A major use of univariate forecasting methods lies in production control where there is a large number of series to be forecast. The appropriate choice of forecasting method has the potential for major cost savings through improved accuracy. Where a new method is to be compared to one already established two distinct approaches to selecting between the two can be considered: aggregate selection, where a single method is applied to all the time series, versus individual selection, where the particular method appropriate for each series is identified and used to forecast future observations for that series. This paper evaluates these two selection rules using 263 data series from a single organization. The results show the potential of ‘individual selection’ and also the difficulty of attaining it. For short lead times ‘aggregate selection’ achieves similar accuracy. For longer leads it is outperformed by ‘individual selection’ and also undermined by sampling variability. As a consequence, ‘aggregate selection’ must be carried out across a wide cross-section of series and across time. When this is done the results of this case study show that ‘aggregate selection’ is both simpler than and of comparable accuracy to ‘individual selection’.
