H O M E / I N F O / P R O D U C T S / O R D E R / S U P P O R T

Contents / Getting started / FAQ / Recent changes / Updates

 

Detailed changes to TSP since January 2000

Detailed changes to TSP 4.5 and 5.0 are listed here by "type" (enhancements and bugfixes), most recent first. Workarounds are available for many of the bugs. Updates can be downloaded from our password-protected download area; send email to us requesting a password if you need an update.

If you think you have found a new bug, please inform TSP tech support, but please, do so only after checking the lists below to make sure it has not already been reported and fixed.

ENHANCEMENTS  (5.0)   2006   2005   (4.5)   2004   2003   2002   2001   2000   (4.4 and earlier)

  • 07-04-17 LIML now stores the log likelihood in @LOGL.  This may be useful for likelihood ratio tests when the set of instruments and endogenous variables remains the same.  The LOGL is a function of the log eigenvalue and the reduced form residual covariance matrix.  It is the same as that computed using FIML with linear equations for all the RHS endogenous variables as functions of the instruments.
  • 07-04-17 FORM(SUM,RESID) S3 X1-X3;  now creates  FRML S3 X1+X2+X3;  (an unnormalized equation).  Previously the RESID option was ignored when SUM was used.
  • 07-04-15 When running TSP in interactive mode, TSP will now read up to 500 characters per input line.  Formerly the limit was 80.
  • 07-02-09 ANALYZ(PRMEAN,PRSERIES) - new options when ANALYZ computes series.  PRMEAN is true by default, and prints summary statistics for computed series, such as mean, std.dev., min, max and median. PRSERIES is true when @NOB <= 100, and prints computed series  directly.  Previously PRSERIES was always true, and this caused occasional problems when series were computed and printed by mistake (when the user intended to analyze a scalar function). So now, if you want to print the computed series (and standard error and t-statistic for each observation) when there are more than 100 observations in the current sample, specify the PRSERIES option.
  • 07-02-24 FIML will detect a SUR model (Jacobian is the Identity matrix), which will speed up several of the computations.  This may be helpful  when FIML is used instead of LSQ on large models to reduce memory requirements.
  • 07-01-20 MODEL will suggest using SIML instead of SOLVE when the equations are not uniquely normalized, or when a LHS variable is not an  endogenous variable.

2006

  • 06-12-06 STEPMEM=n  a new nonlinear option for stepsize memory. The initial stepsize lambda is remembered from the previous iteration, and reset to lambda=1 every nth iteration.  The default is STEPMEM=1, which resets lambda=1 every iteration.  This option may be helpful if the natural stepsize for a model is different from 1 and tends to be about the same for successive iterations.  For example, in the ILLUS41 FIML testrun, the stepsize used is often 2.  Using STEPMEM=5 instead of STEPMEM=1 yields convergence in 17 iterations and 73 function evaluations, instead of 20 iterations and 96 function evaluations.
  • 06-10-29 SHOW SERIES  has more concise output, which leaves more room for printing documentation.
  • 06-10-29 READ Stata v8-v9 .DTA files
  • 06-10-28 GMM(MASK=matrix)  check that the number of active orthogonality conditions from the MASK is large enough to identify all the Params. Otherwise it might be difficult for a user to understand the cause of the nonidentified Params.
  • 06-09-05 LP command added for Linear Programming. LP(MAX/[MIN], NLE=<number of <= constraints>, NGE=<number of >= constraints>,     NEQ=<number of = constraints>, TOL=<eps>,SILENT)  <b_obj> <b_constraints> <rhs>;  LP solves for x to minimize phi = b_obj'x, with constraints  b_constraints*x <= >= = rhs  and  x >= 0 .  rhs >= 0  as well. The solution is stored as @COEF.  @PHI (objective function) and @IFCONV are also stored. The Simplex method subroutine from Numerical Recipes is used, which has a nice explanation of the method and some examples which were used for testing.  The LP command was also tested on LAD estimation with the Stackloss dataset. Here is an example of using the LP command to solve the first example  in the Numerical Recipes description:

  ------------

  ? Maximize:  2*x2 - 4*x3

  ? Subject to:  x1 + 6*x2 - x3 = 2

  ?           -3*x2 + 4*x3 + x4 = 8

  ? The solution is:  x2 = 1/3, x4 = 9  (x1=0, x3=0),

  ?  with objective function = 2/3 .

  read(nrow=1,ncol=4) objf1;

  0 2 -4 0 ;

  read(nrow=2,ncol=1) dep1;

  2 8 ;

  read(nrow=2,ncol=4) constr1;

  1  6 -1  0

  0 -3  4  1 ;

  LP(NEQ=2,MAX) objf1 constr1 dep1;

