Email: Sephton@unb.ca
Voice: 506 447 3210
Fax: 506 453 4514
(This review was published in the Journal of Applied Econometrics 13, 1998, 203-207)
EasyReg is written for novice, intermediate, and advanced users. It offers a wide variety of estimators, routines, and tests spanning the content of courses in introductory econometrics to graduate research seminars in some areas of applied time series analysis. Given the current fiscal climate at many post-secondary institutions, EasyReg is an inexpensive way to provide students with legal copies of software that can meet many of their needs. While it does not threaten the market for programmable packages like RATS and TSP, it may bode ill for other menu-driven econometrics programs.
EasyReg is programmed to allow dynamically declared arrays so there are no hard-coded limits to the number of variables or observations that can be entered. The memory capacity of the user's computer will determine the largest data set that can be accommodated. On a Pentium 133 with 32 MB of RAM I was able to read 10,000 observations on 100 variables and perform various transformations.
There are three econometrics levels in the Student version: novice, intermediate, advanced. Each successive level includes the options in the previous level in addition to an increasingly sophisticated menu. For example, the novice level includes only OLS with standard t-values, out of sample forecasting, lagged dependent or independent variables,dummy variables, the Breusch-Pagan heteroskedasticity test, heteroskedasticity consistent t-values, the Durbin-Watson test, and the Jarque-Bera/Salmon-Kiefer normality test. The intermediate level adds F and Wald tests of joint significance and linear parameter restrictions, two stage least squares, tests for ARCH, ADF unit root tests, linear time series models with ARMA errors, and probit/logit analysis, poisson regression, and binomial logit/probit regression. The advanced level adds a variety of cointegration and unit root tests, non-linear regression, quantile regression, non-parametric kernel estimation, multinomial logit, Tobit, structural and non-structural VAR analysis, non-linear non-parametric cointegration analysis, cotrending analysis, and Bierens' integrated conditional moment tests of functional form.
Once the choice of econometrics level has been made, data entry is relatively straightforward. The default data input structure for T observations on K variables is as follows:
K M
Variable Name 1
Variable Name 2
.
.
Variable Name K
X(1,1) .... X(1,K)
..... .
X(T,1) .... X(T,K)
where M is the "missing value code" in the data (allowed only at the beginning or the end of a time series data set). EasyReg also accommodates three other data formats. In one, the first row of the input data file must include the series names, followed by rows of data, with each row containing an observation on K series (in this case EasyReg prompts the user for the missing value code). The program also allows the user to input raw data without series labels, either stacked across observations, by variables, or by variables, across observations. In the latter cases the program prompts the user to enter series labels as well as missing value codes.
EasyReg contains its own database which includes the following data sets:
Data transformations can be done after the data have been read. Point and click options offer a variety of transformations, including linear and multiplicative combinations of variables, logarithmic and exponential transformations, customized functions, min/max, as well as a host of time series transformations (m-period lags, m-period differences, percentage changes, moving averages, partial sums, time trends, dummy variables, and more).
Unit root and stationarity tests are also located under the Data Analysis menu. EasyReg offers a number of tests and testing options depending on the previous choice of econometrics level (which can be changed at any time by returning to the appropriate menu). For example, one can choose the lag structure to be employed, the sample space the test should span, and a variety of other conditions under which tests are performed. The user is also able to simulate the distribution of all unit root tests in short samples to accommodate size distortion, a particularly attractive feature. At the end of the output of each test there is a short summary of the inferences to be drawn (based on a choice of significance level). However, these inferences are related to critical values published in large sample simulations of the test statistics rather than response surface estimates provided by, for example, MacKinnon (1996) or Sephton (1995).
The multiple equation menu provides three options: VAR innovation response analysis, cointegration analysis, or non-linear cotrending analysis. VAR models are limited to nine variables. As is common across the entire program, the user has the option to choose the sample size and whether to include constant, trend, and other terms in the model. The VAR options also allow one to impose zero restrictions on the design matrix. In the case of a restricted model, a SUR option allows efficient estimation after initial OLS results are reported. FIML estimates allow the construction of confidence bands around the innovation responses. VAR models may be structural or non-structural, and the software provides a clear, concise description of the differences, as well as references to classic papers by Sims (1980, 1986) and Bernanke (1986) to guide the uninitiated. As is the case throughout the program, all plots can be saved and subsequently edited and printed or exported to alternative formats.
The cointegration menu offers Johansen and Bierens type tests (for a maximum of five variables). Johansen tests allow intercepts and time trends, both with and without cointegrating restrictions imposed. Both Lambda-max and LR tests are constructed, with inferences again drawn on asymptotic critical values rather than estimates drawn from response surface studies. The Bierens' cointegration test is a non-parametric cointegration test based on Bierens (1997). Both flavours of cointegration tests are described in the cointegration window, with the choice of various options over sample size, number of cointegrating vectors, and various "smoothing parameters" presented to the user. Both cointegration menus offer a very accessible way for us to expose students to topical issues in applied time series econometrics. For those who demand their students learn how to code various routines and tests, EasyReg offers a simple way for them to validate their programming exercises.
The final option under the multiple equation menu is non-linear cotrending analysis. This is experimental research based on Bierens (1996) which examines a non-linear trend stationary vector time series process z(t) such as z(t) = g(t) + u(t) with u(t) a zero mean stationary process and g(t) = E[z(t)] a vector of non-linear deterministic trends. Non-linear cotrending is the phenomenon that one or more linear combinations of g(t) are linear in t. Since the paper is "in progress" , the approach is experimental. However, it allows researchers to secure access to an interesting and timely routine without bearing the burden of coding it themselves (although that would be a useful learning experience). The inclusion of non-linear cotrending analysis points to the currency of the EasyReg program.
Academics working in a Windows environment will find it particularly useful in getting students to use a common platform for empirical work, as well as a check against which to compare their own coded routines written in other languages/packages. While data input from spreadsheets and proprietary formats (such as RATS, WKS ) is not supported, it is easy to transform data sets into a form accessible by EasyReg. The variety of econometrics levels will allow students to use the package as they progress from elementary to intermediate econometrics courses, meeting some of the financial constraints facing institutions of higher learning. Professor Bierens deserves great kudos for sharing his work with us.
Bierens, H.J. (1996), Nonparametric nonlinear co-trending analysis, with an application to inflation and interest in the U.S.', mimeo.
Bierens, H.J. (1997), Nonparametric cointegration analysis', Journal of Econometrics 77, 379-404.
Bierens, H.J. and L. Broersma (1993), The relation between unemployment and interest rate: some empirical evidence', Econometric Reviews 12, 217-256.
Bierens, H.J. and J. Hartog (1988), Non-linear regression with discrete explanatory variables, with an application to the earnings function', Journal of Econometrics 38, 269-299.
Fishback, P.V. And J.V. Terza (1989), Are estimates of sex discrimination by employers robust? The use of never-marrieds", Economic Inquiry 27, 271-285.
MacKinnon, J.G. (1996), Numerical distribution functions for unit root and cointegration tests', Journal of Applied Econometrics 11, 601-618.
Sephton, P.S. (1995), Response Surface Estimates of the KPSS Stationarity Test', Economics Letters 47, 255-261.
Sims, C.A. (1980), Macroeconomics and reality', Econometrica 48, 1-48.
Sims, C.A. (1986), Are forecasting models usable for policy analysis?', Quarterly Review of the Federal Reserve Bank of Minneapolis (Winter), 2-16.
___________________________
[*] The EasyReg web page has moved to URL: http://econ.la.psu.edu/~hbierens/EASYREG.HTM.