high performance computing on graphics processing units: hgpu.org

(Counterparty) Credit Valuation Adjustments (CVA) has become a prevailing form of pricing default risk on over-the-counter (OTC) contracts. Due to the large size of portfolios included in the CVA calculation and its computational complexity, large computing grids are needed for the evaluation. The main purpose of this thesis is to investigate an even more computationally demanding problem, namely computing the sensitivities of CVA to the market and model parameters, a topic which was hardly addressed in the literature so far. We show that the pathwise sensitivities method can be applied for CVA and that it gives significant speed improvement over the conventional finite-differencing techniques. Additionally, we take advantage of the GPU technology to obtain the Greeks fast enough for daily hedging and risk management activities.