Classic Computer Magazine Archive CREATIVE COMPUTING VOL. 11, NO. 1 / JANUARY 1985 / PAGE 128

Business forecasting software in brief. (evaluation)

Available for both the Apple and IBM, Visi Trend/Plot is a combined time series analysis and paragraph plotting system. Without a doubt, it is the most attractive package for use by those who need both forecasting tools and presentation graphics. (Statpro also excels in graphics but is much more expensive.)

As the comparison chart details, VisiTrend/Plot is a complete aid to forecasting. It includes menu-selected options for regression, moving averages, trend analysis, smoothing, and descriptive statistics among others.

The data editor (called a "storage management program") is simple to use and complete. It allows many transformations of the data selected directly from the menu (sums, ratios, leads, lags, logs, or any mathematical or logical transformation defined by a user) and displays the results (with scrolling) of those operations for a quick check on accuracy. The storage management program can read and save DIf files, but it is unable to read or store standard text files (this may make it more difficult to use VisiTrend/Plot if you routinely download data from a mainframe in a standard text file format).

One of the strong features of VisiTrend/Plot is the plotting program which produces all the standard business graphics: pie charts, scatter diagrams, bar charts, single and multiple line charts, and overlays. All this is done in color if you desire and all the graphics may be saved to disk or printed on a wide range of supported graphics printers (Epson, IDSM NEC, etc.). Since you may either automatically scale the plots or use your own scale and since you may also place labels anywhere on the plots, VisiTrend/Plot offers a superior method for presenting forecasts in visual format. Daisy

Daisy stands for Data Analysis Interactive System. It is a complete statistics package with some limited graphics capability available only for the Apple II computers. At $199.95 it is a bargain for serious forecasters and others who require statistical processing.

Its most attractive feature is perhaps the ability to store and display data in spreadsheet format. With an Apple IIc or IIe with an extended 80-column card the spreadsheet is 10 columns by 830 rows and can be adjusted in size to incorporate more columns and fewer rows (e.g., 20 columns by 480 rows). Of course, you view only a portion of the large spreadsheet at one time, but you may scroll the sheet with the cursor control keys. As in VisiCalc this is the easiest way to view large datasets; a splitscreen mode allows non-adjacent columns to be placed next to one another for visual comparison.

Entry of data into the spreadsheet may take place by way of the keyboard or from disk using any DIF database. This means information residing in VisiCalc files can be directly used by Daisy. Data may be saved in standard Daisy files (which load quite rapidly), DIF files, or standard ASCII text files. The latter is quite useful if the data are to be transmitted by modem.

Transforming data in any way is quite simple with many selections available directly from a menu (lagging data, taking logs, etc.). Unusual transformations can be programmed by writing an extension to Daisy (complete instructions are in an appendix to the well written manual). Missing data are also handled.

While Daisy does not handle moving averages or exponential smoothing, it does handle just about every variation of regression analysis. Complete summary statistics are supplied for each regression. Forecasts using the regression equation are also available.

All output from regression (as well as the other statistical routines provided) may be printed. Some limited graphics may also be produced and viewed or printed (scatter diagrams, histrograms, sequence plots, etc.).

In addition, Daisy provides the following statistical routines: descriptive statistics, hypothesis testing, correlation analysis, crosstabulation, non-parametric corellation, and ANOVA.

The manual includes a tutorial, and there is a Help command available on disk for on-line instruction. The 80-column display is used in many instances if your machine has that capability; otherwise, the program reverts to 40-columns. MicroTSP

MicroTSP will be recognized by some business school graduates as a microcomputer version of the TSP (Time series processor) package available on many business school mainframe computers. MicroTSP is specifically designed for statistical analysis of data when forecasting is the objective.

The Micro TSP manual includes specific examples of forecasts computed using the packages as well as a data disk with all the data from the various examples. This is a welcome extra for users new to forecasting who wish to work through the examples in the manual. The two major case studies included on the data disk are a study of telephone demand using basic regression techniques and a sales forecasting model using more sophisticated regression techniques. Also included are examples of forecasting auto sales and the treasury bill rate.

