Modeling and Forecasting Exchange Rates Using Econometric Models and Neural Networks

Abstract

Considering the importance of the exchange rate in economic policy, a variety of patterns are presented to explain how to determine the behavior of the exchange rate and how the modeling and forecasting is. In this context, the present study with a new approach to this problem has investigated the time-series’ nature of exchange rate and performed a non-linear test for daily data of exchange rate in the period of 2003 to 2006 to explain the behavior of exchange rate using time-series regression pattern. So in this study, it was used an artificial neural network modeling in addition to daily modeling and forecasting the exchange rates and minimizing the forecasted error by this method and the results were compared with forecasted values by ARIMA model based on measuring criteria of forecast accuracy. It has been used 80% of date equivalent to 1160 days from March 25th of 2003 to June 5th of 2006 for training the models and in order to evaluate the sensitivity of the results of model in proportion to the exchange rate, the model was estimated with the same way for three categories of data as the exchange rate of the dollar, euro and pound. The results showed that the used neural network had the more forecasting power than ARIMA model and the exchange rates’ price of the euro and pound were as a function of their previous days and the exchange rates’ price of the dollar was as a function of the price in last 6 days. What distinguishes this study compared to other studies is the unique design of artificial neural network that can approximate any arbitrary function and obtain any amount of accuracy that is needed in addition to minimizing the forecasting error by concerning the activation function which is applied in it