Forecasting of printing and writing paper consumption in Iran using artificial neural network and classical methods

The objective of the research is to forecast the trend of the printing and writing paper consumption in Iran for a five-year period using both modern and classical methods. In order to do the forecasting, predictability of time series was primarily studied using Durbin-Watson and Runs tests. Then, artificial neural network model (multilayer perceptrons (MLP)) and univariate and multivariate classical forecasting models such as univariate single exponential smoothing (SES), double exponential smoothing (DES), holt-winters exponential smoothing (HWES) and Box- Jenkins (ARIMA) models, and multivariate econometric model all together were compared in terms of the standard statistical measures. Finally, the consumption of printing and writing paper in Iran was forecasted up to the year 2017 using the most appropriate model. The results of both the parametric test of Durbin-Watson and non-parametric test of Runs show that, the printing and writing consumption series is non-random and predictable. The results of comparing different forecast methods showed that the artificial neural network model has higher forecasting accuracy than the classical models and it is more appropriate for the five-year forecast period. Also, the results of forecasting by using neural network model (MLP), revealed that the printing and writing paper consumption in Iran is forecasted to increase by 5.3%, from around 375 thousand tons in 2012 to 420 thousand tons in 2013, but it falls over the five-year forecast period, from 5.3% in 2013 to 0.07% in 2017.