EasyReg International, 2002 versions

Differences between the versions
December 31, 2001, and February 28, 2002:

Repaired bugs:

CSV files with some variable names enclosed between
quotation marks were not correctly read.

If two or more variable names are the same and you delete
one of them, EasyReg deleted all of them.

The limited dependend variable module did not display
the asymptotic variance matrix.

The module NEXTMENU may give an out of memory error
if the output grows too large. This is due to the limited capacity of Visual
Basic text boxes. If the output exceeds the capacity of the text box, the
output will now be split up in two parts. The first part is written to
output file OUTPUT.TXT, and only the second part is displayed. Moreover,
if the file OUTPUT.TXT exists, a button becomes visible which enables you
to view it via Notepad or Wordpad. Furthermore, in order to free memory,
the ARMA and GARCH error options have been separated from module NEXTMENU
in two new modules, NEXTARMA and NEXTGARCH, respectively.
Thanks to John Baratta for pointing this problem
out to me.

In the OLS module, the option "Chebishev time polynomials"
did not work properly if a linear time trend is included.

Changes and additions:

The "2001" has been dropped from the EasyReg International
main window, because EasyReg is a project in progress, and the current
version does not yet warrant the label "2002".

In the nonlinear maximum likelihood module ML it is
no longer needed to specify a dependent variable, because the dependent
variables are implicitly defined by the model specification.

Two new transformations have been added to the userdefined
nonlinear models ML, NLLS and NLGMM: "Sample sum" and "Sample mean", because
I needed them for my research. These transformations are only available
for crosssection data.

The following new transformation has been added to the
TRANS module:
Given a selected variable X, this transformation
creates two NEW variables:

Dummy missing[X]: Dummy = 1 if the value of X is a missing
observation, Dummy = 0 if not,

Zero missing[X]: Missing values of X are replaced by
0.
These new variables will enable you to specify and estimate
Heckman's sample selection model via userdefined maximum likelihood (module
ML).

A new item "EasyReg links" as been added to the WWW
menu, containing a list of web pages with links to EasyReg. This replaces
the "EasyReg fanclub" web page, which will no
longer be updated.

Differences between the versions
February 28, 2002, and April 7, 2002:

Repaired bugs:

There was an error in the nonlinear GMM module (NLGMM):
The weight matrix was stored as an integer array. It is now stored as a
double precision array.

There were some missing words in the text of some of
the Help items in module ABOUT.

In the case of crosssection data (only) the linear
least squares (OLS) module did not recognize missing values in between
nonmissing values. Thanks to Dr. Woodard R. Springstube (Concordia University
at Austin, TX) for pointing this problem out to me.

If you reestimate a time series model with ARMA or
(G)ARCH errors, and the effective sample size is very small, it is possible
that the nonseasonal ARMA and (G)ARCH coefficients are not displayed.
Now you will get a warning message, and the minimum number of ARMA and
(G)ARCH coefficients is set to 1.

Changes and additions:

Some improvements have been made in the (generalized)
eigenvalue procedure. These improvements may affect the cointegration and
cotrending modules, the matrix calculation module, and the correlation
matrix module.

The correlation matrix module as been extended with
the variance matrix and its eigenvalues and eigenvectors.

If you start nonlinear GMM from consistent starting
values, you will now have the option to skip the first iteration round
with weight matrix W = I.

Differences between the versions
April 7, 2002, and July 4, 2002:

Changes and additions:

The modules that use the simplex method of Nelder and
Mead for function minimization or maximization (ARIMA, NLLS, ML, NLGMM,
NEXTARMA, NEXTGARCH, ICMQUANT) have been endowed with a new window which
explains how the simplex method works, and enables you to adjust the stopping
rules.

The userdefine nonlinear models modules (NLLS, ML,
NLGMM) now allow you to add comments to the EasyReg code lines.

If you make a series of linear or multiplicative combinations
of variables (via the "transformation of variables" module TRANS) involving
the same number of variables, you will now have the option to use the same
coefficients as before rather than typing them in each time.

In the modules OLS, LIMDEP, ARIMA, NLLS, ML, and NLGMM
the variance matrix is now displayed in floating point format.

A new module (EMAILME) has been added explaining the
conditions under which you are welcome to email me with questions. This
module is accessible via "WWW" > "Send me an email". I have gotten too
many questions for which the answers can be found on the EasyReg web page,
or given in the guided tours and the Help menu.

Quite a few new EasyReg users run EasyReg before downloading
the guided tours, and then email me with questions which have already been
answered in the guided tours (see the previous item). Therefore, the guided
tours module (TOURS) now also shows the available guided tours that have
not yet been downloaded.

The modules for which a guided tour is available open
with a window saying that this module comes with a guided tour. This window
will now link to the corresponding guided tour on the web if it is not
downloaded.

Two new guided tours have been put on the EasyReg
International download page:

ARIMA estimation and forecasting;

Discrete dependent variable models.

A new button "About the ICM test" has been added to
all the windows of the Integrated Conditional Moment (ICM) test module
(ICMMEAN), which activates a PDF file containing a review
of the ICM test.

The "Prepare time series for forecasting" module FORCPREP
now allows to remove missing values at the end of the file, by "adding"
a negative number of missing values.

The main module EASYREG.EXE has been redesigned, in
order to make it faster to load. In particular the "large menu" option
has been deleted, and the Help menu items have been replaced with a link
to a web based help file HELP.HTM in the EasyReg folder. The large menu
option was obsolete, because you can set the Windows fonts large or extra
large, which will make the menus readable if you run EasyReg in full screen
mode.


