Artificial Neural Networks rectified which are inspired by human brain, in other words, simulate biologic neural neural networks have significant place in artificial intelligence studies. In recent years, artificial neural networks have been used commonly in the economic and financial areas. The purpose of this study is to evaluate forecast results of exchange rate that are one of the macroeconomic variables by means of an application of artificial neural networks and VAR model. The twelve month data for interval 1992-2008 are separated as test data and train data, the eleven macro economic variables which affect the fluctuations of exchange rate are taken into consideration. The multi-layered feed forward neural network architecture and back propagation algorithms are employed. The data used as input for the network is normalized by "linear normalization method". The forecasting results by the artificial neural network technique are compared with the forecasting results of VAR method. It is shown that the conclusion in this way, the artificial neural network technique is better forecast modeling one than the other. The casual relations between exchange rate and the other parameters that are used in the thesis are surveyed. The results indicate that there are unidirectional causalities from exchange rate to republic gold, current account deficit, IMKB index, M2 money supply and interest rate. The variables of impulse-response functions in forecast modeling are also analyzed that one unit shock given to exchange rate upon the response of the other parameters appear unstable tendency, so sudden changes in exchange rate can cause instability on the market, in this context, the conclusion reached that the excessive fluctuations have to be prevented in order to decrease instability on the market. Keywords : Exchange Rate, Forecasting, Artificial Neural Network, VAR Modelling, Granger Causality Test, Impulse-Response Function.
Tez (Doktora) - Süleyman Demirel Üniversitesi, Sosyal Bilimler Enstitüsü, İşletme Anabilim Dalı, 2009.
Kaynakça var.
Artificial Neural Networks rectified which are inspired by human brain, in other words, simulate biologic neural neural networks have significant place in artificial intelligence studies. In recent years, artificial neural networks have been used commonly in the economic and financial areas. The purpose of this study is to evaluate forecast results of exchange rate that are one of the macroeconomic variables by means of an application of artificial neural networks and VAR model. The twelve month data for interval 1992-2008 are separated as test data and train data, the eleven macro economic variables which affect the fluctuations of exchange rate are taken into consideration. The multi-layered feed forward neural network architecture and back propagation algorithms are employed. The data used as input for the network is normalized by "linear normalization method". The forecasting results by the artificial neural network technique are compared with the forecasting results of VAR method. It is shown that the conclusion in this way, the artificial neural network technique is better forecast modeling one than the other. The casual relations between exchange rate and the other parameters that are used in the thesis are surveyed. The results indicate that there are unidirectional causalities from exchange rate to republic gold, current account deficit, IMKB index, M2 money supply and interest rate. The variables of impulse-response functions in forecast modeling are also analyzed that one unit shock given to exchange rate upon the response of the other parameters appear unstable tendency, so sudden changes in exchange rate can cause instability on the market, in this context, the conclusion reached that the excessive fluctuations have to be prevented in order to decrease instability on the market. Keywords : Exchange Rate, Forecasting, Artificial Neural Network, VAR Modelling, Granger Causality Test, Impulse-Response Function.