History of Changes to Betahat


 
July 21, 2005

New demo of version 6.0.

 
May 26, 2005

Version 6.0 of Betahat is released. 

 

April 3, 2005


Development of the version 6.0 of Betahat is nearly complete and testing is underway. New features include

  1. The ability to open multiple command files at once, and to have one command file call another command file.

  2. The Fair-Taylor algorithm for simulating Rational Expectations models.

  3. A number of small changes to make importing models from other programs easier.

December 26, 2004

Version 5.0 of Betahat is released.

  1. The IMF Mark III model in Betahat format is available on the models page.

  2. The FRB/US model in Betahat format from the US Federal Reserve is available on the models page.

  3. Development of the next version of Betahat is well under way.
     


April 15, 2004


Development of Version 5.00 of Betahat is complete and testing is underway. In addition to the new features mentioned below in the December 20, 2003 update, version 5.00 includes

  1. Rich Text Format support  for command and results editors, including a toolbar, selectable font, and OLE.

  2. A new method for handling long lines in command mode.  In addition to using the "&" character as a line continuation character, Betahat now allows a single command to run to any number of lines, and interprets a comma as the end of line character.  This allows models from other packages to be read in and run with fewer modifications.

  3. A way to include the name of a CSV file on the SIM command that contains details for Monte Carlo runs.

  4. Four new ways to initialize the values when using the Newton or Stacked Newton algorithms.  

  5. The ability to define, use and save Namelists, which are lists of equations, identities, or series names.  This allows the user to use a namelist in place of a long list of identifiers.

  6. A new tab to compare different simulations.  This tab can read in 1 or 2  CSV files and display a table or chart of the levels, differences, or % differences. 

February 6, 2004

Many more benchmarks on the Benchmark page. The new benchmarks include Multivariate t with uncorrelated errors, Multivariate t with independent errors, Logit, Probit, and Poisson regression. New benchmarks for equations used in structural models include Linear & Nonlinear Two-Stage Least Squares, with and without AR1 errors, LIML, and GMM. New benchmarks for system estimators include Pooled Estimation, Pooled Estimation with Groupwise Heteroskedasticity, Pooled Estimation with Groupwise Heteroskedasticity and Cross Sectional Correlation, Linear & Nonlinear Seemingly Unrelated Regression, and Linear & Nonlinear Three-Stage Least Squares.

December 20, 2003

  
We've made much progress on version 5.00 of Betahat. The focus so far has been to continue to add features to the simulation part of the program. We've been working with a variety of large scale Rational Expectations models are are getting very good results.

The new features include  


  1. A new command CONVERTUNIX, which makes some formatting changes to UNIX files so they can be read into Betahat.

  2. Added syntax that allows sums and products to be computed using a loop variable.  The loop variable can also be used to refer to parameters.  These changes were needed to get the Bank of Canada QPM model running in Betahat.

  3. Added SETDATES command so the date and periodicity can be set from command mode.

  4. A new solver that is used in the Newton and Stacked-Time Newton algorithm.  The new solver is much faster than the older one we've been using.

  5. The ability to export/import a simulation in a standardized CSV format for further analysis, or for passing in the terminal conditions for a Rational Expectations model.

  6. New functions to the calculus engine (<, >, <>, >=, <=,=) which means they can now be used in Newton & Stacked-Time algorithms.

  7. A new IF THEN ELSE function using the EXCEL syntax.

  8. The ability to save datasets with over 255 columns of data to the Excel and Access format by using more than one table.

  9. A new option on the SIM command that allows the data in the spreadsheet to be updated with the endogenous series in a simulation.

  10. A better spreadsheet on the spreadsheet tab. 

December 19, 2003 

 

V 4.00 of Betahat is released. There is a new demo and help file on download page.  
The new features from V 3.2 include  

  1. Stacked-Time implementation of the Newton algorithm.

  2. A new function, named DEL(x:y), that will take the delta (difference) of formula y, using a lag of x time periods.

  3. Tabs on the sim form to make room for all the options, including under/over relaxation,  saving the analytic derivatives, and options for setting the initial values when using the Newton or Stacked-Time Newton algorithms.

  4. Excel can now be embedded in the spreadsheet tab so the user can use Excel to edit data from within Betahat.

  5. Resolution of a problem where the augmented Dickey-Fuller tests were not displayed.

  6. Added scalar variables as a data type.  This helps with models such as the Multimod Mark III model which use scalar parameters. 

  7. Support for reading TROLL(TM) FRM data files.

  8. An increase in the number of seasonal AR and seasonal MA parameters to 6 in ARIMA models.  In previous versions only 1 seasonal AR and 1 seasonal MA was allowed.

