Abstract
Demand fluctuation can be separated into trends(linear and non-linear), cycles(including seasonal fluctuation), and white noises. The trends and cycles can be subtracted by demand forecasting, and it is possible to take action previously for trends and cycles in procurement, production, delivery, etc. Meanwhile, safety stock covers uncertain demand caused by white noises. However, if the patterns of trends or cycles vary dynamically, it is difficult to substract them using a forecasting model. Two approaches are considered to cope with this problem. The first approach is to revise the forecasting model progressively, and the second approach is to vary safety stock dynamically for a given forecasting model. Since the first approach is time consuming and it is some times difficult to identify changes in trend patterns, we propose a simple second approach or method, varying safety stock dynamically according to forecasting error and bias during the latest several periods. This proposed method might also be an assistance of the first approach. To justify the proposed method, the average inventory level and stock-out ratio of four different methods are compared in numerical simulation under the condition of (R, Q) policy and simple moving-average forecasting. Method A determines safety stock based on the mean and variance of demand without demand forecasting, and methods B, C, D determine safety stock using forecasting error. Forecasting error in method B is supposed to be zero. Method C ignores the bias of forecasting error, and method D, the proposed method, determines the safety stock using the bias and variance of forecasting error. The simulation results showed that the proposed method has quite good performance for all cases of demand patterns, including stable demand, increasing trend, decreasing trend and Bass model trend, while method B, which is supposed to have zero bias of forecasting error, gives the best performance.