  ------------

  • 06-03-24 Batch input files can have different types of End-of-Line characters, to handle all types of text files.  For example, when downloading  examples from the TSP webpage (unix system) onto a PC, in the past TSP on the PC would not recognize the unix End-of-Line character (LF).  This is also helpful with Mac TSP, where the older MacOS used CR and the new one uses LF, so that both are supported. The input line length is also unlimited now (previous limit was 500 characters).  An upward incompatibility is that READ with a Fortran FORMAT is no longer allowed with data in the batch input file. The Fortran FORMAT still works on external files, though.
  • 06-03-24 Long programs with loops should use memory more efficiently now, because the number of variables which can decrease and increase  in size without wasting memory has been increased from to 10 to 100. When memory runs short, an automatic COMPRESS is used, which deletes @variables unexpectedly.  This should occur less frequently now. It is still helpful to create variables at their largest size first.
  • 06-02-10 OLSQ, and regression diagnostics - a zero R-squared is detected more accurately.

2005

  • 05-04-05 ANALYZ - more concise labelling for scalar functions with NDRAW=;  the new function names are used to label the rows in a combined table. Also, the label for the last column (number of draws used) is changed from num.good to num.draw .
  • 05-04-05 OPTIONS CRT;  is now the default style for echoing the commands in the output file.  This is more concise, since it does not have a large left margin, and makes long lines less likely to wrap around.

TSP 5.0 released March 2005 (list below applies to version 4.5)

  • 05-01-17 GRAPH(SURFACE) x y z; (GiveWin2 graphics version) gives a 3D surface plot (triangular).
  • 05-01-16 BJIDENT(PREVIEW) plots raw and differenced series, grouped with up to 4 plots per window. PREVIEW is the default in interactive mode; NOPREVIEW is the default in batch mode. NOPREVIEW yields the old character style plots in the output file or output window. PREVIEW works in conjunction with the default PLOT option (NOPLOT turns off the plots). The PREVIEW option is remembered for future BJIDENT, BJEST, BJFRCST, OLSQ, 2SLS and PLOTS commands.
  • 05-01-16 BJEST(PREVIEW) plots the residuals in a window.
  • 05-01-16 BJFRCST(PREVIEW) plots the forecast, actual value, and 95% confidence bounds in a window. Also works with the EXP option for exponential forecasts of log models.
  • 05-01-16 PLOTS(PREVIEW) turns on graphics plots for subsquent commands: 1. actual/fitted/residuals for OLSQ, 2SLS (2 plots, actual+fitted and residuals, grouped in one window) 2. CUSUM/CUSUMSQ for OLSQ (1 or 2 plots, CUSUM and/or CUSUMSQ, grouped in one window)
  • 05-01-16 PLOT, GRAPH, HIST - The default window name is now PLOT, GRAPH, HIST instead of "TSP Graphics".
  • 05-01-09 ANALYZ with series output. ANALYZ can now handle equations whose variable arguments vary over the sample - it will output results as series in this case. Some applications of this are elasticities, forecasts, and probability derivatives.
  • 05-01-09 ANALYZ(NDRAW=n, HALTON) computes asymmetric confidence intervals for nonlinear functions by drawing n simulated parameter vectors. These functions can vary over time as well. This is an alternative to the default "delta method" which uses derivatives and is exact for linear functions. HALTON specifies that a (shuffled) Halton sequence is used for the random draws. This provides more uniform coverage of the range of values, so it may yield more accurate integration for a given number of draws.

