Large-Scale Credit Risk Loss Simulation
Simon J. Rees and Joseph Walkenhorst
In this chapter, we present a GPU algorithm to perform a Monte Carlo simulation of a portfolio of credit risk exposures. Multiple design paths are explored with discussion around the relative costs and benefits. The kernel implementation itself is described in detail, identifying and improving several areas of optimization to maximize the performance for demanding problem sizes. The algorithm allows us to generate a loss distribution for the portfolio, and the success of the application motivates discussion into re-simulation of the scenarios in the tail for more detailed analysis.
In the wake of the credit crisis, ...