CDO PRICING IN PRACTICAL TIME: GPU Accelerated Monte Carlo Methods

Abstract

The objective of this thesis is to explore the viability of Monte Carlo based methods of pricing Collateralized Debt Obligations (CDOs). From the early 2000s up until the global financial crisis, the single-factor Gaussian copula was the de-facto means of pricing these securities products. In recent years, usage of the copula has fallen out of favor due to the unreliability of measuring association of securities using correlation. Monte Carlo pricing methods have historically been too computationally expensive for generating timely quotes. The 2007-2008 global financial crisis and subsequent implosion of the CDO market coincided with massive growth in the field of parallel computing, namely the advent of NVidia’s GPU-based CUDA. In light of the partial recovery of the CDO market over the last two years, it is worth reexamining the feasibility of Monte Carlo methods in producing practical valuations.