Banking Crisis Prediction Modeling with Bayesian Model Averaging Approach

Banking crises are occurring intermittently. This indicates that pre-current warning models have not been successful in identifying these crises. Examination of existing models specifies that the failure of these models is mainly due to the identification of explanatory variables and experimental design of the model, which the researchers of the present study aimed at improving. In order to moderate the problem of model uncertainty by averaging all models (Bayesian averaging) the present research attempted to determine the factors affecting the banking crisis in Iran. In this study, 49 variables affecting the banking crisis were included in the model. Finally, using the Bayesian averaging model approach, 12 non-fragile variables affecting the financial crisis were identified consisting of cost of funding, none performing loan (NPL), deposit to loan (DTL), spread, capital adequacy, earning assets to total assets ratio, net LTD (after deducted Legal reserves), cash coverage ratio, net stable funding ratio (NSFR) in the presence of all variables, duration of assets and liabilities, interest rate duration, and increase in properties' possession. According to the results, it could be deduced that the banking crisis index in the Iranian economy is a problem with wide dimensions as the variables related to monetary and financial sector policy makers affect this index. The banks studied in this study are 10 banks listed on the Tehran Stock Exchange (Kar Afarin, Eghtesad-e Novin, Parsian, Sina, Mellat, Tejarat, Saderat, Post Bank, Mellat, Dey) in an 11-year period from 2008 to 2019.