Determining the optimal forecasting combination of the four-level supply chain to minimize the bullwhip effect

Bullwhip effect that occurs in the chain, leads to inefficiencies such as excess inventory and overdue orders during the chain. These problems can be reduced by appropriate predictions. Forecasting must be done in all levels of a supply chain. This paper addresses the problem of optimal combination of forecasting to reduce the bullwhip effect in the four-level supply chain. For this purpose, a four-level supply chain is considered. One of the methods such as moving average, exponential smoothing, linear regression and multilayer perceptron artificial neural network can be considered for predicting in each level. First, the desired supply chain is simulated for this means. The different combinations of aforementioned forecasting methods are calculated. Then a combination of forecasting methods according to minimized bullwhip effects is selected. Finally, the results are analyzed by variance analysis model. Two combinations have the lowest bullwhip effects. Moving average, neural networks, exponential smoothing and linear regression for levels: retailer, wholesaler, manufacturer and supplier respectively as an answer and the other is: moving averages, exponential smoothing, neural network and linear regression in the same mentioned levels and other combinations have less utility.