A Hansl Primer

The scripting language of gretl in 53 minutes

Allin Cottrell
Department of Economics
Wake Forest University

Riccardo (Jack) Lucchetti

Dipartimento di Scienze Economiche e Sociali
Università Politecnica delle Marche

November, 2020
Permission is granted to copy, distribute and/or modify this document under the terms of the
GNU Free Documentation License, Version 1.1 or any later version published by the Free Software
Foundation (see http://www.gnu.org/licenses/fdl.html).
 Contents

1 Introduction 1

2 For the impatient 3

I Without a dataset 5

3 Hello, world! 6
3.1 Manipulation of scalars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.2 Manipulation of strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4 Matrices 11
4.1 Matrix indexing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.2 Matrix operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.3 Matrix pointers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

5 Nice-looking output 15
5.1 Formatted output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.2 Graphics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

6 Structured data types 19

6.1 Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6.2 Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

7 Numerical methods 23
7.1 Numerical optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
7.2 Numerical differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

8 Control flow 28
8.1 The if statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
8.2 Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
8.3 The catch modifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
8.4 The quit statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

9 User-written functions 36
9.1 Parameter passing and return values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
9.2 Recursion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

i
Contents ii

II With a dataset 41

10 What is a dataset? 42
10.1 Creating a dataset from scratch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
10.2 Reading a dataset from a file . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
10.3 Saving datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
10.4 The smpl command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
10.5 Dataset accessors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

11 Series and lists 45

11.1 The series type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
11.2 The list type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

12 Estimation methods 49
12.1 Canned estimation procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
12.2 Generic estimation tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
12.3 Post-estimation accessors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
12.4 Formatting the results of estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
12.5 Named models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

III Reference 52

13 Rules regarding white space 53

13.1 White space in commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
13.2 Spaces in function calls and assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

14 Operators 54
14.1 Precedence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
14.2 Assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
14.3 Increment and decrement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

15 Greek-letter identifiers 57
 Chapter 1

Introduction

What hansl is and what it is not

Hansl is a recursive acronym: it stands for “Hansl’s A Neat Scripting Language”. You might there-
fore expect something very general in purpose. Not really. Hansl was born as the scripting lan-
guage for the econometrics program gretl and its role is unlikely to change. As a consequence,
hansl should not be viewed as a fully fledged programming language such as C, Fortran, Perl or
Python. Not because it lacks anything to be considered as such,1 but because its aim is different.
Hansl should be considered as a special-purpose or domain-specific language, designed to make
an econometrician’s life easier. Hence it incorporates a series of conventions and choices that may
irritate purists and have some marginal impact on raw performance, but that we, as professional
econometricians, consider “nice to have”. This makes hansl somewhat different from plain matrix-
oriented interpreted languages, such as the Matlab/Octave family, Ox and so on.
On the other hand, hansl is by no means just a tool to automate rote tasks. It has several features
which support advanced work: structured programming, recursion, complex data structures, and
so on. As for style, the language which hansl most resembles is probably that of the bash shell.

The intent and structure of this document

The intended readers of this document are those who already know how to write code, and are
familiar with the associated do-s and don’t-s. Such people may wish to add hansl to their toolbox,
alongside languages like C or FORTRAN, or programs such as R, Ox, Matlab, Stata or Gauss, some
of which they are already confident with. Here, therefore, the focus is not so much on “How do I do
this?”, but rather on “How do I do this in hansl?”.
As a consequence, this document aims at making the reader a reasonably proficient hansl user in
a (relatively) short time; however, not all the features of hansl are illustrated; for those, interested
readers should consult the Gretl Command Reference and the Gretl User’s Guide.
This guide comprises two main parts. Part I (“Without a dataset”) concentrates on hansl as a pure
matrix-oriented programming language. Part II (“With a dataset”) exploits the fact that hansl scripts
are run through gretl, which has very nice facilities for handling statistical datasets (interactively
if necessary). This provides hansl with a series of extra constructs and features which make it
extremely easy to write hansl scripts to perform all sorts of statistical procedures.
In order to use hansl, you will need a working installation of gretl. We assume you have one. If you
don’t, please refer to chapter 1 of the Gretl User’s Guide.

Other resources
If you are serious about learning hansl then after working through this primer — or in the process
of doing so—you’ll want to take a look at the following additional resources.

• The Gretl Command Reference. This contains a complete listing of the commands and built-in
functions available in hansl, with a full account of their syntax and options. Examples of usage
1 Hansl is in fact Turing-complete.

1
Chapter 1. Introduction 2

are provided in some instances. This is available in an “online” version for handy reference as
well as in PDF, both accessible via the Help menu in the gretl GUI.

• The Gretl User’s Guide. Chapters 10 to 16, in particular, go into more detail on some of the
programming topics discussed here (data types, loops, the definition and use of functions). In
addition Part II of the Guide, on Econometric Methods, gives many examples of hansl usage.
The Guide is available via gretl’s Help menu; the latest version can also be found online at
http://sourceforge.net/projects/gretl/files/manual/.

• Sample scripts. The gretl package comes with a large number of sample or practice scripts,
which can be found under the menu item /File/Script files/Practice file. Many of these are
simple replication exercises for textbook problems but you will find some more interesting
examples under the Gretl tab.

• Function packages. Relatively ambitious examples of hansl coding can be found in the various
contributed “function packages”. You can download these packages via the gretl menu item
/Tools/Function packages/On server. Once a package is downloaded it appears in the listing
under /Tools/Function packages/On local machine; in that context you can right-click and
select View code to examine the hansl functions.

• The gretl-users mailing list. Most well-considered questions get answered quite quickly and
fully. See https://gretlml.univpm.it/postorius/lists/.
 Chapter 2

For the impatient

OK, so you’re impatient. Then perhaps you’re thinking “Do I really need to go through the whole
thing? After all, I’ve been coding econometric stuff for a while, and I’m pretty confident I can pick
a new scripting language if it’s not too obscure. I just need a few tips to get me started”. If that’s
not what you’re thinking at all, we suggest you move along to chapter 3; but if it is, well then, we’ll
give you a hansl script which exemplifies a hefty share of the topics discussed in the rest of this
primer. We will use for our example a Vector AutoRegressive model, or VAR for short.
As you probably know, a finite-order VAR can be estimated via conditional maximum likelihood
simply by applying OLS to each equation individually. That amounts to solving a least-squares
problem and its solution can be easily written, in matrix notation, as Π̂ = (X 0 X)−1 X 0 Y , where Y con-
tains your endogenous variables and X contains their lags plus other exogenous terms (typically, a
constant term at least). But of course, you may choose to find the maximum of the concentrated
likelihood L = −(T /2) ln |Σ̂| numerically if you so wish.
The following example contains a hansl script which performs these actions:

1. Reads data from a disk file.

2. Performs some data transformation and simple visualization.

3. Estimates the VAR via

(a) the native hansl var command

(b) sequential single-equation OLS
(c) matrix algebra (in 3 different ways)
(d) numerical maximization of the log-likelihood.

4. Prints out the results.

The script also contains some concise comments.

open AWM.gdt --quiet # load data from disk

/* data transformations and visualisation */

series y = 100 * hpfilt(ln(YER)) # the "series" concept: operate on

series u = 100 * URX # vectors on an element-by-element basis
series r = STN - 100*sdiff(ln(HICP)) # (but you also have special functions)

scatters y r u --output=display # command example with an option: graph data

/* in-house VAR */

scalar p = 2 # strong typing: a scalar is not a

# matrix nor a series

var p y r u # estimation command

3
Chapter 2. For the impatient 4

A = $coeff # and corresponding accessor

/* by iterated OLS */

list X = y r u # the list is yet another variable type

matrix B = {} # initialize an empty matrix

loop foreach i X # loop over the 3 var equations

ols $i const X(-1 to -p) --quiet # using native OLS command
B ~= $coeff # and store the estimated coefficients
endloop # as matrix columns

/* via matrices */

matrix mY = { y, r, u } # construct a matrix from series

matrix mX = 1 ~ mlag(mY, {1,2}) # or from matrix operators/functions
mY = mY[p+1:,] # and select the appropriate rows
mX = mX[p+1:,] # via "range" syntax

C1 = mX\mY # matlab-style matrix inversion

C2 = mols(mY, mX) # or native function
C3 = inv(mX’mX) * (mX’mY) # or algebraic primitives

/* or the hard, needlessly complicated way --- just to show off */

function scalar loglik(matrix param, const matrix X, const matrix Y)

# this function computes the concentrated log-likelihood

# for an unrestricted multivariate regression model

scalar n = cols(Y)
scalar k = cols(X)
scalar T = rows(Y)
matrix C = mshape(param, k, n)
matrix E = Y - X*C
matrix Sigma = E’E

return -T/2 * ln(det(Sigma))

end function

matrix c = zeros(21,1) # initialize the parameters

mle ll = loglik(c, mX, mY) # and maximize the log-likelihood
params c # via BFGS, printing out the
end mle # results when done

D = mshape(c, 7, 3) # reshape the results for conformability

/* print out the results */

# note: row ordering between alternatives is different

print A B C1 C2 C3 D

If you were able to follow the script above in all its parts, congratulations. You probably don’t need
to read the rest of this document (though we don’t mind if you do). But if you find the script too
scary, never fear: we’ll take things step by step. Read on.
 Part I

Without a dataset

5
 Chapter 3

Hello, world!

We begin with the time-honored “Hello, world” program, the obligatory first step in any program-
ming language. It’s actually very simple in hansl:

# First example
print "Hello, world!"

There are several ways to run the above example: you can put it in a text file first_ex.inp and
have gretl execute it from the command line through the command

gretlcli -b first_ex.inp

or you could just copy its contents in the editor window of a GUI gretl session and click on the
“gears” icon. It’s up to you; use whatever you like best.
From a syntactical point of view, allow us to draw attention on the following points:

1. The line that begins with a hash mark (#) is a comment: if a hash mark is encountered,
everything from that point to the end of the current line is treated as a comment, and ignored
by the interpreter.
2. The next line contains a command (print) followed by an argument; this is fairly typical of
hansl: many jobs are carried out by calling commands.
3. Hansl does not have an explicit command terminator such as the “;” character in the C lan-
guage family (C++, Java, C#, . . . ) or GAUSS; instead it uses the newline character as an implicit
terminator. So at the end of a command, you must insert a newline; conversely, you can’t put
a newline in the middle of a command — or not without taking special measures. If you need
to break a command over more than one line for the sake of legibility you can use the “\”
(backslash) character, which causes gretl to ignore the following line break.

