DECOMPOSITION - A STRATEGY FOR JUDGMENTAL FORECASTING

This paper reports the results of studies concerning the accuracy and efficiency of time‐series extrapolation decisions made with the assistance of an interactive graphical tool called GRAFFECT. The tool facilitates the decomposition of the extrapolation task by permitting the serial decomposition of the cue data as the task proceeds. GRAFFECT uses an interactive graphical interface controlled substantially with the use of a mouse. The extrapolation task is divided into the following: (1) trend modelling and extrapolation, (2) seasonal pattern modelling, and (3) extrapolation from the noise residual series. As each component is modelled its effect is stored and the information is washed out of the cue series. The ultimate forecast is produced by automatic recomposition of the judgementally determined components. The results show a significant improvement in forecast accuracy over unaided judgment, resulting in a subjective extrapolation that betters deseasonalized single exponential smoothing. Copyright © 1990 John Wiley & Sons, Ltd.