Accelerating Quadrature Methods for Option Valuation

Abstract

Abstract

Automated application acceleration using software to hardware transformation

“…In the third case study the reference PC is an Intel Pentium 4 CPU at 3.6GHz with 1GB of RAM, and the software implementation involves C code compiled with maximum speed optimisation options. This is to normalise our result to that reported in [3].…”

“…Based on high level user function calls in software, the framework is able to generate dedicated hardware accelerators automatically. The results show that the automatically generated hardware accelerators can achieve approximately the same speed-ups compared to manually optimised versions reported in [3] to solve European option pricing problem by quadrature method. In particular, our automatically generated hardware accelerators in an xc4vlx160 FPGA can generally run more than 18 times faster and can be up to 143 times more power efficient than a Pentium 4 processor in single precision arithmetic, and 7 times faster and 77 times more power efficient in double precision arithmetic.…”

“…We generate FPGA accelerator implementations automatically using our framework and compare the results to the manually optimised implementations proposed by Tse et al [3].…”

“…In this case, the provider decides the numerical method the integration will be using (i.e. Trapezoidal rule or Simpsons rule [3]) and the platform on which the integration will be running (i.e. software or hardware).…”

Option pricing with multi-dimensional quadrature architectures

“…Some recent studies have shifted the focus to pipelined tree based methods [1], finite-difference methods [2] and quadrature methods [3]. All of these studies have provided solutions for the pricing of certain options (such as American options) which cannot be handled easily by Monte Carlo methods.…”