Predicting US Banks Bankruptcy: Logit versus Canonical Discriminant Analysis

Abstract

Using a large panel of US banks over the period 2008-2013, this paper proposes an early warning framework to identify bank heading to bankruptcy. We conduct a comparative analysis based on both Canonical Discriminant Analysis and Logit models to examine and to determine the most accurate one. Moreover, we analyze and improve suitability of models by comparing different optimal cut-off score (ROC curve vs theoretical value). The main conclusions are: i) Results vary with cut-off value of score ii) the logistic regression using 0.5 as critical cut-off value outperforms DA model with very high correct classification. On the downside, it produces the highest error type 1 rate iii) ROC curve validation improves the quality of the model by minimizing the error of misclassification of bankrupt banks. Also, it emphasizes better prediction of failure of banks because it delivers in mean the highest error type II.

Keywords: Bankruptcy prediction, Canonical Discriminant Analysis, Logistic regression, Principal Component Analysis, CAMELS, ROC curve, Early-warning system

JEL Classification: G21, G28, G33, C25, C38, C53