Grabit: Gradient tree-boosted Tobit models for default prediction

In this paper, a novel approach for default prediction is presented. An often encountered problem in default prediction is the fact that there is relatively little default data since bankruptcies are usually rare events. We show how this class imbalance issue can be alleviated by the use of a novel model, the so called Grabit model which is obtained by applying gradient tree boosting to the Tobit model, and additional data. We use the Grabit model for predicting defaults on loans made to Swiss small and medium-sized enterprises (SME) and obtain a large improvement in predictive performance compared to other state-of-the-art approaches. In contrast to the Tobit model, the Grabit model can account for general forms of non-linearities and interactions, it is robust against outliers in covariates and scale invariant to monotonic transformations for the covariates, and its predictive performance is not impaired by multicollinearity.