2020 Elements of econometrics

Manual for doing the computer exercises in

Dougherty, C. Introduction to Econometrics
(Oxford: Oxford University Press, fifth edition 2016) using

gretl
gretl
gretl (GNU Regression, Econometrics and Time-Series Library) is
the official econometrics application for 2020 Elements of
Econometrics. It is free (obviously, a major consideration for
distance learning students) and very powerful. It does everything
that is required in the computer regression exercises in the course
text Introduction to Econometrics (C. Dougherty, Oxford
University Press, fifth edition 2016). In fact, we use only a small
fraction of its features. It is similar to major econometrics
applications such as Stata, EViews, SPSS, and SAS, with the crucial
difference that you do not have to pay for it. If you already have
access to one of the commercial applications just mentioned, there
is no need to change. However, if you are struggling to fit
regressions with a regression feature built into a spreadsheet
application such as Microsoft Excel, quit now and switch to gretl.
Spreadsheet applications can do only very basic regressions and are
quite unsuitable for this course.
gretl should be downloaded from the its home page
http://gretl.sourceforge.net/. See that page for its history and
current status.
This manual provides just the information needed to undertake the
regression exercises in the text. The information is provided as you
need it, so that you are not overwhelmed by too much information
too soon. By the time that you have completed the exercises, you
will have a good basic working knowledge of the application.
Comprehensive reference information can be found in the gretl
manual gretl-guide.pdf located in the Help section of gretl. The
Help section also provides direct access to information about the
commands and functions used by gretl.
It is assumed that you have downloaded and installed gretl and that
you have downloaded the necessary data sets from the OUP website
at
http://global.oup.com/uk/orc/busecon/economics/dougher
ty5e/
(The OUP link is likely to appear if you type ‘Dougherty 5e’
into your search engine.)

GUI/CLI/Batch mode
In common with other sophisticated econometrics packages, gretl
may be operated in three ways: via a graphical user interface (GUI),
via a command line interface (CLI), and in batch mode.
The GUI involves menus and windows, is intuitively designed, and
provides an excellent starting point for experimenting with gretl.
However, it will mostly be ignored here, for two reasons. One, the
less important, is that operations with menus and windows can be
cumbersome to describe. The other is that serious researchers
always primarily use batch mode.
The CLI involves using a special command line window, called the
‘command console’ in the gretl manual, instead of the GUI. The
command line window may be opened by clicking on the Tools tab
University of London International Programmes 1
20 Elements of econometrics

in the main menu and then ‘Gretl console’. Operations, such as

opening a data file or fitting a regression, are effected by typing in a
line of text into this window and then pressing the Enter key.
Batch mode is essentially a well-organized use of the CLI. Instead
of entering the commands one by one, they are all written together,
line by line, in a text file with extension .inp. The text file is then
run by gretl, the commands in it being executed sequentially.
Batch mode has huge advantages over the interactive use of the
GUI or the CLI:
You have an exact and permanent record of your operations
If you make a mistake somewhere in your operations, all you
have to do is to correct the faulty line in the batch file rather
than repeat all your pointing and clicking, and
If at some future time you wish to modify the analysis or
develop it further, you do not have to start from scratch. You
modify the existing batch file or create a new one using it as a
base. This is how research is done.
There is a further small advantage of batch mode over the GUI. If
you are skilled at word processing, you may actually find it quicker
to use a batch file than the GUI.
Note that gretl calls batch files ‘script files’. To run one, you launch
gretl, go to Files/Script files/User files and then browse to the
location of your file. Double click on its name and a window listing
its contents will open. To run the commands, click the cogs icon in
the menu of this window.
Since the exercises in the text are relatively simple, each involving
only a few commands, the instructions in this manual assume that
you use the CLI directly. Indeed, if you prefer, you could use the
GUI. However the explanation here is in terms of using the CLI
because I can describe exactly what you need to do for each of the
exercises that you are going to undertake. Short of leaving out a
space where there should be one, or using a number ‘1’ instead of a
letter ‘l’, you should not be able to make mistakes.

Common operations
Here are a few commands that will be useful in the exercises: You
will learn other commands as you work your way through the
exercises. In fact, doing the exercises is a relatively painless way of
acquiring a basic understanding of how to use gretl.
open followed by a path and then the name of a data set. This
opens the data set.
store followed by a path and the name of the current data set.
This saves the data set. The command should be used
after generating new variables, to save the enlarged data
set for future use. In general, there is no need to give the
enlarged data set a different name.
ols followed by a variable name, then const, and further
variable names. The first variable is regressed on the
rest. Omit const if there is no intercept in the model.

