Parallel Algorithm for BSDEs Based High Dimensional American Option Pricing on the GPU
School of Computer Science and Technology, Shandong University, Jinan 250101, China
Journal of Computational Information Systems (JCIS), Vol. 10 (2), 763-771, 2014
@article{peng2014parallel,
title={Parallel Algorithm for BSDEs Based High Dimensional American Option Pricing on the GPU},
author={Peng, Ying and Liu, Hui and Yang, Shuzhen and Gong, Bin},
year={2014}
}
In this paper, we explore the opportunity for solving high dimensional Backward Stochastic Differential Equations (BSDEs) on the GPU with application in high dimensional American option pricing. A Least Square Monte Carlo method based numerical algorithm for solving the BSDEs is studied and summarized in four phases. For the parallel GPU algorithms of different phases, multiple factors which affect the performance, such as the task allocation and data store/access strategies and the thread synchronization features, are taken into consideration. Experimental results for different dimensionality show a maximum speedup of 75 for the backward calculation phase and that of 29 for the global algorithm compared with the CPU serial version.
January 10, 2014 by hgpu
