This thesis explores three important issues in credit risk modeling: the nonlinear credit risk stress testing models, the recovery term structure of point-in-time (PIT) loss given default (LGD), and the estimation of LGD by mixture beta regression model.
In the first essay of this thesis, we study the credit risk stress testing models. By incorporating the regime-switching and quantile regression techniques into credit risk stress testing models, we propose two new dynamic models that outperform the traditional linear regression model according to both the point estimate accuracy and the confidence interval breaches. This confirms the importance of nonlinear regression approaches in the estimation and the prediction of credit risk determinants. The proposed models perform especially well in capturing the extreme values on the tail of the estimated distribution of the credit risk measure. The proposed models could be used for both the International Financial Reporting Standard 9 (IFRS9) expected loss calculation and Basel’s Advanced Internal Rating-Based (AIRB) regulatory capital calculation purposes.
In the second essay, we examine and model the time-series pattern of recovery throughout the bankruptcy and workout process of a retail credit portfolio; whereas other researchers are concerned with predicting the overall recovery rates of debt instruments, we model the amounts a creditor can recover at different points in time subsequent to the default event. This is of practical interest to commercial banks in managing the risk of their default loan portfolios. Like managing performing loan portfolios, banks must assign loss provision and determine the capital requirement associated with non-performing (i.e., defaulted) loan portfolios. Given the fact that it usually takes two to three (up to five or more) years to complete the recovery process for a typical defaulted retail (corporate) loan, it is important to understand the time-varying risk characteristic of the defaulted portfolio as a function of its vintage in the recovery process. An accurate point-in-time (PIT) risk assessment enables financial institutions to manage their defaulted loan portfolios in a timely fashion.
In the third essay, we further extend our understanding of the distribution of LGD. For credit risk management purposes, the LGD of credit instruments is one of the key determinants in estimating capital requirements for financial institutions. To address the practical problems encountered in implementing LGD prediction model (e.g., in capturing the bimodal characteristic of the LGD distribution), we propose to develop a mixture beta regression LGD model. By using the maximum likelihood estimation and the method of moment approaches, the parameters of the mixture beta regression model can be estimated. Furthermore, we examine the impact of the systematic factors and model the time-series variation of the LGD distribution as a function of these systematic factors. Finally, through a number of empirical analyses, we demonstrate the superior performance of our proposed mixture beta models in comparison with the single-beta logit-linked model commonly considered in the literature. / Thesis / Doctor of Philosophy (PhD)
| Identifer | oai:union.ndltd.org:mcmaster.ca/oai:macsphere.mcmaster.ca:11375/26245 |
| Date | January 2021 |
| Creators | Jin, Yuchuan |
| Contributors | Miu, Peter, Finance |
| Source Sets | McMaster University |
| Language | English |
| Detected Language | English |
| Type | Thesis |
Page generated in 0.0019 seconds