2 University of London International Programmes

 Manual for the computer exercises

summary followed by a list of variable names. This produces a

table giving the mean, maximum, minimum, and
standard deviation of each variable listed.
corr followed by a list of variables. The correlation
coefficients are calculated.
genr followed by an equation. This creates a new variable
defined as the dependent variable of the equation.
smpl followed by a condition in parentheses. This restricts the
observations to the subset that satisfy the condition.
smpl is cumulative, so that a second use of it will select
a subset of the subset selected by its first use. The
command smpl full must be used to return to the full
sample.
Examples of the use of these commands, and a few others, will be
found in the following list of exercises.
Exercise 1.7 Opening a data file, fitting a simple regression, and
saving and printing the output
Exercise 3.14 Using the corr command for computing correlation
coefficients
Exercise 3.15 Creating new variables from existing ones
Exercise 4.4 Limiting the sample to those observations that
satisfy some condition
Restoring the full sample
Creating a new variable as a function of an existing
one. See Help/Function reference from the gretl
main menu for a complete list of available functions.
Exercise 5.27 Using a backslash in a batch file to extend a long
command that has wrapped to a second line.
Exercise 7.3 Sorting a data set
Exercise 8.16 Instrumental variables (IV) regression and
Hausman test
Exercise 10.4 Defining a dummy variable using a condition
Exercise 10.4 logit regression
Exercise 10.6 probit regression
Exercise 10.9 tobit regression
Exercise 10.11 heckit (sample selection) regression
Exercise 11.2 Plotting a time series
Exercise 11.9 Creating lagged versions of existing variables
Exercise 13.14 Creating first differences of variables using the diff
command
Exercise 13.14 Testing for a unit root (augmented Dickey–Fuller
test) with the adf command
Exercise 14.2 Testing for the presence of fixed effects with an F
test
Exercise 14.6 Testing for the presence of random effects with a
Breusch–Pagan test

University of London International Programmes 3

20 Elements of econometrics

Exercise 14.6 Discriminating between fixed effects and random

effects using a Hausman test

Exercise 1.7 (EAWE)

Opening a data set
This involves using one of the EAWE data sets described in detail in
Appendix B of the text. There are 20 for the use of students working
together with an instructor. The students should use different data
sets and therefore get slightly different results. If you are working
on your own, it does not matter which one you use (other than set
21 which is used for examples in the text). You should download
your data set in Stata format from the OUP website. I shall
represent the path to the subfolder with your data set as path\. To
open the dataset, you should type the following command in the
console box
open path\eawe**.gdt

where ** is the number of your data set. Thus if you are using data
set 7 and have stored the data set in subfolder ‘data’ of folder
‘gretl’ on your C: drive, path will be C:\gretl\data and the
command will be
open c:\gretl\data\eawe07.gdt

Fitting a simple ordinary least squares regression

The command is ols. It should be followed by the name of the
dependent variable, then const (assuming you want an intercept
in the regression) and then the name of the explanatory variable. In
this exercise the command is
ols S const ASVABC

Note that gretl is case sensitive. ols and const must always be
lower case. S and ASVABC must be upper case because they are
defined as upper case in the data set.

Saving and printing the output

An easy way to print the output is to block off the regression output
that appears in the console window and paste it into a Word
document (or other word-processor document). Change the
typeface to Courier so that the output lines up properly. If the lines
wrap, reduce the size of the font. Print this document in the usual
way.
Alternatively, you could save the whole console window as a text
file and edit it afterwards.

Exercise 1.8 (EAWE)

The regression command is
ols EARNINGS const S

4 University of London International Programmes

 Manual for the computer exercises

Exercises 1.20 and 1.21 (EAWE)

R2 is part of the output (unadjusted R-squared) in Exercises
1.7 and 1.8.

Exercises 2.18, 2.19, 2.23, and 2.24 (EAWE)

The t statistics are part of the output in Exercises 1.7 and 1.8.

Exercise 2.29 (EAWE)

You will need to use the standard error of the coefficient. It is part
of the output in Exercise 1.8.

Exercise 2.32 (EAWE)

F statistics are generally calculated using data on the sums of the
squares of the residuals. Occasionally they may equivalently be
calculated from data on R2, but since this is not always the case it is
better to give priority to the sums of squares approach. In the
present case, using equation (2.85) in the text, the expression for F
is
ESS / k  1
F k  1, n  k  
ESS

RSS / n  k  RSS / n  2

since k = 2. However the output from a gretl regression reports RSS

but does not report ESS. Given equation (2.83), we could calculate
it as (TSS – RSS), where TSS is the total sum of squares:

 Y Y 
n
2
TSS  i
i 1

but TSS is not reported either. To get around this, regress the
dependent variable on a constant only, fitting the model
Y  1  u
The estimator of 1 will be Y . The fitted value of Y in all
observations will be Y . The residual in observation i will be
Yi  Y , and hence RSS for this regression will be  Y  Y  . By
n 2
1 i

definition, this is TSS for any regression with Y as the dependent

variable. So fit this regression first, and then compute ESS for the
main regression as TSS – RSS (RSS for the main regression). Thus
the necessary commands are
ols EARNINGS const
ols EARNINGS const S

To compute F from R2, use equation (2.87) in the text.

University of London International Programmes 5

20 Elements of econometrics

Exercise 2.33 (EAWE)

Use the output from Exercise 1.8.

Exercise 3.2 (EAWE)

ols S const ASVABC SM SF

Exercise 3.3 (EAWE)

ols EARNINGS const S EXP

Exercise 3.4 (EAWE)

ols S const ASVABC SF
$uhat is an invisible temporary variable that contains the
residuals from the most recent regression. The next command
saves them permanently as ES.
genr ES = $uhat
ols SM const ASVABC SF
genr ESM = $uhat
ols ES const ESM
The exercise asks for a plot of ES on ESM. I have not managed to do
this using console box commands.

