A Markov Chain Approach to Model Credit Rating Dynamics: Creditinfo Case

• Main Author: Andrea Valtorta 1989-
• Other Authors: Háskóli Íslands
• Format: Thesis
• Language: English
• Published: 2021
• Subjects: Tölfræði; Johansen; Iceland
• Online Access: http://hdl.handle.net/1946/39936

With the use of the Markov chain framework this work investigates the dynamics between the scores generated by the credit bureau Creditinfo. These scores, also called ratings, are assigned to companies established in Iceland and reflect their probability of default within the next twelve months. The first part implements and compares different methods to calculate transition matrices, namely: the cohort method based on discrete-time observations and two other methods based on continuous-time observations, one assuming time homogeneity, the duration method, and the other relaxing such constraint, the Aalen-Johansen estimator. In the second part, non-Markovian behaviors are tested. The likelihood ratio test suggests that a Markov chain of third order is the most appropriate model for the data at hand. In addition, autoregression was used as an alternative approach to double-check such results. This leads to the next analysis, checking whether rating drift is present. The momentum matrix seems to indicate that, even if some scores show a path dependence, the momentum concept is not there and a company whose rating dropped is not more likely to drop even further or vice versa. It was also important to verify time homogeneity, the chi-squared test suggested that time is not homogeneous and that the transition probabilities vary from one year to the next.