One decided advantage of MicroTSP, which is available in both Apple and IBM versions, is its ability to use DIF files which are compatible with programs like VISICalc. The MicroTSP manual even includes an extended example of how to take information from the Holiday Inn forecasting demonstration and transfer the results to VisiCalc to compute projected occupancy rates, projections of total revenue, and forecasts of related variables. Of course, the real power of VisiCalc becomes available when you begin reconsidering the assumptions underlying the model constructed with the aid of MicroTSP.

Data for use in forecasting models can be obtained by telephone using your micro to capture the data and then converting that data to a form usable by MicroTSP. Once in the MicroTSP package, a one-key command converts the file to DIF format if you wish to use the data in other programs.

Some very simple graphics (scatter diagrams, residual plots, etc.) are produced by MicroTSP, and transformations of any data are very simple to invoke the GENR (generate) command. A comprehensive data editor allows for easy housekeeping in the data files. The package can be set up to run "batch" jobs if you wish to leave the machine unattended while running many routines. Also included in the software is the capability of running quite sophisticated regression techniques (two stage-least squares and Cochrane-Orcutt two-stage iteration models). References for advanced techniques are provided in the manual. Statpro

Statpro has been around in an Apple version for some time but is now available for the IBM PC and PC XT.

The software is actually an integrated statistics, database, and graphics package with comprehensive coverage in each area. A smaller, more directed offshoot of Statpro will soon be available for the IBM. Its title is The Forecaster, and it is promised to include the most often used forecasting tools from Statpro.

For the forecaster, Statpro is the most complete package of those we have reviewed. It is the closest thing to using a mainframe statistical package like SAS or SPSS, but it also includes some bells and whistles that no mainframe package has.

Every conceivable way to transform the data is included in the database section of Statpro: leading and lagging (called "offsetting" in Statpro), missing data provisions, and range checking to eliminate errors. If you use a hard disk, information retrieval and storage time is greatly reduced.

Data are not stored in DIF files in the Apple and IBM PC versions but rather in a Pascal format making it virtually impossible to exchange information between Statpro and other programs. Standard text files would also be difficult to read into Statpro. Wadsworth has promised a module in early 1985 to allow the IBM version to read DIF files. The PC XT version has none of these drawbacks. It is already able to read and process both ASCII files and the popular DIF files.

The available statistics in Statpro covers everything listed in the comparison chart as well as a list of statistical procedures usually found only is statistics textbooks. If you have used SAS or SPSS on a mainframe, you will be at home in the Statpro statistical menu.

For simple regression alone you can choose between fitting a linear equation (as we did with the program listing in this article in the carpet sales example) and one of up to 16 other forms of regression (exponential, Gompertz curve, logistics curve, etc.). For time series analysis, Statpro offers moving averages, multistage leas tsquares, polynomial and trig functions, and exponential smoothing. even seasonal indices can be computed by Statpro.

The third part of the package, "graphics workstation," can display most of what you have calculated in various ways. It is the most complete graphics module in any of the packages listed here and is actually more like a separate, and quite complete, graphics generator. DSM

DSM stands for Decision Support Modeling. It is a rather comprehensive statistics and operations research package for IBM and MS-DOs compatible computers. For forecasting purposes, the package has a complete regression module capable of handling up to 15 variables at one time. A forecast command allows easy computation of forecasts after an equation has been estimated.

File management is quite simple with a full range of transformation allowed on the variables: log, reciprocal, square root, square, and formula. DSM does have the ability to read data from some database and spreadsheet programs. There is an editor for data files, and you can print any screen (data or otherwise) at will.

In addition to regression, DSM provides moving averages and exponential smoothing. Other capabilities of DSM, not directly related to forecasting, are: linear porgramming, inventory modelling (using simulation), PERT (program evaluation and review technique), queueing, and breakeven analysis (using simulation).

