TSP is an econometrics/statistical package available in the Department of Economics in two basic forms. First, the Department has a site license for the PC version of TSP, allowing the package to be placed on departmental computers for use by students and faculty. A UNIX version TSP is also available on the Department Project Vincent machines. The purpose of this document is to provide a brief introduction on both how one can access TSP and the basic TSP command. Additional information can be found in the TSP manuals (available in the PC lab) and using TSP's online help system.

TSP ON UNIX

1. Invoking TSP interactively
   • After logging into Vincent add the econ locker by typing add econ at the prompt.
   • At the prompt type tsp. The program will now accept commands for execution.
2. Invoking TSP in batch mode.
   • Create a file containing your commands. Save it with a meaningful name such as myfile.
   • Now type: tsp myfile. The output will go to a file named myfile.out.
3. Executing a previously prepared file interactively.
   • At the tsp prompt type
     
     input "filename.tsp"
   • You must use quotation marks around the file name. Remember that UNIX is case sensitive. Answer whether you want the output displayed at the terminal or written to a file, and then give a file name if you choose to write to a file.
4. Printing output
   • If you save the output to a file you can then print the file by using the lpr command. For example to print the file myfile.out you would type
     
     lpr -P he64 myfile.out

TSP ON THE PC

The PC version of TSP works fundamentally in the same manner as the UNIX version. One difference is that, upon typing tsp on the PC, you are given the option of either entering interactive mode or providing TSP with a batch file already containing your commands. You may also run batch files by typing

tsp myfile

as you would on the UNIX system. The output will be saved to a file named myfile.out.

TSP COMMAND LANGUAGE BASICS

All TSP commands end with a semicolon (;). You can have more than one command on a given line as long as they are separated by semicolons. In interactive mode, a carriage return is the same as a semicolon. Any information after a question mark (?) on a given line is viewed as a comment. Here are some commonly required steps in a TSP program and the corresponding commands.

1. Specifying the type of data
   
   FREQ A;
   
   The frequency types are A:annual, Q:quarterly, M:monthly and N:undated. Undated is the default specification
   
   Example:
   FREQ Q;
   
2. Setting the sample size
   
   SMPL beg end;
   
   • If the data is annual the sample can refer to years by using the last two digits. For example 68 refers to 1968.
   • Quarters and months are represented by the last two digits of the year, colon, and the quarter and month. For example January of 1954 is 54:1 is the data is monthly while 54:1 refers to the first quarter of 1954 is the data is quarterly.
   • When the FREQ is undated, use the sample numbers.
   • Examples:
     
     SMPL 62,90;
     SMPL 72:1,88:4;
     SMPL 1,25;
     
3. Reading data from the screen
   
   READ (options) list of series or matrices or constants;
   • Variable names may be up to 8 characters long and may only consist of letters and numbers. They must start with a letter.
   • An identical command to read is load
     
     LOAD (options) list of series or matrices or constants;
   • The data then follow with one observation for each variable on a line. The numbers should be separated by blanks.
   • Example:
     
     READ y, x1, x2;
     11 10 34
     23 25 45
     43 34 20
     12 45 100
     43 34 72;
   • An alternative would be to use the load statement with no argument and then include the data at the end of the program after the end statement. This data section could include a SMPL, a LOAD statement and an END statement.
   • Example:
     
     ? Main program
     LOAD;PRINT a1 a2 b1 b2 b3;
     STOP;
     END;
     ? Data section
     SMPL 1,3;
     LOAD
     a1 a2 b1 b2 b3;
     3 45.1 23 56.5
     12 23 3.4 31.1
     21 21.34 56.6 32
     END;
4. Reading in the data from a file
   • The data are first typed into a file. The data should be entered with one observation for each variable on a line (default). They can also be entered by variable using the byvar option on the load command. The numbers should be separated by blanks. The file is then saved with a name such as test1.dat. Then the load command must include the file name in quotation marks as an option. The format is as follows
     
     LOAD (file='test1.dat') series names;
     Example:
     
     READ(file='zel.dat') y, k, l;
5. Obtaining basic statistics on the variables
   • Basic statistics can be obtained using the MSD command.
     
     MSD(options) list of series;
     
     MSD with no options will compute means, standard deviations, sums, maxima, minima, and variances for all variables in the list. Useful options include COVA (covariances) and CORR (correlations).
   • Example:
     
     MSD(CORR) y, x1, x2;
6. Generating new variables
   • New variables can be created using previously defined ones. Data transformations can also be performed. The GENR command is used to transform series as follows:
     
     GENR newvar = algebraic formula;
     
     or
     
     newvar = algebraic formula;
   • The equation may be any arithmetic equation involving variables, constants and mathematical functions.
   • Examples:
     
     GENR ly = log(y);
     GENR ey = exp(y);
     GENR y = y/100;
     GENR x12 = x1*x1;
     GENR x12 = x1**2;
     GENR x1lag = x1(-1);
   • The SET command is used to set the values of scalars or single elements of series. The form is
     
     SET scalar = algebraic expression
     
     or
     
     subscripted variable = algebraic expression
   • Examples:
     
     SET beta = 5;
     SET x1 = GNP(72:3);
     SET beta = beta**2;
7. Plotting or graphing variables
   • Variables can be plotted against the observation number using plot command.
     
