Benchmarks
As part of our quality control we compare the results of Betahat with a variety of references. Some of these references are from textbooks, other programs, and published statistical benchmarks. We do not publish all of our hundreds of benchmarks since that would take time away from development, but we do publish a few for those that like to be reassured of the accuracy of our results.
Published benchmarks include
- Linear Regression
- Nonlinear Regression
- Multivariate t
- Logistic
- Probit
- Poisson Regression
- Two-Stage Least Squares - Linear, Nonlinear, Linear AR(1), Nonlinear AR(1), LIML
- Linear GMM, 6 different lag types
- Nonlinear GMM, 6 different lag types
- Pooled Estimation
- Pooled Estimation with Groupwise Heteroskedasticity
- Pooled Estimation with Groupwise Heteroskedasticity and Cross Sectional Correlation
- Linear Seemingly Unrelated Regression
- Nonlinear Seemingly Unrelated Regression
- Linear Three-Stage Least Squares
- Nonlinear Three-Stage Least Squares
- Random Number Generator
Linear Regression
The following table provide links that compare the results of Betahat to the benchmarks provided by National Institute of Standards.
Nonlinear Regression
The following table provide links that compare the results of Betahat to the benchmarks provided by National Institute of Standards.
| Betahat Results | Betahat command file | Data | NIST reference |
| Misral1a.out | Misral1a.prg | Misral1a.prn | Misral1a |
| Chwirut2.out | Chwirut2.prg | Chwirut2.prn | Chwirut2 |
| Chwirut1.out | Chwirut1.prg | Chwirut1.prn | Chwirut1 |
| Lanczos3.out | Lanczos3.prg | Lanczos3.prn | Lanczos3 |
| Gauss1.out | Gauss1.prg | Gauss1.prn | Gauss1 |
| Gauss2.out | Gauss2.prg | Gauss2.prn | Gauss2 |
| DanWood.out | DanWood.prg | DanWood.prn | DanWood |
| Misralb.out | Misralb.prg | Misralb.prn | Misralb |
Multivariate t
Multivariate t with uncorrelated errors, and with independent errors
Betahat command file Betahat Results
Discrete Dependent Variable Models
Logistic Betahat command file Betahat Results
Probit Betahat command file Betahat Results
Poisson regression Betahat command file Betahat Results
Two-Stage Least Squares, LIML and GMM
Linear, Nonlinear, Linear AR(1), Nonlinear AR(1) Betahat command file Betahat Results
Limited Information Maximum Likelihood Betahat command file Betahat Results
Linear GMM, with lag types White's, Bartlett(Newey-West), Truncated, Parzen, Quadratic Spectral, and Tukey-Hanning. Betahat command file Betahat Results
Nonlinear GMM, with lag types White's, Bartlett(Newey-West), Truncated, Parzen, Quadratic Spectral, and Tukey-Hanning. Betahat command file Betahat Results
System Estimators
The benchmarks for the pooled estimators use the data file POOLstr.DT7 that is supplied with the demo. The Betahat values and the benchmark values are in the result files.
Pooled Estimation Betahat command file Betahat Results
Pooled Estimation with Groupwise Heteroskedasticity Betahat command file Betahat Results
Pooled Estimation with Groupwise Heteroskedasticity and Cross Sectional Correlation Betahat command file Betahat Results
Linear Seemingly Unrelated Regression Betahat command file Betahat Results
Nonlinear Seemingly Unrelated Regression Betahat command file Betahat Results
Linear Three-Stage Least Squares Betahat command file Betahat Results
Nonlinear Three-Stage Least Squares Betahat command file Betahat Results
Random Number Generator
The default random generator in Betahat is called the Mersenne Twister. The source code can be freely downloaded from http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt.html, along with some details about the algorithm. To make sure the algorithm was implemented properly we wrote a 10 megabyte file of random numbers from Betahat so we could test the generator using a test suite created by George Marsaglia. The test program he wrote is freely available at his website. The results of the test suite are quite lengthy, but the numbers passed all of the tests.