Note that two of the models in DSM use simulation as the technique for modeling or forecasting inventory and predicting breakeven points. These models are menu-controlled, so no active programming experience is required. They are in the fist sophisticated uses of simulation we have seen on a microcomputer for business users, and they put the awesome power of simulation in the hands of non-programmers.

Since DSM is provided to educational institutions as an educational tool, the company also provides a student workbook to help first-time modelers use the program. The workbook was not included in the package but is said to be available separately for $8.95. A demonstration disk is also available for $15. TK! Solver

Without question, this is the most innovative software we have seen recently. In its narrow application corridor is king.

TK! Solver is an equation solver or, more accurately, an equation system solver. It is the only package of its kind on the market. But just exactly what does it do?

You enter equations on a "rule sheet" like those in Table 6 which define a situation--in this case, the derivation of gross national product for an economy--and press the exclamation point key; TK! Solver finds the unknown values and displays them on a "variables sheet." You can solve for all unknown variables by exchanging known with unknown variables without reformulating the equations.

If a model or system of equations cannot be solved directly, TK! Solver can be instructed to approach the answer in successive iterations until it converges on the solution. The manual describes this feature as "a method of problem solution that uses successive approximations of the answer to converge to a desired mathematical result."

This could be useful in situations where the same unknown is on both sides of the equation. Other uses of this feature include the solution of systems of simultaneous equations where the relationships between the unknowns are not adequately defined to allow direct solution.

Moving around on any of the TK! Solver "sheets" is much like scrolling around a VisiCalc spreadsheet. Changes to existing models are easy to make, and recalculation is just an exclamation point away. The package does not handle numerical integration techniques or matrix operations (rarely used in forecasting) and is only for the purpose of finding real-number solutions; it will not handle "general" or "symbolic" solutions.

A users group with its own journal is supported by Software Arts and some TK! SolverPacks are available for specific applications (such as financial management techniques). Among the formulas included in the TK! Solver Pack for financial management are compound interest, net present value/internal rate of return, level debt service analysis, bond swap analysis, bond refunding decisions, convertible debt analysis, capital asset pricing model applications, and Black-Sholes option pricing.

TK! Solver has the ability to produce plots of the outcomes of its calculations. When calculations have been made, the results can be plotted on an X-Y line graph showing one list against the values in all other specified variable lists. The plots can be viewed on the screen or printed using standard ASCII characters on most printers.

While the package is not a general tool for forecasting, in some specific applications it may be the greatest thing since peanut butter. The documentation for TK! Solver assumes the user has no prior experience with the computer hardware and includes a complete tutorial with practice in moving from one "sheet" to the next and setting up various types of models (some examples are included).

A Help command on-like is loaded by typing?, which then asks for the topic on which help is needed. A mini-tutorial on that topic is then scrolled across the screen. The software is available for the IBM PC (and other MS-DOS machines), Wang PC, DEC Rainbow, DEC Professional 350, Apple IIe with an extended 80-column card, and Macintosh.

TK! Solver is well suited to the knowledge processing philosophy on which Macintosh is based. The program uses many of the Macintosh enhancements: the mouse as a pointer, very high resolution plotting (and printing), 12-digit precision, recognition of very small numbers, expanded error diagnosis, cut and paste, and access to Macintosh desk accessories.

The TK! Solver Packs are also available for Macintosh, but, surprisingly, the Macintosh price for the program and the add-on packages is below both the IBM and Apple II prices ($249 for Macintosh TK! Solver and $59 for each TK! Solver Pack). Graph N' Calc

Graph N' Calc uses a spreadsheet to allow data entry and transformations. Available only for the Apple III, the spreadsheet is only 10 rows by 100 columns and cannot be reconfigured to include fewer columns and more rows.

This can be a distinct disadvantage, because it severely limits the number of variables and transformations you can have available for analysis at one time (even though you may be using only a few at a time). The spreadsheet can be reviewed by scrolling to the left or right (all ten rows are shown at all times).

Transformations are easily handled from the menu, but the transformations available are limited to logs, square roots, arithmetic functions, and absolute values. Other calculations can be inserted into "user modules" by programmers following the instructions in the manual.