2004

  • 04-11-27 BJEST(EXACTML,NOCONST,MAXIT=0,MAXSQZ=123) U; These options will make BJEST simply compute @LOGL using parameter values in @START, where U is a regression residual.
  • 04-11-27 AR1(FEI), AR1(REI) now store @RES, @FIT. AR1(FEI) with the default OBJFN=ML computes @FIT and @RES the same as a regular AR1 command with individual dummy variables.
  • 04-11-08 SIML, SOLVE print partial results when a numeric error occurs. For example, if dynamic simulation has explosive growth, after a certain number of periods the values become too large to continue. Results are then printed, with further periods set to missing values.
  • 04-10-17 LAD (RESAMPLE=[BILIAS]/DIRECT). This implements the Bilias et al Journal of Econometrics (2000) method for computing bootstrap SEs for censored quantile/LAD estimation. It zeros observations where the predicted Y (@FIT) is censored. The number of observations zeroed is reported in the label for the SEs. RESAMPLE=DIRECT resamples from the original data and does censored estimation.
  • 04-10-08 BJFRCST(SILENT) - same as NOPRINT,NOPLOT.
  • 04-10-04 LIML(BEKKER) Bekker standard errors are now computed by default in LIML. These SEs are slightly larger than the original SEs from the k-class matrix, and they are more likely to be correctly sized in small samples and when there are a large number of instruments. They are also helpful when using the FULLER option. The estimated concentration parameter (mu^2) and Cragg-Donald F-statistic (CDF) are now computed and printed by default in LIML. For a single RHS endogenous variable, CDF is an F-statistic which tests if the (K2) excluded exogenous variables (Z2) have zero coefficients in the reduced form. It is also valid for multiple RHS endogenous variables. mu^2 = F*K2. They are stored as @MU2 and @CDF. These statistics can be used to assess if the model has a "weak instruments" or "many instruments" problem- AR1(FEI,FEPRINT)
  • 04-09-27 AR1(OBJFN=GLS) on panel data and irregularly spaced observations. AR1(FEI) is a fixed individual effects model, for panel data using FREQ(PANEL). Both OBJFN=ML (default) and OBJFN=GLS are available. When OBJFN=GLS is used, a Common Factor test is computed, as a likelihood ratio test vs. an unrestricted OLS regression of y on x, x(-1), y(-1) and fixed effects. The ML command can also be used to estimate an AR(1) model for irregularly spaced time series, and an example is available for this.
  • 04-09-27 PROBIT(REI,HCOV=QP), PROBIT(FEI,HCOV=QP). HCOV=P is a panel-grouped BHHH matrix. P = A", A = sum( G(i)*G(i)' ), where G(i) is the gradient for individual i. A reference is equation (13.53) on p.407 of Wooldridge, Econometric Analysis of Cross Section and Panel Data, 2002. HCOV=Q is exactly equivalent to HCOMEGA=BLOCK (with no df correction) for linear models. In linear models, A = (1/s4)*X'Omega*X, N = s2*(X'X)" .
  • 04-08-20 PANEL(HCOMEGA=[BLOCK]/DIAGONAL,HCTYPE=0/1); OLSQ(HCOMEGA=[BLOCK]/DIAGONAL,HCTYPE=0/1/2/3); 2SLS(HCOMEGA=[BLOCK]/DIAGONAL). The HCOMEGA option specifies the form of the Omega matrix E[u*u'] for use with ROBUST standard errors. When FREQ (PANEL) is in effect, the default is HCOMEGA=BLOCK, which is a block diagonal matrix; this yields SEs which are consistent to both heteroskedasticity and autocorrelation. The HCTYPE=0/1 option controls whether or not a degrees of freedom correction is applied as well. When FREQ(PANEL) is not in effect, HCOMEGA=DIAGONAL is the default, which yields SEs robust to heteroskedasticity. Additional labels have been added to robust SEs to indicate which type of Omega matrix is used, and if a degrees of freedom correction has been applied.
  • 04-08-20 FIML(HCOV=C,EPSMIN=value) - the EPSMIN option is now used to control the numeric step size when computing HCOV=C and HCOV=U. If you have parameters with values smaller than 1E-5 in magnitude, it will be helpful to use EPSMIN with a value somewhat smaller than your smallest parameter value. Otherwise, a step size that is too large is used and small parameters will appear to have zero standard errors.
  • 04-03-04 INTERVAL (LOWER=lowerlimit,UPPER=upperlimit) y_series list_of_X_series; does interval regression, similar to ordered probit but with known bound values between the categories.
  • 04-02-24 AR1(REI) - random effects - individual (ML estimator). Similar to PANEL(REI), but with an added AR(1) component. Follows Baltagi and Li, Journal of Econometrics, 1991.
  • 04-02-05 LIST(SUFFIX=name_or_value) listname arguments; - creates a list by applying a suffix to arguments.
  • 04-01-27 Text and Format strings can now contain semicolons.
  • 04-01-25 SORT(RANK,AVETIES/[NOAVETIE],MINTIES/[NOMINTIE]) series rank; The new AVETIES and MINTIES option control computation of the rank when there are tied values in the series. The default is to assign a random (and unique integer) rank for tied values.

