Betahat has all the standard methods to estimate, test, and simulate econometric models.  The equations can be time series or structural, linear or nonlinear, estimated as a single equation or in a system.  The features that are distinctive to the program include robust methods for reading data, a dual mode interface, and econometrics methods including a very general nonlinear maximum likelihood model solver, Vector ARIMA, multivariate stochastic variance systems, and stochastic simulation.

  Starting with version 2.0 a command/batch interface was added to the program.  The command mode interface made larger scale macroeconometric models more feasible using Betahat since they usually have a very large number of data series, equations, and identities, and they are commonly run as a large set of instructions.

With Betahat you get a choice of interfaces

Betahat has both a menu interface, and a batch/command mode interface.  The menu interface works well for for exploratory type work with datasets that have a fairly small number of series and the models do not have a huge number of variables.  The batch/command interface works very well where the user needs to perform a series of tasks on a repetitive basis or if there are a large number of variables in the regressions or data set.  The batch/command interface makes larger scale macroeconometric models feasible using Betahat.

Flexible Data Handling

Betahat can read and merge data from Lotus, Excel, Access, dBase III, dBase IV, FoxPro 2.0, FoxPro 2.5, Visual FoxPro, TROLL(TM),  DIF, ASCII(text), and its own file format DT5.  Once data is read into the program it is presented in a spreadsheet editor where it can be viewed, edited, sorted, the periodicity can be changed, transformations can be performed with a wide variety of mathematical functions, and new series can be added.  The spreadsheet features cut and paste from the clipboard allowing for very easy importing/exporting data to/from other Windows programs.  If you have MS Excel on your machine, you can use Excel to edit data from within Betahat.

The full Census X-11 seasonal decomposition routines are available from within the program.  These are called from the spreadsheet and have 2 full screen entry forms where you select the options that control the type of computations performed.  The results are saved in the editor and are also placed in Betahat's spreadsheet so you can immediately use the de-seasonalized series in your regressions.

Data Bank

Betahat has a data bank where equations can be saved and recalled later.  This is how single equations are prepared  to be used in some systems.  If the equation is estimated the coefficients are saved as well so the equation could be simulated later without re-estimating it.

Graphics

Betahat also has a graphics section.  There are 2 different graphics engines available.  The first is native to Betahat and includes pie, bar, line, XY and other types of graphs and plots.  The second is only available to users of Microsoft Excel.  If you select this option the Betahat will use the graphics engine that is in Excel.  All of the graphics techniques and options in Excel are available in Betahat.

Model Types

1.                    Linear multiple regression(OLS), with options to

a)                   Save the residuals and predicted values.

b)                   Save the residuals if forecasting in sample, and save the predicted values.

c)                   Present a  Runs test for normality of the residuals.

d)                   Present the elasticities at the means and at the last observations of the series.

e)                   Present the variance-covariance matrix and correlation matrices.

f)                    Perform a ridge trace or do recursive residuals.

g)                   Perform heteroskedasticity tests.

h)                   Estimate the coefficient standard errors using the Newey-West or White's covariance matrix.

i)                     Perform conditional regression where some observations are excluded from the sample.

j)                     Test and impose linear restrictions on the coefficients.

k)                   Estimate mixed regressions.

l)                     Impose Polynomial distributed lags.

m)                  Simulate the equation with user supplied coefficients.

n)                   Estimate the model with weighted least squares.

o)                   Generate forecast statistics by holding out some observations and then forecasting in-sample.

p)                   Bootstrap of the residual and data. 

q)                   Display the X'X and X'Y matrices.

r)                    Display the eigenvalues, the sum of the eigenvalues, and the sum of the reciprocal of the eigenvalues of the X'X matrix.

s)                   Display a regression ANOVA.

t)                    Automatically run auxiliary regressions and regression error specification tests.

2.                    AR-1, including Cochrane-Orcutt, with or with out the Prais-Winsten transformation, grid search maximum likelihood, and Beach-MacKinnon.  The AR-1 models can be estimated with most of the options shown above for the linear model, however a few of the options for linear model are not relevant for AR-1 type models.  The AR-1 models also have options to control the convergence criteria.