While the moving average technique is not available, Graph N' Calc seems made for forecasters because of the inclusion of three versions of exponential smoothing and a regression routine. The regression routine supplies few summary statistics, however, and there is no provision for logging data which is often important to forecasting.

Graphics are nicely portrayed by Graph N' Calc from any data in the spreadsheet. In addition to displaying regressio lines and scatter plots, the program is also capable of producing bar charts, pie charts, and high-low volume charts. There is also a module which allows you to create "slides" and show them in slide projector fashion. There is no provision, however, for creating "text slides" to bridge the displays of graphics material and supply forecast results in tabular form.

The manual contains a tutorial which explains the operation of the system. In addition, a second book, Fundamentals of Forecasting, is packaged with the program. Taken together, the two books should allow even a neophyte to create decent forecasts, assuming adequate data. SmartForecasts

SmartForecasts, available only for the IBM PC, is a software package aimed directly at forecasters. It offers both graphical and statistical tools from a command type system similar to Daisy and MicroTSP. As in Daisy, data are stored and viewed in a spreadsheet format capable of holding up to 150 observations of 15 different variables.

SmartForecass can read DIF files and thus is compatible with VisiCalc and Lotus 1-2-3. The spreadsheet is viewed with a SEE command which moves the window to the desired position much like the GO TO command in spreadsheet programs. Scrolling in not available.

Transformations of the data are effected by using the DEFINE command to create the new variable by using some arithmetic operation, a functional transformation, or time shifting. Functional transformations include logs, exponentials, rounding, and square roots.

The statistical analysis portion of the program includes several features that make the SmartForecasts environment particularly enjoyable to use:

* Function keys and short commands eliminate nested menus which are time consuming (Daisy, ESP, and MicroTSP also use the command form of interface).

* An OOPS command can immediately undo any mistake in manipulation.

* The entire program is on one disk.

* There is an on-line help command which senses the context of your inquiry.

One unique feature is an "eyeball forecasting" routine which allows you to use interactive graphics to make forecasts. The EYEBALL command permits you to complete two alternative forecasts of some variable, say sales. The software treats one forecast as the "champion," while the other is the "challenger." You can refine the forecasts until you make one you prefer to the current champion; at that time the labels switch since you have identified a new champion. You even have the ability (although we question its value) to draw a "manual" forecast directly on the screen. By moving a cursor around the screen and identifying points (or by using the numeric keypad) data are manually entered to create a challenger.

"Multiseries forecasting" is a feature that is potentially handy for users who wish to forecast more than a single series at once. Suppose you have three different product lines and wish to see the forecasts for all three at once over the next three months. That requires only a short set of commands in SmartForecasts.

For forecasters who must deal with data which exhibit seasonality (variations which occur regularly from season to season), SmartForecasts may be the vest available package. It calculates seasonal adjustment factors, and the manual includes a complete explanation of how to use them (ESP also does a nice job in this area). Since seasonality is at the root of many business uses of forecasting, it would seem to make sense to use a package that addresses the situation directly. It is surprising to us that many so-called forecasting packages do not deal directly with seasonal data. Failure to do so in most forecasting situations can lead to serious misinterpretation of the forecasts.

A technical reference section in the manual is quite useful for statistically-inclined users who wish to see the actual solution routines displayed for each analysis procedure in the software. For those with a statistics background this will make clear whether the correct technique has been chosen for a given situation. Expert Choice

Consider the situation of forecasting the most cost effective computer for your office to use. Now this is not quite like the forecasting situations presented earlier, but it is a decision involving future consequences for which many alternatives are open. Expert Choice is a unique program designed to handle a special case of decision making. The Expert Choice tree (much like a decision tree but upsidedown) has the root at the top with branches on the next level. The criteria for the decision (chosen by you) are listed just under the goal: cost, performance, maintenance, obsolescence, software, and portability.

