A comparative analysis of the ARMA and Neural Network Models: A case of Turkish economy

Abstract

The aim of this paper is to provide a detailed econometric analysis of the changes in the nominal exchange rate, inflation rate, nominal interest rate and the real gross domestic product in Turkey for the period from January 1987 to December 2007, based on the monthly data. To this end, both ARMA and Neural Network modeling techniques have been employed in order to present a comparative analysis for their estimation and forecast performances. The results indicate that the NN predictions are consistent with those of ARMA models in the sense that the NN models can perform as good as the ARMA models in the estimation process. However when evaluated for their forecast performances, they differ considerably depending upon the movements in the variables and the length of the sample period.