An Adaptive Forecasting Algorithm
Hugh F. VanLandingham, Richard L. Moose · 2005
The forecasting of discrete-time random processes with nonstationary statistics has long been recognized as an important problem [1,21. In particular, inventory control in e.g. retail stores with a large variety of items relies on the manager's ability to forecast future demands based on periodically spaced ac- counts of the actual demand which themselves show wide variation. Many forecasting algorithms have been presented in the literature [1-41. The varying degrees of complexity depend to a great extent on the model used for the demand process. One popular assumption on the demand history for items held in inventory is that the underlying process is a continuous, piecewise linear process, i.e. reflecting trend changes in de- mand. Perhaps, a more realistic model would allow for inaccuracies and complexities of the actual demand process by including the effects of both correlated and uncorrelated errors in the measured sequence of demand data. With this background the next section presents a reasonable stochastic model for the demand process and an explicit statement of the estimation problem to be investigated in this paper.