EasyReg: 1998 versions

Difference between Version 1.13
(January 1998) and Version 1.12 (December 1997):

If you take impossible transformations, such as a log
of a negative number, EasyReg incorrectly replaced the missing value code
by zero (no missing values). This problem also occurred if you take an
exp transformation of too big a number, or if a transformation is ineffective
or just too effective (min/max, dummies). These transformations will no
longer be allowed: EasyReg will then give a warning, indicating the problem,
and abort all transformations.

The VAR innovation response analysis was limited to
systems of no more than 9 variables. This upperbound has been removed.
However, large VAR models may cause an "outofmemory" error, depending
on the memory capacity of your PC. Moreover, for VAR systems with more
than 9 variables it is no longer possible to impose Granger causality restrictions.

Bierens' nonparametric cointegration test was limited
to cointegrated systems of no more than 5 variables. This upperbound has
been increased to 10. The critical values of the test, and the optimal
number "m" of Chebishev time polynomials, have been recalculated and build
in. See the extended Tables 14 of: Bierens,
H.J., "Nonparametric Cointegration Analysis", Journal of Econometrics
77 (1997), 379404.
Johansen's cointegration procedure, however, is
still confined to systems of no more than 5 variables.

The computation of auto and crosscorrelation functions
required a minimum of 30 time series observations. This lowerbound has
been reduced to 10, because there was a demand for it. However, with such
short time series the estimate of the auto/crosscorrelation function is
highly unreliable!

Difference between Version 1.14
(February 1998) and Version 1.13 (January 1998):

There were several bugs in the module for testing the
unit root with drift hypothesis against nonlinear trend stationarity. Thanks
to Dave Cushman
for pointing this out to me.

Difference between Version 1.15
(February 1998) and Version 1.14 (February 1998):

The options for the deterministic parts (intercept,
linear time trend, seasonal dummies) in the modules for linear time series
regression models and VAR innovation response analysis have been augmented
with nonlinear time trends, using Chebishev time polynomials, in order
to capture data heterogeneity. See: Bierens H.J. (1997): "Testing the Unit
Root with Drift Hypothesis Against Nonlinear Trend Stationarity, with an
Application to the US Price Level and Interest Rate", Journal of Econometrics
81, 2964 (This option was already available in SIMPLREG,
the DOS predecessor of EasyReg). However, if you choose this new option
then outofsample forecasting with linear time series regression models
is no longer possible, because Chebishev time polynomials cannot be extrapolated.
I needed this option for the revision of my paper "Nonparametric Nonlinear
CoTrending Analysis, with and Application to Interest and Inflation in
the US".

Some minor cosmetic upgrades.

Difference between Version 1.16
(March 1998) and Version 1.15 (February 1998):

A new module, "Teaching tools", has been build in. This
module demonstrates

the difference between the concepts of convergence in
probability and convergence in distribution,

the size and power of the ttest, and

spurious time series regression results.
I have used "stand alone" versions of this module in
undergraduate econometrics and graduate statistics classes at Southern
Methodist University and PennState University.

Some minor cosmetic upgrades.

Difference between Version 1.17
(March 1998) and Version 1.16 (March 1998):

The file EASYREG.EXE is the "shell" around the other
EXE files (which are the "work horses" of EasyReg, conducting specific
tasks). The file EASYREG.EXE itself contains all the menus, but previously
also the modules for choosing the working directory, searching for data
files, choosing the econometrics levels, and the database tasks. EASYREG.EXE
has now been substantially reduced in size, by creating the following new
EXE files for the latter tasks:

DATA.EXE: Choosing the working directory, seaching for
data files, or retrieving data from the EasyReg database;

DATABASE.EXE: Adding, retrieving or deleting data from
the EasyReg database;

ESLEVELS.EXE: Selection of the econometrics level.
The effect of this reorganization is twofold:

There is now more memory available for data;

EasyReg now runs faster.

The startup window has been redesigned

Some minor cosmetic upgrades.

Difference between Version 1.18
(May 1998) and Version 1.17 (March 1998):

A new module has been added which conducts General Method
of Moments (GMM) estimation for systems of linear equations with possibly
common parameters. The GMM models involved also include seemly unrelated
regressions (SUR), and fixed effect and pooled panel data models.

A new module has been added which computes the correlation
matrix and its eigenvalues and eigenvectors of a set of variables. (Due
to a request from an EasyReg user.)

The crosssection data set used by Bierens and PottButer
(1990), Econometric Reviews 9, 123184, has been added to the EasyReg
database.

A bug in the artificial crosssection data generator
has been repaired. If you specify a probit or logit model,the intercept
is automatically adjusted in order to prevent the Y variable from becoming
constant zero or one. The bug involved was that also in the standard regression
case the intercept was automatically adjusted such that the Y variable
is centered around zero, without notification. Moreover, the option has
been added to declare the artifical data involved as (annual) time series
data.

A nonessential bug which incorrectly disabled the option
of merging crosssection data files has been repaired.

The module for renaming variables is now called from
the input menu instead of the transformation menu.

Some minor cosmetic upgrades.

Difference between Version 1.19
(June 1998) and Version 1.18 (May 1998):

A new module has been added which computes either the
cumulative distribution function F(x) and corresponding pvalue in a given
x, or the critical value at a given significance level, for the following
distributions or tests: Standard normal, standard Cauchy, Student t, chisquare,
and F.

All test statistics which have one of these null distributions
are now automatically endowed with pvalues. (These additions have been
suggested by an EasyReg user.)

A nonessential bug in the Tobit module has been repaired
(The reported frequency of the zero Y values was wrong).

If you try to estimate a Tobit model and you get the
message that a Tobit model is not appropriate, EasyReg will now explain
why. (I got too many queries from EasyReg users who were denied access
to Tobit estimation.)

If the parameters of a multinomial logit model are not
identified, EasyReg now provides an example how to solve the problem. (Also
this is due to queries from EasyReg users.)

A button "View INPUT file" in the "Quit" window has
been added.

Some minor cosmetic upgrades.

Difference between Version 1.20
(July 1998) and Version 1.19 (June 1998):

It is now possible to generate artificial ARIMA and
VAR time series data.

When computing the autocorrelation function of a single
time series, the BoxPierce and LjungBox Q statistics (for degrees of
freedom up to 10) are now computed as well.