2003

  • 03-12-18 Win32 version - if the batch input file does not exist, prompt the user to Press Enter, so that they have time to read the error message before the window closes.
  • 03-12-18 KERNEL(BANDWIDTH=bandwith,RELBAND=relativeband,IQR) x; or KERNEL y x; With a single argument, a Gaussian kernel density of x is computed and stored in @DENSITY.With two arguments, a Gaussian kernel regression of y on x is computed; the smoothed values of y are stored in @FIT.
  • 03-11-12 MSD - when all observations are missing, or the sample is empty, @MEAN and @MSD are now stored as missing values, instead of generating an error message.
  • 03-03-12 LAD (LOWER=lowerlimit,UPPER=upperlimit,NBOOT=bootstraps) - LAD can now handle lower or upper censoring, but not both in the same model. NBOOT is used to control the number of bootstrap replications for computing standard errors of the parameters.
  • 03-11-18 TOBIT(LOWER=lowerlimit,UPPER=upperlimit) - lowerlimit and upperlimit can be scalars or series. Can handle both lower and upper censoring in the same model.
  • 03-11-14 HIST (GiveWin2 graphics version) Same as GiveWin1-2 below, but with more control over number of bins and centering of bins for integer variables.
  • 03-11-14 HIST (GiveWin1-2 compatible graphics version) Default is now to do graphics version and skip text version.To get text output instead of graphics, use the NOPREVIEW option. If graphics-only options like CDF are used for the text version, they are ignored silently. Same for text-only options like WIDTH=. @HIST and @HISTVAL are always stored. MIN= and MAX= are now implemented. They have the effect of trimming away data which lies outside their range. TITLE='title string' now works. The default title is the series name. WINDOW='window title' works. The default window titles are 'TSP Graphics 1' , 'TSP Graphics 2' , etc. Multiple HIST arguments create multiple output windows (one histogram per window).
  • 03-11-14 HIST (text version) Automatically detects integer series and switches to an INTEGER mode, which is similar to DISCRETE mode. (INTEGER mode shows cells with zero counts in the range, while DISCRETE leaves them out).
  • 03-10-28 PANEL (REI,[ALL]/NOALL, nonlinear_options) does Random Effects Individual (One-way) ML estimation. VARCOMP is the GLS version of this. The variable which indexes Individual is given in a FREQ (PANEL, ID=individual_var, TIME=time_var) command. The data is assumed sorted by ID. If you want to estimate a model with random Time or Group effects, and no standard Individual effects, you can use FREQ (PANEL, ID=time_or_group_variable), but you need to make sure the data is sorted by this variable.
  • 03-10-28 PANEL(REIT) does Random Effects Individual and Time (Two-way) ML estimation. The variables which index Individual and Time are given in a FREQ (PANEL, ID=individual_var, TIME=time_var) command. The data is assumed sorted by ID. time_var is a non-negative integer variable which indexs the second variance component. It does not have to index time periods; it can index groups, so that models with nested structures can be estimated. The group could be the same value in all observations for an individual, which would be a nested value, or it can have a partially nested or non-nested (true time) structure. The Log likelihood parameterization is based on Peter Davis's article in Journal of Econometrics, January 2002, with simplifications from the 3-component model to the 2-component model
  • 03-10-28 PANEL(REI,REIT) now prints and stores the Ahrens-Pincus Unbalancedness measures (@APUI for I dimension - always, and @APUT for T dimension - if REIT is used). Titles for the different estimators within PANEL have been changed slightly and underlined in the output. LOGL and SBIC for the BYID model are now printed and a summary table of all the estimated ML models with LOGL, SBIC and title is printed after the last estimated model. A * is put next to the model with the lowest SBIC.
  • 03-10-22 SBIC is now computed from LOGL, using the number of identified parameters (@NCID) in the model. That is, the parameters with zero standard errors are not counted.
  • 03-08-10 RANDOM (NOREPLACE,DRAW=var) will draw from a series or matrix without replacement.
  • 03-06-28 LOGIT now handles missing values. When there are multiple records per case, one record with missing values will drop the entire case. The SUFFIX option is not described correctly in the manual. SUFFIX does not allow the user to select the set of alternatives being used; use a SELECT command for this instead. SUFFIX is used to give short names to the alternatives. SUFFIX does not imply COND. If some of the alternatives are never chosen, be sure to use SUFFIX or NCHOICE= to make sure the full set of conditional variables is used (corresponding to all available alternatives).
  • 03-06-06 BJIDENT(ESACF,NAR=nar,NMA=nma,[BARTLETT]/NOBART,PRINT/[NOPRINT]) series; computes extended Sample ACF of Tsay and Tiao, JASA, March 1984, p.84-96. Useful for identifying stationary and nonstationary ARMA models. Output normally includes the ESACF correlations, their P-values, and a table of Indicators (9 indicates P-value < .01, 1 indicates P-value between .01 and .05). These are stored as @ESACF, %ESACF, and @ESACFI.
  • 03-05-04 GMM(INST=(list1 | list2 | ... | listG)) eq1-eqG; new syntax for different instruments in each equation (like MASK=matrix, which still works and is improved now). It now uses a smaller COVOC matrix, which only uses the selected orthogonality conditions (the old implementation had a full sized COVOC matrix with zeroed rows and columns for the unused instrument x equation pairings). The instrument names are printed for each equation, unless the TERSE or SILENT options are used. The new implementation takes less memory and is faster than the old version
  • 03-04-17 DOT - DOT loops can now be nested up to 10 deep (so there can be up to ten dots in a name).
  • 03-03-14 MSD(ALL,WEIGHT=w) - the weight is now used for computing the quartiles
  • 03-03-14 long names - it is now possible to have more than 9999 long names. The new upper limit is 9999999. Usually you need a special large version of TSP to get even 9999 long names, and it can be slow to use this many different variable names. If TSP runs out of space to store long names, it will now print up to 10 of the longest names, so you can try to reduce their length if most names are shorter.
  • 03-02-03 CONVERT list_of_series - can now convert more than one series in a single command. Each series is replaced by its converted version. To store them under different names, either use the new = old form of the command, or use the COPY command to store the old series under the new names before running CONVERT.
  • CONVERT(MAP=mapseries,[SMPL]/NOSMPL) - computes SUM (default) or AVERAGE from old series to new, using a MAP of pointers. This is helpful for aggregating grouped data, such as industries, states, or individuals with panel data. The rows of the map correspond to the rows in the old series. The values in the map correspond to the rows of the new series. Zero values mean the observation is not mapped. The SMPL option is the default when MAP is used, and it puts the map and old series under the control of the current SMPL, while the new output series will be FREQ N, starting at observation 1.
  • 03-10-08 FORM - it now creates FRMLs after a VAR estimation. The PARAM, COEFPR, and VARPR options are implemented as well.