Note also that the print command automatically appends a line break, and does not recognize
“escape” sequences such as “\n”; such sequences are just printed literally. The printf command
can be used for greater control over output; see chapter 5.
Let’s now examine a simple variant of the above:

/*
Second example
*/
string foo = "Hello, world"
print foo

In this example, the comment is written using the convention adopted in the C programming lan-
guage: everything between “/*” and “*/” is ignored.1 Comments of this type cannot be nested.
1 Each type of comment can be masked by the other:
• If /* follows # on a given line which does not already start in ignore mode, then there’s nothing special about /*,
it’s just part of a #-style comment.
• If # occurs when we’re already in comment mode, it is just part of a comment.

6
Chapter 3. Hello, world! 7

Then we have the line

string foo = "Hello, world"

In this line, we assign the value “Hello, world” to the variable named foo. Note that

1. The assignment operator is the equals sign (=).

2. The name of the variable (its identifier) must follow the following convention: identifiers can
be at most 31 characters long and must be plain ASCII. They must start with a letter, and can
contain only letters, numbers and the underscore character.2 Identifiers in hansl are case-
sensitive, so foo, Foo and FOO are three distinct names. Of course, some words are reserved
and can’t be used as identifiers (however, nearly all reserved words only contain lowercase
characters).

3. The string delimiter is the double quote (").

In hansl, a variable has to be of one of these types: scalar, series, matrix, list, string, bundle
or array. As we’ve just seen, string variables are used to hold sequences of alphanumeric charac-
ters. We’ll introduce the other ones gradually; for example, the matrix type will be the object of
the next chapter.
The reader may have noticed that the line

string foo = "Hello, world"

implicitly performs two tasks: it declares foo as a variable of type string and, at the same time,
assigns a value to foo. The declaration component is not strictly required. In most cases gretl
is able to figure out by itself what type a newly introduced variable should have, and the line
foo = "Hello, world" (without a type specifier) would have worked just fine. However, it is
more elegant (and leads to more legible and maintainable code) to use a type specifier at least the
first time you introduce a variable.
In the next example, we will use a variable of the scalar type:

scalar x = 42
print x

A scalar is a double-precision floating point number, so 42 is the same as 42.0 or 4.20000E+01.

Note that hansl doesn’t have a separate variable type for integers or complex numbers.
An important detail to note is that, contrary to most other matrix-oriented languages in use in the
econometrics community, hansl is strongly typed. That is, you cannot assign a value of one type to
a variable that has already been declared as having a different type. For example, this will return
an error:

string a = "zoo"
a = 3.14 # no, no, no!

If you try running the example above, an error will be flagged. However, it is acceptable to destroy
the original variable, via the delete command, and then re-declare it, as in

scalar X = 3.1415
delete X
string X = "apple pie"
2 Actually one exception to this rule is supported: identifiers taking the form of a single Greek letter. See chapter 15

for details.
Chapter 3. Hello, world! 8

There is no “type-casting” as in C, but some automatic type conversions are possible (more on this
later).
Many commands can take more than one argument, as in

set echo off

set messages off

scalar x = 42
string foo = "not bad"
print x foo

In this example, one print is used to print the values of two variables; more generally, print can
be followed by as many arguments as desired. The other difference with respect to the previous
code examples is in the use of the two set commands. Describing the set command in detail
would lead us to an overly long diversion; suffice it to say that this command is used to set the
values of various “state variables” that influence the behavior of the program; here it is used as a
way to silence unwanted output. See the Gretl Command Reference for more on set.
The eval command is useful when you want to look at the result of an expression without assigning
it to a variable; for example

eval 2+3*4

will print the number 14. This is most useful when running gretl interactively, like a calculator, but
it is usable in a hansl script for checking purposes, as in the following (rather silly) example:

scalar a = 1
scalar b = -1
# this ought to be 0
eval a+b

3.1 Manipulation of scalars

Algebraic operations work in the obvious way, with the classic algebraic operators having their
traditional precedence rules: the caret (^) is used for exponentiation. For example,

scalar phi = exp(-0.5 * (x-m)^2 / s2) / sqrt(2 * $pi * s2)

in which we assume that x, m and s2 are pre-existing scalars. The example above contains two
noteworthy points:

• The usage of the exp (exponential) and sqrt (square root) functions; it goes without saying
that hansl possesses a reasonably wide repertoire of such functions. See the Gretl Command
Reference for the complete list.
• The usage of $pi for the constant π . While user-specified identifiers must begin with a letter,
built-in identifiers for internal objects typically have a “dollar” prefix; these are known as
accessors (basically, read-only variables). Most accessors are defined in the context of an open
dataset (see part II), but some represent pre-defined constants, such as π . Again, see the Gretl
Command Reference for a comprehensive list.

Hansl does not possess a specific Boolean type, but scalars can be used for holding true/false
values. It follows that you can also use the logical operators and (&&), or (||), and not (!) with
scalars, as in the following example:

a = 1
b = 0
c = !(a && b)
Chapter 3. Hello, world! 9

In the example above, c will equal 1 (true), since (a && b) is false, and the exclamation mark is the
negation operator. Note that 0 evaluates to false, and anything else (not necessarily 1) evaluates to
true.
A few constructs are taken from the C language family: one is the postfix increment operator:

a = 5
b = a++
print a b

the second line is equivalent to b = a, followed by a++, which in turn is shorthand for a = a+1,
so running the code above will result in b containing 5 and a containing 6. Postfix subtraction is
also supported; prefix operators, however, are not supported. Another C borrowing is inflected
assignment, as in a += b, which is equivalent to a = a + b; several other similar operators are
available, such as -=, *= and more. See the Gretl Command Reference for details.
The internal representation for a missing value is NaN (“not a number”), as defined by the IEEE 754
floating point standard. This is what you get if you try to compute quantities like the square root
or the logarithm of a negative number. You can also set a value to “missing” directly using the
keyword NA. The complementary functions missing and ok can be used to determine whether a
scalar is NA. In the following example a value of zero is assigned to the variable named test:

scalar not_really = NA
scalar test = ok(not_really)

Note that you cannot test for equality to NA, as in

if x == NA ... # wrong!

because a missing value is taken as indeterminate and hence not equal to anything. This last
example, despite being wrong, illustrates a point worth noting: the test-for-equality operator in
hansl is the double equals sign, “==” (as opposed to plain “=” which indicates assignment).

3.2 Manipulation of strings

Most of the previous section applies, with obvious modifications, to strings: you may manipulate
strings via operators and/or functions. Hansl’s repertoire of functions for manipulating strings
offers all the standard capabilities one would expect, such as toupper, tolower, strlen, etc., plus
some more specialized ones. Again, see the Gretl Command Reference for a complete list.
In order to access part of a string, you may use the substr function,3 as in

string s = "endogenous"
string pet = substr(s, 3, 5)

which would result to assigning the value dog to the variable pet.
The following are useful operators for strings:

• the ~ operator, to join two or more strings, as in4

string s1 = "sweet"
string s2 = "Home, " ~ s1 ~ " home."
3 Actually,there is a cooler method, which uses the same syntax as matrix slicing (see chapter 4): substr(s, 3, 5) is
functionally equivalent to s[3:5].
4 On some national keyboards, you don’t have the tilde (˜) character. In gretl’s script editor, this can be obtained via its

Unicode representation: type Ctrl-Shift-U, followed by 7e.

Chapter 3. Hello, world! 10

• the closely related ~= operator, which acts as an inflected assignment operator (so a ~= "_ij"
is equivalent to a = a ~ "_ij");

• the offset operator +, which yields a substring of the preceding element, starting at the given
character offset. An empty string is returned if the offset is greater than the length of the
string in question.

A noteworthy point: strings may be (almost) arbitrarily long; moreover, they can contain special
characters such as line breaks and tabs. It is therefore possible to use hansl for performing rather
complex operations on text files by loading them into memory as a very long string and then
operating on that; interested readers should take a look at the readfile, getline, strsub and
regsub functions in the Gretl Command Reference.5
For creating complex strings, the most flexible tool is the sprintf function. Its usage is illustrated
in Chapter 5.

5 We are not claiming that hansl would be the tool of choice for text processing in general. Nonetheless the functions

mentioned here can be very useful for tasks such as pre-processing plain text data files that do not meet the requirements
for direct importation into gretl.
 Chapter 4

Matrices

Matrices are one- or two-dimensional arrays of double-precision floating-point numbers. Hansl

users who are accustomed to other matrix languages should note that multi-index objects are not
supported. Matrices have rows and columns, and that’s it.

4.1 Matrix indexing

Individual matrix elements are accessed through the [r,c] syntax, where indexing starts at 1. For
example, X[3,4] indicates the element of X on the third row, fourth column. For example,

matrix X = zeros(2,3)
X[2,1] = 4
print X

produces

X (2 x 3)

0 0 0
4 0 0

Here are some more advanced ways to access matrix elements:

1. In case the matrix has only one row (column), the column (row) specification can be omitted,
as in x[3].

2. Including the comma but omitting the row or column specification means “take them all”, as
in x[4,] (fourth row, all columns).

3. For square matrices, the special syntax x[diag] can be used to access the diagonal.

4. Consecutive rows or columns can be specified via the colon (:) character, as in x[,2:4]
(columns 2 to 4). But note that, unlike some other matrix languages, the syntax [m:n] is
illegal if m > n.

5. It is possible to use a vector to hold indices to a matrix. E.g. if e = [2, 3, 6], then X[,e]
contains the second, third and sixth columns of X.

Moreover, matrices can be empty (zero rows and columns).