Exercise 3.10 (EAWE)

Use the results from Exercises 3.2 and 3.3

Exercise 3.12 (EAWE)

An example might be
ols WEIGHT04 const HEIGHT S AGE
(What kinds of coefficient would you anticipate for S and AGE?)

Exercise 3.14 (EAWE)

ols S const SM SF ASVABAR ASVABWK ASVABPC
ols S const SM SF ASVABC
corr ASVABAR ASVABWK ASVABPC

Exercise 3.15 (EAWE)

The genr command generates a new variable given the specified
relationship. The store command saves the modified data set. The
asterisks should be replaced by the number of the EAWE data set,
as opened in Exercise 1.8. See Exercise 1.7 for further information

6 University of London International Programmes

 Manual for the computer exercises

on path\. You should always save a data file after defining new
variables if you think that you will need them again.
genr CHILDREN = SIBLINGS + 1
ols CHILDREN const SM SF
corr SM SF
genr SP = SM + SF
ols CHILDREN const SP
store path\eawe**.gdt

Exercise 3.17 (EAWE)

genr PWE = AGE - S - 5
ols EARNINGS const S PWE
ols EARNINGS const S PWE AGE
corr S PWE AGE
store path\eawe**.gdt

Exercise 3.18 (EAWE)

See the explanation in Exercise 2.32 of the computation of the F
statistic.
ols S const
ols S const ASVABC SM SF

Exercise 3.19 (EAWE)

ols S const ASVABC SM SF ASVABNO ASVABCS

Exercise 3.20 (EAWE)

ols S const ASVABC SM SF ASVABNO

Exercise 4.4 (CES2013)

First you should download the CES2013 data set CES2013.gdt
from the OUP website, launch gretl, and open the data set with the
open command (see Exercise 1.6). You should work with only one
of the different categories of expenditure. (If the exercises are being
done by a group, the instructor should assign a different category to
each student.) The commands use CAT as a placeholder for the
three or four letter name of your category. The first command
restricts the sample to those observations with positive values for
your category of expenditure. The last but one restores the full
sample. The last saves the enlarged data file.
smpl CAT > 0 –-restrict

University of London International Programmes 7

20 Elements of econometrics

ols CAT const EXP

Just to avoid any doubt, if your category were CLOT (clothing), the
commands would be
smpl CLOT > 0 --restrict
ols CLOT const EXP

To fit the logarithmic regression, you first need to create the

logarithmic variables
genr LGCAT = log(CAT)
genr LGEXP = log(EXP)
ols LGCAT const LGEXP
Note that log is log to base e, the natural logarithm.
smpl full
store path\CES2013.gdt

In the last command, you should replace path\ by the path to the
location where you have downloaded the CES2013 data file. See
Exercise 3.15 for remarks on saving data files.

Exercise 4.5 (CES2013)

genr LGSIZE = log(SIZE)
ols LGCAT const LGEXP LGSIZE
store path\CES2013.gdt

Exercise 4.6 (EAWE)

genr LGWT04 = log(WEIGHT04)
genr LGHT = log(HEIGHT)
ols LGWT04 const LGHT

Exercise 4.9 (EAWE)

genr LGEARN = log(EARNINGS)
ols LGEARN const S EXP
store path\EAWE**.gdt

Exercise 4.10 (EAWE)

genr EARNSTAR = EARNINGS/exp(mean(LGEARN))
genr LGEARNST = log(EARNSTAR)
ols EARNSTAR const S EXP
ols LGEARNST const S EXP
mean(LGEARN)produces the mean of LGEARN and
exp(mean(LGEARN)) produces its exponential.

8 University of London International Programmes

 Manual for the computer exercises

Exercise 4.12 (CES2013)

smpl CAT > 0 --restrict
genr CATSTAR = CAT/exp(mean(LGCAT))
genr LGCATST = log(CATSTAR)
ols CATSTAR const EXP SIZE
ols LGCATST const LGEXP LGSIZE

Exercise 4.17 (EAWE)

ols WEIGHT04 const HEIGHT
genr YHAT = $yhat
genr YHATSQ = YHAT*YHAT
ols WEIGHT04 const HEIGHT YHATSQ

Exercise 5.1 (EAWE)

ols S const ASVABC SM SF MALE

Exercise 5.3 (EAWE)

ols LGEARN const S EXP MALE
See Box 5.1 in the text for a guide to the interpretation of dummy
variable coefficients in semilogarithmic regressions.

Exercise 5.6 (EAWE)

ols S const ASVABC SM SF MALE ETHBLACK ETHHISP

Exercise 5.9 (EAWE)

ols LGEARN const S EXP MALE ETHBLACK ETHHISP

Exercise 5.11 (EAWE)

Use the results from Exercises 5.1 and 5.6.

Exercise 5.12 (EAWE)

Use the results from Exercises 5.3 and 5.9.

Exercise 5.13 (EAWE)

ols S const ASVABC SM SF MALE ETHHISP ETHWHITE

University of London International Programmes 9

20 Elements of econometrics

Exercise 5.14 (EAWE)

ols LGEARN const S EXP MALE ETHHISP ETHWHITE

Exercise 5.15 (EAWE)

ols LGEARN const S EXP MALE FEMALE

Exercise 5.18 (EAWE)

genr MALEASVC = MALE*ASVABC
ols S const ASVABC SM SF MALE ETHBLACK ETHHISP MALEASVC
store path\EAWE**.gdt
In the last command, you should replace ** by the number of your
particular data set and path\ by the path to it. See Exercise 3.15
for remarks on saving data files.

