Robust statistical interactive forecasting for Windows-based PCs
Interest in forecasting packages is on the upswing for several reasons:
The number of procedures for generating forecasts is increasing, the packages
are becoming easier to use, and faster, cheaper computing power and storage
capability allows users to tackle larger data sets and get the results back
faster than ever before. Finally, as lead times shorten (because of time-based
competition and Just-in-Time practices), the general accuracy of the result
forecasts is increasing.
In this month's column, guest reviewer Rex Sprietsma examines Forecast Pro
for Windows Batch Edition, from Business Forecast Systems, Inc.
-- Steven Melnyk, Software Review Editor
By Rex Sprietsma
Forecast Pro Batch for Windows is a fairly sophisticated software package
that facilitates the forecasting of hundreds, even thousands, of items automatically
by incorporating a variety of statistical methods, including moving average,
exponential smoothing, Box-Jenkins models, events models, multiple-level
models and Croston's intermittent demand model. The first three are most
frequently used to capture and extrapolate patterns found in historical
data. Event models are used to capture responses to promotions and other
unusual events, while Croston's model is used to forecast items where the
demand is often zero. Finally, multiple-level models are used to organize
the data into hierarchies and exploit relationships between the various
levels.
Augmenting these various methods is a built-in expert system that helps
in selecting the most appropriate forecasting technique for each variable
or individual item. This expert system works by applying a combination of
rule-based logic and testing to each item to be forecast. This package is
particularly well suited to the batch processing requirements often found
in corporate and manufacturing environments. Some very nice features make
it a versatile forecasting tool for users with varying levels of expertise.
It allows the user to manually select different forecast models, make individual
forecast adjustments, and perform data experimentation.
Product performance:
This review is based on the writer's experience with the software at Babson
Bros. Co., which began using the DOS version of Forecast Pro Batch program
back in 1986. Our primary purpose at that time was to find a cost-effective
program to create a 12-month forecast for each of about 4,000 service parts,
every month, which could be used by the planners and buyers in making purchasing
decisions.
Performance was tested using a Unisys MDP 5605 equipped with a Pentium 66
MHz chip and 32 MB RAM. The software was also evaluated using a Unisys 486DX
with 16 MB RAM. The program ran well on both, and obviously the number crunching
is performed in less time as the computer processor speed is increased.
A run of 10,000 monthly time series with three years of history took 74
minutes to forecast on a 486/66. The same run took seven minutes on a Pentium/133.
Overall, I have been pleased with the performance of this program. If possible,
I highly recommend attending one of the seminars offered by BFS. They are
informative and provide a sound base of understanding for the use and capabilities
of the forecasting software package.
Documentation
Eric Stellwagen and Robert Goodrich, authors of the manual, have done a
nice job in explaining how to set the program up on a PC, the application
and use of the program, and the theory behind the statistical models, diagnostics
and methodology used in Forecast Pro Batch.
The manual is divided into three parts. Part 1 covers system requirements
and product support, the installation procedure and an overview of forecasting.
Part 2 includes a series of six lessons that demonstrate basic and key features
of the program. Part 3 includes references and appendixes, and covers the
selection of an appropriate data format, setting up a forecast script, a
complete reference of the menus, options and commands, and finally a detailed
statistical reference, which may prove a bit challenging for non-statisticians.
Fortunately, it is not necessary to understand the statistical theory to
use this software effectively.
The Forecast Pro Batch manual includes a glossary, bibliography and index.
In addition, ease of use is enhanced through the use of a graphical user
interface with pull-down menus, a tool bar with icons for commonly performed
tasks (define the script, read in the data, generate the forecast, adjust
the forecast, save the result, and so on) and a complete, context-sensitive
help system.
Technical support
My experience with technical support has been very positive. In dealing
with forecasting software, many times the data set being used is the source
of the problem, so in some cases it is easier for BFS to diagnose a problem
by trying to replicate the error on their end. While this may have required
a several day turn-around a few years ago, it can be performed much more
quickly today by sending data set files via the modem or the Internet.
Business Forecast Systems maintains a support staff that is trained in the
use of the software and also in statistical forecasting. This enables them
to respond quickly and efficiently to both operational and statistical questions.
Support is available by phone, fax, mail and over the Internet.
Best features
The software boasts many fine features. I'll note three of my favorites.
First is its built-in expert system that automatically matches the best
statistical forecasting technique with the historical data set of each variable
to come up with a forecast. This software looks for data properties that
lead to a specific technique and then, if the results are ambiguous, try
the remaining viable techniques and measure their performance. The objective
of this expert system is to pick the optimal technique.
Second among the noteworthy features is the ability to view graphically
the historical data and forecast and the statistical detail provided. Third,
is the flexibility built into many aspects of this program, from manual
forecast selection override and setting the forecast horizon to creating
forecast scripts and building event models.
Suggested improvements
Although the number of variables this program can forecast is unlimited,
if you want to make manual adjustments to the forecast of specific variables
and then view the results on a graph, it is necessary to limit the size
of the variables in a file to 1,024 or less. In practice, it makes sense
anyway, because it takes time to step through each graph. If you are completely
satisfied with the automated forecast output and do not have to make manual
adjustments, it is possible to forecast files with a larger number of variables.
I still prefer to use one input data file and create one forecast output
file. When necessary, we split the historical data into four or five separate
files and forecast each separately, so that adjustments can be made to an
individual forecast. Another feature I would like to see is the ability
to freeze a forecast, which would prevent portions of a forecast from being
overwritten.
Rex Sprietsma is the market research manager at Babson Bros. Co., Naperville,
Ill., a manufacturer and distributor of dairy farm equipment, chemicals,
and supplies. He has an MBA from The University of Illinois at Chicago.
Product Summary
Forecast Pro Batch Edition for Windows
Type of program:
Statistical interactive forecasting
Vendor:
Business Forecast Systems 68 Leonard Street Belmont,
MA 02178
