Exploring judgemental forecasting

Most of the present knowledge of the accuracy or goodness of human judgement has been gained from studies carried out in a setting of multivariate non-serially correlated cues. This is not representative of the task of time series forecasting where there is typically a single series of serially correlated cues. As judgement is widely used in this setting, this study seeks to investigate the extent to which some of the widely documented judgemental biases and heuristics apply to time series forecasting. The research design varied the series presentation, series length and the type of series to investigate the influences of presentation scale, length of series, recency and anchoring and adjustment in estimating a judgemental forecast. The time series used were modelled from a stationary ARMA process. The study found that while scale did not influence accuracy, series length and the most recent segment slope did influence it. Subjects’ forecasts could be modelled as exponential smoothing or anchoring and adjustment, where the anchor point corresponded to the long term average of the stationary series.