Affiliation:
1. Warsaw School of Economics
2. Vistula University of Warsaw
Abstract
In some credit portfolios the number of observed defaults is always very limited. This is particularly evident in the Loss Given Default (LGD) estimation based on the new definition of default (the new definition of default was introduced in European banks in recent years) where only a small sample of empirical data is observed. The basic proposed LGD model is based on splitting recoveries into two classes of recoveries: value close to 0 or close to 1. This paper addresses also the problem with unresolved cases using the Bayesian approach, which assumes a distribution of further recoveries for unresolved cases. The Bayesian approach is considered with a combination of two binary models. The modelling approach for LGD is illustrated on real data for a long time period for mortgage loans. The proposed methodology takes into account the specificity of LGD data for both bimodal LGD distribution and uncertainty about unresolved cases, which lead to reduce a model bias.
Reference35 articles.
1. Yao X., Crook J., Andreeva G. (2017), Enhancing two-stage modelling methodology for loss given default with support vector machines, European Journal of Operational Research, 263(2), 679–689, DOI: 10.1016/j.ejor.2017.05.017.
2. Düllmann K., Trapp M. (2004), Systematic risk in recovery rates – an empirical analysis of U.S. corporate credit exposures, Discussion Paper Series 2, Deutsche Bundesbank.
3. Rapisarda G., Echeverry D. (2013), A non-parametric approach to incorporating incomplete workouts into loss given default estimates, Journal of Credit Risk, 9(2), DOI: 10.21314/JCR.2013.159.
4. Christensen E., Henrik J. (2006), Joint default and recovery risk estimation: an application to CDS data, Working Paper, Copenhagen Business School.
5. Bakshi G., Madan D., Zhang F. (2001), Investing the sources of default risk: lessons from empirically evaluating credit risk models, Working Paper, 15, University of Maryland.