Abstract

In the context of Basel II, credit migration matrices have become major components in credit risk management. This thesis is an empirical investigation of various issues that arise in the analysis of credit rating migration. Using the unique and rich internal rating data of a Canadian SME loan portfolio, the thesis investigates credit rating migration from different perspectives. More specifically, the thesis compares and analyzes three multi-state Markov chain estimation methods of migration matrices to accommodate the special features of the internal rating data. The drivers of differences among the three competing methods are investigated theoretically. The statistical and economical differences are examined through non-parametric bootstrap and credit VaR techniques respectively. The underlying assumptions of the three competing methods, time homogeneous first-order Markov chain, are then examined. In the final chapter, the thesis adopts a one-factor model and applies it to our unique internal rating data to forecast migration matrices conditioning on macroeconomic variables. The forecasts are based on different estimation methods of migration matrices. We illustrate the extent to which a one-factor migration matrix could improve the accuracy of the credit loss distribution. Then hypothetical scenarios are selected to evaluate stressed migration matrices with corresponding stressed economic capital.