Repaired bugs:

If you run the ARIMA modeling and forecasting module
and the time series is short, it is possible that the nonseasonal ARMA
coefficients are not displayed. This bug was repaired in the NEXTARMA and
NEXTGARCH modules, which are activated if you reestimate a time series
model with ARMA or (G)ARCH errors, respectively, but I had forgotten to
repair it in the ARIMA module as well. Thanks to Nick Kiefer for pointing
this problem out to me.

The option in the modules ARIMA, NEXTARMA and NEXTGARCH
to change the start values of the simplex iteration manually did not work.
It does now.

A few nonessential cosmetic bugs in module ARIMA have
been repaired.

The transformations involving a constant in the userdefined
nonlinear models modules NLLS, ML, and NLGMM did not work. They do now.

Differences between the versions
July 4, 2002, and September 6, 2002:

Changes and additions

The Tobit model module (TOBIT) has been redesigned,
and a new guided tour on Tobit models has been added. This guided tour
can be downloaded from the EasyReg download page.

A new module (HECKMAN) conducting maximum likelihood
estimation of Heckman's sample selection model has been added. The corresponding
separate guided tour can be downloaded from the EasyReg
download page.

The 'Fonts' menu option has been removed from the EasyReg
main Window, because

It was actually a design tool to figure out the best
fonts for EasyReg;

I don't think anybody else used it;

It caused unexpected font changes, and I have not been
able to figure out why.

New integral transformations have been added to the
userdefined nonlinear models modules, ML, NLLS and NLGMM, and some other
obsolete integral transformations have been removed. Also, some of the
names of the transformations have been changed (Your templates still work,
though). Moreover, it is now possible to execute code lines conditionally
on the value of a dummy variable. Furthermore, quite a few new debugging
options and features have been added. Click here
to see what is new.

Some cosmetric upgrades.

Repaired bugs

If you run a Poisson regression, via module LIMDEP,
and the dependent variable Y takes large (integer) values, it was possible
to get the error message "RunTime error 6 Overflow". The reason is that
the dependent variable was treated as an integer (which it is), but integers
in Visual Basic have a limited range. The dependent variable in Poisson
regression is now treated as a double precision variable, so that this
error will no longer occur.

The option "Write residuals to the input file" in the
"What to do next?" module (NEXTMENU) did not work (It gave an error message).
It works now.

Differences between the versions
September 6, 2002, and October 11, 2002:

Changes and additions

A new module has been added which converts monthly time
series to quarterly time series. This module is accessible via Menu > Input.

The menu item "Exit > Remove EASYREG.DAT and exit" now
shows the contents of the input file first, and provides three options
for removing the folder EASYREG.DAT:

Remove it completely.

Move it from its current location to a folder "TrashBin"
so that it can be recovered.

Do not remove it.

A new separate guided tour on VAR innovation response
analysis is now available. See the EasyReg download
page.

Repaired bugs:

The option to select a subsample in module ARIMA and
the userdefined nonlinear models modules ML, NLLS and NLGMM did not work:
it gave an error message. It works now.

A bug in the network version of EasyReg has been repaired.

The module DATAOLD.EXE which reads data files in former
EasyReg default format did not work properly if the missing value code
is not an integer. It does now.

There was a nonessential bug in the structural VAR
part of the VAR innovation response module (VAR.EXE): The two methods to
compute standard error bands were reversed.

If you import a bitmap file in Paint, via module PICVIEW,
EasyReg calls PBRUSH.EXE. This works fine under Windows 95 and 98, but
not under Windows 2000 and XP, because in the latter Windows versions the
Paint program is now called MSPAINT.EXE. EasyReg will now call PBRUSH.EXE
first, and if this fails it will call MSPAINT.EXE.
The first three bugs are recent ones (July 2002), due
to minor changes I made in the modules involved.

Differences between the versions
October 11, 2002, and December 2, 2002:

Changes and additions:
 It is now possible to import pictures directly in MS Paint,
rather than saving them first and viewing them via module PICVIEW. The latter
option still exists, though.
 The discrete dependent variable module (LIMDEP) has been
extended with negative binomial probit and logit count data models.
 The nonparametric density and regression estimation modules have been
redesigned. In particular, kernel density estimation and SMINK density estimation have
been separated in different modules (KDENSITY and SMINKDEN), which both come with
PDF and HTM files explaining these procedures. Moreover, the 3 dimensional plot procedure
has been put in a separate module (PLOT3DIM).
 The unit root module (UNITROOT) has been redesigned and extended with
Jörg Breitung's nonparametric unit root tests:
 Breitung, J. (2002): "Nonparametric Tests for Unit Roots and
Cointegration", Journal of Econometrics 108, 343364.
 Also, Breitung's nonparametric cointegration tests have been included, under
'Menu > Multiple equation models'.
 The VAR module has been augmented with forecast error variance
decomposition, and the guided tour involved has been adjusted accordingly.
 Bierens' nonparametric SMINK regression estimation (SMINKREG) as been included, under
'Menu > Single equation models'. This module comes with
PDF and HTM files explaining this procedure. The latter file, SMINKREG.HTM, and module
SMINKREG itself, contain links to the original paper:
 Bierens, H.J (1983): "Sample
Moments Integrating Normal Kernel Estimators of Density
and Regression Functions", Sankhya 45, Series
B, 160192.

Repaired bugs:
 A bug in the Chebishev polynomial part of the OLS module has been
repaired.
 The appearance of EasyReg under Windows XP was not optimal,
because Windows XP changes the default background color of Windows applications.
This problem has been repaired. The appearance
of EasyReg under Windows XP is now about the same as under Windows 98, except for the push
buttons and the menus.