Re: [Quantlib-users] RelinkableHandle<> and impact on run time for multi-step Monte Carlo

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

Hello Amine,
    you're using C++, not Python, is that correct?  (Just to rule out <
https://github.com/lballabio/QuantLib-SWIG/issues/212>).

Luigi


On Fri, Jun 5, 2020 at 6:12 PM Amine Ifri <ami...@gm...> wrote:

> Hi Luigi, all,
>
> I have been working on leveraging QL to create a small multi-step Monte
> Carlo framework which simulates market data at each future time step and
> calls discounting swap engine on vanilla swaps.  A full Monte Carlo run
> with 1000 scenarios and 350 time points takes about 4min to run. I run my
> application through the time profiler and noticed that quite a bit of time
> is spent in the scenario loop where I continuously relink the yield curve
> handle to the appropriately simulated curve.
>
> Are there any ways/tips one could provide on where I could look for extra
> time savings? I found out that by using the discount(Time ) instead of the
> discount(Date&) variant of the method for example, one can save on the year
> fraction calculation, which improves the time a bit. I also de-activated a
> check on MaxTime() in TermStructure::checkRange() which is continuously
> called at each scenario.
>
> In particular, I would like to know if I could switch off the notify
> observers() in the Handle::linkTo() method without affecting the inner
> workings, as I am not interested at this stage in using the
> observable/observer pattern.
>
> Thanks and Regards,
> Amine