GPU-Accelerated Method for Simulating Efficient Portfolios in the Mean-Variance Analysis

This paper considers portfolio optimization whose goal consists in finding a set of efficient portfolios regarding the framework of the Mean-Variance Analysis. Our work utilizes the GPU's computing capabilities to accelerate the computation within portfolio optimization. In particular, we present a nontrivial GPU-accelerated method to produce a set of minimum variance portfolios for a given target range of expected returns under the basic constraints of the framework. We evaluate the experimental performance of the method by synthesizing an implementation using CUDA. The experimental results show that our implementation performs substantially faster than its implementation counterpart using CLAPACK with respect to the task of simulating the efficient frontier on large data sets with the number of assets in the range of hundreds.