In the example above, the matrix X was constructed using the function zeros(), whose meaning
should be obvious, but matrix elements can also be specified directly, as in

scalar a = 2*3
matrix A = { 1, 2, 3 ; 4, 5, a }

The matrix is defined by rows; the elements on each row are separated by commas and rows are
separated by semicolons. The whole expression must be wrapped in braces. Spaces within the
braces are not significant. The above expression defines a 2 × 3 matrix.

11
Chapter 4. Matrices 12

Note that each element should be a numerical value, the name of a scalar variable, or an expression
that evaluates to a scalar. In the example above the scalar a was first assigned a value and then
used in matrix construction. (Also note, in passing, that a and A are two separate identifiers, due
to case-sensitivity.)

4.2 Matrix operations

Matrix sum, difference and product are obtained via +, - and *, respectively. The prime operator
(’) can act as a unary operator, in which case it transposes the preceding matrix, or as a binary
operator, in which case it acts as in ordinary matrix algebra, multiplying the transpose of the first
matrix into the second one.1 Errors are flagged if conformability is a problem. For example:

matrix a = {11, 22 ; 33, 44} # a is square 2 x 2

matrix b = {1,2,3; 3,2,1} # b is 2 x 3

matrix c = a’ # c is the transpose of a

matrix d = a*b # d is a 2x3 matrix equal to a times b

matrix gina = b’d # valid: gina is 3x3

matrix lina = d + b # valid: lina is 2x3

/* -- these would generate errors if uncommented ----- */

# pina = a + b # sum non-conformability

# rina = d * b # product non-conformability

Other noteworthy matrix operators include ˆ (matrix power), ** (Kronecker product), and the “con-
catenation” operators, ~ (horizontal) and | (vertical). Readers are invited to try them out by running
the following code

matrix A = {2,1;0,1}
matrix B = {1,1;1,0}

matrix KP = A ** B
matrix PWR = A^3
matrix HC = A ~ B
matrix VC = A | B

print A B KP PWR HC VC

Note, in particular, that A3 = A · A · A, which is different from what you get by computing the cubes
of each element of A separately.
Hansl also supports matrix left- and right-“division”, via the \ and / operators, respectively. The
expression A\b solves Ax = b for the unknown x. A is assumed to be an m × n matrix with full
column rank. If A is square the method is LU decomposition. If m > n the QR decomposition is
used to find the least squares solution. In most cases, this is numerically more robust and more
efficient than inverting A explicitly.
Element-by-element operations are supported by the so-called “dot” operators, which are obtained
by putting a dot (“.”) before the corresponding operator. For example, the code

A = {1,2; 3,4}
B = {-1,0; 1,-1}
eval A * B
eval A .* B
1 In fact, in this case an optimized algorithm is used; you should always use a’a instead of a’*a for maximal precision

and performance.
Chapter 4. Matrices 13

produces

1 -2
1 -4

-1 0
3 -4

It’s easy to verify that the first operation performed is regular matrix multiplication A · B, whereas
the second one is the Hadamard (element-by-element) product A B. In fact, dot operators are
more general and powerful than shown in the example above; see the chapter on matrices in the
Gretl User’s Guide for details.
Dot and concatenation operators are less rigid than ordinary matrix operations in terms of con-
formability requirements: in most cases hansl will try to do “the obvious thing”. For example, a
common idiom in hansl is Y = X ./ w, where X is an n × k matrix and w is an n × 1 vector. The
result Y is an n × k matrix in which each row of X is divided by the corresponding element of w.
In proper matrix notation, this operation should be written as

Y = hwi−1 X,

where the h·i indicates a diagonal matrix. Translating literally the above expression would imply
creating a diagonal matrix out of w and then inverting it, which is computationally much more
expensive than using the dot operation. A detailed discussion is provided in the Gretl User’s Guide.
Hansl provides a reasonably comprehensive set of matrix functions, that is, functions that produce
and/or operate on matrices. For a full list, see the Gretl Command Reference, but a basic “survival
kit” is provided in Table 4.1. Moreover, most scalar functions, such as abs(), log() etc., will
operate on a matrix element-by-element.

Function(s) Purpose
rows(X), cols(X) return the number of rows and columns of X, respectively
zeros(r,c), ones(r,c) produce matrices with r rows and c columns, filled with zeros
and ones, respectively
mshape(X,r,c) rearrange the elements of X into a matrix with r rows and c
columns
I(n) identity matrix of size n
seq(a,b) generate a row vector containing integers from a to b
inv(A) invert, if possible, the matrix A
maxc(A), minc(A), meanc(A) return a row vector with the max, min, means of each column of
A, respectively
maxr(A), minr(A), meanr(A) return a column vector with the max, min, means of each row of
A, respectively
mnormal(r,c), muniform(r,c) generate r ×c matrices filled with standard Gaussian and uniform
pseudo-random numbers, respectively

Table 4.1: Essential set of hansl matrix functions

The following piece of code is meant to provide a concise example of all the features mentioned
above.

# example: OLS using matrices

# fix the sample size

scalar T = 256

# construct vector of coefficients by direct imputation

Chapter 4. Matrices 14

matrix beta = {1.5, 2.5, -0.5} # note: row vector

# construct the matrix of independent variables

matrix Z = mnormal(T, cols(beta)) # built-in functions

# now construct the dependent variable: note the

# usage of the "dot" and transpose operators

matrix y = {1.2} .+ Z*beta’ + mnormal(T, 1)

# now do estimation
matrix X = 1 ~ Z # concatenation operator
matrix beta_hat1 = inv(X’X) * (X’y) # OLS by hand
matrix beta_hat2 = mols(y, X) # via the built-in function
matrix beta_hat3 = X\y # via matrix division

print beta_hat1 beta_hat2 beta_hat3

4.3 Matrix pointers

Hansl uses the “by value” convention for passing parameters to functions. That is, when a variable
is passed to a function as an argument, what the function actually gets is a copy of the variable,
which means that the value of the variable at the caller level is not modified by anything that goes
on inside the function. But the use of pointers allows a function and its caller to cooperate such
that an outer variable can be modified by the function.
This mechanism is used by some built-in matrix functions to provide more than one “return” value.
The primary result is always provided by the return value proper but certain auxiliary values may be
retrieved via “pointerized” arguments; this usage is flagged by prepending the ampersand symbol,
“&”, to the name of the argument variable.
The eigensym function, which performs the eigen-analysis of symmetric matrices, is a case in
point. In the example below the first argument A represents the input data, that is, the matrix
whose analysis is required. This variable will not be modified in any way by the function call. The
primary result is the vector of eigenvalues of A, which is here assigned to the variable ev. The
(optional) second argument, &V (which may be read as “the address of V”), is used to retrieve the
right eigenvectors of A. A variable named in this way must be already declared, but it need not be
of the right dimensions to receive the result; it will be resized as needed.

matrix A = {1,2 ; 2,5}

matrix V
matrix ev = eigensym(A, &V)
print A ev V
 Chapter 5

Nice-looking output

5.1 Formatted output

A common occurrence when you’re writing a script — particularly when you intend for the script to
be used by others, and you’d like the output to be reasonably self-explanatory — is that you want to
output something along the following lines:

The coefficient on X is Y, with standard error Z

where X, Y and Z are placeholders for values not known at the time of writing the script; they will
be filled out as the values of variables or expressions when the script is run. Let’s say that at run
time the replacements in the sentence above should come from variables named vname (a string),
b (a scalar value) and se (also a scalar value), respectively.
Across the spectrum of programming languages there are basically two ways of arranging for this.
One way originates in the C language and goes under the name printf. In this approach we (a)
replace the generic placeholders X, Y and Z with more informative conversion specifiers, and (b)
append the variables (or expressions) that are to be stuck into the text, in order. Here’s the hansl
version:

printf "The coefficient on %s is %g, with standard error %g\n", vname, b, se

The value of vname replaces the conversion specifier “%s,” and the values of b and se replace the
two “%g” specifiers, left to right. In relation to hansl, here are the basic points you need to know:
“%s” pairs with a string argument, and “%g” pairs with a numeric argument.
The C-derived printf (either in the form of a function, or in the form of a command as shown
above) is present in most “serious” programming languages. It is extremely versatile, and in its
advanced forms affords the programmer fine control over the output.
In some scripting languages, however, printf is reckoned “too difficult” for non-specialist users.
In that case some sort of substitute is typically offered. We’re skeptical: “simplified” alternatives
to printf can be quite confusing, and if at some point you want fine control over the output, they
either do not support it, or support it only via some convoluted mechanism. A typical alternative
looks something like this (please note, display is not a hansl command, it’s just illustrative):

display "The coefficient on ", vname, "is ", b, ", with standard error ", se, "\n"

That is, you break the string into pieces and intersperse the names of the variables to be printed.
The requirement to provide conversion specifiers is replaced by a default automatic formatting of
the variables based on their type. By the same token, the command line becomes peppered with
mutiple commas and quotation marks. If this looks preferable to you, you are welcome to join one
of the gretl mailing lists and argue for its provision!
Anyway, to be a bit more precise about printf, its syntax goes like this:

printf format, arguments

The format is used to specify the precise way in which you want the arguments to be printed.

15
Chapter 5. Nice-looking output 16

The format string

In the general case the printf format must be an expression that evaluates to a string, but in
most cases will just be a string literal (an alphanumeric sequence surrounded by double quotes).
However, some character sequences in the format have a special meaning. As illustrated above,
those beginning with a percent sign (%) are interpreted as placeholders for the items contained in
the argument list. In addition, special characters such as the newline character are represented via
a combination beginning with a backslash (\).
For example,

printf "The square root of %d is (roughly) %6.4f.\n", 5, sqrt(5)

will print

The square root of 5 is (roughly) 2.2361.

Let’s see how:

• The first special sequence is %d: this indicates that we want an integer at that place in the
output; since it is the leftmost “percent” expression, it is matched to the first argument, that
is 5.

• The second special sequence is %6.4f, which stands for a decimal value with 4 digits after the
decimal separator1 and at least 6 digits wide; this will be matched to the second argument.
Note that arguments are separated by commas. Also note that the second argument is neither
a scalar constant nor a scalar variable, but an expression that evaluates to a scalar.

• The format string ends with the sequence \n, which inserts a newline.

