Predictive accuracy of simple versus complex econometric market share models
While econometric market share models have been shown to be useful to managers as descriptive tools, controversy exists over their use in forecasting. For instance, Brodie and de Kluyver (1987) showed, using data for 15 brands in three markets, that naive forecasting will often do better than econometric models when predicting market share. In the discussion of the paper Hagerty (1987) showed theoretically that these results were not surprising. This paper extends the analysis of Hagerty (1986, 1987) by deriving conditions under which naive econometric models are expected to do better than complex models when predicting market share. The results show that the naive model is preferred when the number of parameters used in the econometric model is too large or when the number of points used to fit the models is too small. We also show that the decision to go for the naive or econometric model is not greatly influenced by the number of points withheld for model validation, under the assumption of similarity of correlations between predictor variables in the estimation and validation data.