Demand forecast accuracy and performance of inventory policies under multi-level rolling schedule environments

Our incentive is to study the behaviour of lot-sizing rules in a multilevel context when forecast demand is subject to changes within the forecast window. To our knowledge, only Bookbinder and Heath (1988) proposed a lot-sizing study in a multi-echelon rolling schedule with probabilistic demands. But their simulation study was limited to two arborescent structures with 6 nodes. By means of an extensive simulation study we show that it is always worth decreasing the error magnitude. This should encourage companies to develop Electronic Data Interchange to ameliorate demand forecast. Although the presence or absence of forecast errors matters more than the error level, we show that lot-sizing rules exhibit significant differences in their behaviour as the level of error is augmented. This paper also provides a clear description of the rolling procedure when applied to general product structures, probabilistic demand within the forecast window and positive lead times.