The conversion specifiers in the square-root example are relatively fancy, but as we noted earlier
%g will work fine for almost all numerical values in hansl. So we could have used the simpler form:

printf "The square root of %g is (roughly) %g.\n", 5, sqrt(5)

The effect of %g is to print a number using up to 6 significant digits (but dropping trailing zeros); it
automatically switches to scientific notation if the number is very large or very small. So the result
here is

The square root of 5 is (roughly) 2.23607.

The escape sequences \n (newline), \t (tab), \v (vertical tab) and \\ (literal backslash) are recog-
nized. To print a literal percent sign, use %%.
Apart from those shown in the above example, recognized numeric formats are %e, %E, %f, %g, %G
and %x, in each case with the various modifiers available in C. The format %s should be used for
strings. As in C, numerical values that form part of the format (width and or precision) may be
given directly as numbers, as in %10.4f, or they may be given as variables. In the latter case, one
puts asterisks into the format string and supplies corresponding arguments in order. For example,

scalar width = 12
scalar precision = 6
printf "x = %*.*f\n", width, precision, x

If a matrix argument is given in association with a numeric format, the entire matrix is printed
using the specified format for each element. A few more examples are given in table 5.1.
1 The decimal separator is the dot in English, but may be different in other locales.
Chapter 5. Nice-looking output 17

Command effect
printf "%12.3f", $pi 3.142
printf "%12.7f", $pi 3.1415927
printf "%6s%12.5f%12.5f %d\n", "alpha",
3.5, 9.1, 3 alpha 3.50000 9.10000 3
printf "%6s%12.5f%12.5f\t%d\n", "beta",
1.2345, 1123.432, %11 beta 1.23450 1123.43200 11
printf "%d, %10d, %04d\n", 1,2,3 1, 2, 0003
printf "%6.0f (%5.2f%%)\n", 32, 11.232 32 (11.23%)

Table 5.1: Print format examples

Output to a string
A closely related effect can be achieved via the sprintf function: instead of being printed directly
the result is stored in a named string variable, as in

string G = sprintf("x = %*.*f\n", width, precision, x)

after which the variable G can be the object of further processing.

Output to a file
Hansl does not have a file or “stream” type as such, but the outfile command can be used to
divert output to a named text file. To start such redirection you must give the name of a file;
by default a new file is created or an existing one overwritten but the --append can be used to
append material to an existing file. Only one file can be opened in this way at any given time. The
redirection of output continues until the command end outfile is given; then output reverts to
the default stream.
Here’s an example of usage:

printf "One!\n"
outfile "myfile.txt"
printf "Two!\n"
end outfile
printf "Three!\n"
outfile "myfile.txt" --append
printf "Four!\n"
end outfile
printf "Five!\n"

After execution of the above the file myfile.txt will contain the lines

Two!
Four!

Three special variants on the above are available. If you give the keyword null in place of a real
filename along with the write option, the effect is to suppress all printed output until redirection
is ended. If either of the keywords stdout or stderr are given in place of a regular filename the
effect is to redirect output to standard output or standard error output, respectively.
This command also supports a --quiet option: its effect is to turn off the echoing of commands
and the printing of auxiliary messages while output is redirected. It is equivalent to doing

set echo off

set messages off
Chapter 5. Nice-looking output 18

before invoking outfile, except that when redirection is ended the prior values of the echo and
messages state variables are restored.

5.2 Graphics
The primary graphing command in hansl is gnuplot which, as the name suggests, in fact provides
an interface to the gnuplot program. It is used for plotting series in a dataset (see part II) or
columns in a matrix. For an account of this command (and some other more specialized ones, such
as boxplot and qqplot), see the Gretl Command Reference.
 Chapter 6

Structured data types

Hansl possesses two kinds of “structured data type”: associative arrays, called bundles and arrays
in the proper sense of the word. Loosely speaking, the main difference between the two is that
in a bundle you can pack together variables of different types, while arrays can hold one type of
variable only.

6.1 Bundles
Bundles are associative arrays, that is, generic containers for any assortment of hansl types (in-
cluding other bundles) in which each element is identified by a string. Python users call these
dictionaries; in C++ and Java, they are referred to as maps; they are known as hashes in Perl. We
call them bundles. Each item placed in the bundle is associated with a key which can used to retrieve
it subsequently.
To use a bundle you first either “declare” it, as in

bundle foo

or define an empty bundle using the null keyword:

bundle foo = null

These two formulations are basically equivalent, in that they both create an empty bundle. The
difference is that the second variant may be reused — if a bundle named foo already exists the
effect is to empty it—while the first may only be used once in a given gretl session; it is an error to
declare a variable that already exists.
To add an object to a bundle you assign to a compound left-hand value: the name of the bundle
followed by the key. The most common way to do this is to join the key to the bundle name with a
dot, as in

foo.matrix1 = m

which adds an object called m (presumably a matrix) to bundle foo under the key matrix1. The key
must satisfy the rules for a gretl variable name (31 characters maximum, starting with a letter and
composed of just letters, numbers or underscore)
An alternative way to achieve the same effect is to give the key as a quoted string literal enclosed
in square brackets, as in

foo["matrix1"] = m

When using the more elaborate syntax, keys do not have to be valid as variable names — for example,
they can include spaces—but they are still limited to 31 characters.
To get an item out of a bundle, again use the name of the bundle followed by the key, as in

matrix bm = foo.matrix1
# or using the long-hand notation
matrix m = foo["matrix1"]

19
Chapter 6. Structured data types 20

Note that the key identifying an object within a given bundle is necessarily unique. If you reuse an
existing key in a new assignment, the effect is to replace the object which was previously stored
under the given key. It is not required that the type of the replacement object is the same as that
of the original.
A quicker way, introduced in gretl 2017b, is to use the defbundle function, as in

bundle b = defbundle("s", "Sample string", "m", I(3))

in which every odd-numbered argument must evaluate to a string (key) and every even-numbered
argument must evaluate to an object of a type that can be included in a bundle.
Note that when you add an object to a bundle, what in fact happens is that the bundle acquires
a copy of the object. The external object retains its own identity and is unaffected if the bundled
object is replaced by another. Consider the following script fragment:

bundle foo
matrix m = I(3)
foo.mykey = m
scalar x = 20
foo.mykey = x

After the above commands are completed bundle foo does not contain a matrix under mykey, but
the original matrix m is still in good health.
To delete an object from a bundle use the delete command, with the bundle/key combination, as
in

delete foo.mykey
delete foo["quoted key"]

This destroys the object associated with the key and removes the key from the hash table.1
Besides adding, accessing, replacing and deleting individual items, the other operations that are
supported for bundles are union and printing. As regards union, if bundles b1 and b2 are defined
you can say

bundle b3 = b1 + b2

to create a new bundle that is the union of the two others. The algorithm is: create a new bundle
that is a copy of b1, then add any items from b2 whose keys are not already present in the new
bundle. (This means that bundle union is not necessarily commutative if the bundles have one or
more key strings in common.)
If b is a bundle and you say print b, you get a listing of the bundle’s keys along with the types of
the corresponding objects, as in

? print b
bundle b:
x (scalar)
mat (matrix)
inside (bundle)

Bundle usage
To illustrate the way a bundle can hold information, we will use the Ordinary Least Squares (OLS)
model as an example: the following code estimates an OLS regression and stores all the results in
a bundle.
1 Internally, gretl bundles in fact take the form of GLib hash table.
Chapter 6. Structured data types 21

/* assume y and X are given T x 1 and T x k matrices */

bundle my_model = null # initialization

my_model.T = rows(X) # sample size
my_model.k = cols(X) # number of regressors
matrix e # will hold the residuals
b = mols(y, X, &e) # perform OLS via native function
s2 = meanc(e.^2) # compute variance estimator
matrix V = s2 .* invpd(X’X) # compute covariance matrix

/* now store estimated quantities into the bundle */

my_model.betahat = b
my_model.s2 = s2
my_model.vcv = V
my_model.stderr = sqrt(diag(V))

The bundle so obtained is a container that can be used for all sort of purposes. For example, the
next code snippet illustrates how to use a bundle with the same structure as the one created above
to perform an out-of sample forecast. Imagine that k = 4 and the value of x for which we want to
forecast y is
x0 = [10 1 − 3 0.5]
The formulae for the forecast would then be

ŷf = x0 β̂
q
sf = σ̂ 2 + x0 V (β̂)x
CI = ŷf ± 1.96sf

where CI is the (approximate) 95 percent confidence interval. The above formulae translate into

x = { 10, 1, -3, 0.5 }

scalar ypred = x * my_model.betahat
scalar varpred = my_model.s2 + qform(x, my_model.vcv)
scalar sepred = sqrt(varpred)
matrix CI_95 = ypred + {-1, 1} .* (1.96*sepred)
print ypred CI_95

6.2 Arrays
A gretl array is a container which can hold zero or more objects of a certain type, indexed by
consecutive integers starting at 1. It is one-dimensional. This type is implemented by a quite
“generic” back-end. The types of object that can be put into arrays are strings, matrices, bundles
and lists; a given array can hold only one of these types.