2002

  • 02-11-30 LAD - remove the LMHET test and add the Machado & Santos-Silva (modified Glejser) hetereoskedasticity test. 02-10-16 dashed lists - expand leading zeros better, like A01-A05, A01-A10 .
  • 02-01-30 SHOW series - shows PANEL "frequency" when FREQ(PANEL) is in effect. SHOW equation - shows # arguments, # operations, so the size of the equation can be assessed.

2001

  • 01-10-04 LAD(CENSOR=series) - new Method title, corrected SILENT and TERSE output, handle different types of non-convergence, prevent crash due to zero divide for pivot.
  • 01-07-16 READ(FORMAT=STATA) - read Stata v7 .DTA files.
  • 01-06-15 PROBIT(REI,NHERMITE=[20/last]) - new NHERMITE option to control number of points in Hermite quadrature. The default number of points is 20. Higher values (up to 92 or higher) may be required for accuracy if RHO is relatively high. If the NHERMITE option is used, the new value is remembered and is used in subsequent commands unless it is changed again.
  • 01-06-07 SORT, MSD(ALL), REGOPT SWILK, GRAPH - use a faster sorting algorithm (hybrid QuickSort). Noticable speed difference for 1000 or more observations. Also works well on presorted series.
  • 01-06-07 LMS - reorder coefs/VCOV so that C is in original position. Prior to this, C was always put last.
  • 01-06-07 IN/OUT - remove limit of 16 different databanks in a run. There is still a limit of 8 active IN and 8 active OUT databanks at any particular point in a run.
  • 01-05-18 WRITE(FORMAT=FREE) series; start each obs. on new line. Clean up rounding, and use minimal spacing (more compact output).
  • 01-05-16 LAD(CENSOR=series) - Censored Quantile Regression, using the BRCENS algorithm by Bernd Fitzenberger.
  • 01-05-10 GRAPH(PREVIEW) - PANEL graphics (separate lines for each individual). SORT/NOSORT option to control ordering of line segments. SYMBOL/NOSYMBOL option for drawing lines with symbols at each point. Missing values are now handled independently for each X,Y series pair.
  • 01-05-10 DOS/Win version - PLOT(SYMBOL) - for drawing lines with symbols. PLOT/GRAPH(SYMBOL) were modified slightly so that all lines have symbols. Originally, the first line was without symbols. Having symbols on all lines makes it easier to identify the individual points when the lines are smooth.
  • 01-06-07 unix/gnuplot version (Sun; Linux pending)
  • 01-05-10 2SLS - print/store a pseudo F-statistic for zero slopes.
  • 01-04-30 READ(FORMAT=EXCEL) - now handles all current versions of Excel files (5, 7/95, 97/98/2000/2002). Reading the Excel 97 and higher files directly may be helpful, because these files can have more than 16384 rows (up to 65536 rows), and files with this many rows cannot be easily saved to the Excel 4 format (which TSP has long been able to read).
  • 01-04-30 LAD - store @UNIQUE=0 if the solution is probably not unique. Store @IFCONV=0 if there is a loss of precision in the simplex iterations.
  • 01-04-04 PROBIT(FEI,FEPRINT) - Fixed Effects for Individuals. FREQ(PANEL) must be in effect. Hundreds or thousands of fixed effects can be handled, and balanced or unbalanced data are fine. A very efficient iteration algorithm is used to estimate the fixed effects. It is ok if some individuals have all zeroes or all ones for the dependent variable, although their data would not be informative for the slope coefficients. The fixed effects estimator is known to have a finite-T bias, but the size and direction of the bias are not known.
  • 01-03-15 REGOPT - Run the diagnostic regressions (for CHOW, CHOWHET, RESET2, etc.) in the default precision (usually NOFAST). Previously, the diagnostics were run always in FAST precision to save time (FAST is about 3 times faster than NOFAST for each regression). The precision can make a difference for some diagnostics like RESET2 in some models with a large number of RHS variables.
  • 01-03-12 PLOT(PREVIEW) - With FREQ(PANEL) on stacked panel data, break lines between individuals. Also handle SMPL gaps for any FREQ by making a continuous SMPL and treating data within gaps as missing. (Previously, a character-style plot was used if there were any SMPL gaps).
  • 01-03-08 PROBIT(REI) - Random Effects for Individuals. Like the REI option in other commands, FREQ(PANEL) must be in effect. For a benchmark with bivariate probit for the case T=2, see http://www.stanford.edu/~clint/bench/probre2.tsp .
  • 01-03-01 REGOPT @CHOWHET - optionally compute a Chow test which is robust to simple heteroskedasticity. This is the MAC2 test from Thursby(1992) Journal of Econometrics.
