مقايسه قدرت پيش بيني روش شبكه عصبي مصنوعي با ساير روش هاي پيش بيني: مورد قيمت چغندرقند

Comparison of forecasting ability of artificial neural network with other forecasting methods: case of sugar beet price

The aim of this study was to forecast nominal and real price of sugar beet and to compare forecasting ability of artificial neural network method with other forecasting methods. The stationary of the series was tested and then, in order to investigate whether series are stochastic, nonparametric test of Vald-Wulfowitz and parametric test of Durbin-Watson were applied. Based on the above tests results, nominal price of sugar beet were recognized non-stochastic and predictable, while the real price series was found stochastic. The study period covers 1971-2005. The models used for forecasting were autoregressive, moving average, ARIMA, Single and Double exponential smoothing, harmonic, ARCH and artificial neural network. Based on the lowest forecasting error criterion, harmonic model forecasted nominal price of sugar beet with lowest forecasting error. The amount of nominal series forecasted by different models was at range of 344000-396000 and 398000-448504 rials per ton for 2004 and 2005, respectively. The happened values of nominal price series for 2004 and 2005 were 387200 and 447000 rials per ton, respectively.