Prediction of Inflation Rates in Iran Using Dynamic Artificial Neural Network (Time Series Approach)

Prediction of inflation is of great importance for policy making. This need has led to extensive application of various models to forecast inflation rate. The competition between alternative applied forecasting models has led to further development in this area. The present study has applied artificial neural networks for prediction of inflation rate. The applied neural network is a multi-layer based on time series inflation approach during 1959-2007 using different algorithms of back propagation. To evaluate the artificial neural network performance, the prediction error criteria has been used. The findings of the research showed that the most appropriate networks are those based on Levenberg-Marquard training algorithm in which neurons in the hidden layers use nonlinear activation functions, and in the output layer use linear activation functions, and the number of neurons in each layer is chosen optimally. According to this preferred network, the inflation rates are predicted to be 10.59-21.99 during 2008-20012.