Dipartimento di Scienze Statistiche

This study integrates cures, partial recoveries, and write-offs in modeling Loss Given Default (LGD) and investigates the performance of different algorithms in estimating each component of the decomposed approach. We use a unique database of defaulted real estatebacked loans in European countries. The aim of this study is to accurately estimate the ultimate recovery rate, hence the LGD, by using various machine learning methods including random forest, k-nearest neighbor, extreme gradient boosting, and multivariate adaptive regression splines. We find that the new models we used to estimate each component of the equation, outperform the traditional statistical models such as logistic regression or OLS, and in particular, random forest leads with the highest performance among all models in terms of both in-sample and out-of-sample results. The results confirm that using the random forest in this multiple-step modeling of the recovery rate could improve the whole recovery rate estimation performance.