Joint optimisation of demand forecasting and stock control parameters

Exponential smoothing methods are very commonly used for forecasting demand in a supply chain context. When estimating the parameters used in these methods, a common practice is to optimise only the smoothing constants and not the initial parameter values. In this paper we show that if we treat initial values as well as smoothing constants as decision variables, a considerable reduction in forecast error can be achieved. Additionally, the optimisation of the forecasting method should not be treated separately from the production or inventory model in which forecasts are used. The case of a centralised supply chain with an order-up-to inventory policy shows that calculated forecasts of demand, determined by minimising mean absolute error (MAE) or mean squared error (MSE), are not optimal. Finally, a method for simultaneous optimisation of demand forecasting and a stock control policy is described. Initial and smoothing parameters in the forecasting methods can be determined to minimise the total costs.