Not logged in.
Quick Search - Contribution
|Title||Point-in-Time Loss Given Default modelling for Banking products|
|Institution||University of Zurich|
|Faculty||Faculty of Business, Economics and Informatics|
|Abstract Text||This thesis covers a large scale modelling analysis dealing with the prediction of the loss given default for Banking products, with a focus on loans to multinational corporations. The work enhances the available literature on the topic by a comprehensive study of both, the analysis of a large number of different parametric, semi-parametric and non-parametric statistical modelling approaches, and the selection of predictor variables at the same time. Hence, the work merges the two main choices to be made in practice when building a loss given default model. The resulting models are designed to be point-in-time, in the sense that they describe the variation of the loss given default with changing economic conditions. This is especially relevant for downturn conditions under the Basel II regulatory framework, and for the whole economic cycle in the new IFRS 9 accounting standard becoming binding for financial institutions starting 2018. Departing from different combinations of modelling approaches and explanatory variables, almost 500 different model specifications are analyzed. The decision for a final model specification is based on the most important characteristics a loss given default needs to show to be used in practice. Looking at accuracy of the models using mean squared error and mean absolute error, rank-ordering of predicted loss given default, economic-intuition and sensitivity of the model outcomes, as well as the most important model assumptions inherent in different modelling techniques, it is found that the zero- and one-inflated beta regression model is the most promising approach to model the loss given default for loans to multinational companies. Different reasonable choices of macroeconomic variables are found. These are the AAA corporate bond spread, the credit cycle index, the unemployment rate and the annual %-differences in the real GDP. Accounting for the desired sensitivity of the model outcomes, the AAA corporate bond spread is selected as the macro-economic variable for the champion model. There still exist important caveats for the resulting champion model. A practitioner however, can use the results of the model analysis and the caveats in order to develop a suitable point-in-time loss given default model for her or his specific purpose.|