Steven P. Schnaars

How to develop and use scenarios

Abstract Scenario analysis is an increasingly popular way to look at the future business environment. This paper provides a critical assessment of the literature on scenario analysis. It summarizes what is currently known about this approach to forecasting, and offers some guidelines regarding the construction and use of scenarios. It also offers a comparison and evaluation of many of the techniques that have been proffered to generate scenarios, suggesting which are worth while and which are not.

The use of multiple scenarios in sales forecasting: An empirical test

Abstract Nearly 500 annual sales forecasts were generated from the responses of 82 subjects who were presented with either a time-series plot of historical sales data by itself or with the same plus three scenarios, and were then asked to make forecasts. Sales forecasts were made in either a stable or an unstable environment. The findings did not support the claims made by scenario advocates. Scenarios did not make unexpected outcomes less surprising. Instead, scenarios were found to increase confidence in a favored forecast. Furthermore, no support was found for the contention that scenarios improved upon ‘eyeball’ extrapolations or made judgmental sales forecasts more accurate than quantitative extrapolations. Scenarios were found to be tainted by many of the same biases previously identified by cognitive psychologists.

Extrapolation models on very short-term forecasts

This study compares the predictive accuracy of seven popular extrapolation models on 180 unconditional forecasts of 15 economic indicators reported at very short intervals (i.e., weekly). In total, over 1200 forecasts are generated. The models tested range in complexity from a simple random walk model to generalized adaptive filtering. It is shown that movements in series collected at such short intervals are dominated by random fluctuations. They appear to follow a random walk. As a result, extrapolation models, especially complex approaches, provide forecasts that are less accurate than are those provided by a simple random walk model. It is argued that shortening the interval at which a data series is collected (e.g., from quarterly to weekly) is an inappropriate means of dealing with the uncertainties encountered in longer-range forecasts. Instead, it merely trades one problem for another.

An Application of a Segmentation Design Based on a Hybrid of Canonical Correlation and Simple Crosstabulation

This research illustrates a practical application of a segmentation scheme based on a combination of canonical correlation and crosstabulation. A sample of readers of a popular genre of novels is segmented on the basis of psychographic, demographic, and activity measures. Four distinct consumer segments are postulated-movers and shakers, isolated readers, young swingers, and laggards. The characteristics that differentiate each of these segments are also presented.