Judgments of the relative position between the pairs of computers covering each decision criterion are entered by answering questions presented in verbal fashion (but you can insert numerical values directly once you get the hang of using the program to speed up data entry). Once your preferences are entered, Expert Choice calculates the priorities and displays results along with an inconsistency index (which shows how consistent your judgments between pairs of computers on each criterion were--if you receive less than 1.00, the tre is considered satisfactory).

Computers can be compared on all six criteria at once or on one criterion at a time; more detailed trees can also be produced with more levels from top to bottom and more criteria across (with virtually no size limit).

Office relocation, house selection decisions, employee evaluations, make or buy decisions, risk evaluation in lending--all these are suggested uses of Expert Choice. While many decisions about future actions could be considered with Expert Choice, its most important feature may be that it forces you to be explicit about the goals of any decision and the criteria for the final selection. ESP

For a comprehensive forecasting program little could be added to ESP, a software package on six disks (plus three tutorial disks) which has been adopted by Chase Econometrics as the software package they will deliver and supply to their customers for use with their PC PLANR Workstation.

The Econometrics Software Program, available only for the IBM, can read ASCII text files, DIF files, Lotus 1-2-3 NWKS) files nad Microsoft's Multitools (SYLK) files. Additionally, data from Chase Econometrics can be downloaded (using a communications package) in a form usable by ESP. The software uses the Intel 8087 math coprocessor, if installed, and supports all hard sisk (including the XT disk).

While it is among the most expensive of the forecasting packages at $795, three characterisctics set it apart from the competition:

* The ability to read virtually any type of data file, no just its own special configuration files. Files from any word processor (ASCII files) or almost any spreadsheet program like Lotus 1-2-3 or VisiCalc (DIF files) can be used and data resident on a mainframe can be downloaded and used by ESP because of this feature (note that ESP does not include the software to actually perform the downloading function).

* Large databases or many different databases can be handled with ease and speed by using a hard disk. Calculation speed can also be increased with a math coprocessor chip (not included).

* ESP includes all the econometric tools forecasters are likely to use routinely: transformations, basic statistical functions, regression (many types), specialized forecasting techniques, graphics displays, and on-line help and tutorial functions.

ESP started its life as a mainframe package and is remarkably similar to MicroTSP in its format. The way in which commands are entered (and even some of the command names) are similar. Both use the command form of program control calling routines with one-word commands like SMPL 81M1 83M6 and OLSQ SALES PRICE ADVERT. Both MicroTSP and ESP are able to read a wide range of data file types as well.

ESP also includes on-line help, menus for various program sections, and a much more extensive graphic capabilities than MicroTSP. ESP can also use all forms of data: time-series, cross-sectional, vectors, and even matrices. Any data saved by ESP are recalled simply by referring to them by name.

How large a database can ESP handle? A large database with 75,000 observations on 500 variables will fit on a single disk and you can access up to 1000 variables at a time. For most people, in terms of capacity ESP will be like using a mainframe.

The ESP manual is not a statistics tutorial (i.e., it will not teach you how to forecast), but it is among the best of manuals for learning how to operate the software with dexterity. Three tutorial disks are included and provide you with your first experiences on ESP. On-line tutorials provide a quick run through of most of the capabilities of the program (there are actually five different levels of on-line help; you use just what is appropriate).

Like MicroTSP, ESP can run "batch" jobs and save these run commands as a separate file allowing you to rerun routine projects with a single command. Perhaps you might update a forecast each week or month as new data become available--the batch mode is made for this of use.

One of the most popular advanced forecasting techniques is specifically available in ESP: Box-Jenkins Time Series Analysis. The identification of a Box-Jenkins madel is covered in the reference section of the manual.

Three separate software packages are produced by a company called Stratix. All three packages, which are available both for Apple II line and for the IBM-PC, are of interest to individuals interested in forecasting and business economics.

Xtrapolator performs trend analysis (regression where the independent variable is time or some transformation of time) in a variety of different forms. Each of the Stratix packages is more-or-less single purpose, and the purpose of Xtrapolator is to perform twelve distinct types of time series analysis: linear trend, moving average, exponential curve, Gompertz curve, geometric curve, double moving average, second order polynomial, exponential smoothing, third order polynomial, double exponential smoothing autoregressive trend, and triple exponential smoothing.