     PLOT series plotting character, series plotting character ...;
   • Examples:
     
     PLOT x1 *, x2 +, x3 a;
   • Variables can be graphed against each other using the graph command.
     
     Graph series for y-axis, series for x-axis;
   • Examples:
     
     GRAPH x1, x2;
8. Running a linear regression
   • To run an ordinary least squares regression, use the OLSQ command. It is defined as:
     
     OLSQ (options) depvar independent variables;
   • If you want a constant term you include either C or Constant in the list.
   • Example:
     
     OLSQ y, C, x1, x2;
   • OLSQ stores most of the results of a regression in data storage for later use. They results have names that begin with @. A partial list follows:
     
     @SSR sum of squared residuals
     @S2 variance of residuals
     @S standard deviation of residuals
     @NOB number of observations
     @NCOEF number of coefficients
     @COEF coefficient vector
     @VCOV covariance matrix of estimated coefficients
     @RES estimated residuals
     @LOGL log of likelihood function
     @FIT fitted values of dependent variable
9. Running a nonlinear regression
   • Nonlinear regression is straightforward in TSP. The model is specified using the FRML statement, initial parameter values are provided using the PARAM statement and the model is estimated using the LSQ statement. The form of the FRML statement is
     
     FRML equation name variable name = algebraic expression;
     
     or
     
     FRML equation name algebraic expression;
     
     where the second form is implicitly zero.
   • The parameter statement has the following form
     
     PARAM parameter name value parameter name value ... ;
     
     If the value is not given, it is assumed to be zero. These are simply initial guesses for the parameters.
   • The form of the LSQ statement is
     
     LSQ (options) list of equation names;
     
     The equation names must be previously specified in a PARAM statement.
   • Example:
     
     FRML EQ1 Y = B0 + B1*(X1**B2) + b3*(X2**B4);
     PARAM B0 2 B1 3 B2 0.7 B3 1 B4 0.5;
     LSQ EQ1;
10. Manipulating series or matrices using matrix commands..
    • Sometimes we are interested in quantities created from regression output. We can create vectors and matrices using the MMAKE command. MMAKE makes a matrix from a set of series or a vector from a set of scalars. The number of rows in a matrix will be the number of observations and the number of columns will be the number of series. The form is
      
      MMAKE matrix name list of series;
      
      or
      
      MMAKE vector name list of scalars;
    • Examples:
      
      MMAKE vec1 0 2 3 4;
      MMAKE mat2 x1 x2 x3;
    • Matrices can be manipulated using the MATRIX command. The form is
      
      MATRIX matrix = matrix equation;
    • Matrices must be conformable. Some common operators are given below
      
      MAT b = m*n; matrix product
      MAT mt = m'; matrix transpose
      MAT a = m'c; transpose with implied product
      MAT ainv = a"; matrix inverse
      MAT xinvx = x"x; inverse with implied product
      MAT xtxinv = (x'x)"; inverse of x'x
    • Examples:
      
      MMAKE delta 0 1 1 0;
      MAT forcst = delta'@coef;
      MAT varfcst = delta'@vcov*delta;
11. Stop or end
    • In order to tell the program to stop, you need to include the statement
      
      STOP;
      
      or
      
      END;
    • END is also used to separate the program and data sections if a data section is used.
12. Obtaining on-line help
    • Help is available using the HELP command. HELP with no argument gives a menu. HELP "Name of command" will give the user additional information about the indicated command.
    • Examples;
      
      HELP OLSQ;
      HELP PLOT;

AN EXAMPLE TSP PROGRAM
A scientist conducted experiments on corn yields. Various plots received different levels of nitrogen and had different seeding rates (plant population). The dependent variable for the regression is corn yield while the independent variables are nitrogen and plant population in thousands. There are 25 observations. A variety of functional forms could be used to estimate this relationship. Five possible functional forms are given below.

