A parallel and pipelined implementation of a pascal-simplex based multi-asset option pricer on FPGA using OpenCL

Abstract

Multi-asset option pricing is a financial tool that offers a competitive advantage in the global financial markets. To achieve the commercial advantage, performance and accuracy are two key desiderata of such tools. Techniques for financial option pricing have been actively researched for some time. Combined with recent advances in heterogeneous computing, new computing methods of lower latency option pricing have become available. Field-Programmable Gate Array technology offers a powerful framework for building architectures for highly parallel option pricing. This paper presents a versatile, high performance implementation of an accurate multi-asset option pricer on an Intel Stratix 10 GX device. The implementation is highly flexible for the number of assets supported, the maximum tree depth, and the pipelining parameters. The theory that this architecture is based on is that of a recombining multinomial tree approach, which is a generalization of the binomial tree model. This novel architecture was implemented on Field-Programmable Gate Array technology. The implementation presented in this paper enables the computation of an option price on up to 7 assets and can compute up to 43 times faster than a software implementation in a general-purpose processor.