EasyReg International 2010

Previous versions of EasyReg

• SimplReg, the DOS predecessor of EasyReg.
• Versions 1.08-1.12 (1997)
• Versions 1.13-1.20 (1998)
• Versions 1.21-1.30 (1999)
• EasyReg 2000
• EasyReg International, 2001 versions
• EasyReg International, 2002 versions
• EasyReg International, 2003 versions
• EasyReg International, 2004 versions
• EasyReg International, 2005 versions
• EasyReg International, 2006 versions
• EasyReg International, 2007 versions
• EasyReg International, 2008 versions
• EasyReg International, 2009 versions
• EasyReg International, 2010 versions

Current and future versions of EasyReg

• Current version of EasyReg International
• Upgrade procedure
• Upgrades not yet included in the current version
• Planned future upgrades

Differences between the versions October 1, 2009 and February 20, 2010:

Improvements

The annoying message that the current version of EasyReg is more than three months old has been removed because I don't have the time to upgrade EasyReg often. Instead, a new item, "Check for upgrades", has been added to the WWW menu which opens the current web page.

The previous versions of EasyReg assumed the presence of a C drive. However, in rare cases there may not be a C drive, or may not be accessible by Windows, for example on computers with multiple operating systems. In that case EasyReg gave an error message and crashed. This problem has now been solved. Instead of looking for the C drive, EasyReg will now look for the first accessible hard drive. Thanks to Greg Barbieri for testing this solution. Consequently, EasyReg can now run on computers with multiple operating systems.

Since the upgrade procedures via the WWW menu requires access to folder c:\Temp, these options are no longer available if the C drive is missing or folder c:\Temp is not accessible by Windows.

Bug repairs

The transformation of variables module TRANS did not allow for negative values of the variable "a" in the transformations max(x,a), min(x,a), the dummy variables and missing value transformations. It does now.

A cosmetic bug in the linear and multiplicative transformation windows of module TRANS has been repaired.

A bug in the quantile regression module QUANTILE has been repaired. The effect of this bug was that multicollinearity was not detected. It is now. Thanks to Shaun McPeck for pointing this out to me.

Differences between the versions April 9, 2010 and March 28, 2011:

Additions and improvements

The maximum and minimum quantile indices in the quantile regression module QUANTILE have been relaxed to 99% and 1%, respectively. However, you will get warning messages if you set the maximum beyond 90% or the minimum below 10%.

The module COINTJ which conducts Johansen's cointegration tests has been updated, as follows.

• The options that cointegration restrictions on the intercept or time trend parameters are assumed but not imposed have been removed because they are confusing.
• A new help window has been added which guides the user in choosing one of the remaining options, i.e., cointegrating restrictions on the intercept parameters imposed, no cointegrating restrictions on the intercepts parameters (in both cases with or without seasonal dummy variables), and cointegrating restrictions on the time trend parameters imposed.
• A minor bug has been repaired. This bug gave occasionally an incorrect warning that the test of these cointegrating restrictions rejects the latter.
• The Bierens-Martins test for time-varying cointegration has been added as an option.
  Reference:
  • Bierens, H. J., and L. F. Martins (2010), "Time Varying Cointegration", Econometric Theory 26, 1453–1490.
    (http://econ.la.psu.edu/~hbierens/TVCOINT.PDF)

Modules ARIMA and ARIMAMODSEL have been updated. If you difference more than 1 time then you get now a warning message saying that time series that need more than once to be differenced in order to make them stationary are very rare. The windows involved have been redesigned.

The simplex iteration in modules NEXTARMA and NEXTGARCH is now automatically restarted until the log-likelihood does not change anymore.

The online registration for new EasyReg users via a "Devious visitors" e-mail has been discontinued because it takes me too much time to keep up with these emails.

Bug repairs

When estimating a small N - large T panel data model, the GMM module gave an error message. After recompiling module GMM it worked fine.

A cosmetric bug in module PVAL has been repaired.

Differences between the versions March 28, 2011 and December 3, 2011:

Improvements

The help file (HELP.HTM) has been updated.

In specifying a linear regression model, there is no need to add the constant term 1 to the data first, as the OLS module will ask you whether you want to include an intercept. If you add the constant 1 to the data (via Menu > Input > Transform variables > Constant = 1) and include it as a regressor in another position than the last, EasyReg did not recognize the intercept and therefore did not conduct the overall F test and the Breusch-Pagan test because these tests require an intercept. EasyReg now recognizes the presence of an intercept regardless how it is added to the model, and compute the overall F test and the Breusch-Pagan test if the intercept is present.

The Newton iteration for the Probit model in module LIMDEP has been improved, so that it now runs faster.

Bug repairs

The recursive out-of-sample forecasting option in module NEXTGARCH has been removed because it did not work properly.

A bug in module HECKMAN has been repaired.

Upgrade EasyReg to the current version

Note that EasyReg can also be upgraded to the current version via the EasyReg menu item "WWW > Upgrade EasyReg to the latest version." However, before you use this option, open

• ERIPATCH1.EXE.
  [Updated on July 13, 2011]

Upgrades not yet included in the current version

• January 29, 2012
  The option in the Johansen's cointegration module to conduct the Bierens-Martins test for time-varying cointegration now also creates plot data for the time-varying cointegrating vector(s). This plot data has to be imported in EasyReg as time series, so that the plots can be made via the time series plot module.

• February 6, 2012
  The problem under Windows 7 that some text and label boxes are blacked-out has been fixed.

• March 28, 2012
  In the auto/cross correlation module the computation of the Box-Pierce Q statistics is now optional, with default option not to compute them.

Planned future upgrades

I have gotten many requests from EasyReg users to include new features in EasyReg, such as tests for long memory, or double-censored Tobit models. However, EasyReg is a one-man's project, so I may not have the time to do that before I retire from PennState. Therefore, I will give priority to building in new features related to my own research, as the main purpose of keeping EasyReg free is to promote my own research. On the other hand, any bugs reported to me will be fixed straight away.

There are a few imperfections in EasyReg that have bothered me and which I am (occasionally) working on to improve:

1. The upgrade procedure via WWW does not work well under Windows 7. It will be replaced in due course with a direct downloading procedure instead of the current procedure via Internet Explorer.
2. Allowing for sample sizes over 32,767 observations. This maximum is due to the maximum integer value in Visual Basic. However, this is going to be a major operation, affecting almost all modules.
3. The 3-dimensional (3D) plots look sometimes ragged in the case of narrow shapes. This is likely due to the fact that the EasyReg 3D plots are based on a grid of rectangles rather than on a grid of triangles.
4. As reported in
   • Choi, Hwan-sik and Nicholas M. Kiefer (2005), "Software evaluation: EasyReg International", International Journal of Forecasting 21, 609-616,
   the accuracy of the numerical function optimizer in EasyReg (based on the Nelder and Mead simplex method), which is used for models with possibly non-unimodal objective functions, in particular the user-defined nonlinear models and ARMA models, is inferior to the accuracy of the numerical optimizer "nlinfit" in Matlab 7. However, as Choi and Kiefer mentioned, most numerical optimizers have the same problem with the particular nasty data set involved. On the other hand, in my experience the accuracy of the EasyReg numerical optimizer is adequate for most actual and simulated data.

Back to the EasyReg page

Back to the EasyReg download page