Array operations
The following is, we believe, rather self-explanatory:

strings S1 = array(3)
matrices M = array(4)
strings S2 = defarray("fish", "chips")
S1[1] = ":)"
S1[3] = ":("
M[2] = mnormal(2,2)
print S1
eval inv(M[2])
Chapter 6. Structured data types 22

S = S1 + S2
print S

The array() takes an integer argument for the array size; the defarray() function takes a variable
number of arguments (one or more), each of which may be the name of a variable of the given type
or an expression which evaluates to an object of that type. The corresponding output is

Array of strings, length 3

[1] ":)"
[2] null
[3] ":("

0.52696 0.28883
-0.15332 -0.68140

Array of strings, length 5

[1] ":)"
[2] null
[3] ":("
[4] "fish"
[5] "chips"

In order to find the number of elements in an array, you can use the nelem() function.
 Chapter 7

Numerical methods

7.1 Numerical optimization

Many, if not most, cases in which an econometrician wants to use a programming language such
as hansl, rather than relying on pre-canned routines, involve some form of numerical optimization.
This could take the form of maximization of a likelihood or similar methods of inferential statistics.
Alternatively, optimization could be used in a more general and abstract way, for example to solve
portfolio choice or analogous resource allocation problems.
Since hansl is Turing-complete, in principle any numerical optimization technique could be pro-
grammed in hansl itself. Some such techniques, however, are included in hansl’s set of native
functions, in the interest of both simplicity of use and efficiency. These are geared towards the
most common kind of problem encountered in economics and econometrics, that is unconstrained
optimization of differentiable functions.
In this chapter, we will briefly review what hansl offers to solve generic problems of the kind

x̂ ≡ Argmax f (x; a),

x∈<k

where f (x; a) is a function of x, whose shape depends on a vector of parameters a. The objective
function f (·) is assumed to return a scalar real value. In most cases, it will be assumed it is also
continuous and differentiable, although this need not necessarily be the case. (Note that while
hansl’s built-in functions maximize the given objective function, minimization can be achieved
simply by flipping the sign of f (·).)
A special case of the above occurs when x is a vector of parameters and a represents “data”. In
these cases, the objective function is usually a (log-)likelihood and the problem is one of estimation.
For such cases hansl offers several special constructs, reviewed in section 12.2. Here we deal with
more generic problems; nevertheless, the differences are only in the hansl syntax involved: the
mathematical algorithms that gretl employs to solve the optimization problem are the same.
The reader is invited to read the “Numerical methods” chapter of the Gretl User’s Guide for a
comprehensive treatment. Here, we will only give a small example which should give an idea of
how things are done.

function scalar Himmelblau(matrix x)

/* extrema:
f(3.0, 2.0) = 0.0,
f(-2.805118, 3.131312) = 0.0,
f(-3.779310, -3.283186) = 0.0
f(3.584428, -1.848126) = 0.0
*/
scalar ret = (x[1]^2 + x[2] - 11)^2
return -(ret + (x[1] + x[2]^2 - 7)^2)
end function

# ----------------------------------------------------------------------

set max_verbose 1

23
Chapter 7. Numerical methods 24

matrix theta1 = { 0, 0 }
y1 = BFGSmax(theta1, "Himmelblau(theta1)")
matrix theta2 = { 0, -1 }
y2 = NRmax(theta2, "Himmelblau(theta2)")

print y1 y2 theta1 theta2

We use for illustration here a classic “nasty” function from the numerical optimization literature,
namely the Himmelblau function, which has four different minima; f (x, y) = (x 2 + y − 11)2 + (x +
y 2 − 7)2 . The example proceeds as follows.

1. First we define the function to optimize: it must return a scalar and have among its arguments
the vector to optimize. In this particular case that is its only argument, but there could have
been other ones if necessary. Since in this case we are solving for a minimum our definition
returns the negative of the Himmelblau function proper.

2. We next set max_verbose to 1. This is another example of the usage of the set command; its
meaning is “let me see how the iterations go” and it defaults to 0. By using the set command
with appropriate parameters, you control several features of the optimization process, such
as numerical tolerances, visualization of the iterations, and so forth.

3. Define θ1 = [0, 0] as the starting point.

4. Invoke the BFGSmax function; this will seek the maximum via the BFGS technique. Its base
syntax is BFGSmax(arg1, arg2), where arg1 is the vector contining the optimization variable
and arg2 is a string containing the invocation of the function to maximize. BFGS will try
several values of θ1 until the maximum is reached. On successful completion, the vector
theta1 will contain the final point. (Note: there’s much more to this. For details, be sure to
read the Gretl User’s Guideand the Gretl Command Reference.)

5. Then we tackle the same problem but with a different starting point and a different optimiza-
tion technique. We start from θ2 = [0, −1] and use Newton–Raphson instead of BFGS, calling
the NRmax() function instead if BFGSmax(). The syntax, however, is the same.

6. Finally we print the results.

Table 7.1 on page 27 contains a selected portion of the output. Note that the second run converges
to a different local optimum than the first one. This is a consequence of having initialized the
algorithm with a different starting point. In this example, numerical derivatives were used, but
you can supply analytically computed derivatives to both methods if you have a hansl function for
them; see the Gretl User’s Guide for more detail.
The optimization methods hansl puts at your disposal are:

• BFGS, via the BFGSmax() function. This is in most cases the best compromise between perfor-
mance and robustness. It assumes that the function to maximize is differentiable and will try
to approximate its curvature by clever use of the change in the gradient between iterations.
You can supply it with an analytically-computed gradient for speed and accuracy, but if you
don’t, the first derivatives will be computed numerically.

• Newton–Raphson, via the NRmax() function. Actually, the name is misleading. It should be
called something like “curvature-based”, since it relies on the iterations

xi+1 = −λi C(xi )−1 g(xi )

where g(x) is the gradient and C(xi ) is some measure of curvature of the function to op-
timize; if C(x) is the Hessian matrix, you get Newton–Raphson. Again, you can code your
own functions for g(·) and C(·), but if you don’t then numerical approximations to the gra-
dient and the Hessian will be used, respectively. Other popular optimization methods (such
Chapter 7. Numerical methods 25

as BHHH and the scoring algorithm) can be implemented by supplying to NRmax() the appro-
priate curvature matrix C(·). This method is very efficient when it works, but is rather fragile:
for example, if C(xi ) happens to be non-negative definite at some iteration convergence may
become problematic.

• Derivative-free methods: the only method that hansl offers presently is simulated annealing,
via the simann() function, but an implementiation of the Nelder–Mead algorithm (also known
as the “amoeba” method) should be just a matter of time. These methods work even when
the function to maximize has some form of disconinuity or is not everywhere differentiable;
however, they may be very slow and CPU-intensive.

7.2 Numerical differentiation

For numerical differentiation we have fdjac. For example:

set echo off

set messages off

function scalar beta(scalar x, scalar a, scalar b)

return x^(a-1) * (1-x)^(b-1)
end function

function scalar ad_beta(scalar x, scalar a, scalar b)

scalar g = beta(x, a-1, b-1)
f1 = (a-1) * (1-x)
f2 = (b-1) * x
return (f1 - f2) * g
end function

function scalar nd_beta(scalar x, scalar a, scalar b)

matrix mx = {x}
return fdjac(mx, beta(mx, a, b))
end function

a = 3.5
b = 2.5

loop for (x=0; x<=1; x+=0.1)

printf "x = %3.1f; beta(x) = %7.5f, ", x, beta(x, a, b)
A = ad_beta(x, a, b)
N = nd_beta(x, a, b)
printf "analytical der. = %8.5f, numerical der. = %8.5f\n", A, N
endloop

returns

x = 0.0; beta(x) = 0.00000, analytical der. = 0.00000, numerical der. = 0.00000

x = 0.1; beta(x) = 0.00270, analytical der. = 0.06300, numerical der. = 0.06300
x = 0.2; beta(x) = 0.01280, analytical der. = 0.13600, numerical der. = 0.13600
x = 0.3; beta(x) = 0.02887, analytical der. = 0.17872, numerical der. = 0.17872
x = 0.4; beta(x) = 0.04703, analytical der. = 0.17636, numerical der. = 0.17636
x = 0.5; beta(x) = 0.06250, analytical der. = 0.12500, numerical der. = 0.12500
x = 0.6; beta(x) = 0.07055, analytical der. = 0.02939, numerical der. = 0.02939
x = 0.7; beta(x) = 0.06736, analytical der. = -0.09623, numerical der. = -0.09623
x = 0.8; beta(x) = 0.05120, analytical der. = -0.22400, numerical der. = -0.22400
x = 0.9; beta(x) = 0.02430, analytical der. = -0.29700, numerical der. = -0.29700
x = 1.0; beta(x) = 0.00000, analytical der. = -0.00000, numerical der. = NA
Chapter 7. Numerical methods 26

Details on the algorithm used can be found in the Gretl Command Reference. Suffice it to say
here that you have a fdjac_quality setting that goes from 0 to 2. The default value is to 0,
which gives you forward-difference approximation: this is the fastest algorithm, but sometimes
may not be precise enough. The value of 1 gives you bilateral difference, while 2 uses Richardson
extrapolation. As you go up, you gain in accuracy, but the method becomes considerably more
CPU-intensive.
Chapter 7. Numerical methods 27

? matrix theta1 = { 0, 0 }
Replaced matrix theta1
? y1 = BFGSmax(theta1, "Himmelblau(11, theta1)")
Iteration 1: Criterion = -170.000000000
Parameters: 0.0000 0.0000
Gradients: 14.000 22.000 (norm 0.00e+00)

Iteration 2: Criterion = -128.264504038 (steplength = 0.04)

Parameters: 0.56000 0.88000
Gradients: 33.298 39.556 (norm 5.17e+00)

...

--- FINAL VALUES:

Criterion = -1.83015730011e-28 (steplength = 0.0016)
Parameters: 3.0000 2.0000
Gradients: 1.7231e-13 -3.7481e-13 (norm 7.96e-07)

Function evaluations: 39
Evaluations of gradient: 16
Replaced scalar y1 = -1.83016e-28
? matrix theta2 = { 0, -1 }
Replaced matrix theta2
? y2 = NRmax(theta2, "Himmelblau(11, theta2)")
Iteration 1: Criterion = -179.999876556 (steplength = 1)
Parameters: 1.0287e-05 -1.0000
Gradients: 12.000 2.8422e-06 (norm 7.95e-03)

Iteration 2: Criterion = -175.440691085 (steplength = 1)

Parameters: 0.25534 -1.0000
Gradients: 12.000 4.5475e-05 (norm 1.24e+00)

...

--- FINAL VALUES:

Criterion = -3.77420797114e-22 (steplength = 1)
Parameters: 3.5844 -1.8481
Gradients: -2.6649e-10 2.9536e-11 (norm 2.25e-05)

Gradient within tolerance (1e-07)

Replaced scalar y2 = -1.05814e-07
? print y1 y2 theta1 theta2

y1 = -1.8301573e-28

y2 = -1.0581385e-07

theta1 (1 x 2)

3 2

theta2 (1 x 2)

3.5844 -1.8481

Table 7.1: Output from maximization

 Chapter 8

Control flow

The primary means for controlling the flow of execution in a hansl script are the if statement
(conditional execution), the loop statement (repeated execution), the catch modifier (which enables
the trapping of errors that would otherwise halt execution), and the quit command (which forces
termination).