  9. Another interface for specifying linear models that are nonlinear in variables.  A regression equation can be specified by entering the formulas to use as independent and dependent variables, such as "LOG(Y) CONSTANT LOG(X)" to regress the log of Y on a constant and the log of X.

  10. Automatic normalization.  This allows a regression equation to be estimated (or a parameterized equation entered) and then you can specify which variable is endogenous.

  11. A new type of equation, called parameterized equations, which make it easy to use equations in simulations that were not estimated in Betahat.  These also make it easy to add policy variables to a simulation.

  12. A new command named ENDOGENOUS that allows the user to list the endogenous variables in a simulation when using the Newton or Stacked-Time Newton algorithm.


July 07, 2003

We have changed the name of our econometric program from Beta to Betahat.  
V 3.20 of Betahat is released. There is a new demo and help file on download page.  
The new features from V 3.1 include  

  1. The addition of a sparse Newton (Newton-Raphson) algorithm for simulations of structural systems.  Analytic derivatives are used to evaluate the Jacobian.  Details of the implementation can be found in the help file.

  2. A new simulation option that allows you to write to disk the values of each equation and identity at every iteration.  This can be used to help you track down instability in your model.

  3. A new simulation option that allows you to save the jacobian for each iteration when using the Newton algorithm. 


June 18, 2003   

V 3.10 of Beta is released. There are a number of new additions to the web site associated with this.  

  1. New demo and help file on download page.  

  2. Updated nonlinear benchmarks and nonlinear summary on the Econometric Benchmarks page. 

  3.  New benchmark on the recently added random number generator.   

The new features from V 3.0 include  

  1. The ability to add a tab of data to an existing Excel file, and a table to an existing Access database.

  2. The option to use the Excel graphics engine on the Graph tab. 

  3. A change to the command line code so that the most recently estimated model could be simulated without naming each equation.

  4. More options for multivariate stochastic variance models.

  5. Mersenne Twister algorithm for random numbers.

  6. The ability to add a residual to stochastic equations for stochastic simulation.

  7. The ability when doing a sim to easily use lagged values for undefined endogenous variables on the first iteration.  

  8. A second algorithm, Jacobi, for simulations , and Ragged Edge.

  9. Analytic derivatives for nonlinear least squares.



Version 3.0

  1. Moved from 16 bit code for Windows 3.1 to 32 bit for  32 bit Windows versions. 

  2. Added ability to handle > 32000 observations per series, which also required a new database format named DT3.  Tested dataset with 2,000,000 observations per series with no problem. 

  3. Added ability to read Excel Files.



Version 2.10

  1. Linear/Nonlinear Full Information Maximum Likelihood with linear/nonlinear constraints within/across equations. 

  2. Options that define estimation methods to eliminate some pop-up menus. 

  3. The ability to save the regression and simulation output to Lotus(tm) file format in a standardized way. This is to enable the user to automate routine forecasts with a minimum of effort. 

  4. The addition of another algorithm for models that require general optimization.

  5. The LOGIT model type can now allow multinomial LOGIT and a coding other than 0-1 for the dependent variable. 

  6. A new data file format with the file extension DT2 is used which allows: 
         Identities to be named. This is useful for users of the command/batch mode interface who previously had to rely on the USEALLIDENT option to specify identities. 
         The addition of comments about a data series. The comments are saved with the data and can be up to 32,000 characters per series.

  7.  The ability to write various database formats as well as read. This is useful if you need to perform SQL on data that is not in a database format. You can now read in from, say, text format, save the data in dBase, then read in the dBase format using SQL to select the observations to use. 

  8. The command/batch mode can now read database formats with SQL in a single command. 

  9. A much faster and memory efficient method for computing nonlinear 3SLS with the ability to impose nonlinear restrictions across equations. 

  10. Generalized Method of Moments for systems. 

  11. New estimators for panel data include Fixed Effects(One Way), Fixed Effects Two Factor, Random Coefficients, and Random Effects(One Way).

  12. The ability to use stratified data for the pooled and panel estimators.

  13. Additional methods to correct pooled models for AR-1 errors. 

  14. Static simulations in addition to the previously available dynamic simulations.



Version 1.30

  1. Multivariate stochastic variance.

  2. Various types of Generalized Method of Moments, linear and nonlinear.

  3. The addition of exogenous variables to the vector ARIMA model type.

  4. Many more unit root and co-integration tests.

  5. Multivariate t regression for robust regression, with uncorrelated or independent errors. 

  6. The ability to create price indexes.

  7. A very easy to use method to compare forecast values from one simulation to another, whether using the same model type or not.