Credit risk models for mortgage loan loss given default

Arguably, the credit risk models reported in the literature for the retail lending<br/>sector have so far been less developed than those for the corporate sector,<br/>mainly due to the lack of publicly available data. Having been given access to a<br/>dataset on defaulted mortgages kindly provided by a major UK bank, this work<br/>first investigates the Loss Given Default (LGD) of mortgage loans with the<br/>development of two separate component models, the Probability of Repossession<br/>(given default) Model and the Haircut (given repossession) Model. They are then<br/>combined into an expected loss percentage. Performance-wise, this two-stage<br/>LGD model is shown to do better than a single-stage LGD model (which directly<br/>models LGD from loan and collateral characteristics), as it achieves a better Rsquare<br/>value, and it more accurately matches the distribution of observed LGD.<br/>We next investigate the possibility of including macroeconomic variables into<br/>either or both component models to improve LGD prediction. Indicators relating<br/>to net lending, gross domestic product, national default rates and interest rates<br/>are considered and the interest rate is found to be most beneficial to both<br/>component models. Finally, we develop a competing risk survival analysis model<br/>to predict the time taken for a defaulted mortgage loan to reach some outcome<br/>(i.e. repossession or non-repossession). This allows for a more accurate<br/>prediction of (discounted) loss as these periods could vary from months to years<br/>depending on the health of the economy. Besides loan- or collateral-related<br/>characteristics, we incorporate a time-dependent macroeconomic variable based<br/>on the house price index (HPI) to investigate its impact on repossession risk. We<br/>find that observations of different loan-to-value ratios at default and different<br/>security type are affected differently by the economy. This model is then used<br/>for stress test purposes by applying a Monte Carlo simulation, and by varying the<br/>HPI forecast, to get different loss distributions for different economic outlooks.

Name dc:type.qualificationname

Ph.D.

Level dc:type.qualificationlevel

doctoral

Grantor dc:publisher.institution

University of Southampton

Year dc:date.issued

2010

Author dc:creator

• Leow, Mindy

Advisors dc:contributor.advisor

• Mues, Christophe
• Thomas, Lyn