  • 01-02-14 AR1(OBJFN=GLS) - print/store @COMFAC - common factor test. This is a likelihood ratio test of AR(1) vs. OLS with added RHS variables (lagged dependent and lagged RHS variables). It is similar to the Wald NL AR test sometimes done in OLS. COMFAC is not computed for the usual default OBJFN=ML, because there is no comparable treatment of the first observation (even if presample data exists). The test is also not computed if there are instruments. The COMFAC test will only be seen by default when there is a lagged dependent variable; otherwise the OBJFN=GLS option will be needed to print/store it.
  • 01-02-14 PANEL(TOTAL,WITHIN) - print/store @LOGLT, @SBICT, @AICT, @LOGLW, @SBICW, @AICW.
  • 01-02-14 VAR, LSQ, FIML, SAMPSEL - use LOG(NOB*NEQ) when computing @SBIC for multiequation models. This is the correct DF, counting the total number of residuals, not just the number of observations. It makes it possible to use SBIC to compare multiequation models like SUR with stacked panel models.
  • 01-02-09 OPTIONS [ARGSUB]/NOARGSUB; - controls substition of actual arguments for formal arguments, for command executed within a PROC. OPTIONS NOARGSUB; turns off the substitution, which is the way PROCs worked in TSP 4.3 and earlier. NOARGSUB is helpful if the PROC has LOCAL variables which have the same names as global variables which are passed as arguments to the PROC (it prevents the local variables from being used instead of the PROC arguments).
  • 01-02-08 LSQ/3SLS/GMM/FIML(FEI) - FEI option for Fixed Effects Individual, for linear models. Checks are made for linearity in the parameters, and linearity in endogenous variables for FIML. It is assumed the equations are linear in all variables. Individual means are removed from all variables, including instruments. The estimated fixed effects are not computed at present, because it would be complicated to figure out which variables enter each equation, and computing the SEs for the fixed effects might be difficult as well.
  • 01-02-06 2SLS/LIML(FEI,FEPRINT) - The new FEI option computes 2SLS or LIML estimates with Fixed Effects for Individuals. These are computed by removing individual means from all variables and instruments. The FREQ(PANEL) command must be in effect prior to using the FEI option, so that the individuals are well defined.
  • 01-01-26 READ(FORMAT=FREE) - If the the number of values found in the file does not match the expected NOB*NVAR, print a clearer message, and an attempt to guess values of NOB or NVAR which match the number of values read.
  • 01-01-26 PARAM - If 2 consecutive numbers are found in the arguments, issue a (single) warning, instead of an error message. Ignore ( ) * ** in the argument list. Useful for pasting results from a coef/se/t/*/p table back into a PARAM command, to restart iterations from the previous estimates.
  • 01-01-24 FORM(SUM) - FORM(SUM) EQ S X1-X3; creates FRML EQ S = X1+X2+X3; Useful when creating a sum with a lot of terms (or an arbitrary number of terms that is not known in advance). It can be used as part of a LogL for ML, when summing across the T dimension in Panel data (T does not have to be known in advance).
  • 01-01-23 RANDOM(MEAN=series,STDEV=series,VAR=series) - these options can now take the names of series values, as well as scalar names or values.They are most useful for drawing Poisson or NegBin random variables with different means in each observation.
  • 01-01-19 ACTFIT - Compute changes and percent changes versions of Theil U-66. Store @RMSE, @RMSPE, ... @U66, @U66P. Upper/lowercase output, better aligned. Add SILENT/TERSE options
  • 01-01-18 OPTIONS LIMWMISS=[10] - new option for controlling the number of warning messages printed for missing values. Often with cross section or panel data, missing values are scattered through the observations, and the warnings for missing values in transformations are not every informative, since the object is to run a regression where the missing values will ultimately be dropped. OPTIONS LIMWMISS is better than OPTIONS LIMWARN for this, because LIMWARN will suppress other types of warnings which may still be informative.
  • 01-01-18 WRITE(FORMAT=FREE) - The values written in free format are more concise, especially if they are integers. Double and single precision variables are written to all significant digits.
  • 01-01-15 Nonlinear iteration output - print more digits for F= and FNEW= . The new default is to print 11 digits (vs. 5). OPTIONS SIGNIF=n; can be used to print more digits if desired.

