What’s new in TSP 5.0
(planned release: February 2005)

Click here for a PDF file with a complete list of new and updated features along with detailed documentation.
This PDF is currently dated August 2004; changes to LIML, BJEST, ANALYZ and PROBIT(REI,HCOV=Q) have been made since then.

 

Completely revised User’s Guide and Reference Manual, in a new compact format.

A completely revised HELP system for Givewin/TSP (context-based) and Windows TSP.

 

New PANEL estimation commands:

PANEL(REI) Random Effects (Individual), using ML estimation. ML provides results which are easier to reproduce (versus the many choices of variance components estimators for 2-step GLS). Automatic checking for multiple optima. ML also yields t-statistics for the variance components, and direct Likelihood Ratio tests.

PANEL(REIT) Random Effects (Individual and Time), using ML estimation. It handles unbalanced data, for large numbers of individuals. It can also handle nested two-way random effects, such as Individual and Group.

AR1(REI or FEI) Random or Fixed Effects (Individual), via exact ML estimation. FE also via conditional ML (NLS) estimation, which can handle irregularly spaced data and a Common Factor test.

2SLS(FEI), LIML(FEI), SUR/3SLS/GMM/FIML(FEI) Fixed Effects (Individual) for panel data (linear equations only). No limit on the number of Fixed Effects. Standard errors are computed for the fixed effects in 2SLS and FEI. Computed by individual mean-differencing the data and instruments prior to estimation.

PROBIT(REI or FEI) Random or Fixed Effects (Individual) for panel data. No limit on the number of Effects. The user may specify the number of Hermite quadrature points for random effects (default 20, no upper limit). Standard errors for FEs. QMLE SEs can be computed using a panel-grouped BHHH matrix (HCOV=Q).

Improved graphics:

PLOT, GRAPH(LINE) handle panel data, by inserting gaps between individuals. GRAPH(NOSORT,LINE) allows lines which are joined in time order, rather than being sorted by the X values. SYMBOL option puts large symbolts at each point. unix versions have graphics, using gnuplot interface and .GIF output.


Panel graphics using GiveWin

HIST graphics output (Windows/GiveWin version only). Improved to remove text output when there is color graphics.

BJIDENT,BJEST,BJFRCST,PLOTS(PREVIEW) (Windows/GiveWin version only). Replace the old text plots (differenced series, residuals, forecasts, CUSUM, CUSUMSQ) with graphics plots.

GRAPH(SURFACE) x y z; (Windows/GiveWin2 version only). 3D surface plot.

Other new estimation procedures:

LAD(LOWER=lowerlimit or UPPER=upperlimit) Censored Quantile Regression. SEs/VCOV via bootstrapping; also available for uncensored LAD and quantile regression.

TOBIT(LOWER=lowerlimit,UPPER=upperlimit) General 2-Limit Tobit. (Can handle both LOWER and UPPER in the same model). Default is still LOWER=0.

INTERVAL Interval regression. This is like an ordered probit, but the bound values between the categories are known.

KERNEL(BANDWIDTH=bandwidth,RELBAND=relativeband,IQR) x; or KERNEL y x; Gaussian Kernel density and univariate regression.

Improvements to existing commands:

ANALYZ extensions: (1) expressions which vary over observations (such as elasticities for several time periods), (2) asymmetric confidence intervals, using Monte Carlo. [Completed, 1/9/05].

BJEST(HCOV=U), ML PROC (HCOV=U) parameter Standard Errors from numeric second derivatives (new default). Usually much better than the BFGS or DFP rank 1 update methods. BJEST can now be used with ML PROC to estimate a linear or nonlinear regression with ARMA errors.

FIML(HCOV=C,HITER=C) parameter Standard Errors from Discrete Hessian (very close approximation to analytic Hessian, based on numerical differences of the analytic first derivatives). Can also be used for iteration.

LIML(BEKKER) new default Bekker standard errors, which are more robust to problems with many/weak instruments. The estimated concentration parameter and Cragg-Donald F-statistic are printed, for detecting models where the results are sensitive to many/weak instruments.

Regression diagnostics added: (1) Chow test robust to heteroskedasticity.
(2) AR1(OBJFN=GLS) - Common Factor test ( AR(1) vs. OLS including y(-1) and X(-1) ).

READ now handles all current Excel files (2/3/4/5/7/98/2000/2002). This is especially helpful for datasets with more than 32768 rows, although there is still a limitation of 65535 rows. READ can also handle Stata v7 files.

Note that many (but not all) of these features are already shipping on the current TSP 4.5 CD-ROM).

Countdown to the release of TSP 5.0: