Lonstaff-Schwartz Method for pricing American options

[Old posts from the commercial version of ArrayFire] Discussion of ArrayFire using CUDA or OpenCL.

Moderator: pavanky

Lonstaff-Schwartz Method for pricing American options

Postby melonakos » Wed Jan 12, 2011 5:48 pm

We recently received an email asking:

I would like to parallelize an algorithm that is close to the Lonstaff-Schwartz Method for pricing American options via simulation / least squares.


Thanks for this question. I had to google this to understand what you were asking. I found http://demonstrations.wolfram.com/PricingABermudanOptionWithTheLongstaffSchwartzMonteCarloMeth/. It looks like this is a Monte Carlo type simulation and the functions used are mainly exponentials and basic arithmetic. Jacket should do really well for you here. I'd try to solve this in 3 steps:

1) Just using GDOUBLEs (or GSINGLEs) in the biggest vectorized statements that you can make. http://wiki.accelereyes.com/wiki/index.php/Jacket_Basics
2) Play with GFOR if there are FOR-loops that you can't vectorize. http://wiki.accelereyes.com/wiki/index.php/GFOR_Usage
3) Play with GCOMPILE if there are expressions that meet the GCOMPILE criteria and if those functions are consuming a lot of time. http://wiki.accelereyes.com/wiki/index.php/GCOMPILE_Usage

Good luck and we look forward to hearing your progress on these forums.

-John
John Melonakos (john@arrayfire.com)
User avatar
melonakos
 
Posts: 503
Joined: Tue Jun 10, 2008 9:49 am

Return to [archive-commercial] Programming & Development with ArrayFire

cron