Does Non-linearity Matter in Retail Credit Risk Modeling?

In this research we propose a new method for retail credit risk modeling. In order to capture possible non-linear relationships between credit risk and explanatory variables, we use a learning vector quantization (LVQ) neural network. The model was estimated on a dataset from Slovenian banking sector. The proposed model outperformed the benchmarking (LOGIT) models, which represent the standard approach in banks. The results also demonstrate that the LVQ model is better able to handle the properties of categorical variables.