3.                    Full ARIMA and ARMAX modeling with autocorrelation and partial autocorrelation functions.  The ARIMA and ARMAX section also has several options such as the ability to use weighted data in all calculations.

4.                 Other single equation model types include stepwise, logistic, models with Box-Cox transformations, exponential smoothing, descriptive statistics, nonlinear least squares, cross correlation, vector correlation, binomial and multinomial logit, Probit, Tobit, and Poisson. Unit root tests Dickey-Fuller, Augmented Dickey-Fuller, and Phillips-Perron are available, and robust regression methods least average error, multivariate t with uncorrelated errors, and multivariate t with independent errors.

5.                    Programmable nonlinear maximum likelihood.  The likelihood function can be broken down into several parts so the function is easier to read, modify, and specify.  For example an ARCH model could have one definition to specify the regular regression component, and another definition to specify the variance component such as:                                    

            e = Y - B1*X1
VAR = B2 + B3*e(-1)^2 
log likelihood = -.5*(1.837877 + LOG(VAR) + e^2/VAR )

            The program has “push-button programming” for popular model types Probit, Logit, ARCH, ARCH-M, GARCH, EGARCH, AARCH (augmented ARCH), QARCH (Quadratic ARCH), Log ARCH, NARCH (Nonlinear ARCH), TARCH (Threshold ARCH), GARCH t-distribution, dependent variable heteroskedasticity, multiplicative heteroskedasticity, variance is linear function of exogenous variables, standard deviation is linear function of exogenous variables, Box-Cox transformation with unique gamma on each independent variable, Box-Cox transformation with unique gamma on the dependent variable and each independent variable, Tobit, Exponential Regression, Log-Normal Regression, Gamma Regression, Weibull Regression, and Generalized Gamma Regression.  All you have to do is select the model type and the programming is done for you, or you could program it yourself.

6.                    Methods for use with RHS(right hand side) endogenous variables include Two-Stage Least Squares, linear or nonlinear, with or without ARMA errors, Limited Information Maximum Likelihood (LIML),  linear and nonlinear Generalized Method of Moments with options for 6 different lag patterns.

7.                    For pooled or panel data there are model types pooled estimation, pooled estimation with groupwise heteroskedasticity, pooled estimation with groupwise heteroskedasticity and cross-sectional correlation, random coefficients, fixed effects, two way fixed effects, random effects, and two way random effects.  AR-1 correction for each model type is available. 

8.                    Linear systems model types Seemingly Unrelated Regression, and Three-Stage Least Squares.  These models use the single equation model types that have been specified by the user.  The individual  equations may use weighted estimation, restrictions, conditional estimation, PDL lags, or other options available for that type of model.  For example when estimating a SUR, any, or each equation in the system, could use weighted least squares or have its' residuals or predicted values saved.

9.                    Nonlinear systems model types Seemingly Unrelated Regression, Three-Stage Least Squares, generalized method of moments, and Full Information Maximum Likelihood (FIML). These models use the single equation model types, linear or nonlinear, that have been specified by the user.  The individual  equations may use weighted estimation, conditional estimation, ARMA errors, or other options available for that type of model.  Linear or nonlinear restrictions are allowed within or across equations.

10.                 Other system estimators include vector autoregressions, vector ARIMAX, and multivariate stochastic variance( SUR systems of nonlinear equations with various ARCH type variances, estimated by maximum likelihood).

Simulations

Betahat has comprehensive simulation capabilities.  Estimated equations from a wide variety of model types can be freely intermixed.  There are options for add factors and identities for both single equations and systems.  Stochastic simulation is easy to implement for both single equation and systems where the data, coefficients, and residuals are allowed to vary.  There is automatic comparison of any simulation to a base simulation to compare different model or system types.  The output is selectable so you can display only equations and identities that are of interest.  Simulation results may be saved to text, CSV, or Excel format, and may be saved in the Betahat database.

Algorithms include Gauss-Seidel, Jacobi, Newton and Stacked-Time Newton.  Automatic normalization allows a regression equation to be estimated (or a parameterized equation entered) and then you can specify which variable is endogenous.