2000

  • 00-11-16 RANDOM(VCOV=V) - print an error message if V is not positive definite, because it will not usually be possible to draw a set of normal variables with VCOV approximately equal to V.
  • 00-11-10 2SLS, LSQ(INST=) eqn; will store & print @FOVERID = @PHI/(@S2*(NX-NZ)) if NX>NZ. This is one form of the test of overidentifying restrictions. Essentially, it is a test for whether the excluded instruments enter the right hand side of the equation with nonzero coefficients (i.e. whether their exclusion significantly degrades the fit).
  • 00-11-10 DBLIST(SILENT), SHOW(SILENT) SERIES; - store @RNMS silently.
  • 00-09-25 CDF(F) - allow non-integer DF for F. Useful for evaluating the incomplete beta function (beta CDF).
  • 00-09-25 LIST(first=4,last=1) nums; or LIST nums 4-1; - allow numeric list in decreasing order.
  • 00-07-26 SOLVE - improve labelling of model blocks.
  • 00-06-27 READ - properly handles trailing comments in free format data files.
  • 00-06-22 MODEL(DONGALLO) - new option, which uses the Don-Gallo method for ordering simultaneous blocks, to find a near-minimal feedback set. Labels the feedback variables with an F in the incidence matrix, and prints a block summary. The feedback information is stored in the model variable, for use when SOLVE is upgraded to handle feedback sets in the future. This ordering is helpful in some cases with Gauss-Seidel as well.
  • 00-06-22 SOLVE - prints an iteration count for Gauss-Seidel.
  • 00-06-14 CDF(CHISQ) - handles non-integer DF. This is useful for evaluating the incomplete gamma function (gamma CDF).
  • 00-05-20 SIML - use just a single iteration on any linear model. Previously, it used an extra (unnecessary) iteration if the model did not have a time-invariant Jacobian, or if SIML was used on a single time period.
  • 00-05-20 SIML(PRNRES) - print Jacobian for first 2 time periods.
  • 00-05-19 EQSUB([LAGS]/NOLAGS) - new option to control substitution of lagged dependent variables. NOLAGS is good for SOLVE, SIML, and FIML models, to leave the lagged endogenous variables alone. The default LAGS is good for LSQ and ML, to substitute expressions for lagged residuals.
  • 00-05-19 DBCOPY - create LOAD instead of READ commands, for improved SIMPC compatibility.
  • 00-05-19 SOLVE - give iteration count for Gauss-Seidel.
  • 00-05-18 SIML, SOLVE - give ERRORs for all missing exogenous variables (instead of just mentioning the first variable).
  • 00-05-11 DBLIST - now stores @RNMS (list of all series in databank). Useful if you want to do an operation on all the series, such as writing them to a spreadsheet file.
  • 00-05-11 SHOW SERIES; - now stores @RNMS (list of all series).
  • 00-05-10 TSP/FAME interface - now handles OPTIONS DOUBLE; (double precision series). (The TSP/FAME interface is available on unix systems at present).
  • 00-05-09 BJEST - use all obs for backforecasting for pure MA models, to match the calculations in Box and Jenkins (1976).
  • 00-05-05 SUR(WNAME=Omega) - Minimum Distance - suppress @COVU, @COVT, REGOUT by default for Minimum Distance. Title of Minimum Distance for output.
  • 00-05-05 LSQ(HITER=D) - allow this HITER option (although it is not recommended).
  • 00-04-04 SORT(ALL) - skip warning for missing values, in all but the first series.
  • 00-03-29 BJEST(HCOV=U) - (numeric second derivatives) is now the default for EXACTML. Avoids problems with inappropriate zero SEs from HCOV=F. HITER=U is also useful if HITER=F has false convergence with CRIT=0.
  • 00-03-29 ML(HCOV=U) - now the default for ML PROC. ML(HCOV=C) is available for ML on a FRML; might be faster than HCOV=N for models with many Params.
  • 00-03-29 FIML(HITER=C,HCOV=C) - HCOV=C produces SEs that are almost identical to what HCOV=N would make (HCOV=N is not implemented for FIML). HITER=C usually has fast convergence (like what HITER=N would provide). These are not the defaults, but they are available.
  • 00-03-15 LSQ - speed up slightly on single equation if NOB is high.
  • 00-03-06 GRAPH, PLOT - gnuplot graphics on unix.
  • 00-02-18 Changed LINE number to COMMAND number in output, to support a GiveWin edit feature. The GiveWin edit feature prints the physical line number from the input file when there is an error message, so a double-click on that error message will jump to Edit the appropriate line in the input file.
  • 00-02-16 HELP HIST - add the graphical HIST options available in TSP/GiveWin.
  • 00-02-07 EPSMIN=[.0001] - new nonlinear option. Useful with HITER=F, to make more accurate numeric first derivatives, if some Param values are close to zero.
  • 00-01-28 GiveWin - remove page headers in default output.
  • 00-01-17 BJEST - print moduli of polynomial roots, if regular roots are being printed with ROOTS or PRINT options. The moduli are useful for detecting near unit roots. Always store REal parts of roots, IMaginary parts, and MOduli as @ARRTRE, @ARRTIM, @ARRTMO, etc. for @MA @SAR @SMA.
  • 00-01-14 MATRIX - recognize composite (A*B)'(A*B) as symmetric.
  • 99-11-18 SIML - warn if an ENDOG variable never appears in any equation.

