AR Autoregressive errors ARCH Autoregressive conditional heteroscedastic errors OLS Ordinary least squares PLS Partial least squares POISSON Poisson regression QR Quantal response (logit, probit, ordered) STEPWISE Stepwise regression SURE Seemingly unrelated regression estimation VAR Vector autoregressive 2SLS Two stage least squares 3SLS Three stage least squares
Non-linear models include:
FIML Full information maximum likelihood GMM Generalized method of moments ML Maximum likelihood NLS Non linear least squares Step-algorithms include:
BFGS Broyden-Fletcher- Goldfarb-Shanno algorithm BHHH Berndt-Hall-Hall- Hausman algorithm DFP Davidon-Fletche-Powell algorithm DW Dennis-Wolkowicz algorithm GA Genetic algorithm GAUSS Linearized in variables algorithm GN Gauss Newton algorithm GO Global optimization NM Nelder-Meade search algorithm NM Newton-Raphson algorithm SA Simulated annealing Step-size methods include:
LS Line search QP Quadratic programming TRUSTTrust region Both the White and Newey-West robust estimators are supported, as well as the Murphy-Topel two step estimation. The defaults used during non-linear estimation can be altered heuristically during execution, or through a script file. Gradients, Hessian, and Jacobian are estimated numerically as the default; however they can be written as a procedure by the user. Coefficient restrictions can be imposed with PARAM, and investigated using ANALYZ, which can be used following either linear or non-linear estimation. Descriptive statistics, automatic treatment of missing values, and weighted analysis are standard.
Non-linear Processes
The maximum likelihood (ML) procedure permits the estimation of any specified likelihood - Gaussx includes examples for a number of non-linear processes:
AGARCH Asymmetric GARCH process ANN Artificial neural network ARCH Autoregressive conditional heteroscedastic process ARFIMA Autoregressive fractional integrated moving average process ARIMA Autoregressive integrated moving average process ARMA Autoregressive moving average process DBDC Double-bounded dichotomous choice process EGARCH Exponential GARCH process FIGARCH Fractionally integrated GARCH process FMNP Feasible multinomial probit FPF Frontier production function process GARCH GARCH process IGARCH Integrated GARCH process KALMAN Kalman filter LOGIT Binomial logit process MGARCH Multivariate GARCH process MNL Multinomial logit MNP Multinomial probit MSM Markov switching models MVN Multivariate normal process NEGBIN Negative binomial process NPE Non parametric estimate ORDLGT Ordered logit process ORDPRBT Ordered probit process PGARCH Power GARCH process POISSON Poisson process PROBIT Binomial multivariate probit process SV Stochastic volatility process TGARCH Truncated GARCH process TOBIT Tobit process VARMA Vector autoregressive moving average process WHITTLE Local Whittle process Constrained Optimization
A choice of numeric, analytic, or symbolic derivatives is available for FIML, ML and NLS. The default method of deriving gradients and Hessians is numeric, using finite differencing. Analytic derivatives are specified by the user, while symbolic derivatives are calculated using the automatic differentiation capability of Maple 9. Symbolic gradients and Hessians can be saved as procedures and reused. Analytic differentiation work only for Gaussx for Windows, and requires Maple 9 or higher.
PANEL estimates the coefficients of a linear regression model for panel data.. Both fixed effects and random effects (error components model) for balanced and unbalanced models are supported. The Hausman test for testing the orthogonality of the random effects and the regressors, and the Bresuch Pagan test for random effects are implemented.
Survival models (life data regression) are estimated using ML for both uncensored or censored data. Supported parametric models include:
BETA Beta process COX Cox proportional hazards process EXPON Exponential process GAMMA Gamma process GOMPERTZ Gompertz process GUMBEL Gumbel (largest extreme value) process INVGAUSS Inverse Gaussian process LOGISTIC Logistic process LOGLOG Loglogistice process LOGNORM Log normal process NORMAL Normal process PARETO Pareto process PEARSON Pearson process SEV Smallest extreme value process WEIBULL Weibull process Survival measures, based on the last survival model estimation, include survival rate, inverse survival rate, hazard rate, cumulative hazard rate and cumulative failure rate. For each measure, the rate, the standard error, and the lower and upper confidence band are reported for each observation. Residuals include ordinary, standardized, Cox-Snell, deviance, martingale, Schoenfeld, scaled Schoenfeld and score. The indices can be linear or non-linear.
The Cox proportional hazards model supports the same survival and residual measures; ties are treated using the Breslow, Efron, discrete and exact methods.
Non-parametric processes (support the same survival measures, using either the Kaplan-Meier or Nelson-Aalan algorithms.
