No Title

4 Application to a large-scale model

The results reported in this section were obtained by IMIS from a data set containing 150 observations of price and volumes for each of some 60 brands of alcoholic beverage sold through supermarkets in one metropolitan area. The linear statistical analysis alone, conducted in a spreadsheet environment, takes a day in the hands of a domain expert with extensive experience in developing and using these techniques. Further time and effort are obviously required for the interpretation of the results and the identification of effects captured by IMIS's non-linear analysis and knowledge-based interpretation. The results reported here were generated by IMIS running on a Sun SPARCstation 5 in less than 30 minutes.

In this case, IMIS found the most important date of all to be date 50 which, from figure 2 is clearly seen to be the date when the stable coefficient pattern at the start of the observation period became turbulent. The transcript note was:

It turned out that at about that date there was a reduction in price competitiveness in the market. IMIS also identified the longest period of high turbulence as being from dates 58 to 60 inclusive.

The changing competitive structure was identified in the following transcript segment:

This amounts to a consistent set of shifts within the competitive space. The focus brand has moved closer to Brand T and away from three other brands (H, G and B). It has also moved closer to Brand S and away from Brand C as well as Brand G. Finally, it moved towards Brand A and away from Brands H and B. The movement away from Brands H, G and B is thus confirmed by virtue of two independent findings. Whether these changes can be explained depends on the domain experts' judgements about the characteristics of these brands.

No Title - 23 AUG 96

[Next] [Previous] [Top] [Contents]

Generated with CERN WebMaker