The estimation of transition probabilities between external rating classes is of vital importance in a bank’s risk management. The following Bachelor’s Thesis presents a wide range of questions arising during the modeling of these rating migrations while focusing on probabilities of transitions into the default state. We choose the commonly used framework of continuous-time Markov chains for the modeling, where so-called generators denoted by Q play an important role. Transition matrices over the time horizon [0, t] are given by e^(Q t) The entries of these transition matrices give the probabilities of migrating from one rating class to another one within the considered time period. As rating agencies usually provide one-year migration matrices, these have to be embedded into the continuous-time framework first. This means that one has to find a valid generator Q for which e^(Q 1) equals the original one-year transition matrix. Unfortunately, this is usually not possible. Therefore, one tries to calibrate a sufficiently similar continuous-time process by finding a valid generator Q for which e^(Q 1) gives approximately the original observed migration matrix. As the major new result of this thesis we propose to calibrate a continuous-time process by solving a quadratic optimization problem with linear constraints for the underlying generator. As an advantage of the optimization, by setting additional constraints one can even enforce to receive a ”monotonous” generator and, hence, receive a calibrated ”monotonous” Markov process. This monotonicity ensures for example that transition rates to the default state are smaller for better-rated obligors. As another main result of this thesis a new testing procedure is proposed that allows to investigate, if differences between two migration matrices, for example the original one and through the optimization calibrated one, are significant. While developing these new ideas, many other arising problems are considered and an overview about the corresponding literature is provided. In addition, a case study provides illustrations while the implementation of the existing and new proposals in R allows the reader to try out the proposals on new datasets.
«
The estimation of transition probabilities between external rating classes is of vital importance in a bank’s risk management. The following Bachelor’s Thesis presents a wide range of questions arising during the modeling of these rating migrations while focusing on probabilities of transitions into the default state. We choose the commonly used framework of continuous-time Markov chains for the modeling, where so-called generators denoted by Q play an important role. Transition matrices over th...
»