The log of zero is undefined so the analysis changes the sample size for creating the log variables, and runs regressions 4 and 5 with only 20 observations. An alternative would be to use a very small positive number close to zero. A copy of the computer code follows with comments in parentheses.

freq n;	(sets the frequency to undated)
smpl 1 25;	(sets the sample size to 25)
read (file="corn2.dat") y pop n;	(reads the variables y,pop,n from the file corn2.dat)
genr n2=n**2;	(creates the variable n2)
genr pop2=pop**2;	(creates the variable pop2)
genr pn=pop*n;	(creates the variable n*pop)
print y n pop;	(prints the variables y,n,pop)
msd(corr) y n pop;	(computes interesting statistics)
olsq y C n pop;	(runs regression #1)
olsq y C n n2 pop pop2;	(runs regression #2)
olsq y C n n2 pop pop2 pn;	(runs regression #3)
smpl 6 25;	(changes the sample size to eliminate the zero observations)
genr ly = log(y);	(creates the variable log(y))
genr ln = log(n);	(creates the variable log(n))
genr lp = log(pop);	(creates the variable log(pop))
olsq ly C ln lp;	(runs regression #4)
frml eq1 y = b1*(b2*n**(­b3)+(1­b2)*pop**(­b3))**(­1/b3);	(forms equation eq1)
param b1 5 b2 .3 b3 .2 ;	(sets parameters)
lsq eq1;	(estimates equation eq1)
stop;	stops

Example Output

LINE 0 TSP 4.2A (09/09/92) DECStation 01/29/93 4:33 PM PAGE 1

TSP Version 4.2A

Copyright (C) 1992 TSP International

ALL RIGHTS RESERVED

In case of questions or problems, see your local TSP

consultant or send a description of the problem and the

associated TSP output to:

TSP International

P.O. Box 61015, Station A

Palo Alto, CA 94306

USA

PROGRAM

LINE ******************************************************************

| 1 freq n;

| 2 smpl 1 25;

| 3 read (file="corn2.dat") y pop n;

| 4 genr n2=n**2;

| 5 genr pop2=pop**2;

| 6 genr pn=pop*n;

| 7 print y n pop;

| 10 msd(corr) y n pop;

| 11 olsq y C n pop;

| 12 olsq y C n n2 pop pop2;

| 13 olsq y C n n2 pop pop2 pn;

| 14 ly = log(y);

| 15 smpl 6 25;

| 16 genr ln = log(n);

| 17 genr lp = log(pop);

| 18 olsq ly C ln lp;

| 19 frml eq1 y = b1*(b2*n**(­b3)+(1­b2)*pop**(­b3))**(­1/b3);

| 20 param b1 5 b2 .3 b3 .2 ;

| 21 lsq eq1;

| 22 stop;

