Journal of Credit Risk

The first paper in this September issue of The Journal of Credit Risk is a technical report from Luis Otero-González, Pablo Durán-Santomil, Rubén Lado-Sestayo and Milagros Vivel-Búa from Universidad de Santiago de Compostela. The authors use a unique data set on the Spanish RMBS market to analyzes the validity of using the loan to-value (LTV) ratio to explain the behavior of mortgage borrowers at an empirical level. Whilst the findings confirm the adequacy of the new Basel III proposal, the authors note the significance shown in the regression models estimated with the "seasoning" variable could be considered in order to improve the models used to measure capital requirements.

The issue's second paper, ‘Modeling joint defaults in correlation-sensitive instruments' by Dariusz Gątarek and Juliusz Jabłecki, presents a simple model for joint defaults and shows how it can be applied to pricing and risk-managing instruments that are sensitive to credit correlation. The presented model strives to be intuitive, analytically tractable, flexible, and one that could serve as a viable alternative to copula models.

In ‘Estimating credit risk parameters using ensemble learning methods: an empirical study on loss given default', Han Sheng Sun and Zi Jin propose a novel ensemble learning method for credit risk parameter estimation. The case study investigates two well-established ensemble learning methods (stochastic gradient boosting and random forest) and propose two new ensembles (ensemble by partial least squares and bag-boosting) in the application of predicting the loss given default.

Finally, Parastoo Rafiee Vahid and Abbas Ahmadi in their paper, ‘Modeling corporate customers' credit risk considering the ensemble approaches in multiclass classification: evidence from Iranian corporate credits', present a neural network based model to classify the credit risk of bank corporate customers into four groups based on definitions of Central Bank of the Islamic Republic of Iran. In particular, the authors find that model trained using the "Self Organizing Map and Radial Function Approach" performed better than the more commonly considered model trained using "Support Vector Machines".