8.1 The if statement

Conditional execution in hansl uses the if keyword. Its fullest usage is as follows

if <condition>
...
elif <condition>
...
else
...
endif

Points to note:

• The <condition> can be any expression that evaluates to a scalar: 0 is interpreted as “false”,
non-zero is interpreted as “true”; NA generates an error.

• Following if, “then” is implicit; there is no then keyword as found in, e.g., Pascal or Basic.

• The elif and else clauses are optional: the minimal form is just if . . . endif.

• Conditional blocks of this sort can be nested up to a maximum of depth of 1024.

Example:

scalar x = 15

# --- simple if ----------------------------------

if x >= 100
printf "%g is more than two digits long\n", x
endif

# --- if with else -------------------------------

if x >= 0
printf "%g is non-negative\n", x
else
printf "%g is negative\n", x
endif

# --- multiple branches --------------------------

if missing(x)
printf "%g is missing\n", x
elif x < 0

28
Chapter 8. Control flow 29

printf "%g is negative\n", x

elif floor(x) == x
printf "%g is an integer\n", x
else
printf "%g is a positive number with a fractional part\n", x
endif

Note, from the example above, that the elif keyword can be repeated, making hansl’s if statement
a multi-way branch statement. There is no separate switch or case statement in hansl. With one
or more elifs, hansl will execute the first one for which the logical condition is satisfied and then
jump to endif.

+ Stata users, beware: hansl’s if statement is fundamentally different from Stata’s if option: the latter se-
lects a subsample of observations for some action, while the former is used to decide if a group of statements
should be executed or not; hansl’s if is what Stata calls “branching if”.

The ternary query operator

Besides use of if, the ternary query operator, ?:, can be used to perform conditional assignment
on a more “micro” level. This has the form

result = <condition> ? <true-value> : <false-value>

If <condition> evaluates as true (non-zero) then <true-value> is assigned to result, otherwise

result will contain <false-value>.1 This is obviously more compact than if . . . else . . . endif.
The following example replicates the abs function by hand:

scalar ax = x>=0 ? x : -x

Of course, in the above case it would have been much simpler to just write ax = abs(x). Consider,
however, the following case, which exploits the fact that the ternary operator can be nested:

scalar days = (m==2) ? 28 : maxr(m.={4,6,9,11}) ? 30 : 31

This example deserves a few comments. We want to compute the number of days in a month, coded
in the variable m. The value we assign to the scalar days comes from the following pathway.

1. First we check if the month is February (m==2); if so, we set days to 28 and we’re done.2

2. Otherwise, we compute a matrix of zeros and ones via the operation m.={4,6,9,11} (note
the use of the “dot” operator to perform an element-by element comparison — see section 4.2);
if m equals any of the elements in the vector, the corresponding element of the result will be
1, and 0 otherwise;

3. The maxr function gives the maximum of this vector, so we’re checking whether m is any one
of the four values corresponding to 30-day months.

4. Since the above evaluates to a scalar, we put the right value into days.

The ternary operator is more flexible than the ordinary if statement. With if, the <condition>
to be evaluated must always come down to a scalar, but the query operator just requires that the
condition is of “suitable” type in light of the types of the operands. So, for example, suppose you
have a square matrix A and you want to switch the sign of the negative elements of A on and above
its diagonal. You could use a loop,3 and write a piece of code such as
1 Some readers may find it helpful to note that the conditional assignment operator works in exactly the same way as
the =IF() function in spreadsheets.
2 OK, we’re ignoring leap years here.
3 The loop keyword is explained in detail in the next section.
Chapter 8. Control flow 30

matrix A = mnormal(4,4)
matrix B = A

loop r = 1 .. rows(A)
loop c = r .. cols(A)
if A[r,c] < 0
B[r,c] = -A[r,c]
endif
endloop
endloop

By using the ternary operator, you can achieve the same effect via a considerably shorter (and
faster) construct:

matrix A = mnormal(4,4)
matrix B = upper(A.<0) ? -A : A

+ At this point some readers may be thinking “Well, this may be as cool as you want, but it’s way too
complicated for me; I’ll just use the traditional if”. Of course, there’s nothing wrong with that, but in some
cases the ternary assignment operator can lead to substantially faster code, and it becomes surprisingly
natural when one gets used to it.

8.2 Loops
The basic hansl command for looping is (doh!) loop, and takes the form

loop <control-expression> <options>

...
endloop

In other words, the pair of statements loop and endloop enclose the statements to repeat. Of
course, loops can be nested. Several variants of the <control-expression> for a loop are sup-
ported, as follows:

1. unconditional loop

2. while loop

3. index loop

4. foreach loop

5. for loop.

These variants are briefly described below.

Unconditional loop
This is the simplest variant. It takes the form

loop <times>
...
endloop

where <times> is any expression that evaluates to a scalar, namely the required number of itera-
tions. This is only evaluated at the beginning of the loop, so the number of iterations cannot be
changed from within the loop itself. Example:
Chapter 8. Control flow 31

# triangular numbers
scalar n = 6
scalar count = 1
scalar x = 0
loop n
scalar x += count
count++
print x
endloop

yields

x = 1.0000000
x = 3.0000000
x = 6.0000000
x = 10.000000
x = 15.000000
x = 21.000000

Note the usage of the increment (count++) and of the inflected assignment (x += count) operators.

Index loop
The unconditional loop is used quite rarely, as in most cases it is useful to have a counter variable
(count in the previous example). This is easily accomplished via the index loop, whose syntax is

loop <counter>=<min>..<max>
...
endloop

The limits <min> and <max> must evaluate to scalars; they are automatically turned into integers if
they have a fractional part. The <counter> variable is started at <min> and incremented by 1 on
each iteration until it equals <max>.
The counter is “read-only” inside the loop. You can access either its numerical value through the
scalar i or use the accessor $i, which will perform string substitution: inside the loop, the hansl
interpreter will substitute for the expression $i the string representation of the current value of
the index variable. An example should made this clearer: the following input

scalar a_1 = 57
scalar a_2 = 85
scalar a_3 = 13

loop i=1..3
print i a_$i
endloop

has for output

i = 1.0000000
a_1 = 57.000000
i = 2.0000000
a_2 = 85.000000
i = 3.0000000
a_3 = 13.000000

In the example above, at the first iteration the value of i is 1, so the interpreter expands the
expression a_$i to a_1, finds that a scalar by that name exists, and prints it. The same happens
through the rest of the iterations. If one of the automatically constructed identifiers had not been
defined, execution would have stopped with an error.
Chapter 8. Control flow 32

While loop
Here you have

loop while <condition>

...
endloop

where <condition> should evaluate to a scalar, which is re-evaluated at each iteration. Looping
stops as soon as <condition> becomes false (0). If <condition> becomes NA, an error is flagged
and execution stops. By default, while loops cannot exceed 100,000 iterations. This is intended as
a safeguard against potentially infinite loops. This setting can be overridden if necessary by setting
the loop_maxiter state variable to a different value.

Foreach loop
In this case the syntax is

loop foreach <counter> <catalogue>

...
endloop

where <catalogue> can be either a collection of space-separated strings, or a variable of type list
(see section 11.2). The counter variable automatically takes on the numerical values 1, 2, 3, and so
on as execution proceeds, but its string value (accessed by prepending a dollar sign) shadows the
names of the series in the list or the space-separated strings; this sort of loop is designed for string
substitution.
Here is an example in which the <catalogue> is a collection of names of functions that return a
scalar value when given a scalar argument.

scalar x = 1
loop foreach f sqrt exp ln
scalar y = $f(x)
print y
endloop

This will produce

y = 1.0000000
y = 2.7182818
y = 0.0000000

For loop
The final form of loop control emulates the for statement in the C programming language. The
syntax is loop for, followed by three component expressions, separated by semicolons and sur-
rounded by parentheses, that is

loop for (<init>; <cont>; <modifier>)

...
endloop

The three components are as follows:

1. Initialization (<init>): this must be an assignment statement, evaluated at the start of the
loop.
Chapter 8. Control flow 33

2. Continuation condition (<cont>): this is evaluated at the top of each iteration (including the
first). If the expression evaluates as true (non-zero), iteration continues, otherwise it stops.
3. Modifier (<modifier>): an expression which modifies the value of some variable. This is
evaluated prior to checking the continuation condition, on each iteration after the first.

Here’s an example, in which we find the square root of a number by successive approximations:

# find the square root of x iteratively via Newton’s method

scalar x = 256
d = 1
loop for (y=(x+1)/2; abs(d) > 1.0e-7; y -= d/(2*y))
d = y*y - x
printf "y = %15.10f, d = %g\n", y, d
endloop

printf "sqrt(%g) = %g\n", x, y

Running the example gives

y = 128.5000000000, d = 16256.3
y = 65.2461089494, d = 4001.05
y = 34.5848572866, d = 940.112
y = 20.9934703720, d = 184.726
y = 16.5938690915, d = 19.3565
y = 16.0106268314, d = 0.340172
y = 16.0000035267, d = 0.000112855
y = 16.0000000000, d = 1.23919e-11

Number of iterations: 8

sqrt(256) = 16

Be aware of the limited precision of floating-point arithmetic. For example, the code snippet below
will iterate forever on most platforms because x will never equal exactly 0.01, even though it might
seem that it should.

loop for (x=1; x!=0.01; x=x*0.1)

printf "x = .18g\n", x
endloop

However, if you replace the condition x!=0.01 with x>=0.01, the code will run as (probably) in-
tended.

Loop options
Two options can be given to the loop statement. One is --verbose. This has simply the effect of
printing extra output to trace progress of the loop; it has no other effect and the semantics of the
loop contents remain unchanged.
The --progressive option is mostly used as a quick and efficient way to set up simulation studies.
When this option is given, a few commands (notably print and store) are given a special, ad hoc
meaning. Please refer to the Gretl User’s Guide for more information.

