Time series characteristics and the widths of judgemental confidence intervals

Abstract

This study investigates the task based determinants of confidence interval estimates in judgemental time series forecasting. It examines the properties of time series, develops metrics to quantify these properties and regresses the metrics against confidence intervals that have been judgementally estimated in two separate studies. Although the noise or randomness in the series should be the sole determinant of the width of the confidence intervals, results of the regression studies indicate that the levels of seasonality in the series provided the greatest explanation for the differing widths. Furthermore, the trend of the series also seemed to influence the size of the interval widths. Over both studies, the level of noise in the series only explained about 15% of the widths of the judgemental confidence intervals.