path.  But you
would need to print out the values during optimization to be sure of what's
happening.

Sorry this wasn't a lot of help,

    Luigi


On Fri, May 15, 2020 at 2:25 PM Frederick Delaroche <
fre...@ri...> wrote:

> Hi all,
>
>
>
> I am using FittingBondDiscountCurve, Svensson fitting method. Fitting is
> working nicely; however, I wanted to add a specific control over change in
> quotes (i.e. no constraint on Svensson parameters) to keep the fitted quote
> within a given range (e.g. a bid / ask).
>
>
>
>    - *Is there any way to do this with QuantLib?*
>
>
>
> I have tried a basic solution (modifying the FittingCost::values function
> by changing the error when error itself was above (or below) a given
> value), but it doesn’t work as expected: depending on the limit, I don’t
> find the optimal values for the fitted quotes.
>
>
>
> “Not find the optimal values” means:
>
>    - If I don’t have any limit for quotes move, I have a set of fitted
>    quotes. If I check the max error compared to orginal quotes and if I set
>    this move as the limit, I will find a different set of fitted quotes which
>    give a “worse” discount curve.
>
>
>
> Thanks for your help.
>
>
>
> Rgds,
>
>    -
>
> _______________________________________________
> QuantLib-users mailing list
> Qua...@li...
> https://lists.sourceforge.net/lists/listinfo/quantlib-users
>