Model-free computation of risk contributions in credit portfolios

Abstract

Abstract In this work, we propose a non-parametric density estimation technique for measuring the risk in a credit portfolio, aiming at efficiently computing the marginal risk contributions. The novel method is based on wavelets, and we derive closed-form expressions to calculate the Value-at-Risk (VaR), the Expected Shortfall (ES) as well as the individual risk contributions to VaR (VaRC) and ES (ESC). We consider the multi-factor Gaussian and t-copula models for driving the defaults. The results obtained along the numerical experiments show the impressive accuracy and speed of this method when compared with crude Monte Carlo simulation. The presented methodology applies in the same manner regardless of the used model, and the computational performance is invariant under a considerable change in the dimension of the selected model. The speed-up with respect to the classical Monte Carlo approach ranges from twenty-five to one-thousand depending on the used model.