The effects of forecasting errors and mis-specification

This paper represents a long-standing collaboration between Robert Fildes and Brian Kingsman who sadly died before it could be completed. This paper proposes a methodology for examining the effect of demand uncertainty and forecast error on lot sizing methods, unit costs and customer service levels in MRP type manufacturing systems. A number of cost structures were considered which depend on the expected time between orders. A simple two-level MRP system where the product is manufactured for stock was then simulated. Stochastic demand for the final product was generated by two commonly occurring processes and with different variances. Various lot sizing rules were then used to determine the amount of product made and the amount of materials bought in. The results confirm earlier research that the behaviour of lot sizing rules is quite different when there is uncertainty in demand compared to the situation of perfect foresight of demand. The best lot sizing rules for the deterministic situation are the worst whenever there is uncertainty in demand. In addition the choice of lot sizing rule between 'good' rules such as the EOQ turns out to be relatively less important in reducing unit cost compared to improving forecasting accuracy whatever the cost structure. The effect of demand uncertainty on unit cost for a given service level increases exponentially as the uncertainty in the demand data increases. The paper also shows how the value of improved forecasting can be analysed by examining the effects of different sizes of forecast error in addition to demand uncertainty. In those manufacturing problems with high forecast error variance, improved forecast accuracy should lead to substantial percentage improvements in unit costs. Key Words: forecast error / MRP/ lot sizing/ buffer stocks and service levels/ demand uncertainty/ value of forecasting/ manufacturing systems/ forecast mis-specification 1