Decision support for the decomposition of event information in judgemental time series forecasting

Substantial recent emphasis has been placed on the importance of discontinuities and contextual information in time series forecasting. To accommodate these factors, statistical forecasts are often judgementally adjusted in an ad hoc, informal manner that is prone to bias, especially when the model is not understood. The paper reports an alternative approach. An interactive graphical decision aid was developed, enabling the forecaster to: decompose discontinuities for contextual ´event´ information; forecast from the resultant simplified series; and anticipate future events in the forecast. Empirical tests showed a significant improvement in forecast accuracy over unaided judgement. This effect occurred for series with upward and downward slopes, and with small and large numbers of special events.<>