Exercise 5.20 (EAWE)

genr MALES = MALE*S
ols LGEARN const S EXP MALE ETHHISP ETHWHITE MALES
store path\EAWE**.gdt

Exercise 5.21 (EAWE)

genr MALEBLAC = MALE*ETHBLACK
genr MALEHISP = MALE*ETHHISP
ols S const ASVABC SM SF MALE ETHBLACK ETHHISP MALEBLAC MALEHISP
store path\EAWE**.gdt

Exercise 5.25 (EAWE)

ols S const ASVABC SM SF ETHBLACK ETHHISP
smpl MALE = 1 --restrict
ols S const ASVABC SM SF ETHBLACK ETHHISP
smpl full
smpl MALE = 0 --restrict
ols S const ASVABC SM SF ETHBLACK ETHHISP
smpl full

The smpl command is cumulative. smpl MALE = 1 --

restrict eliminates all the female observations. If it is followed
by smpl MALE = 0 --restrict, all the male observations are
eliminated as well, leaving a null data set. To avoid this, any smpl
command that is temporary and no longer needed should be
cancelled with the smpl full command. This restores the sample
to the full sample.

10 University of London International Programmes

 Manual for the computer exercises

Exercise 5.26 (EAWE)

ols LGEARN const S EXP MALE ETHBLACK ETHHISP
smpl MALE = 1 --restrict
ols LGEARN const S EXP MALE ETHBLACK ETHHISP
smpl full
smpl MALE = 0 --restrict
ols LGEARN const S EXP MALE ETHBLACK ETHHISP
smpl full

Exercise 5.27 (EAWE)

genr MALESM = MALE*SM
genr MALESF = MALE*SF
ols S const ASVABC SM SF MALE ETHBLACK ETHHISP MALEASVC MALESM MALESF
MALEBLAC MALEHISP
ols S const ASVABC SM SF ETHBLACK ETHHISP
corr ASVABC SM SF MALE ETHBLACK ETHHISP MALEASVC MALESM MALESF
MALEBLAC MALEHISP
store path\EAWE**.gdt

Note that the command for the first regression is so long that it has
wrapped to a second line. This is fine if you are using the command
line (console box). However, if you are writing a batch file, gretl
treats each line as a separate command. To indicate that a second
line is a continuation of the first, one needs to insert a backslash at
the end of the first line.
The corr command computes the correlations between the
variables listed. It has also wrapped and would need a backslash at
the end of the first line if you were writing a batch file.

Exercise 5.28 (EAWE)

genr MALEEXP = MALE*EXP
ols LGEARN const S EXP MALE ETHBLACK ETHHISP MALES MALEEXP MALEBLAC
MALEHISP
ols LGEARN const S EXP ETHBLACK ETHHISP
corr S EXP MALE ETHBLACK ETHHISP MALES MALEEXP MALEBLAC MALEHISP
store path\EAWE**.gdt

Exercise 6.1 (EAWE)

ols S const ASVABC SM
ols S const ASVABC
ols S const SM
corr ASVABC SM

University of London International Programmes 11

20 Elements of econometrics

Exercise 6.2 (EAWE)

ols LGEARN const S EXP
ols LGEARN const S
ols LGEARN const EXP
corr S EXP

Exercise 6.3 (EAWE)

ols LGEARN const S EXP MALE ETHBLACK ETHHISP
ols LGEARN const S EXP MALE ETHBLACK ETHHISP ASVABC
corr S EXP MALE ETHBLACK ETHHISP ASVABC

Exercise 6.8 (EAWE)

ols LGEARN const S EXP ASVABC MALE ETHBLACK ETHHISP
ols LGEARN const S EXP ASVABC MALE ETHBLACK ETHHISP AGE
corr S EXP ASVABC MALE ETHBLACK ETHHISP AGE

Exercise 6.10 (EAWE)

If you did Exercise 3.17 and saved your data set afterwards, you do
not need to define PWE again here.
genr PWE = AGE - S - 5
ols LGEARN const S ASVABC MALE ETHBLACK ETHHISP
ols LGEARN const S ASVABC MALE ETHBLACK ETHHISP PWE
ols LGEARN const S ASVABC MALE ETHBLACK ETHHISP EXP
corr S EXP PWE
Variation
smpl MALE = 1 --restrict
ols LGEARN const S ASVABC MALE ETHBLACK ETHHISP
ols LGEARN const S ASVABC MALE ETHBLACK ETHHISP PWE
ols LGEARN const S ASVABC MALE ETHBLACK ETHHISP EXP
corr S EXP PWE
smpl full
smpl MALE = 0 --restrict
ols LGEARN const S ASVABC MALE ETHBLACK ETHHISP
ols LGEARN const S ASVABC MALE ETHBLACK ETHHISP PWE
ols LGEARN const S ASVABC MALE ETHBLACK ETHHISP EXP
corr S EXP PWE
smpl full

12 University of London International Programmes

 Manual for the computer exercises

Exercise 6.12 (EAWE)

