Artificial Neural Network (ANN) and Co-integration Methods (ARDL & Johansen-Juselius) Approach for Price Forecasting of Chicken in Iran

Regarding the importance of price forecasting of the protein products including chicken, this research uses methods of ARDL, Johnson-Juselius and ANNs to forecast the chicken price in Iran for the various time paths consisting of one month, six months and twelve months. Accordingly, the main hypothesis relies on the more efficiently of the ANNs than those of the other econometric methods. Monthly data are collected for the domestic resources related to the agricultural sector for the period March-1991 to February-2006. The data from March-1991 to February-2005 is used for models estimating and networks training and the rest data is used to past prediction power evaluating. The empirical results obtained confirm that the performance of the three layers Elman ANN with eight neurons in input layers, three neurons in hidden layers and sigmoid activation function (for the time path of twelve months) and a three layers Elman ANN including seven neurons in hidden layers with hyperbolic tangent activation function (for the time path of one month) in forecasting has been more precise than that of the Co-integration methods. But in the time path of six months, ARDL method is more precise than that of the Elman ANN. Implication is that the use of modern methods such as ANNs in prediction of the chicken price is able to affect policymakers in the poultry industry toward making better decisions in the market.