Re: [Quantlib-users] LD Path Generator in QL-Python for multi-factor processes

[Jack G <jac...@gm...>]

Sorry, ignore me - this change does seem to work (I think I was seeing an
OOM error before)

I'll make a pull request into SWIG repo with this change once I've tested
that it's doing the right thing.

On Fri, Nov 27, 2020 at 5:47 PM Jack G <jac...@gm...> wrote:

> Luigi,
>
> Thanks for the pointer. It seems as though the only multi-path generator
> that is currently exposed via SWIG is for pseudorandoms only (
> https://github.com/lballabio/QuantLib-SWIG/blob/master/SWIG/montecarlo.i).
>
> I have the feeling that I should be adding the following to expose a
> multi-path generator for low discrepancy numbers, but for some reason this
> is upsetting my SWIG build (the error I get is "gcc: fatal error: Killed
> signal terminated program cc1plus" which I usually see when I add something
> subtly bad)... am I thinking on the right lines here?
>
> *%{*
> *typedef
> QuantLib::MultiPathGenerator<GaussianLowDiscrepancySequenceGenerator>*
> *    GaussianMultiPathSobolGenerator;*
> *%}*
> *class GaussianMultiPathSobolGenerator {*
> *  public:*
> *    %extend {*
> *      GaussianMultiPathSobolGenerator(*
> *                   const ext::shared_ptr<StochasticProcess>& process,*
> *                   const std::vector<Time>& times,*
> *                   const GaussianLowDiscrepancySequenceGenerator&
> generator,*
> *                   bool brownianBridge = false) {*
> *          return new GaussianMultiPathSobolGenerator(process,*
> *                                                     QuantLib::TimeGrid(*
> *                                                         times.begin(),*
> *                                                         times.end()),*
> *                                                     generator,*
> *                                                     brownianBridge);*
> *      }*
> *    }*
> *    Sample<MultiPath> next() const;*
> *    Sample<MultiPath> antithetic() const;*
> *};*
>
> Best,
> Jack
>
> On Fri, Nov 27, 2020 at 4:35 PM Luigi Ballabio <lui...@gm...>
> wrote:
>
>> Hello Jack,
>>     the Heston process is two-dimensional (it models both underlying and
>> vol) so you'll need to use a multi-path generator.  The error you're
>> getting is the generator trying to cast the process to a one-dimensional
>> one and failing.
>>
>> Luigi
>>
>>
>> On Fri, Nov 27, 2020 at 4:13 AM Jack G <jac...@gm...> wrote:
>>
>>> Dear QL Users,
>>>
>>> I'm getting a slightly cryptic error when I try to manually generate
>>> paths for Heston using LD numbers in Python.
>>>
>>> It works fine for BS:
>>>
>>> *import QuantLib as ql*
>>>
>>> *initialValue = ql.QuoteHandle(ql.SimpleQuote(100))*
>>> *sigma = 0.2*
>>>
>>> *today = ql.Date().todaysDate()*
>>> *riskFreeTS = ql.YieldTermStructureHandle(ql.FlatForward(today, 0.0,
>>> ql.Actual365Fixed()))*
>>> *dividendTS = ql.YieldTermStructureHandle(ql.FlatForward(today, 0.0,
>>> ql.Actual365Fixed()))*
>>> *volTS = ql.BlackVolTermStructureHandle(ql.BlackConstantVol(today,
>>> ql.NullCalendar(), sigma, ql.Actual365Fixed()))*
>>> *process = ql.BlackScholesProcess(initialValue, riskFreeTS, volTS)*
>>>
>>> *times = ql.TimeGrid(1, 12)*
>>>
>>> *rng = ql.UniformLowDiscrepancySequenceGenerator(12)*
>>> *seq = ql.GaussianLowDiscrepancySequenceGenerator(rng)*
>>> *path_gen = ql.GaussianSobolPathGenerator(process, times, seq, False)*
>>>
>>> *x = path_gen.next()*
>>>
>>>
>>> However, when I try the same with Heston, I get the following error: "
>>> RuntimeError: Boost assertion failed: px != 0"
>>>
>>> *v0 = 0.09 kappa = 0.8 theta = 0.09 rho = 0.4 sigma = 0.1 process =
>>> ql.HestonProcess(riskFreeTS, dividendTS, initialValue, v0, kappa, theta,
>>> sigma, rho) times = ql.TimeGrid(1, 12) rng =
>>> ql.UniformLowDiscrepancySequenceGenerator(12) seq =
>>> ql.GaussianLowDiscrepancySequenceGenerator(rng) path_gen =
>>> ql.GaussianSobolPathGenerator(process, times, seq, False) x =
>>> path_gen.next()*
>>>
>>>
>>> My working hypothesis is that the rng needs to be dimension 24 instead
>>> of 12 because of the second vol process that needs to be simulated, but
>>> when I change the dimension to 24 I get the following error: "
>>> RuntimeError: sequence generator dimensionality (24) != timeSteps (12)"
>>>
>>> I've looked through the QuantLib-SWIG code but not got too far... any
>>> pointers about how I might deal with this? Or have I done something stupid
>>> somewhere?
>>>
>>> Thanks,
>>> Jack
>>>
>>>
>>> _______________________________________________
>>> QuantLib-users mailing list
>>> Qua...@li...
>>> https://lists.sourceforge.net/lists/listinfo/quantlib-users
>>>
>>