The text should have specified using ASVABC as an explanatory
variable in the second regression as well as in the first.
genr PREVEXP = EXP - TENURE
ols LGEARN const S ASVABC EXP MALE ETHBLACK ETHHISP
ols LGEARN const S ASVABC PREVEXP TENURE MALE ETHBLACK ETHHISP
smpl MALE = 1 --restrict
ols LGEARN const S ASVABC EXP MALE ETHBLACK ETHHISP
ols LGEARN const S ASVABC PREVEXP TENURE MALE ETHBLACK ETHHISP
smpl full
smpl MALE = 0 --restrict
ols LGEARN const S ASVABC EXP MALE ETHBLACK ETHHISP
ols LGEARN const S ASVABC PREVEXP TENURE MALE ETHBLACK ETHHISP
smpl full

Exercise 6.13 (EAWE)

ols LGEARN const S ASVABC EXP TENURE MALE ETHBLACK ETHHISP
smpl MALE = 1 --restrict
ols LGEARN const S ASVABC EXP TENURE MALE ETHBLACK ETHHISP
smpl full
smpl MALE = 0 --restrict
ols LGEARN const S ASVABC EXP TENURE MALE ETHBLACK ETHHISP
smpl full

Exercise 6.16 (EAWE)

ols S const SM SF ASVABC
ols S const SM SF ASVABAR ASVABWK ASVABPC

Exercise 7.3 (EAWE)

As should be obvious, the first command sorts the observations in
(increasing) size of S. The second command restricts the sample to
the first 203 of the sorted observations (approximately three-eights
of 540). The fourth command redefines the sample as those
observations with the 203 largest values of S.
The specification has deliberately been kept rather simple, in
anticipation of Exercise 7.4. Obviously it would be a good idea to
add ASVABC, ETHBLACK, and ETHHISP.
dataset sortby S
smpl 1 188
ols EARNINGS const S EXP MALE

University of London International Programmes 13

20 Elements of econometrics

smpl full
smpl 313 500
ols EARNINGS const S EXP MALE
smpl full

Exercise 7.4 (EAWE)

It is assumed that MALES and MALEEXP have been defined and
saved in previous exercises. The residuals from the first regression
are saved as EEARN. As in Exercise 7.3, it would be a good idea to
add ASVABC, ETHBLACK, and ETHHISP to the specification ... and
their squares and interactions with each other and the other
variables to the test equation. The final command is a special gretl
function that produces the test statistic, nR2, for the White test.
ols EARNINGS const S EXP MALE
genr EEARN = $uhat
genr EEARNSQ = EEARN*EEARN
genr SSQ = S*S
genr EXPSQ = EXP*EXP
genr SEXP = S*EXP
ols EEARNSQ const S EXP MALE SSQ EXPSQ SEXP MALES MALEEXP
genr WHITE = $trsq

Exercise 7.6 (EDUC)

It is assumed that you have downloaded the EDUC data set and
have opened it within gretl.
dataset sortby GDP
smpl 1 14
ols EDUC const GDP
smpl full
smpl 25 38
ols EDUC const GDP
smpl full

Exercise 7.8 (EAWE)

It is assumed that the variables defined in Exercise 7.4 have been
saved.
Repeat of Exercise 7.3 with LGEARN
dataset sortby S
smpl 1 188
ols LGEARN const S EXP MALE
smpl full

14 University of London International Programmes

 Manual for the computer exercises

smpl 313 500

ols LGEARN const S EXP MALE
smpl full
Repeat of Exercise 7.4 with LGEARN
ols LGEARN const S EXP MALE
genr EELG = $uhat
genr EELGSQ = EELG*EELG
genr SSQ = S*S
genr EXPSQ = EXP*EXP
genr SEXP = S*EXP
ols EELGSQ const S EXP MALE SSQ EXPSQ SEXP
MALES MALEEXP
genr WHITELG = $trsq

Exercise 7.9 (EDUC)

Scaling by population
genr EDUCPOP = EDUC/POP
genr GDPPOP = GDP/POP
genr RECPOP = 1/POP
dataset sortby GDPPOP
smpl 1 14
ols EDUCPOP RECPOP GDPPOP
smpl full
smpl 25 38
ols EDUCPOP RECPOP GDPPOP
smpl full
Scaling by GDP
genr EDUCGDP = EDUC/GDP
genr RECGDP = 1/GDP
dataset sortby RECGDP
smpl 1 14
ols EDUCGDP const RECGDP
smpl full
smpl 25 38
ols EDUCGDP const RECGDP
smpl full
Logarithmic specification
genr LGEDUC = log(EDUC)
genr LGGDP = log(GDP)

University of London International Programmes 15

20 Elements of econometrics

dataset sortby LGGDP

smpl 1 14
ols LGEDUC const LGGDP
smpl full
smpl 25 38
ols LGEDUC const LGGDP
smpl full
Whole sample regressions, for comparison
ols EDUCPOP RECPOP GDPPOP
ols EDUCGDP const GDPREC
ols LGEDUC const LGGDP

Exercise 8.16 (EAWE)

