Application of NNARX to agriculture sector value added forecasting: A case of Iran’s agriculture sector

Abstract

This study firstly uses the Cobb-Douglas production function and Auto-Regression Distributed Lag (ARDL) approach for estimating the long-run function of Iran's agriculture sector value added and then compares the forecasting performance of specified ARDL model with Neural Network Auto-Regressive model with eXogenous inputs (NNARX) using forecasting performance criteria (R 2 , MAD and RMSE). The results of ARDL specification indicated that 1% increase in labor, capital and energy factors will increase Iran's agriculture sector value added 0.36, 0.23 and 0.32%, respectively. Also, the results of forecast performance criteria show that NNARX nonlinear model forecasting performance for Iran's agriculture sector value added is better in contrast with the ARDL linear model because (1) The Root Mean Square Error (RMSE) and Mean Absolute Deviation (MAD) divided are less than 1 and (2) The R2 divided is more than 1. Therefore, according to the importance of the agriculture sector as the main alimentary source for mankind, accurate prediction of agriculture sector value added for its using new NNARX model is strongly recommended to the agriculture sector policy makers.