Re: [Quantlib-dev] Serialising YieldTermStructure

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

It should be it.  Logarithms can't handle negative values, and rates extend
into that range when searching.

I don't see any such constraint in other interpolations.


On Fri, Apr 26, 2019 at 5:11 PM R S <raj...@ho...> wrote:

> Ok - thanks for that.
>
> Is that the only “rule” or do you think there is some sort of mapping
> between the Interpolator’s and the Trait’s?
>
> (or is it a question of trial and error?)
>
>
> Regards
> -RS
> On 26 Apr 2019, 16:09 +0100, Luigi Ballabio <lui...@gm...>,
> wrote:
>
> Ok, I checked.  Log interpolation is indeed not usable with forward or
> zero rates.
>
> Luigi
>
>
> On Fri, Apr 26, 2019 at 4:04 PM R S <raj...@ho...> wrote:
>
>> I am using the very old version - nothing negative there
>>
>> BusinessDayConvention depoBizDayConv = ModifiedFollowing;
>> DayCounter depoDCC = Actual360();
>> vector<Rate> rates;
>> rates.push_back(0.0382);
>> rates.push_back(0.0372);
>> rates.push_back(0.0363);
>> rates.push_back(0.0353);
>> rates.push_back(0.0348);
>> rates.push_back(0.0345);
>> /**
>> * Derfine deposit ladder
>> */
>> vector<Period> qlperiods;
>> qlperiods.push_back(1 * Weeks);
>> qlperiods.push_back(1 * Months);
>> qlperiods.push_back(3 * Months);
>> qlperiods.push_back(6 * Months);
>> qlperiods.push_back(9 * Months);
>> qlperiods.push_back(1 * Years);
>> /**
>> * Populate Deposit Helpers
>> */
>> rateUtility.depoHelpers(rates, qlperiods, calendar, fixingDays,
>> true, depoDCC, depoBizDayConv, helper);
>>
>> /**
>> * Define swap rates
>> */
>> Frequency swapFixedLegFrequency = Annual;
>> BusinessDayConvention swapFixedLegConvention = Unadjusted;
>> DayCounter swFixedLegDCC = Thirty360(Thirty360::European);
>> boost::shared_ptr<IborIndex> swapFloatingLegIndex(new Euribor6M);
>>
>> rates.clear();
>> rates.push_back(0.037125);
>> rates.push_back(0.0398);
>> rates.push_back(0.0443);
>> rates.push_back(0.05165);
>> rates.push_back(0.055175);
>> /**
>> * define swap ladder
>> */
>> qlperiods.clear();
>> qlperiods.push_back(2 * Years);
>> qlperiods.push_back(3 * Years);
>> qlperiods.push_back(5 * Years);
>> qlperiods.push_back(10 * Years);
>> qlperiods.push_back(15 * Years);
>> /**
>> * Populate Swap Helpers
>> */
>> rateUtility.swapRateHelpers(rates, qlperiods, calendar,
>> swapFixedLegFrequency,
>> swapFixedLegConvention, swFixedLegDCC, swapFloatingLegIndex, helper);
>>
>>
>> Regards
>> -RS
>> On 26 Apr 2019, 14:34 +0100, Luigi Ballabio <lui...@gm...>,
>> wrote:
>>
>> What version of the examples?  In the most recent one, a few of the input
>> rates are negatives and log-linear interpolation can't handle them.
>>
>> Luigi
>>
>>
>> On Tue, Apr 9, 2019 at 9:31 AM R S <raj...@ho...> wrote:
>>
>>> Yes - I get an error on certain combinations of Trait and. Interpolator
>>>
>>> For example, with a declaration of PiecewiseYieldCurve<ForwardRate,
>>> BackwardFlat>, I can get a dump of dates and rates from the curve
>>>
>>> However, if I. change the declaration
>>> to PiecewiseYieldCurve<ForwardRate, LogLinear>, I get this error:
>>>
>>> terminate called after throwing an instance of 'QuantLib::Error'
>>> what(): 1st iteration: failed at 1st alive instrument, pillar September
>>> 29th, 2004, maturity September 29th, 2004, reference date September 22nd,
>>> 2004: invalid value (-1) at index 0
>>>
>>> Process finished with exit code 134 (interrupted by signal 6: SIGABRT)
>>>
>>> The data is based on the DepoSwap example from the Quantlib Examples
>>>
>>>
>>> Regards
>>> -RS
>>> On 8 Apr 2019, 16:42 +0100, Luigi Ballabio <lui...@gm...>,
>>> wrote:
>>>
>>> Yes, the bootstrap is done when the curve is first used.  That would be
>>> when it's first asked for rates or discounts during swap valuation.
>>>
>>> What do you mean when you say that some combinations don't work?  Do you
>>> get a compile-time error, or does the bootstrap fail at run-time?
>>>
>>> Luigi
>>>
>>>
>>> On Wed, Mar 27, 2019 at 10:18 PM R S <raj...@ho...> wrote:
>>>
>>>> Thanks for this tip on this serialisation. I attach the results of
>>>> running Swap pricing with PiecewiseYieldCurve<Discount,LogLinear> versus
>>>> InterpolatedDiscount<LogLinear> curve
>>>> (I get similar results when I compare PiecewiseYieldCurve<ForwardRate,BackwardFlat>
>>>> with InterpolatedDiscount<BackwardFlat>)
>>>>
>>>> As you can see, the curve generation time remains unchanged but the
>>>> swap valuation time is halved (all times in nanoseconds)
>>>>
>>>> Is this because the bootstrapping occurs only when NPV on the swap is
>>>> called?
>>>>
>>>> Also, I noticed that certain Trait’s work only with certain
>>>> interpolators. For example, ForwardRate trait only works with BackwardFlat.
>>>>
>>>> Are the valid combinations documented somewhere?
>>>>
>>>>
>>>>