The command for undertaking an instrumental variable regression
is tsls (short for two stage least squares; all IV regressions may
be considered to be special cases of this estimation procedure). The
first item after the command is, as usual, the dependent variable,
and this is followed by const (assuming that you do want a
constant) and the list of the explanatory variables appearing in the
regression. Then follows a semicolon, and finally the list of the
instrumental variables being used. gretl follows the convention that
any variable that does not need an instrument is said to be acting as
its own instrument and so should be included in the list of
instruments following the semicolon. This applies even to the
constant. Hence the list of instruments is the same as the list of the
explanatory variables except (1) those variables needing
instruments do not appear, and the external instruments do
appear. In this case ASVABC is dropped from the list of instruments
because it is being instrumented, and SM, SF, SIBLINGS, and
LIBRARY are included in the list of instruments.
The Durbin–Wu–Hausman test is automatically included in the
tsls regression output. It is implemented by gretl in a way that is
different from that used in Stata and so may give somewhat
different results.
tsls LGEARN const S EXP ASVABC ; const S EXP SM SF SIBLINGS
corr ASVABC SM SF SIBLINGS
The second command has been included to investigate the strength
of the correlation between ASVABC and the instruments. Although
the exercise in the text does not suggest including MALE,
ETHBLACK, and ETHHISP in the regression, obviously this would
improve the specification.
tsls LGEARN const S EXP ASVABC MALE ETHBLACK ETHHISP ; const S EXP MALE
ETHBLACK ETHHISP SM SF SIBLINGS LIBRARY

Exercise 9.10 (EAWE)

ols ASVABC const S

16 University of London International Programmes

 Manual for the computer exercises

tsls ASVABC const S ; const SM

cor S SM

Exercise 10.4 (EAWE)

genr COLLEGE = (S > 12)
ols COLLEGE const ASVABC SM SF MALE
logit COLLEGE const ASVABC SM SF MALE --p-values
logit COLLEGE const ASVABC SM SF MALE
The first command generates COLLEGE as a dummy variable that is
equal to 1 when the condition in parentheses is satisfied and equal
to 0 when it is not. So COLLEGE is equal to 1 for those who had
more than 12 years of schooling (which means that they must have
at least one year of college) and it is 0 for the rest. The second
command performs a linear probability model regression. The third
performs the corresponding logit regression. The fourth does the
same, except that the output reports the marginal effects instead of
the p values. As can be seen, this is actually the default for the logit
regression in gretl. The marginal effects are calculated at the mean
values of all of the explanatory variables.

Exercise 10.6 (EAWE)

probit COLLEGE const ASVABC SM SF MALE --p-values
probit COLLEGE const ASVABC SM SF MALE

Exercise 10.8 (LFP2011)

Note that the probit commands are so long that they may wrap. If this is the case,
a back slash should be placed at the end of the first line to indicate that it
continues to the second.
genr MARL06 = MARRIED*CHILDL06
smpl MALE = 0 --restrict
probit WORKING const S AGE CHILDL06 CHILDL16 MARRIED
MARL06 ETHBLACK ETHHISP --p-values
probit WORKING const S AGE CHILDL06 CHILDL16 MARRIED
MARL06 ETHBLACK ETHHISP

Exercise 10.9 (CES2013)

Remember, CAT is a placeholder for your category of expenditure.
See Exercise 4.4.
ols CAT const EXP
smpl CAT > 0 --restrict
ols CAT const EXP
smpl full
tobit CAT const EXP
University of London International Programmes 17
20 Elements of econometrics

The first command fits the regression using all the observations,
including the zero observations. The next two commands do the
same, restricting the sample to the non-zero observations. The last
performs the tobit regression.

Exercise 10.11 (EAWE)

smpl S > 12 --restrict
genr COLLYEAR = S - 12
genr LGEARNCL = LGEARN
ols LGEARNCL const COLLYEAR EXP ASVABC MALE ETHBLACK ETHHISP
smpl full
genr D = (S > 12)
heckit LGEARNCL const COLLYEAR EXP ASVABC MALE ETHBLACK ETHHISP ;
D const ASVABC MALE ETHBLACK ETHHISP SM SF SIBLINGS
The third command defines the logarithm of earnings for those who
went to college. It is not defined for the rest. The sixth command
defines a dummy variable D that is equal to 1 for observations that
are selected and 0 for the rest. The first part of the heckit
command regresses the logarithm of earnings on the usual
explanatory variables. The second part (after the semicolon) lists
the dummy variable D followed by the variables used to explain
selection. If this command is being used in a batch file, a backslash
should be added at the end of the first line to indicate that the
command continues to a second line.

Exercise 10.13 (LFP2011)

genr MARL06 = MARRIED*CHILDL06
genr LGEARN = ln(EARNINGS)
smpl MALE = 0 --restrict
heckit LGEARN const S ASVABC ETHBLACK
ETHHISP ; WORKING const S ASVABC ETHBLACK
ETHHISP AGE CHILDL06 CHILDL16 MARRIED MARL06
The data set is lfp.gdt. Compare the regression results with those
in Table 10.8 in the textbook. The basic regression specification is
the same. The only difference is in the specification of the selection
model. Note that WORKING is a dummy variable that indicates that
the individual has been selected into the sample. It is not an
explanatory variable. It has the same function as D in Exercise
10.11.

