New forecasting insights on the bullwhip effect in a supply chain

Abstract

This paper investigates the effects of various forecasting methods, such as moving average (MA), exponential smoothing (ES) and linear regression (LR) on the bullwhip effect in a four-echelon supply chain. In order to compare the effectiveness of these forecasting methods, a bullwhip effect measure is utilized for two cases: one where the demand is non-trended and one where it has an increasing trend. The forecasting methods in these two cases are ranked based on the size of bullwhip effect they cause, and a statistical method is utilized to ascertain the significance of the differences among these effects. The results obtained show that the LR, double MA and double ES methods result in the least, the second least and the largest bullwhip effect, respectively, in the case of trended demand. Moreover, the simple MA, single ES and LR methods cause the least, the second least and the largest bullwhip effect, respectively, in the case of non-trended demand. Finally, the results are generalized for other tree-like supply chains using an analytical representation.