Nuametrics is designed for those who wish to use regression (with up to seven independent variables) as the technique for forecasting. Both Nuametrics and Xtrapolator allow the transformation of variables, the use of DIF files (if those files have only a single column of numbers), and the creation of plots.

The final package from Stratix is MicroBJ which has as its sole task the estimation of Box-Jenkins time series models. These are forecasting models with time as the only independent variable that are estimated using a complex procedure developed by Box and Jenkins. The three program modules allow you to identify, estimate, and then forecast with the appropriate Box-Jenkins model. Plots are also available with this software. MicroBJ, however, will not read DIF files like its companion programs.

We have mentioned that certain packages look like SPSS on a mainframe, but the program which replicates the mainframe version almost perfectly is SPSS/PC itself. This is not a program for neophytes or individuals who don't have an IBM PC fully equipped. It is also not your program if you will be forecasting infrequently or always with the same technique.

But SPSS/PC is your program if you fall into any of these categories:

* Need a complete statistical package capable of a wide variety of analytical techniques and an infinite variety of transformations.

* Need to transfer data back and forth to a large mainframe program capable of handling huge datasets.

* Require complete plotting and report writing capability.

* Need the ability to handle up to 200 variables at once.

SPSS/PC requires the use of a hard disk, and it is strongly advised that you have the 8087 math coprocessor (available free to academic sites from Intel Corporation) installed in your IBM. It also requires 320K of memory.

Is it easy to use? It was very easy for me to use because I use SPSS on a mainframe quite frequently and the command structure is almost identical (the manual has a special section just for users of mainframe SPSSx to spell out the differences to them). It was quite a shock to realize that the very same capabilities we have on a large mainframe can now be had on a microcomputer (well, not exactly the same; there are limitations on the size of the dataset).

Will it be an easy system for new users to learn? That is a different question. The SPSS/PC command system takes a little getting used to. First-time users will require some time to learn the command system before becoming productive, but that same command type system has its advantages; namely, it is quite fast as a way of requesting specific analyses, and it is infinitely versatile in what it can request.

For any user (new or experienced) there is a complete help facility on-line which offers descriptions and examples of situations you may encounter. If you forget the variables you have in memory, a SHOW command is available, and if you forget the options you have in effect, a DISPLAY command will remind you.

Keep in mind, too, that SPSS/PC is also a report writer. That is, it has the capability of presenting the information you have analyzed in printed form according to any format you may desire. That can be quite a timesaver since you do not have to transfer any of the information, tables, or plots to another package to complete a report.

A special feature unique to SPSS/PC is the use of a portable file to transport your data and definitions between SPSS/PC and SPSSx on a mainframe without having to redefine them every time. An EXPORT command produces a portable ASCII file that can be read with the IMPORT command in SPSS/PC or mainframe SPSSx. You can upload EXPORT Files to a mainframe using Kermit protocol (an error checking routine used on some mainframe systems) if both the mainframe and the IBM PC have Kermit installed.

SPSS/PC does not include a communications package to perform the uploading, but the manual advises the programs such as Crosstalk and PC-Talk III can transfer such ASCII files. Note that SPSS/PC files containing special graphics characters may not be transferred correctly to mainframes or to other types of personal computers.

It would be impossible to list all the procedures available in the program, but suffice it to say that there are few statistical procedures missing. While SPSS/PC is not sold as only a forecasting package, it could do quite well in that category while providing a powerful general statistics package and a report generator.

Products: VisiTrend-Plot (computer program)
Daisy (computer program)
MicroTSP (Statistical software)
Statpro (computer program)
DSM (computer program)
Graph N' Calc (computer program)
SmartForecasts (computer program)
Nuametrics-Xtrapolator-MicroBJ (computer program)
SPSS-PC+ (Statistical software)
ESP (Computer program)
Expert Choice (Computer program)
TK Solver (Mathematical software)