BUGFIXES

  • 00-06-27 READ - trailing comments in free format data were not properly handled (hard to understand the error messages).
  • 00-05-31 SMPL - crashed sometimes when the second argument was a series (instead of a scalar).
  • 00-05-24 FRML - ignore commas such as FRML EQ, Y = B*X;
  • 00-05-09 BJEST - correct residual observation labels when NDIFF>0.
  • 00-04-25 LSQ - allow NOVAR=NOB for NEQ=1 (usually a perfect fit).
  • 00-04-21 SIML, FIML - do not declare linearity if there are non-differentiable operations. Allow more than one iteration in this case.
  • 00-03-23 OLSQ/REGOPT - CUSUM and CUSUMSQ plots after a regression with missing values included extra invalid values.
  • 00-03-17 SOLVE - in a recursive/dynamic model where one or more endogenous variables never appear on the RHS of any equation. In some cases these equations were not evaluated correctly. This type of model is useful for generating artificial data for a GARCH model, or a VAR/Impulse Response.
  • 00-03-15 MATRIX - if a user give a FRML name as an argument, give a proper error message.
  • 00-03-09 POISSON - sometimes crashed when there were missing values.
  • 00-02-15 FREQ(PANEL) with lags - if a SELECT command removes some observations from the middle of an individual, the later observations for that individual may be improperly set to missing for lagged variables.
  • 00-02-06 ANALYZ - handle lag in equation (with NOB=1) correctly.
  • 00-02-06 PANEL(ROBUST,SILENT) - do not print "Robust" label for SEs.
  • 00-01-30 TITLE(NOPAGE) '...'; - do not insert : before ( .
  • 00-01-26 WRITE(FILE='x',FORM=LAB) , interactive mode - don't pause.
  • 00-01-26 ML with CNORM2(x,y,0) - had problems with derivatives if RHO was a constant (zero in this case).
  • 00-01-25 DIFFER - on an unnormalized equation, make an unnormalized equation. (Not an equation with D1 as the dependent variable).
  • 00-01-25 FRML EQ1, Y=A; TSP didn't like parsing the comma.
  • 00-01-25 TITLE(NOPAGE) 'text'; TSP tried to make TITLE'(...
  • 00-01-25 PROC/DOT - if the first DOT command in a run occurred within a PROC, and a COMPRESS occurs before the second use of the PROC, the second use of DOT did not work correctly.
  • 00-01-25 MATRIX - if dependent variable was a PARAM, it was changed to a const.
  • 00-01-25 MATRIX - if dependent variable was a subscripted variable, it was changed to a scalar variable.
  • 00-01-25 SORT - did not properly handle mixed double & single series (changed them all to current precision).
  • 00-01-25 DOT - give proper ERROR message for lag argument. DOT X(-1); is illegal. (DOT can only handle names or numbers; not lags or subscripted arguments at present).
  • 00-01-14 DIFFER(PRINT) - print long equation names correctly; make output label more compact.
  • 00-01-05 COMPRESS; or automatic compress when SELECT is in effect, followed by a regression or other model with missing data - the data in the regression may include invalid data.
  • 99-12-15 MATRIX - RANK(TxK_matrix), where T>K - was incorrect
  • 99-11-09 ANALYZ - correct DF for Wald test when restrictions are singular/collinear.

 

If you have any questions or comments about TSP please send an email to info@tspintl.com.

Comments or questions about this website should also be sent to the info@tspintl.com. Lost? Please consult the site map.