AR(k) with hetersoscedastic residuals Binomial probit Heteroscedastic binomial probit Multinomial probit OLS, residuals distributed normal OLS, residuals distributed t OLS, heteroscedastic residuals Poisson Tobit SURE
Continuous Distributions
beta | beta4 | boxcox | burr |
cauchy | chisq | chisq_scaled | erf |
expon | f | f_scaled | fatiguelife |
fisk | foldednormal | frechet | gamma |
ged | gengamma | genlogistic | genpareto |
gumbel | halfnormal | invgamma | invgauss |
johnson_sb | johnson_sl | johnson_su | laplace |
levy | loggamma | logistic | loglog |
lognorm | maxwell | ncchisq | ncf |
nct | normal | pareto | pearson |
pert | power | rayleigh | reciprocal |
sev | skewnormal | students_t | t_scaled |
triangular | uniform | vonmises | weibull |
Discrete Distributions
bernoulli | binomial | geometric | hypergeom |
logarithmic | negbin | poisson | rectangular |
step |
AMORT Amortization schedule FRONTIER Markowitz efficient frontier FV Future value GINI Gini coefficients LP Linear programming MCALC Mortgage calculation ME Maximum entropy PV Present value RSM Response surface methodology and multi response optimization SOLVE Solve a system of equations SPECTRAL Power spectrum estimation WELFARE Consumer surplus (CV, EV, MS and deadweight loss)
Parametric AD Anderson-Darling normality test, with censoring ANOVA Analysis of Variance ARCH Engle's ARCH test. BARTLETT Bartlett's Test for equality of variances: BKW Belsley-Kuh-Walsh collinearity test BP Breusch Pagan homoscedasticity test CHISQ Chi squared test. CHOW Chow stability test. DF Dickey-Fuller unit root test. EG Engle-Granger cointegration test F F test FTEST Linear restriction test GRANGER Granger causality test HANSEN Hansen test of overidentifying restrictions. HAUSMAN Hausman specification test JB Jarque-Bera normality test. JTEST Davidson and MacKinnon's J-Test for non-nested estimations. KPSS KPSS stationarity test. LBQ Ljung-Box Q test. LM Lagrange Multiplier test LRT Likelihood ratio test NW Newey West D test PIT Probability integral transformation test PPC Probability plot correlation test RECURS CUSUM and CUSUM-squared tests for stability SF Shapiro-Francia normality test, with censoring SW Shapiro-Wilks normality test, with censoring THEIL Thiel's decomposition of two vectors TTEST T test WALD Wald test WELCH Welch's test for equality of means Non-Parametric BF Brown-Forsythe test of scale CONOVER Conover test for treatment FRIEDMAN Friedman test for treatment KS Kolmogorov-Smirnov test KURTOSIS Kurtosis test KW Kruskal-Wallis test of location LEVENE Levene test of scale MOOD Mood's test of location MW Mann Whitney U test of location OBRIEN O'Briens test of scale RUNS Runs test of randomness SIGN Sign test for treatment SKEWNESS Skewness test WALSH Walsh test for treatment WILCOXON Wilcoxon test for treatment
ANOVA N-way analysis of variance for fixed, random or mixed models. Nested and interaction effects, variance components. CATALOG Descriptive comment for each variable CLUSTER Cluster groups and dendrogram CORDIM Correlation dimension COVA Means, standard deviations, minimum and maximum, sum, covariance and correlation matrices, autocorrelogram, partial autocorrelogram. CROSSTAB Cross tabulation of data DIVISIA Divisia indices FREQ Frequency distributions LYAPUNOV Lyapunov exponent PRIN Principal components SAMA Seasonal adjustment (including Census X12), SVD Singular value decomposition TABULATE Tabulates data across two class variables
ACF Autocorrelation function ACV Autocovariance function COMBS All k combinations of a vector DECONV Deconvolution INTERP Interpolation INVERT Find the inverse values of a function. ISCHAR Tests for a character vector PERMS All permutations of a vector POLYDIV Polynomial division POLYINV Polynomial inversion
Windows Project Control Screen |
GAUSSX for WINDOWS runs under Windows 2000, XP, Vista, Win7 and Win8. Gaussx for Windows requires GAUSS for Windows 6.0 or higher, and about 7 MB of hard drive. GAUSSX supports both 32 bit and 64 bit versions of GAUSS.
GAUSSX for UNIX and MAC runs in Terminal mode. Networking is built in, so that individuals will each have their own configuration file. The econometric specifications for the Unix version is identical to the Windows version. Gaussx for Unix has been designed to be machine independent by writing the entire package in GAUSS. Thus, if your Unix machine runs GAUSS, it will run Gaussx. Gaussx requires Gauss for Unix 4 or higher, and about 1MB of hard drive.
Pricing and Ordering
The package includes: source code, menu driven installation, tutorial, 50 sample command files with index, compiled HTML help for syntax, and a complete 450 page manual in PDF format (with reference section and index). A hard copy version of the manual is an optional extra. Single-user, network and student versions are available. Academic prices start at about $225, with a 30 day, no-question refund policy, and free technical support by phone and the internet. For technical information, contact Econotron Software, and for ordering information, contact Aptech Systems.