EXECUTION

*******************************************************************************

Current sample: 1 to 25

LINE 7 TSP 4.2A (09/09/92) DECStation 01/29/93 4:33 PM PAGE 2

Y N POP

1 50.60000 0.00000 9.00000

2 54.20000 0.00000 12.00000

3 53.50000 0.00000 15.00000

4 48.50000 0.00000 18.00000

5 39.20000 0.00000 21.00000

6 78.70000 50.00000 9.00000

7 85.90000 50.00000 12.00000

8 88.80000 50.00000 15.00000

9 87.50000 50.00000 18.00000

10 81.90000 50.00000 21.00000

11 94.40000 100.00000 9.00000

12 105.30000 100.00000 12.00000

13 111.90000 100.00000 15.00000

14 114.20000 100.00000 18.00000

15 112.20000 100.00000 21.00000

16 97.80000 150.00000 9.00000

17 112.40000 150.00000 12.00000

18 122.60000 150.00000 15.00000

19 128.60001 150.00000 18.00000

20 130.30000 150.00000 21.00000

21 88.90000 200.00000 9.00000

22 107.10000 200.00000 12.00000

23 121.00000 200.00000 15.00000

24 130.60001 200.00000 18.00000

25 135.89999 200.00000 21.00000

RESULTS OF COVARIANCE PROCEDURE

*******************************

NUMBER OF OBSERVATIONS: 25

MEAN STD DEV MINIMUM MAXIMUM

Y 95.28000 28.54143 39.20000 135.89999

N 100.00000 72.16878 0.00000 200.00000

POP 15.00000 4.33013 9.00000 21.00000

SUM VARIANCE

Y 2382.00002 814.61335

N 2500.00000 5208.33333

POP 375.00000 18.75000

CORRELATION MATRIX

Y N POP

Y 1.00000

N 0.85354 1.00000

POP 0.22524 0.00000 1.0000

Equation 1

************

Method of estimation = Ordinary Least Squares

Dependent variable: Y

Current sample: 1 to 25

Number of observations: 25

Mean of dependent variable = 95.2800 Adjusted R­squared = .759201

Std. dev. of dependent var. = 28.5414 Durbin­Watson statistic = 1.07241

Sum of squared residuals = 4315.47 F­statistic (zero slopes) = 38.8342

Variance of residuals = 196.158 Schwarz Bayes. Info. Crit. = 5.53735

Std. error of regression = 14.0056 Log of likelihood function = ­99.8620

R­squared = .779268

Estimated Standard

Variable Coefficient Error t­statistic

C 39.2540 11.0280 3.55947

N .337560 .039614 8.52125

POP 1.48467 .660232 2.24871

Equation 2

************

Method of estimation = Ordinary Least Squares

Dependent variable: Y

Current sample: 1 to 25

Number of observations: 25

Mean of dependent variable = 95.2800 Adjusted R­squared = .918138

Std. dev. of dependent var. = 28.5414 Durbin­Watson statistic = 1.20268

Sum of squared residuals = 1333.72 F­statistic (zero slopes) = 68.2938

Variance of residuals = 66.6862 Schwarz Bayes. Info. Crit. = 4.62063

Std. error of regression = 8.16616 Log of likelihood function = ­85.1841

R­squared = .931781

Estimated Standard

Variable Coefficient Error t­statistic

C ­22.3932 23.4332 ­.955618

N .830703 .081428 10.2017

N2 ­.246571E­02 .390417E­03 ­6.31559

POP 8.63229 3.27617 2.63487

POP2 ­.238254 .108449 ­2.19692

Equation 3

************

Method of estimation = Ordinary Least Squares

Dependent variable: Y

Current sample: 1 to 25

Number of observations: 25

Mean of dependent variable = 95.2800

Std. dev. of dependent var. = 28.5414

Sum of squared residuals = .013029

Variance of residuals = .685752E­03

Std. error of regression = .026187

R­squared = .999999

Adjusted R­squared = .999999

Durbin­Watson statistic = 1.70018

F­statistic (zero slopes) = .570198E+07

Schwarz Bayes. Info. Crit. = ­6.78690

Log of likelihood function = 59.0194

Estimated Standard

Variable Coefficient Error t­statistic

C 14.1268 .079577 177.525

N .465503 .369809E­03 1258.77

N2 ­.246571E­02 .125197E­05 ­1969.46

POP 6.19762 .010650 581.939

POP2 ­.238254 .347770E­03 ­685.091

PN .024347 .174579E­04 1394.59

Current sample: 6 to 25

Equation 4

************

Method of estimation = Ordinary Least Squares

Dependent variable: LY

Current sample: 6 to 25

Number of observations: 20

Mean of dependent variable = 4.65721

Std. dev. of dependent var. = .171201

Sum of squared residuals = .098996

Variance of residuals = .582330E­02

Std. error of regression = .076311

R­squared = .822233

Adjusted R­squared = .801319

Durbin­Watson statistic = 1.35260

F­statistic (zero slopes) = 39.3154

Schwarz Bayes. Info. Crit. = ­4.85905

Log of likelihood function = 24.7053

Estimated Standard

Variable Coefficient Error t­statistic

C 2.76948 .217186 12.7516

LN .240777 .032775 7.34635

LP .283115 .057010 4.96607

*******************************************************************************

NONLINEAR LEAST SQUARES

***********************

EQUATIONS: EQ1

Working space used: 643

STARTING VALUES

B1 B2 B3

VALUE 5.00000 0.30000 0.20000

F= 4.8027 FNEW= 4.3379 ISQZ= 0 STEP= 1.0000 CRIT= 12.803

F= 4.3379 FNEW= 4.1044 ISQZ= 0 STEP= 1.0000 CRIT= 6.3604

F= 4.1044 FNEW= 4.1036 ISQZ= 0 STEP= 1.0000 CRIT= 0.26400E­01

F= 4.1036 FNEW= 4.1036 ISQZ= 0 STEP= 1.0000 CRIT= 0.11075E­05

CONVERGENCE ACHIEVED AFTER 4 ITERATIONS

8 FUNCTION EVALUATIONS.

LOG OF LIKELIHOOD FUNCTION = ­80.4943

NUMBER OF OBSERVATIONS = 20

Standard

Parameter Estimate Error t­statistic

B1 3.01798 1.05080 2.87209

B2 .474150 .369490 1.28326

B3 .236724 .751327 .315074

Standard Errors computed from quadratic form of analytic first

derivatives (Gauss)

Equation EQ1

***************

Dependent variable: Y

Mean of dependent variable = 106.800 Std. error of regression = 14.6881

Std. dev. of dependent var. = 17.9674 R­squared = .845027

Sum of squared residuals = 3667.59 Adjusted R­squared = .826795

Variance of residuals = 215.740 Durbin­Watson statistic = .862494

*******************************************************************************

END OF OUTPUT.

MEMORY ALLOCATED (WORDS) : 1000000

MEMORY ACTUALLY REQUIRED : 2307 ( 0%)

CURRENT VARIABLE STORAGE : 1658