The PC corner.
Those of you who have read my columns over the years are aware of my preference for MicroTSP. Its major strengths include:
1. The ability to use either a menu system or a command line to issue commands to the program;
2. The ability to mix structural and autoregressive/moving average variables in the same model (in essence, modelling the residual of a structural equation as an autoregressive/moving average process);
3. Speed -- it's one of the fastest on the market and doesn't require a math coprocessor (although a coprocessor version exists);
4. An excellent user's manual, which is really more of a tutorial in econometrics.
The latest version of MicroTSP for the IBM PC (7.0) has been released. Among its improvements are a completely revised ARMA (autoregressive moving average) estimation procedure, which can give substantially more accurate results in models with many or long moving-average terms. The FORCST and FIT commands now give you the option of generating forecast standard errors and a graph of the 95 percent confidence intervals. The problems with importing Lotus 1-2-3 data series with embedded NAs (not available) and ERRs (error) have been resolved, and the program can now directly import series that are the results of formula evaluations. Graphics capabilities have been expanded, although they are still not quite presentation quality. Finally, the menu system, already one of the best around, has been simplified even further. The bottom line -- the best package on the market for under $1,000 has gotten even better.
It's not that MicroTSP is without fault. I'm not particularly ecstatic about its data-banking facilities, and you still can't use variable transformations directly in equation estimations; you have to generate a transformed variable first. However, I believe that it offers the best price/performance combination on the market.
By the way, I echo Bill's aversion to the copy protection feature. I've discussed the problem of copy protection with the author of the package, David Lilien, and he is taking the matter under consideration.
X-11 SEASONAL ADJUSTMENT
I continue to get calls from people looking for an inexpensive version of the Census X-11 seasonal adjustment program for the PC. The only stand-alone version I am familiar with is a version put out by Excel Research. I've reviewed it in this column before and recommend it highly. The author, Mike Narcowitz, has released a new version that addresses my main complaints: lack of flexibility in reporting options; additive/multiplicative selection; trading day adjustments; and outlier screening. For $95, it's a great bargain. For more information, write or phone: Mike Narcowitz, Excel Research, 768 River Road, Trenton, NJ 08628, (609) 883-0583.
The Macintosh version of MicroTSP represents the first genuine econometrics program with a full Macintosh interface. There are other good statistics programs with a Macintosh interface -- Systat 5.0 and StatView II are two of the most popular -- but neither is specifically aimed at the econometrician. Until the Macintosh version of MicroTSP was available, Macintosh econometricians who had to conduct vector autoregressions, two-stage least squares, or dynamic simulations had to make do with the arcane DOS-like command syntax of SHAZAM or RATS. Not any more. According to the manual, Macintosh MicroTSP, ". . . eats, drinks, and breathes Apple's Human Interface Guidelines." I agree. MicroTSP is a pleasure to use.
I was shocked when I opened my copy of MicroTSP: The entire program took less than 500k of disk space. This contrasts sharply with Systat 5.0, which takes over 2 megabytes of disk space. I suspect that one of the reasons for the difference in size is that MicroTSP has more modest graphing and charting capabilities than Systat and does not include a drawing layer. Still, the graphs drawn by MicroTSP are more than adequate. When I need presentation quality graphics, even Systat is inadequate. I have to export to DeltaGraph or a full featured drawing program.
MicroTSP reputedly runs on one megabyte of memory, but anyone with a Mac that anemic these days needs help. Still, the low memory requirements simplify things when allocating memory under Multifinder. Data are held in RAM, and as many as 16 million data points can be used on a 64 megabyte machine (not unreasonable with virtual memory under System 7) but data sets are limited to available memory, with a maximum of 500 per work file. The actual number that can be handled simultaneously depends on the amount of RAM available and the number of observations in each series. Text files can be imported almost effortlessly; simply state the number of observations and/or the dates of the observations and paste the file directly into the data editor.
