| The development of the hardware technology changes program structure. Today, the Multi-core technology has gradually improved. Multi-core CPUs and GPUs (Graphic Processing Unit) have become mainstream processors. Driven by the market demand, the performance of Multi-core processors will manage to continue along the path of Moore’s law. Traditional serial programming is facing serious challenges. The computer industry has changed from the central processing of CPUs to the cooperative processing of both CPUs and GPUs. Meanwhile, parallel computing has been greatly developed in life science, military, finance and many other fields. Now the GPU already has good programmability. It has evolved into a highly parallel, multi-core, multi-threaded processor. Comparing with CPUs, GPUs have outstanding advantage in computing power and memory bandwidth. More specifically, GPUs are dedicated to solve the case which can be expressed as a data parallel computing problem. Parallel program executed on many data elements has two characteristics. First, most of the parallel work focuses on performing operations on a data set. The data set is typically organized into a common structure. The program could hide memory access latency well when it has high compute density (the ratio of computing time to memory access time). Second, flow controls do not need to be very accurate. Since the same commands execute on all data elements, the execution units could share control units. At present, many software developers, scientists, and researchers develop lots of acceleration work with GPUs.Option pricing is an Age-old problem. Louis Bachelier published his paper "Theorie de la Speculation" in1900. It is recognized as the milestone of modern finance. Since American Chicago Board Option Exchange established in1973, Options trading had revolutionary change. The option pricing is a very important problem encountered in financial engineering. Option pricing is a contingent claim. Over the years, there is no standard method for option pricing. As more computation has been applied to finance-related problems, finding efficient ways to implement option pricing models on modern architectures has become more important. The price of option changes while market changes in option contracts. It directly affects the profit and loss of the buyers and sellers. Option pricing is the core problem of the options trading.In the field of computer science, financial computing, economics and engineering, the Backward Stochastic Differential Equation (BSDE) is a robust tool. Comparing with the Black-Scholes formula, the BSDE is more robust to fit in the situation of probability model uncertainty, thus can be used not only to perform more approximated calculations for financial derivatives pricing, but also to help a variety of investors making more rational decisions in risk hedging and risk analysis.In the numerical algorithms of BSDE for solving option pricing, two representative algorithms are selected:a high accurate theta scheme method and a binomial lattice based method. These two numerical methods need much more computing efforts. The computation time may be very long to get relatively accurate results, thus acceleration becomes an efficiently way to solve the numerical methods. The aim of this paper is the efficient use of GPU acceleration for option pricing with BSDEs. Therefore, Acceleration is especially important to both models. |
PDF Full Text Request