Exercise 11.2 Construction of a relative price

index (DF)
In the commands in this and the following exercises, CAT is a place
holder and should be replaced by the three or four letter name of
your category of expenditure.
open path\demand.gdt
18 University of London International Programmes
 Manual for the computer exercises

genr PRELCAT = 100*PCAT/PTPE

plot PRELCAT
Thus, if your category were FLOW, the second and third commands
would be
genr PRELFLOW = 100*PFLOW/PTPE
scatters PRELFLOW DATE
store path\demand.gdt
The plot is part of the output file and in ASCII format, so somewhat
unsophisticated. Better plots can be obtained using the gnuplot
facility bundled with gretl. I do not know how to use it from the
command line, but it can be accessed easily through the GUI via the
View/Graph specified vars tab on the main menu.
You will need PRELCAT in many of the following exercises, so save
the data set once it has been defined, overwriting the existing one.

Exercise 11.3 Linear regression (DF)

ols CAT const DPI PRELCAT

Exercise 11.4 Logarithmic multiple regression (DF)

genr LGDPI = log(DPI)
genr LGCAT = log(CAT)
genr LGPRCAT = log(PRELCAT)
ols LGCAT const LGDPI LGPRCAT
store path\demand.gdt

Exercise 11.5 Multicollinearity (DF)

ols LGCAT const LGDPI LGPRCAT TIME
corr LGCAT LGDPI LGPRCAT TIME

Exercise 11.9 Naive attempts to model dynamics

(DF)
The lags command creates lagged versions of existing variables.
The command is followed by the number of lags desired, then a
semicolon, and then a list of the variables in question. Thus
LGDPI_1 is LGDPI lagged one time period (in this case, one year),
and LGDPI_2 is LGDPI lagged two time periods.
lags 2 ; LGCAT LGDPI LGPRCAT
ols LGCAT const LGDPI LGDPI_1 LGPRCAT
LGPRCAT_1
ols LGCAT const LGDPI LGDPI_1 LGDPI_2
LGPRCAT LGPRCAT_1 LGPRCAT_2

University of London International Programmes 19

20 Elements of econometrics

Exercise 11.10 Reparameterized model (DF)

genr X1 = LGDPI - LGDPI_1
genr X2 = LGDPI - LGDPI_2
genr P1 = LGPRCAT - LGPRCAT_1
genr P2 = LGPRCAT - LGPRCAT_2
ols LGCAT const LGDPI X1 X2 LGPRCAT P1 P2

Exercise 11.13 Simple dynamics: a partial

adjustment model (DF)
If the lags command is followed by a variable name or names, it is
assumed that only the first lag is required.
lags CAT LGCAT
ols CAT const DPI PRELCAT CAT_1
ols LGCAT const LGDPI LGPRCAT LGCAT_1

Exercise 12.3 Tests for autocorrelation (DF)

When it was defined, the Demand Functions data set was declared
to be a time series set with annual data from 1959 to 2003. This
accounts for the inclusion of the Durbin–Watson d statistic among
the diagnostic statistics. The last three commands below are for the
Breusch–Godfrey test. Remember that $uhat is a temporary
variable containing the residuals from the most recent regression.
ols LGCAT const LGDPI LGPRCAT
genr ECAT = $uhat
lags ECAT
ols ECAT const LGDPI LGPRCAT ECAT_1

Exercise 12.4 Plot of residuals (DF)

See Exercise 11.2 for remarks about plots.
ols LGCAT const LGDPI LGPRCAT
genr ECAT = $uhat
scatters EFOOD DATE

Exercise 12.5 Tests for autocorrelation, ADL(1,0)

model (DF)
ols LGCAT const LGDPI LGPRCAT LGCAT_1
genr ECAT = $uhat
lags ECAT
ols ECAT const LGDPI LGPRCAT LGCAT_1 ECAT_1

20 University of London International Programmes

 Manual for the computer exercises

Exercise 12.6 Comparison of OLS and AR(1)

specifications (DF)
ar1 implements the Cochrane–Orcutt iterative procedure for fitting
a regression assumed to be subject to AR(1) autocorrelation.
Adding --pwe at the end changes the estimation method to the
Prais–Winsten variation. This retains the use of the first
observation, which is lost under C–O, and may lead to a gain in
efficiency.
ols LGCAT const LGDPI LGPRCAT
ar1 LGCAT const LGDPI LGPRCAT
ar1 LGCAT const LGDPI LGPRCAT --pwe

Exercise 12.9 Common factor test (DF)

The third, fourth, and fifth commands are for performing the Breusch–Godfrey test.
ar1 LGCAT const LGDPI LGPRCAT
ols LGCAT const LGCAT_1 LGDPI LGDPI_1 LGPRCAT LGPRCAT_1
genr ECAT = $uhat
lags ECAT
ols ECAT const LGCAT_1 LGDPI LGDPI_1 LGPRCAT LGPRCAT_1 ECAT_1
ols LGCAT const LGDPI LGPRCAT LGCAT_1

Exercise 13.14 Tests for unit roots (DF)

The diff command creates new variables defined as the first
differences of the variables listed. The names are the names of the
variables preceded by d_followed by an underscore. The adf
command produces the augmented Dickey–Fuller test statistic for
the variable listed. The number after adf gives the number of
lagged differences to be included in the regression specification.
Thus adf 1 presupposes a model of the type given by equation
(13.65) in the text with the addition of a deterministic time trend.
Options include --nc (no constant, no time trend), --c
(constant, no time trend), and --ct (both constant and time
trend). Adding --verbose produces the full output for the
regression.
diff LGCAT LGPRCAT
adf 1 LGCAT --ct
adf 1 d_LGCAT --ct
adf 1 LGPRCAT --ct
adf 1 d_LGPRCAT --ct

