Improving short-term forecasts

This empirical study compares the accuracy of combined forecasts, found by averaging individual forecasts from univariate time-series techniques, with judgemental forecasts actually made daily by experienced practitioners in real business settings. The value of judgment is assessed, used alone and in combination with quantitatively derived forecasts. The key finding is that the value of each forecasting approach depends on the characteristics of the time series, namely data variability. Automated quantitative forecasts are superior for time series that are relatively stable. Complete reliance on quantitative procedures is not only more efficient, but reduces forecast errors. However, as the volatility of the time series increases, a point is reached where judgmental inputs are desirable, either to supplement or even to replace the forecasts provided by quantitative techniques.