A combined statistical framework for forecasting default rates of Greek Financial Institutions' credit portfolios

Credit risk modeling remains an important research topic both for financial institutions and the academic community due to its significant contribution to the issue of a bank’s capital adequacy. In this paper we build macro models for the default rates of Greek bank’s loan portfolios. Modeling is performed at two levels: First we use common techniques: regime switching regression, Bayesian regression averaging and linear regression; subsequently we combine the forecasts of the three statistical techniques. This results in increasing performance accuracy and minimizing model risk. Our main goal is twofold: First we attempt to investigate the determinants and the sensitivities of default rates in the Greek banking system where Non Performing Loans (NPLs) have risen sharply due to the sovereign debt crisis which led to a decrease in GDP from 2007 to 2016 of 25%. Secondly, the suggested statistical models can serve as the basis of projecting Greek portfolio dynamics under various macro scenarios. We find that dynamic forecasting combinations exhibit higher predictive accuracy than individual methods. This may provide practitioners with significant insight and policy tools for the banking supervision division in order to enhance monitoring efficiency and support informed decision making.