计算机科学 ›› 2018, Vol. 45 ›› Issue (5): 291-294.doi: 10.11896/j.issn.1002-137X.2018.05.050
张劼,文敏华,林新华,孟德龙,陆豪
ZHANG Jie, WEN Min-hua, Jame LIN, MENG De-long and LU Hao
摘要: 风险价值(Value at Risk,VaR)是风险管理的基本工具,可对现有头寸的下行风险提供量化衡量方法。基于历史模拟法的VaR(Historical VaR)是最流行的计算方法之一,被广泛应用于世界各大金融机构。对金融产品进行实时或准实时的VaR计算,对于及时规避金融风险具有重要意义。由于金融产品日益复杂,产品数量持续增长,现有CPU计算平台上的计算能力已经难以满足VaR的性能需求。为解决这一问题, 在GPU上使用CUDA 对Historical VaR的计算代码进行了实现和优化。通过改进排序算法、基于Multi-stream 隐藏通讯时间、解耦数据依赖并实现细粒度并行等优化方法,CUDA版本的VaR计算性能比优化后的CPU单核性能提升了42.6倍,为快速计算超大数量债券的VaR提供了有效的解决方案。以上优化方法也可以为金融领域内其他算法的GPU化提供思路。
[1] Group of Thirty.Derivatives:Practices and Principles.http://group30.org/images/uploads/publications/G30_Derivatives-PracticesandPrinciples.pdf. [2] ZHANG B,OOSTERLEE C W.Option pricing with COS me-thod on graphics processing units[C]∥IEEE International Symposium on Parallel & Distributed Processing,2009(IPDPS 2009).IEEE,2009:1-8. [3] DANG D M,CHRISTARA C C,JACKSON K R.An efficient graphics processing unit-based parallel algorithm for pricing multi-asset American options[J].Concurrency and Computation:Practice and Experience,2012,24(8):849-866. [4] BURKE N,RAU-CHAPLIN A,VARGHESE B.Computing probable maximum loss in catastrophe reinsurance portfolios on multi-core and many-core architectures[J].Concurrency and Computation:Practice and Experience,2016,28(3):836-847. [5] ABBAS-TURKI L A,VIALLE S,LAPEYRE B,et al.Pricing derivatives on graphics processing units using Monte Carlo si-mulation[J].Concurrency and Computation:Practice and Expe-rience,2014,26(9):1679-1697. [6] FATICA M,PHILLIPS E.Pricing American bonds with least squares Monte Carlo on GPUs[C]∥Proceedings of the 6th Workshop on High Performance Computational Finance.ACM,2013:5. [7] WESTON S,SPOONER J,RACANIRE S,et al.Rapid computation of value and risk for derivatives portfolios[J].Concurrency and Computation:Practice and Experience,2012,24(8):880-894. [8] ZHANG N,MAN K L,XIE D.Computing Value at Risk inOpenCL on the Graphics Processing Unit[J].Lecture Notes in Electrical Engineering,2015,9:71-78. [9] WOOD T.Applications of GPUs in computational finance[D].Amsterdam:Universiteit van Amsterdam,2010. |
No related articles found! |
|