Exercise 13.19 Test for cointegration (DF)

Remember that $uhat is a temporary variable that contains the
residuals from the most recent regression.
ols LGCAT const LGDPI LGPRCAT
University of London International Programmes 21
20 Elements of econometrics

genr ECAT = $uhat

adf 1 ECAT --nc

Exercise 13.23 Error correction model (DF)

diff LGCAT LGDPI LGPRCAT
lags 1 ECAT
ols d_LGCAT ECAT_1 d_LGDPI d_LGPRCAT

Exercise 14.1 Pooled OLS regression (OECD2000)

open path\OECD2000.gdt
ols E const G TIME2 TIME3

Exercise 14.2 (OECD2000)

There are various ways of indicating to gretl that a data set has the
dual structure required for panel regressions. One, used here, is to
specify the unit indicator and the time indicator, followed by --
panel-vars. Here the units are countries and ID is the country
indicator. The time indicator is simply TIME itself. The panel
command fits a within-groups fixed effects model by default. It
automatically includes a test for the presence of fixed effects, the
null hypothesis being that they do not exist and that a pooled OLS
regression is adequate (and preferable, being more efficient
because it does not involve the estimation of unnecessary dummy
variable coefficients). The panel command does this by
performing an F test of the joint explanatory power of the dummy
variables in the least squares dummy variables (LSDV) version of
the fixed effects model, the specification with an intercept and no
dummy for one of the units of observation. It compares RSS for the
LSDV model with RSS for the pooled OLS regression. At least this
is what is says it is doing. It does not actually have to fit the LSDV
model itself because the within-groups regression is equivalent.
setobs ID TIME --panel-vars
panel E const G TIME2 TIME3

Exercise 14.3 (OECD2000)

D* is interpreted by gretl as shorthand for a list of all the variables
in the data set starting with D and is a convenient way of including
all the country dummies in the specification. The delete
command is supposed to cause the listed variable(s) to be dropped
from the data set. If the specification includes an intercept, it is
necessary to drop one of the dummies. However I have not
succeeded in getting this command to work. So in the second
regression, DAUS still exists and the specification is subject to exact
multicollinearity. To avoid this, gretl drops the final dummy
variable (that for the USA).
ols E G TIME2 TIME3 D*
delete DAUS

22 University of London International Programmes

 Manual for the computer exercises

ols E const G TIME2 TIME3 D*

Exercise 14.4 (OECD2000)

Use RSS from Exercise 14.1 and the second regression in 14.3

Exercise 14.5 (NLSY2000)

open path\NLSY2000.gdt
setobs ID TIME --panel-vars
genr LGEARN = log(EARNINGS)
ols LGEARN const S ASVABC EXP MALE ETHBLACK ETHHISP
panel LGEARN const S ASVABC EXP MALE ETHBLACK ETHHISP
store path\NLSY2000.gdt

Exercise 14.6 (NLSY2000)

The --random-effects switch indicates to gretl that the panel
regression should be random effects, rather than the default
within-groups fixed effects. After performing a random effects
regression, gretl automatically compares the results with those of
pooled OLS and fixed effects regressions. It performs a Breusch–
Pagan test to test for the presence of random effects, the null
hypothesis being that pooled OLS is adequate. It performs a
Hausman test to discriminate between the presence of random and
fixed effect, the null hypotheses being random effects.
ols LGEARN const S ASVABC EXP MALE ETHBLACK ETHHISP UNION
panel LGEARN const S ASVABC EXP MALE ETHBLACK ETHHISP UNION --random-
effects

Exercise 14.7 (NLSY2000)

If a data set has been declared to be panel data, placing the
hausman command after a pooled OLS regression will
automatically cause the corresponding fixed and random effects
models also to be fitted and the three tests already described to be
performed: the F test discriminating between pooled OLS and fixed
effects (see Exercise 14.2), the Breusch–Pagan test discriminating
between pooled OLS and random effects, and the Hausman test
itself discriminating between fixed effects and random effects.
ols LGEARN const S ASVABC EXP MALE ETHBLACK ETHHISP
hausman
panel LGEARN const S ASVABC EXP MALE ETHBLACK ETHHISP
panel LGEARN const S ASVABC EXP MALE ETHBLACK ETHHISP --random-effects

Exercise 14.8 (NLSY2000)

ols LGEARN const S ASVABC EXP MALE ETHBLACK ETHHISP UNION
University of London International Programmes 23
20 Elements of econometrics

hausman
panel LGEARN const S ASVABC EXP MALE ETHBLACK ETHHISP UNION
panel LGEARN const S ASVABC EXP MALE ETHBLACK ETHHISP UNION --random-
effects

Exercise 14.11 (DIFFDIFF)

open path\DIFFDIFF.gdt
genr D = (DATE = 2011)
genr DMALE = D*MALE
smpl S = 16 --restrict
ols EARNINGS const D MALE DMALE
smpl full
smpl S > 12 --restrict
smpl S < 17 --restrict
ols EARNINGS const D MALE DMALE
ols EARNINGS const D MALE DMALE S
smpl full
smpl S > 16 --restrict
ols EARNINGS const D MALE DMALE
ols EARNINGS const D MALE DMALE S
smpl full

24 University of London International Programmes