Academic journal article
By Kim, Junghun; Trivedi, Pravin K.
The American Statistician , Vol. 48, No. 4
Regression Analysis of Time Series 4.02 (Statistical/Mathematical Software)--Evaluation
Time Series Processor 4.2B (Statistical/Mathematical Software)--Evaluation
Micro Time Series Processor 7.0 (Statistical/Mathematical Software)--Evaluation
PC Generalized Instrumental Variable Estimation 7.0 (Statistical/Mathematical Software)--Evaluation
RATS (Regression Analysis of Time Series) Version 4.02
Available from Estima, P.O. Box 1818, Evanston, IL 60204-1818. 386/486 Basic Price $400; volume and upgrade discounts and site license prices available.
TSP (Time Series Processor) Version 4.2 B
Available from TSP International, P.O. Box 61015, Station A, Palo, Alto, CA 94306. 286 extended memory version, 386/486 versions $400. Upgrade and volume discounts available.
MicroTSP (Micro Time Series Processor) Version 7.0
Available from Quantitative Micro Software, 4521 Campus Drive, Suite 336, Irvine, CA 92715. Single user $595; 1-5 network license $1,600; 6-10 licenses $2,000; user's manual for the network license $25.
PcGive (Generalized Instrumental Variable Estimation) Version 7.0
Available from Institute of Economics and Statistics, University of Oxford, Manor Road, Oxford, OX1 3UL, U. K. Single copies $400; PcGive 7/386. $480. Educational and site license discounts of 50%.
The four time-series software packages reviewed here have a strong econometric orientation, though all offer plenty for the general time-series analyst. Of the four, TSP 4.2B has the longest history. Initially a mainframe program, it has evolved into a PC-based program, in parallel with MicroTSP to which it bears some resemblance. PcGive, like Micro TSP, is in its seventh version as an interactive PC-based menu-driven program, with a mainframe ancestor dating back to 1969. RATS too has been around as a PC-based program for nearly 10 years. Over time each program has attracted a committed following, and each has some strong and unique features to differentiate it from the rest. Over time all four have become increasingly more powerful and easier to use.
These latest versions are larger, more versatile, with ever increasing features and vastly improved numerical and graphical capabilities, and much improved in terms of the ease and convenience of their use. Though competition has led to many commonalities between the four packages reviewed here, there are important differences between them. Though all four packages have strong time-series motivation, most also have data-handling and estimation capabilities for analysis of large cross-sectional data sets. MicroTSP and PcGive are menu based with an attractive and friendly user interface, and hence are very suitable for interactive use and teaching purposes; RATS and TSP offer well developed programming language features and procedures that appeal to advanced users.
1.2 General Features
All four packages are widely, probably predominantly, used by econometricians but probably not by statisticians. Though emphasis may vary from one application to another, statistical and econometric time series analysis overlap in many respects. Hence these packages should be interest to statisticians. Time-series packages with greater statistical orientation include SAS/ETS from the SAS Institute, S-Plus from StatSci, SCA from Scientific Computing Associates, SPSS, and MINITAB. There is considerable overlap between these and the econometrically oriented packages in terms of the coverage of autocorrelation and autoregression, univariate ARIMA models, forecasting, and smoothing. However, some other procedures like state-space modeling, Fourier transforms, vector ARMA models, automatic intervention detection, and estimation analysis are examples of topics that are often better covered in statistical rather than econometric packages, though with a few exceptions, the four packages reviewed here also cover such topics. One difference between the computing needs of econometric and statistical users lies in the former's emphasis on estimation and uses (prediction and policy simulation) of linear and nonlinear systems of structural time series relations. The procedures for dealing with these are also useful in other contexts, such as measurement error and latent variable models. …