A comparison between the power of artificial neural network models and dynamic neural network in predicting exchange rate: an application of wavelet transformation

Subject Areas : Futurology

Mohammad Ali  Khatib Semnani ¹ (استادیار)
Manijeh  Hadinejad ² (استادیار)
Roxana  Khoshouie ³ (کارشناس ارشد اقتصاد، دانشگاه آزاد اسلامی واحد علوم و تحقیقات)

Received: 2013-11-15 Accepted : 2014-03-07 Published : 2015-11-23

Keywords: exchange rate, Neural Network, MFNN, NNARX, wavelet decomposition,

Abstract :

The present study is an attempt in applying the combination of dynamic neural network and decomposition of wavelet in order to make possible the selection of an optimized pattern for predicting considered variable. For the purpose of research, monthly time series of exchange rate from April 1998 to December 2012 were used including 177 observations from which 150 observations were used for modeling purpose and 27 observations were used for simulation or in other words for presenting predictions out of samples. The findings of present study imply that firstly, dynamic neural network models compared to feed-forward multilayer neural networks have better performance in predicting exchange rate out of sample, based on both criteria for prediction error calculation: MSE & RMSE and secondly, applying wavelet decomposition technique improves prediction results of mentioned models based on both criteria. The third point is that among mentioned models, the best result belongs to predictions obtained from dynamic neural networks based on decomposed data by wavelet technique. Therefore, applying this combination of models as an optimized combination is suggested to monetary researchers, analysts and decision makers of country.