Re: [Quantlib-users] Source bumping calculation time

[Aleksis A. R. <ale...@go...>]

Hi Peter. If I don't recompute and recalibrate the basket after each bump in the swaption vol cube, then I see no sensitivity impact at all when recomputing the bumped NPVs (so the bump risk returns a zero grid).

I have resorted to using 16 points/3.0 std devs in the integration scheme - this has had made the computation bearable without too much loss of accuracy. 

Thanks, Aleksis

> On 18 Jan 2021, at 16:57, Peter Caspers <pca...@gm...> wrote:
> 
> Hi Aleksis,
> 
> computing the calibration basket via 'MaturityStrikeByDeltaGamma' is
> quite time consuming, so I'd try to compute the basket only once
> (under the base scenario) and reuse it for each sensitivity bump
> scenario.
> 
> Thanks
> Peter
> 
> On Sun, 17 Jan 2021 at 13:52, Aleksis Ali Raza via QuantLib-users
> <qua...@li...> wrote:
>> 
>> Hi, a question on optimizing calculation time: the code I’ve written for bermudan swaption valuation in python (see below) takes forever when I run risks using source bumping (bucketed ATM and skew swaption vega are the really killers). I assume it’s the recalibration step I have added for the NPV attribute that’s the issue but I can’t figure a way around that. Any optimization tips would be appreciated (eg. different parameters for a less time-consuming integration scheme??).
>> 
>> Thanks, Aleksis
>> 
>> 
>> class bermudanswaption():
>> 648
>> 649     def __init__(self, calendar,settlement, used_model, swap, ratecurves, index, swvolcube_clean, swapbase,
>> 650                  mean_reversion,position,NC_periods):
>> 651
>> 652         discount_curve = ratecurves.loc['discountcurve', 'ratecurves']
>> 653         self.swvolcube = swvolcube_clean
>> 654         self.swapbase = swapbase
>> 655         self.used_model = used_model
>> 656         self.discount_curve = discount_curve
>> 657         self.position=position
>> 658
>> 659         fixed_schedule=swap.fixedSchedule()
>> 660         exerciseDates = [calendar.advance(i, -ql.Period('2D')) for i in fixed_schedule][1+NC_periods:-1]
>> 661         exercise = ql.BermudanExercise(exerciseDates)
>> 662         stepDates = exerciseDates
>> 663         self.exerciseDates=exerciseDates
>> 664         sigmas = [ql.QuoteHandle(ql.SimpleQuote(0.01))]*(1+len(exerciseDates))
>> 665         self.used_model = used_model
>> 666
>> 667         if settlement == 'physical':
>> 668             type = 0
>> 669             method = 1
>> 670         else:
>> 671             type = 1
>> 672             method = 2
>> 673
>> 674         self.nsswaption = ql.NonstandardSwaption(swap, exercise, type, method)
>> 675         gsr = ql.Gsr(ratecurves.loc[index, 'ratecurves'],
>> 676                      stepDates, sigmas, [ql.QuoteHandle(ql.SimpleQuote(mean_reversion))])
>> 677         engine = ql.Gaussian1dNonstandardSwaptionEngine(gsr, 64, 7.0, True, False,
>> 678                                                         ql.QuoteHandle(ql.SimpleQuote(0)),
>> 679                                                         discount_curve, 2)
>> 680         self.engine = ql.Gaussian1dSwaptionEngine(gsr, 64, 7.0, True, False, discount_curve)
>> 681         self.nsswaption.setPricingEngine(engine)
>> 682         self.model = gsr
>> 683     def NPV(self):
>> 684         engine = ql.Gaussian1dSwaptionEngine(self.model, 64, 7.0, True, False, self.discount_curve)
>> 685         basket = self.nsswaption.calibrationBasket(self.swapbase, self.swvolcube, 'MaturityStrikeByDeltaGamma')
>> 686         for basket_i in basket:
>> 687             ql.as_black_helper(basket_i).setPricingEngine(engine)
>> 688         method = ql.LevenbergMarquardt()
>> 689         ec = ql.EndCriteria(1000, 10, 1e-8, 1e-8, 1e-8)
>> 690         self.model.calibrateVolatilitiesIterative(basket, method, ec)
>> 691         npv = self.nsswaption.NPV()*self.position
>> 
>> _______________________________________________
>> QuantLib-users mailing list
>> Qua...@li...
>> https://lists.sourceforge.net/lists/listinfo/quantlib-users