This master thesis aims to analyze several approaches for the implementation of stress tests in structural credit portfolio models. In this model class, an obligor defaults if its so-called ability-to-pay variable falls below a certain default threshold. In order to reflect dependencies among the obligors, the ability-to-pay variables are typically described by several systematic risk factors. We give an overview of the existing stress testing literature and focus on the implementation of stress tests through a truncation of systematic risk factors, based on the initial idea of Bonti et al. (2006). In particular, the impact of truncated risk factors on the dependence structure of the obligors, specified by the copula of the ability-to-pay variables, is analyzed depending on the tail thickness of the common factors. In doing so, we found some indications that the results of Bae and Iscoe (2016) also hold true in case of several systematic risk factors. Namely, the dependence among obligors seems to decrease under the truncation in case of Gaussian marginals, while it tends to increase for higher stress levels under t-distributed risk factors. In contrast, if the ability-to-pay variables are specified by a normal variance mixture, the model generally fails to capture an increase in dependence under stress, even in case of heavy tails. Our results are confirmed by stress testing a hypothetical loan portfolio and shed light on the importance of robustness checks and understanding the model behavior in stress tests.
«
This master thesis aims to analyze several approaches for the implementation of stress tests in structural credit portfolio models. In this model class, an obligor defaults if its so-called ability-to-pay variable falls below a certain default threshold. In order to reflect dependencies among the obligors, the ability-to-pay variables are typically described by several systematic risk factors. We give an overview of the existing stress testing literature and focus on the implementation of stress...
»