Managerial intervention in forecasting. An empirical investigation of forecast manipulation

There has been a continuing debate in the forecasting literature concerning the relative advantages of quantitative versus qualitative approaches to forecasting. However, although a considerable number of studies have contrasted the merits of the two approaches, relatively few efforts have investigated the application of human judgement on forecasts generated by quantitative forecasting models. This study is an empirical investigation of the effects of human intervention on forecast accuracy. It examines the consequences of managerial manipulation of sales forecasts initially produced by a smoothing model. A total of 281 products are investigated using multiple measures of forecast accuracy and the effects of subjective revision are discussed in terms of size, directionality, and distribution of errors. The results indicate that human intervention can lead to an overall improvement in forecast performance as reflected in a reduction in the variability and absolute size of forecasting errors. The results also show that more forecasts are improved by manipulation than are degraded. Finally, there is some indication that subjective revision may result in an increase in overall forecasting bias. These findings appear to conflict with previous evidence, however, the empirical setting of the present study is specific to shortterm sales forecasting which is not strictly comparable with the settings of earlier efforts.