BASELINE BANK FAILURE PREDICTION MODEL FOR THE ETHIOPIAN BANKING INDUSTRY: USING LOGISTIC REGRESSION MODEL

Abstract

Failure of a bank and a systemic crisis in one country can easily spill over into other countries and develop into a global crisis. So developing early warning signal models capable of identifying banks with high and increasing failure probabilities ahead of time has a prime importance in preventing or minimizing losses. To develop bank failure prediction model for Ethiopia, the researcher believes that, we don’t have to wait for actual failure to happen. Instead failure situation in other countries can provide us a useful benchmark to easily identify Ethiopian banks with high and increasing probabilities of failure proactively. Against this backdrop, the intent of this paper is to develop baseline bank failure prediction model for the Ethiopian banking industry that might help to prevent any bank failure and financial crises in the future using cross-country experience. The study used banks from Ethiopian, Turkish and U.S. The study was based on secondary data which was collected from the published annual reports of the respective banks. The data are taken on the annual basis from 2008/09 to 2013/14 for Ethiopian banks and from 1997 to 2000 for Turkey banks and from 2008 to 2014 for U.S. banks. The researcher tried to predict financial failure in these banks one year ahead of financial failure date. For this reason, failed banks’ balance sheets and income statements from the period one year prior to failure are used. The researcher used bank specific 19 financial ratios that are calculated from the financial statements of the respective banks as explanatory variables. The study begins with an exhaustive literature review with the purpose of understanding well the topic of bank failure prediction. Most of the models and techniques of failure prediction modeling up to this date are covered here. In analyzing the quantitative data, the study used logistic regression model to ascertain the effects of CAMEL ratios on the likelihood of bank failure. The cross-country suggest that the variables C1 (capital adequacy), E1 (earning), M2 (management), and L1 (liquidity) are statistically significant in predicting bank failure. The cross-country bank failure prediction model displays high percentage of outcomes to be correctly classified, good goodness-of-fit and high specificity. The overall predictability accuracy of the logistic regression model was 92%. The derived cross-country baseline logit model is: Ln [Probability of Failure/ Probability of Non-failure] = -.25 - .507(C1) + .327(M2) - .830(E1) - .093 (L1)