Breaking out of loops

The break command makes it possible to break out of a loop if necessary.4 Note that if you nest
loops, break in the innermost loop will interrupt that loop only and not the outer ones. Here is
4 Hansl does not provide any equivalent to the C continue statement.
Chapter 8. Control flow 34

an example in which we use the while variant of the loop statement to perform calculation of the
square root in a manner similar to the example above, using break to jump out of the loop when
the job is done.

scalar x = 256
scalar y = 1
loop while 1
d = y*y - x
if abs(d) < 1.0e-7
break
else
y -= d/(2*y)
printf "y = %15.10f, d = %g\n", y, d
endif
endloop

printf "sqrt(%g) = %g\n", x, y

8.3 The catch modifier

Hansl offers a rudimentary form of exception handling via the catch keyword. This is not a com-
mand in its own right but can be used as a prefix to most regular commands: the effect is to
prevent termination of a script if an error occurs in executing the command. If an error does occur,
this is registered in an internal error code which can be accessed as $error (a zero value indicat-
ing success). The value of $error should always be checked immediately after using catch, and
appropriate action taken if the command failed. Here is a simple example:

matrix a = floor(2*muniform(2,2))
catch ai = inv(a)
scalar err = $error
if err
printf "The matrix\n%6.0f\nis singular!\n", a
else
print ai
endif

Note that the catch keyword cannot be used before if, elif or endif. In addition, it should not be
used on calls to user-defined functions; it is intended for use only with gretl commands and calls to
“built-in” functions or operators. Suppose you’re writing a function package which includes some
subsidiary functionality which may fail under certain conditions, and you want to prevent such
failure from aborting execution. In that case you should use catch within the particular function
in question, and if an error condition is detected, signal this to the caller by returning a suitable
“invalid” value—say, NA (for a function that returns a scalar) or an empty matrix. For example:

function scalar may_fail (matrix *m)

catch scalar x = ... # call to built-in procedure
if $error
x = NA
endif
return x
end function

function scalar caller (...)

matrix m = ... # whatever
scalar x = may_fail(&m)
if na(x)
print "Couldn’t calculate x"
Chapter 8. Control flow 35

else
printf "Calculated x = %g\n", x
endif
end function

What you should not do here is apply catch to may_fail()

function scalar caller (...)

matrix m = ... # whatever
catch scalar x = may_fail(&m) # No, don’t do this!
...
end function

as this is likely to leave gretl in a confused state.

8.4 The quit statement

When the quit statement is encountered in a hansl script, execution stops. If the command-line
program gretlcli is running in batch mode, control returns to the operating system; if gretl is
running in interactive mode, gretl will wait for interactive input.
The quit command is rarely used in scripts since execution automatically stops when script input
is exhausted, but it could be used in conjunction with catch. A script author could arrange matters
so that on encountering a certain error condition an appropriate message is printed and the script
is halted. Another use for quit is in program development: if you want to inspect the output of an
initial portion of a complex script, the most convenient solution may to insert a temporary “quit”
at a suitable point.
 Chapter 9

User-written functions

Hansl natively provides a reasonably wide array of pre-defined functions for manipulating variables
of all kinds; the previous chapters contain several examples. However, it is also possible to extend
hansl’s native capabilities by defining additional functions.
Here’s what a user-defined function looks like:

function type funcname(parameters)

function body
end function

The opening line of a function definition contains these elements, in strict order:

1. The keyword function.

2. type, which states the type of value returned by the function, if any. This must be one of
void (if the function does not return anything), scalar, series, matrix, list, string or
bundle.

3. funcname, the unique identifier for the function. Function names have a maximum length
of 31 characters; they must start with a letter and can contain only letters, numerals and the
underscore character. They cannot coincide with the names of native commands or functions.

4. The function’s parameters, in the form of a comma-separated list enclosed in parentheses.1

Note: parameters are the only way hansl function can receive anything from “the outside”. In
hansl there are no global variables.

Function parameters can be of any of the types shown below.

Type Description
bool scalar variable acting as a Boolean switch
int scalar variable acting as an integer
scalar scalar variable
series data series (see section 11.1)
list named list of series (see section 11.2)
matrix matrix or vector
string string variable or string literal
bundle all-purpose container
matrices array of matrices
bundles array of bundles
strings array of strings
1 If you have to pass many parameters, you might want to consider wrapping them into a bundle to avoid syntax

cluttering.

36
Chapter 9. User-written functions 37

Each element in the listing of parameter must include two terms: a type specifier, and the name by
which the parameter shall be known within the function.
The function body contains (almost) arbitrary hansl code, which should compute the return value,
that is the value the function is supposed to yield. Any variable declared inside the function is
local, so it will cease to exist when the function ends.
The return command is used to stop execution of the code inside the function and deliver its
result to the calling code. This typically happens at the end of the function body, but doesn’t have
to. The function definition must end with the expression end function, on a line of its own.

+ Beware: unlike some other languages (e.g. Matlab or GAUSS), you cannot directly return multiple outputs
from a function. However, you can return a bundle and stuff it with as many objects as you want.
In order to get a feel for how functions work in practice, here’s a simple example:

function scalar quasi_log (scalar x)

/* popular approximation to the natural logarithm
via Padé polynomials
*/
if x < 0
scalar ret = NA
else
scalar ret = 2*(x-1)/(x+1)
endif
return ret
end function

loop for (x=0.5; x<2; x+=0.1)

printf "x = %4.2f; ln(x) = %g, approx = %g\n", x, ln(x), quasi_log(x)
endloop

The code above computes the rational function

x−1
f (x) = 2 · ,
x+1
which provides a decent approximation to the natural logarithm in the neighborhood of 1. Some
comments on the code:

1. Since the function is meant to return a scalar, we put the keyword scalar after the intial
function.

2. In this case the parameter list has only one element: it is named x and is specified to be a
scalar.

3. On the next line the function definition begins; the body includes a comment and an if block.

4. The function ends by returning the computed value, ret.

5. The lines below the function definition give a simple example of usage. Note that in the
printf command, the two functions ln() and quasi_log() are indistinguishable from a
purely syntactic viewpoint, although the former is native and the latter is user-defined.

In ambitious uses of hansl you may end up writing several functions, some of which may be quite
long. In order to avoid cluttering your script with function definitions, hansl provides the include
command: you can put your function definitions in a separate file (or set of files) and read them in
as needed. For example, suppose you saved the definition of quasi_log() in a separate file called
quasilog_def.inp: the code above could then be written more compactly as
Chapter 9. User-written functions 38

include quasilog_def.inp

loop for (x=0.5; x<2; x+=0.1)

printf "x = %4.2f; ln(x) = %g, approx = %g\n", x, ln(x), quasi_log(x)
endloop

Moreover, include commands can be nested.

9.1 Parameter passing and return values

In hansl, parameters are by default passed by value, so what is used inside the function is a copy
of the original argument. You may modify it, but you’ll be just modifying the copy. The following
example should make this point clear:

function void f(scalar x)

x = x*2
print x
end function

scalar x = 3
f(x)
print x

Running the above code yields

x = 6.0000000
x = 3.0000000

The first print statement is executed inside the function, and the displayed value is 6 because the
input x is doubled; however, what really gets doubled is simply a copy of the original x: this is
demonstrated by the second print statement. If you want a function to modify its arguments, you
must use pointers.

Pointers
Each of the type-specifiers, with the exception of list, may be modified by prepending an asterisk
to the associated parameter name, as in

function scalar myfunc (matrix *y)

This indicates that the required argument is not a plain matrix but rather a pointer-to-matrix, or in
other words the memory address at which the variable is stored.
This can seem a bit mysterious to people unfamiliar with the C programming language, so allow
us to explain how pointers work by analogy. Suppose you set up a barber shop. Ideally, your
customers would walk into your shop, sit on a chair and have their hair trimmed or their beard
shaved. However, local regulations forbid you to modify anything coming in through your shop
door. Of course, you wouldn’t do much business if people must leave your shop with their hair
untouched. Nevertheless, you have a simple way to get around this limitation: your customers can
come to your shop, tell you their home address and walk out. Then, nobody stops you from going
to their place and exercising your fine profession. You’re OK with the law, because no modification
of anything took place inside your shop.
While our imaginary restriction on the barber seems arbitrary, the analogous restriction in a pro-
gramming context is not: it prevents functions from having unpredictable side effects. (You might
be upset if it turned out that your person was modified after visiting the grocery store!)
Chapter 9. User-written functions 39

In hansl (unlike C) you don’t have to take any special care within the function to distinguish the
variable from its address,2 you just use the variable’s name. In order to supply the address of a
variable when you invoke the function, you use the ampersand (&) operator.
An example should make things clearer. The following code

function void swap(scalar *a, scalar *b)

scalar tmp = a
a = b
b = tmp
end function

scalar x = 0
scalar y = 1000000
swap(&x, &y)
print x y

gives the output

x = 1000000.0
y = 0.0000000

So x and y have in fact been swapped. How?

First you have the function definition, in which the arguments are pointers to scalars. Inside the
function body, the distinction is moot, as a is taken to mean “the scalar that you’ll find at the
address given by the first argument” (and likewise for b). The rest of the function simply swaps a
and b by means of a local temporary variable.
Outside the function, we first initialize the two scalars x to 0 and y to a big number. When the
function is called, it is given as arguments &a and &b, which hansl identifies as “the address of” the
two scalars a and b, respectively.
Writing a function with pointer arguments has two main consequences: first, as we just saw, it
makes it possible to modify the function arguments. Second, it avoids the computational cost of
having to allocate memory for a copy of the arguments and performing the copy operation; such
cost is proportional to the size of the argument. Hence, for matrix arguments, this is a nice way to
write faster functions, as producing a copy of a large matrix can be quite time-consuming.3

Advanced parameter passing and optional arguments

The parameters to a hansl function can be also specified in more sophisticated ways than outlined
above. There are three additional features worth mentioning:

1. A descriptive string can be attached to each parameter for GUI usage.

2. For some parameter types, there is a special syntax construct for ensuring that its value is
bounded; for example, you can stipulate a scalar argument to be positive, or constrained
within a pre-specified range.

3. Some of the arguments can be made optional.

A thorough discussion is too long to fit in this document, and the interested reader should refer
to the “User-defined functions” chapter of the the Gretl User’s Guide. Here we’ll just show you a
simple, and hopefully self-explanatory, example which combines features 2 and 3. Suppose you
have a function for producing smileys, defined as
2 In C, this would be called dereferencing the pointer. The distinction is not required in hansl because there is no

equivalent to operating on the supplied address itself, as in C.

