Conditions when market share models are useful for forecasting: Further empirical results

The increased availability of data and access to computers has meant that econometric methods are readily available to model and forecast market share. However, controversy exists over their usefulness. For example R. Brodie and C.A. de Kluyver’s review of empirical studies revealed that the predictive accuracy of causal market share models was not consistently better than that of a naive model. In contrast, V. Kumar and T.B. Heath found that causal models consistently outperformed the naive model when using aggregated weekly scanner data which allowed for more observations. This paper reports the results of a replication and extension study which confirms Kumar and Heath’s findings. However, the increased accuracy from using the causal model is diminished considerably when the more realistic situation of forecasting competitive action is included. The paper concludes by outlining a research agenda aimed at further clarifying the conditions when market share models are useful for forecasting.