Logit-probit models and artificial neural networks for the estimation of currency crisis: The case of Turkey

As economic crises are multi-pronged events, it can be difficult to characterize by using a single indicator. In literature, although some driving factors have been clarified, determining deeper reasons of the crises is still a problem. Even though basic factors like macroeconomic instabilities, internal or external shocks are monitored, there are many questions left for the real reasons of the crises. How economic crises can cost high to the countries where they are faced is getting accepted day by day. Predicting economic crises early plays a significant role in decreasing its problems and cost burden in economy. Turkish economy, which has experienced some crises whose severity and time are different before 1980, has been faced with various crises such as 1994 Crisis, 2000-2001 Crisis and 2008 Global Economic Crisis since 1990s. The aim of this study is to research the predictability of the currency crisis in Turkey by using Logit-Probit methods, and Artificial Neural Networks (ANN) and Adaptive Neuro Fuzzy Inference System (ANFIS) methods which are the two of the artificial intelligence methods; and to identify the variables that affect the currency crises experienced in Turkey. As the result of the study, it has been observed that ANN, which is used for predicting currency crisis for Turkey, shows better results than ANFIS and Logit-Probit methods. When the weighted values of the independent variables are analyzed by the results of the ANN, which has the best performance, it has been found out that three variables that mostly affects the currency crises occurred in Turkey are respectively real effective exchange rate (REER), interest rates on deposits (IROD) and export unit value (XUV). These results indicate that ANN method is quite successful in predicting currency crises. This situation shows the superiority of ANN methods and asserts that it is a strong tool for economic policy and decision-makers.