Forecast Pro Presages Data - 2/21/00

By Geoffrey Hollander, For InfoWorld Test Center

Forecast Pro XE, Version 4.0 BUSINESS CASE Companies that need to accurately predict sales, production, or inventory needs will appreciate Forecast Pro XE's ability to make accurate predictions from past performance figures. TECHNOLOGY CASE Forecast Pro XE produces statistically sophisticated and testable projections for nonstatistician decision-makers while retaining its usefulness to accomplished statisticians who require complete customization. PROS + Easy-to-read graphical displays can be created with one click + Forecasts testable against actual performance + Numerous statistical methodologies supported + Drag-and-drop single-point editing CONS - Runs only on Windows platforms   PLATFORMS Windows 95/98, Windows 2000, Windows NT Business Forecast Systems Inc.Belmont, Mass.; (617) 484-5050 http://www.forecastpro.com/

MOST BUSINESSES rely on guesswork when forecasting future sales, production, or inventory figures. Some sacrifice accuracy simply by tacking a percentage onto last year's figures; others turn to the marketing or sales team, where forecasting is often confused with goal-setting.

On the other hand, statistical forecasting is systematic and objective, but it demands a great deal of time and a skill set that most companies cannot afford to nurture in staff members. Business Forecast Systems puts an end to the forecasting dilemma with the release of Forecast Pro XE, Version 4.0, a system that uses proven, statistical forecasting techniques. Its accuracy and broad support for various statistical models earned it our top score of Excellent.

Forecast Pro XE supports a broad range of forecasting methodologies (including, but not limited to, simple moving averages, curved fitting, Croston's intermittent demand model, exponential smoothing, Box-Jenkins, and dynamic regression) that statistically create quantitative models from past patterns, preferably using several years' worth of data. The accuracy of your forecast depends on the degree to which the statistical model can detect and extract statistical patterns from the historic data. ; the stronger the patterns, the more accurate the forecast.

Forecast XE's Expert Selection feature is the heart of this application and the most useful feature for nonstatistician users. This exceptionally easy-to-use option works the same way an actual statistical analyst would, performing a series of statistical tests, interpreting the results, performing additional tests if necessary, and making recommendations based on the findings.

Expert Selection begins by analyzing the data. Once that analysis is completed, it generates an audit trail -- a report that shows the various mathematical methods the program has applied to the data, what the level of error is likely to be, the final choice of statistical modeling method, and the forecasted results.

Although Forecast Pro can work with as few as five data points, the more data you enter, the more comprehensive and useful your forecasts will be. For instance, Forecast Pro cannot automatically build a seasonal model with fewer than two years' worth of data, because it requires at least two samples for each month or quarter of the year to distinguish seasonality from irregular one-time patterns.

Once the audit trail is complete, you can switch to a graphical view of the forecast. The graphed data appears as a line chart with clearly differentiated lines for historical data, projected data, and upper and lower confidence levels. The user can define the confidence levels, which are meant to give an acceptable range of deviation. They indicate the probable best-and worst-case scenarios in the event that your actual results deviate from the projection.

You can test the accuracy of your projection model by creating a "holdout." In a holdout, the user gives Forecast just a subset of the total data set and asks the product to project for the data that was held back. In this way you could coax the program into calculating "projections" that could then be compared to the actual results of the same period. A graph would clearly display actual performance vs. what Forecast Pro would have predicted for that period of time. This excellent feature allows users to test their data models. If Forecast's projection differed wildly from the actual results, you would know that some hidden causality, such as a labor strike, needed to be built into the data model.

Most forecasting packages make a distinction between interactive forecasting (single variable) and batch forecasting (several variables forecast together). Interactive forecasting is practical when there are a limited number of items to be forecast because it focuses individual attention on each variable. Forecast Pro XE was designed primarily for interactive forecasting, but it does provide batch forecasting for up to 100 variables, which you can view collectively, individually, or in any combination. Batch forecasting is useful for tracking trends across related product lines, but the 100-variable limitation could be a problem for larger companies that wish to make use of batch functionality.

Forecast Pro XE gets an Excellent rating for taking the guesswork out of guessing with elegance and accuracy. Companies that use forecasted data to make purchasing and budgeting decisions will benefit from Forecast Pro's accuracy and moderate price.

Geoff Hollander (gch@nwdatabase.com) owns and operates Northwest Database Services, in Lake Oswego, Ore.