Markov Chains, Clustering, and Reinforcement Learning: Applications in Credit Risk Assessment and Systemic Risk Reduction
| dc.contributor.advisor | Ku, Hyejin | |
| dc.contributor.author | Le, Richard | |
| dc.date.accessioned | 2023-12-08T14:50:44Z | |
| dc.date.available | 2023-12-08T14:50:44Z | |
| dc.date.issued | 2023-12-08 | |
| dc.date.updated | 2023-12-08T14:50:44Z | |
| dc.degree.discipline | Mathematics & Statistics | |
| dc.degree.level | Doctoral | |
| dc.degree.name | PhD - Doctor of Philosophy | |
| dc.description.abstract | In this dissertation we demonstrate how credit risk assessment using credit rating transition matrices can be improved, as well as present a novel reinforcement learning (RL) model capable of determining a multi-layer financial network configuration with reduced levels of systemic risk. While in this dissertation we treat credit risk and systemic risk independently, credit risk and systemic risk are two sides of the same coin. Financial systems are highly interconnected by their very nature. When a member of this system experiences distress such as default, a credit risk event, this distress is often not felt in isolation. Due to the highly interconnected nature of financial systems, these shocks can spread throughout the system resulting in catastrophic failure, a systemic risk event. The treatment of credit risk begins with the introduction of our first-order Markov model augmented with sequence-based clustering (SBC). Once we established this model, we explored its ability to predict future credit rating transitions, the transition direction of the credit ratings, and the default behaviour of firms using historical credit rating data. Once validated, we then extend this model using higher-order Markov chains. This time around, focusing more on the absorbing behaviour of Markov chains, and hence, the default behaviour under this new model. Using higher-order Markov chains, we also enjoy the benefit of capturing a phenomenon known as rating momentum, characteristic of credit rating transition behaviour. Other than the credit rating data set, this model was also applied to a Web-usage mining data set, highlighting its generalizability. Finally, we shift our focus to the treatment of systemic risk. While methods exist to determine optimal interbank lending configurations, they only treat single-layer networks. This is due to technical optimization challenges that arise when one considers additional layers and the interactions between them. These layers can represent lending products of different maturities. To consider the interaction between layers, we extend the DebtRank (DR) measure to track distress across layers. Next, we develop a constrained deep-deterministic policy gradient (DDPG) model capable of reorganizing the interbank lending network structure, such that the spread of distress is better mitigated. | |
| dc.identifier.uri | https://hdl.handle.net/10315/41789 | |
| dc.language | en | |
| dc.rights | Author owns copyright, except where explicitly noted. Please contact the author directly with licensing requests. | |
| dc.subject | Applied mathematics | |
| dc.subject | Mathematics | |
| dc.subject | Finance | |
| dc.subject.keywords | Mathematics | |
| dc.subject.keywords | Finance | |
| dc.subject.keywords | Credit risk | |
| dc.subject.keywords | Systemic risk | |
| dc.subject.keywords | Reinforcement learning | |
| dc.subject.keywords | Clustering | |
| dc.title | Markov Chains, Clustering, and Reinforcement Learning: Applications in Credit Risk Assessment and Systemic Risk Reduction | |
| dc.type | Electronic Thesis or Dissertation |
Files
Original bundle
1 - 1 of 1