随机波动率下的亚式期权定价问题在GPU集群上的实现  

IMPLEMENTATION OF PRICING ASIAN OPTIONS WITH STOCHASTIC VOLATILITY ON GPU CLUSTER

作　　者：徐磊[1] 徐莹[1] 姜广鑫[2] 梁义娟[2] 寇大治[1] 徐承龙[2] 

机构地区：[1]上海超级计算中心,上海201203 [2]同济大学数学系,上海200092

出　　处：《计算机应用与软件》2012年第11期79-82,共4页Computer Applications and Software

基　　金：国家高技术研究发展计划(2009AA012201);上海市科委科研计划项目(08dz1501600);上海浦江人才计划(10PJ1430600)

摘　　要：期权定价作为计算金融领域的核心问题之一,越来越受到关注。随着期权交易的规模和交易量的迅速增长,当前的期权定价平台越来越受到挑战,在尽可能短的时间内对期权进行定价变得越来越困难。传统的计算平台通常使用基于CPU的计算集群,而图形处理器(GPU)具有更高的浮点性能和访存带宽,在价格与功耗方面也优于CPU。尝试使用GPU集群来对具有随机波动率的亚式期权进行定价,同时使用带控制变量的Monte Carlo方法,减小模拟的方差。最终的测试结果表明GPU集群较CPU集群具有更多的优势,适合应用于期权定价领域。Options pricing is one of the core issues in the field of computational finance,which has attracted increasing focus.With the rapid growth of options trading in both scale and volume,there is growing challenge on existing options pricing platforms,and to price an option in shortest possible period of time has become increasingly difficult.Traditional computing platforms often use CPU-based computation clusters,but compared with the tradition CPU,GPU（Graphic Processing Unit） can possess higher floating-point performance and bandwidth,and its cost and power consumption outperform CPU as well.In this paper,we try to use GPU cluster to price Asian options with stochastic volatility,and meanwhile use Monte Carlo method with control variables to reduce the variance simulated.Final testing results show that the GPU cluster has more advantages than the CPU cluster and is well suited for pricing options.

关 键 词：GPU集群 CUDA 亚式期权 随机波动 蒙特卡洛 MPI 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]