Abstract
In this thesis, our aim was establish a framework as CreditMetrics for quantifying credit risk in portfolios of corporate bonds. We depended on assets dependence structures of corporate bonds and Standard & Poor’s credit transition matrices to compute all possible 64 year-end values and all possible 64 year-end joint likelihoods across 64 different states for a two-bond portfolio. The next step was to assessed the credit value-at-risk (Credit VaR). In this thesis, we focued on the problem of modeling the multivariate distributions of several outcomes. To solve this problem we used a promising approach based on Archimedean copulas which is different from the conventional multivariate Normal assumption to focus explicitly on the dependence structure. We used the nonparametric methods of Genest and Rivest to assess the credit value-at-risk for American corporate bond portfolios. Our empirical distributions were heavier tailed than the normal distribution. If investors ignore this phenomenon, they will underestimate the VaR and make a downside loss.