On a 68030 or better machine, computation is reasonably quick -- not as fast as StatView II, but appreciably faster than Systat 5.0. (The current version of Systat, v. 5.1, is reputedly much faster than 5.0.) On a fully configured Macintosh FX or an accelerated CI, vector autoregressions and other nonlinear procedures should be fast enough for even the most jaded speed freaks.
Like all good Macintosh programs, it is possible to boot MicroTSP and begin work immediately with only casual reference to the manual. New users will have to learn only the syntax for defining and retrieving data sets and the one-line commands for specific econometric operations. All of the Macintosh command key conventions (cut, copy, paste, etc.) are applicable to the data editor, fonts and windows are handled normally, and there is extensive use of hierarchal menus. Experienced Macintosh users will feel right at home. More sophisticated operations require a Basic-like programming language, but this too is intuitive and easy to remember.
Some programs (for example, StatView II) allow this kind of data transformation without the need to type a single command, but it takes so many mouse clicks that I quickly tire of the dialogue boxes and yearn for a command line. Systat may have the best of both worlds: New users can opt for the mouse, and experienced users can learn the command structure.
The MicroTSP manual is excellent. It is well written, even with a bit of wit, and there are plenty of worked examples. More so than most statistical software manuals, this one teaches some statistics along the way and includes page references to two of the more popular econometrics texts, R. Pindyck and D. Rubenfield's Econometric Models and Economic Forecasts (McGraw-Hill, 1981) and J. Johnston's Econometric Methods (McGraw-Hill, 1984). A quick skim through the manual left me with only one minor quibble: The authors apparently do not believe in parsimonious parameterization of Box-Jenkins models. In their three-month Treasury bill forecasting example, they try to fit an ARMA model with moving average lags at 1, 2, 3, 6, and 7 periods and autoregressive lags at 1, 2, and 6. I suspect that an ARMA (2,2) would do just as well, and it is much simpler!
I like the MicroTSP and expect that very shortly it will become the only econometrics program I use, but I still have one rather serious complaint: copy protection. The copy protection scheme on MicroTSP is one of the more sensible I've seen -- users are allowed 3 installs and can recover installs by deinstalling the program -- but it is still an insult to have to put up with copy protection at all. It is especially bad for those of us who use DiskExpress II, the most popular disk optimizer for the Macintosh. DiskExpress II is designed to run full time in the background, and, like all optimizers, cannot be used with copy protected software. As a result, DiskExpress II must be deactivated when MicroTSP is installed and that MicroTSP must be deinstalled before optimization. The only solution to this problem is to create a partition for copy-protected software and instruct DiskExpress II to ignore it. But this is possible only on hard disks with enough free space to permit a new partition -- and most disk partition software requires that the disk be reformatted. I understand the need to copy-protect games and perhaps even student versions of some programs. But applying copy protection on a professional level productivity program is insulting.
Software mentioned in this review:
The Macintosh Version 6.6 of MicroTSP is available from Quantiative Micro Software, 4521 Campus Drive, Suite 33, Irvine, CA 92715, (714) 856-3368 (Price: $595).
DeltaGraph. DeltaPoint, Inc., 200 Heritage Harbor, Monterrey, CA 93940, (408) 648-4000.
DiskExpress II. Alsott, Inc., P.O. Box 927, Spring, TX 77383, (713) 353-4090.
StatView II. Abacus Concepts, 1984 Bonita Ave., Berkeley, CA 94704-1038, (800) 666-STAT, ext. 34.
Systat. Systat, Inc., 1800 Central Ave., Evanston, IL 60201, (708) 864-5670.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||MicroTSP statistical software package|
|Author:||Qualls, John H.; Brown, William S.|
|Date:||Oct 1, 1991|
|Previous Article:||The statistics corner: more new products from the Census Bureau.|
|Next Article:||The Process of Change in American Banking: Political Economy and the Public Purpose.|