Monte Carlo Simulation for Financial Derivative Pricing with GPU

Hai Su

(UK Electrical & Computer Engineering Department) 

ABSTRACT: Monte carlo simulation is a method for evaluating the results of algorithms whose outputs depend on some random variables. The accuracy of the evaluation increases as the sample size increases. Therefore, such simulation is usually time consuming when running on CPUs. The graphics processing unit (GPU) can tackle this problem by exploiting many simpler parallel processors to provide more computation power with less energy consumption. However, implementing Monte Carlo simulation on GPUs is distinct from the existing implementation on CPU based machines. This talk explores GPU-based implementation and compares the execution time of GPU-based implementation with its counterpart based on CPU. We use Monte Carlo simulation for financial derivative pricing to verify the GPU implementation and its speedup compared to that of the CPU implementation. As expected, for our selected financial derivative, Monte Carlo simulation based on GPU can achieve up to 25.9X speedup compared to CPU based implementation.