Modelling recovery rates is an essential tool for credit risk management. The aim of this thesis is to perform a comprehensive comparative analysis of various statistical models for predicting recovery rates of defaulted loans, of the real estate industry, in Europe. We use data from Global Credit Data, a non-profit association owned by 50+ member banks. The models under consideration range from traditional approaches to more advanced models. Specifically, these include linear regression, quantile regression, decision tree approaches with regression and neural networks, as well as, mixture regression models. Furthermore, we also build a predictive crisis indicator for Europe, using logistic regression, to include crisis information in these models. Our empirical findings indicate that models built on the entire datasets perform better than those built on separate subsets using decision trees. Notably, the best three performing models are linear regression, neural network and mixture regression model with three components. An in-depth analysis reveals that the mixture regression model exhibits superior predictive accuracy when compared to the latter two models.
«
Modelling recovery rates is an essential tool for credit risk management. The aim of this thesis is to perform a comprehensive comparative analysis of various statistical models for predicting recovery rates of defaulted loans, of the real estate industry, in Europe. We use data from Global Credit Data, a non-profit association owned by 50+ member banks. The models under consideration range from traditional approaches to more advanced models. Specifically, these include linear regression, quan...
»