2
$\begingroup$

I'm currently benchmarking several Monte Carlo computing methods for sensitivities computation purposes.

I've already done some implementation in Python using Numpy and MyGrad libraries and it's fascinating how fast it is.

Let's say that I have a C++ code that computes prices by Monte Carlo, how easy it is to implement AAD for computing sensitivities of said prices ? Which part of the code must be rewriten ?

Are there banks that already implemented it ?

Thank you,

Improve this question
$\endgroup$
11
  • $\begingroup$ A real simple AAD example in case anyone thinks it sounds complicated onlinegdb.com/wxY-6EzXQ :) $\endgroup$
    – Dimitri Vulis
    Commented Nov 26, 2022 at 16:44
  • 1
    $\begingroup$ It is easy but for example considering and entire pricing system such as one of a investment bank it may be daunting to implement this. You'll need to re-write all the basic functions $\endgroup$
    – Hilbert
    Commented Nov 26, 2022 at 17:43
  • 1
    $\begingroup$ Actually, no, you don't have to rewrite all your existing code. May I suggest a good book: amazon.com/dp/1119539455 $\endgroup$
    – Dimitri Vulis
    Commented Nov 26, 2022 at 18:03
  • 1
    $\begingroup$ I know of a bulge bracket US bank in which a few years ago the XVA desk commissioned a review of the XVA library to assess whether AAD could be implemented. The conclusion was that 2-3 people were needed on a full time basis over a 2-5y period to revamp the library and make it AAD. The idea was swiftly discarded. $\endgroup$
    – Daneel Olivaw
    Commented Nov 29, 2022 at 19:05
  • 1
    $\begingroup$ @DaneelOlivaw Is it because of huge amount of legacy code ? Legacy design patterns ? Lack of highly skilled/experienced professionals ? What was the reason behind this ? (of course without breaking industrial secrecy) Thanks $\endgroup$
    – Hilbert
    Commented Dec 9, 2022 at 10:50

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.