Performance evaluation of artificial neural network-autoregressive integrated moving average (ARIMA) hybrid model in forecasting of Iranian saffron export

In terms of quality and quantity, Iranian saffron has a considerable position at the international level and by taking advantage of the existing capacity; we can significantly increase the export earnings from it. On the other hand, sales forecasting based on time series analysis is s a very important element for the designing and implementing of marketing strategies in the international arena. However, the conventional approaches to forecasting, by ignoring the linear (or nonlinear) structure of data, do not provide accurate results. Therefore, the main objective of this study is to design a hybrid model consisting of two methods, artificial neural networks (ANN) and autoregressive integrated moving average (ARIMA), in order to overcome the deficiencies and the use of the unique features of the each of these methods. Using the data related to the export of Iranian saffron during the period 1904-2013, the results of the study showed that the ARIMA–ANN hybrid model is stronger and better performance than ARIMA and ANN individual models in order to forecasting of Iranian saffron export. Therefore, given the considerable performance ARIMA–ANN hybrid model, the use of this model is recommended in setting strategies related to the export and also in the forecasting of the forecasting of time series variables.