Evaluation of intelligent forecasting methods in terms of accuracy and bullwhip effect: An application in chemistry sector

The initial step of production system is demand forecasting which is significant for protecting stock levels, providing better customer service, improving capacity usage and increasing profits. Until today various traditional forecasting methods have been introduced as wells as complex ones for demand forecasting. The accuracy of forecasting mechanisms differs from organization to organization, depending on the structure of sales data of the related product to be forecasted. Robust forecasting methods having high accuracy should be selected to realize the organizational goals. Following the selection of forecasting method, additional enhancement is provided by analyzing the structure of the method.Beside the forecasting accuracy, one another subject to be taken into consideration is bullwhip effect to achieve organizational goals. Bullwhip effect is directly affected by demand forecasting method. In a smooth supply chain, the magnitude of bullwhip effect is desired to be possibly small. In this context, the method should operate to have most accurate forecasts as well as the least bullwhip effect.In this study, historical sales data of a new product in chemical industry is used to forecast future demand by using robust intelligent forecasting methods including artificial neural network, grey theory and fuzzy time series, rather than the traditional ones such as time series and casual methods. In application of these methods, sensitivity analysis are made among their structural features and then parameters and coefficient analysis are investigated to obtain more accurate results. Performance of the methods are determined by mean absolute percentage error which is a common robust measure recognized by scientists and researchers. Afterwards, in a single supply chain consisting of a single manufacturer and single retailer, the magnitude of the bullwhip effect arising from the methods applied are computed using the 1-1 stock policy functioning as if all stages in the chain work on the principle that they will keep in stock one period's and the 2-2 stock policy functioning by keeping two period?s demand. Thus the intelligent forecasting methods are compared in terms of forecasting accuracy and bullwhip effect considering the improvement ways.