Global sensitivity analysis of credit risk portfolios

This paper proposes the use of global sensitivity analysis to evaluate latent factor credit risk models. Our claim is that this type of sensitivity analysis is superior to a local approach in providing the risk modeler with a broader picture of the risk contributions of the key elements to a credit risk model. The main finding is that default probabilities and the correlation of the latent variables are considerably more important than the multivariate distribution and hence the copula of the latent variables.