An Empirical Study on Credit Early Warning Systems

Abstract

Due to its impact on profitability and its potential regulatory consequences financial distress prediction is vitally important for banks. The first generation of prediction models were based on the dichotomous classification of survival versus failure states and utilized balance sheet figures and income statements of bank customers to make predictions. However those models were not designed to accommodate the change in the financial situation of bank customers over time. We define default broadly as the bank declaring a loan as non-performing or initiating the legal process to collect the claimed amounts from the borrower. in this study we use Cox's PH – Proportional Hazard approach to predict the potential defaulters using an unbalanced panel data set from 2005 and 2012. We have 202615 observations on 15593 customers obtained from one of the most reputable participation banks. To our knowledge it is the first application of the Cox PH model to predict financial distress of bank borrowers. it is also important to note that it is also the first such study where only core banking information namely accounting and lending records is used. We did not adopt the traditional approach and thus did not use customer financial statements in our study. We create three different financial distress models and use selectivity ratio and success rate for defaulters terminology to analyze which model's predictive performance is better. We conclude that 72.41% of actual defaulters in the first quarter of 2013 and 58.37% of actual defaulters in 2013 have already been predicted by our Model at the end of 2012.