[Review written with Roderick J. Brodie]
To what extent has research on forecasting improved our ability to make forecasts of market share? Results from Brodie and de Kluyver (1987) were disappointing in that naive models were shown to be as accurate as sophisticated models. The commentary following the Brodie and de Kluyver paper discussed conditions under which sophisticated models might be superior. Alsem et al. draw upon the commentary to make a more detailed examination of this issue. The result is a well designed anokd carefully conducted study that examines ways to improve market share forecasts.
Alsem et al. examined a sophisticated model to forecast market share instead of simply using a naive projection of the latest value. Furthermore, because competitors actions might be expected to affect market share, they developed a sophisticated model to predict competitors actions, rather than merely assuming that competitors will continue to behave as they have in the past. This model predicted competitors actions with respect to price, advertising, store distribution, and shelf-space. They tested these approaches using data for nine brands from three markets. Like Brodie and de Kluyver, these researchers used aggregate bimonthly data. These series consisted of 24, 30, and 36 periods. Forecasts were made for six bimonthly periods for each brand. The results proved to be disappointing. They were as follows: (1) market share forecasts were not very sensitive to the assumptions that are made about competitors future marketing behavior; (2) sophisticated predictions of competitors behavior were not more accurate than naive predictions; (3) knowing the competitors actions did not help to produce more accurate forecasts of market share (that is, the ex post forecasts were not more accurate than the ex ante); and (4) sophisticated methods did not yield forecasts of market share that were more accurate than a naive (no change), prediction. The last conclusion held even for brands with high variation in market share.
It is difficult, then, to find situations where sophisticated methods yield more accurate forecasts of market share than can be obtained from naive forecasts. Kumar and Heath (1990) offer some hope. Rather than using data aggregated over two months they used weekly scanner data Their econometric forecasts tended to yield more accurate forecasts; however, their test was limited to ex ante forecasts of only two product categories, each from a single point in time, and the results were not shown to be statistically significant.