3 However, the const qualifier achieves the same effect. See the Gretl User’s Guide for further details.
Chapter 9. User-written functions 40

function void smileys(int times[0::1], bool frown[0])

if frown
string s = ":-("
else
string s = ":-)"
endif

loop times
printf "%s ", s
endloop

printf "\n"
end function

Then, running

smileys()
smileys(2, 1)
smileys(4)

produces

:-)
:-( :-(
:-) :-) :-) :-)

9.2 Recursion
Hansl functions can be recursive; what follows is the obligatory factorial example:

function scalar factorial(scalar n)

if (n<0) || (n>floor(n))
# filter out everything that isn’t a
# non-negative integer
return NA
elif n==0
return 1
else
return n * factorial(n-1)
endif
end function

loop i=0..6
printf "%d! = %d\n", i, factorial(i)
endloop

Note: this is fun, but in practice, you’ll be much better off using the pre-cooked gamma function
(or, better still, its logarithm).
 Part II

With a dataset

41
 Chapter 10

What is a dataset?

A dataset is a memory area designed to hold the data you want to work on, if any. It may be
thought of a big global variable, containing a (possibly huge) matrix of data and a hefty collection
of metadata.
R users may think that a dataset is similar to what you get when you attach a data frame in R. Not
really: in hansl, you cannot have more than one dataset open at the same time. That’s why we talk
about the dataset.
When a dataset is present in memory (that is, “open”), a number of objects become available for
your hansl script in a transparent and convenient way. Of course, the data themselves: the columns
of the dataset matrix are called series, which will be described in section 11.1; sometimes, you will
want to organize one or more series in a list (section 11.2). Additionally, you have the possibility of
using, as read-only global variables, some scalars or matrices, such as the number of observations,
the number of variables, the nature of your dataset (cross-sectional, time series or panel), and so
on. These are called accessors, and will be discussed in section 10.5.
You can open a dataset by reading data from a disk file, via the open command, or by creating one
from scratch.

10.1 Creating a dataset from scratch

The primary commands in this context are nulldata and setobs. For example:

set echo off

set messages off

set seed 443322 # initialize the random number generator

nulldata 240 # stipulate how long your series will be
setobs 12 1995:1 # define as monthly dataset, starting Jan 1995

For more details see the Gretl User’s Guide, and the Gretl Command Reference for the nulldata
and setobs commands. The only important thing to say at this point, however, is that you can
resize your dataset and/or change some of its characteristics, such as its periodicity, at nearly any
point inside your script if necessary.
Once your dataset is in place, you can start populating it with series, either by reading them from
files or by generating them via appropriate commands and functions.

10.2 Reading a dataset from a file

The primary commands here are open, append and join.
The open command is what you’ll want to use in most cases. It handles transparently a wide variety
of formats (native, CSV, spreadsheet, data files produced by other packages such as Stata, Eviews,
SPSS and SAS) and also takes care of setting up the dataset for you automatically.

open mydata.gdt # native format

open yourdata.dta # Stata format
open theirdata.xls # Excel format

42
Chapter 10. What is a dataset? 43

The open command can also be used to read stuff off the Internet, by using a URL instead of a
filename, as in

open http://someserver.com/somedata.csv

The Gretl User’s Guide describes the requirements on plain text data files of the “CSV” type for
direct importation by gretl. It also describes gretl’s native data formats (XML-based and binary).
The append and join commands can be used to add further series from file to a previously opened
dataset. The join command is extremely flexible and has a chapter to itself in the Gretl User’s
Guide.

10.3 Saving datasets

The store command is used to write the current dataset (or a subset) out to file. Besides writing
in gretl’s native formats, store can also be used to export data as CSV or in the format of R. Series
can be written out as matrices using the mwrite function. If you have special requirements that are
not met by store or mwrite it is possible to use outfile plus printf (see chapter 5) to gain full
control over the way data are saved.

10.4 The smpl command

Once you have opened a dataset somehow, the smpl command allows you to discard observations
selectively, so that your series will contain only the observations you want (automatically changing
the dimension of the dataset in the process). See chapter 4 in the Gretl User’s Guide for further
information.1
There are basically three variants to the smpl command:

1. Selecting a contiguous subset of observations: this will be mostly useful with time-series
datasets. For example:
smpl 4 122 # select observations for 4 to 122
smpl 1984:1 2008:4 # the so-called "Great Moderation" period
smpl 2008-01-01 ; # form January 1st, 2008 onwards

2. Selecting observations on the basis of some criterion: this is typically what you want with
cross-sectional datasets. Example:

smpl male == 1 --restrict # males only

smpl male == 1 && age < 30 --restrict # just the young guys
smpl employed --dummy # via a dummy variable

Note that, in this context, restrictions go “on top of” previous ones. In order to start from
scratch, you either reset the full sample via smpl full or use the --replace option along
with --restrict.

3. Restricting the active dataset to some observations so that a certain effect is achieved au-
tomatically: for example, drawing a random subsample, or ensuring that all rows that have
missing observations are automatically excluded. This is achieved via the --no-missing,
--contiguous, and --random options.

In the context of panel datasets, some extra qualifications have to be made; see the Gretl User’s
Guide.
1 Users with a Stata background may find the hansl way of doing things a little disconcerting at first. In hansl, you first

restrict your sample through the smpl command, which applies until further notice, then you do what you have to. There
is no equivalent to Stata’s if clause to commands.
Chapter 10. What is a dataset? 44

10.5 Dataset accessors

Several characteristics of the current dataset can be determined by reference to built-in accessor
(“dollar”) variables. The main ones, which all return scalar values, are shown in Table 10.1.

Accessor Value returned

$datatype Coding for the type of dataset: 0 = no data; 1 = cross-sectional (un-
dated); 2 = time-series; 3 = panel
$nobs The number of observations in the current sample range
$nvars The number of series (including the constant)
$pd The data frequency (1 for cross-sectional, 4 for quarterly, and so on)
$t1 1-based index of the first observation in the current sample
$t2 1-based index of the last observation in the current sample

Table 10.1: The principal dataset accessors

In addition there are a few more specialized accessors: $obsdate, $obsmajor, $obsminor, $obsmicro
and $unit. These are specific to time-series and/or panel data, and they all return series. See the
Gretl Command Reference for details.
 Chapter 11

Series and lists

Scalars, matrices and strings can be used in a hansl script at any point; series and lists, on the other
hand, are inherently tied to a dataset and therefore can be used only when a dataset is currently
open.

11.1 The series type

Series are just what any applied economist would call “variables”, that is, repeated observations of
a given quantity; a dataset is an ordered array of series, complemented by additional information,
such as the nature of the data (time-series, cross-section or panel), descriptive labels for the series
and/or the observations, source information and so on. Series are the basic data type on which
gretl’s built-in estimation commands depend.
The series belonging to a dataset are named via standard hansl identifiers (strings of maximum
length 31 characters as described above). In the context of commands that take series as arguments,
series may be referenced either by name or by ID number, that is, the index of the series within
the dataset. Position 0 in a dataset is always taken by the automatic “variable” known as const,
which is just a column of 1s. The IDs of the actual data series can be displayed via the varlist
command. (But note that in function calls, as opposed to commands, series must be referred to by
name.) A detailed description of how a dataset works can be found in chapter 4 of the Gretl User’s
Guide.
Some basic rules regarding series follow:

• If lngdp belongs to a time series or panel dataset, then the syntax lngdp(-1) yields its first
lag, and lngdp(+1) its first lead.

• To access individual elements of a series, you use square brackets enclosing

– the progressive (1-based) number of the observation you want, as in lngdp[15], or

– the corresponding date code in the case of time-series data, as in lngdp[2008:4] (for
the 4th quarter of 2008), or
– the corresponding observation marker string, if the dataset contains any, as in GDP["USA"].

The rules for assigning values to series are just the same as for other objects, so the following
examples should be self-explanatory:

series k = 3 # implicit conversion from scalar; a constant series

series x = normal() # pseudo-rv via a built-in function
series s = a/b # element-by-element operation on existing series

series movavg = 0.5*(x + x(-1)) # using lags

series y[2012:4] = x[2011:2] # using individual data points
series x2000 = 100*x/x[2000:1] # constructing an index

+ In hansl, you don’t have separate commands for creating series and modifying them. Other popular
packages make this distinction, but we still struggle to understand why this is supposed to be useful.

45
Chapter 11. Series and lists 46

Converting series to or from matrices

The reason why hansl provides a specific series type, distinct from the matrix type, is historical.
However, is also a very convenient feature. Operations that are typically performed on series in
applied work can be awkward to implement using “raw” matrices — for example, the computation
of leads and lags, or regular and seasonal differences; the treatment of missing values; the addition
of descriptive labels, and so on.
Anyway, it is straightforward to convert data in either direction between the series and matrix
types.

• To turn series into matrices, you use the curly braces syntax, as in
matrix MACRO = {outputgap, unemp, infl}

where you can also use lists; the number of rows of the resulting matrix will depend on your
currently selected sample.
• To turn matrices into series, you can just use matrix columns, as in
series y = my_matrix[,4]

But note that this will work only if the number of rows in my_matrix matches the length of
the dataset (or the currently selected sample range).

Also note that the lincomb and filter functions are quite useful for creating and manipulating
series in complex ways without having to convert the data to matrix form (which could be compu-
tationally costly with large datasets).

The ternary operator with series

Consider this assignment:

worker_income = employed ? income : 0

Here we assume that employed is a dummy series coding for employee status. Its value will be
tested for each observation in the current sample range and the value assigned to worker_income
at that observation will be determined accordingly. It is therefore equivalent to the following much
more verbose formulation (where $t1 and $t2 are accessors for the start and end of the sample
range):

series worker_income
loop i=$t1..$t2
if employed[i]
worker_income[i] = income[i]
else
worker_income[i] = 0
endif
endloop

11.2 The list type

In hansl parlance, a list is an array of integers, representing the ID numbers of a set (in a loose
sense of the word) of series. For this reason, the most common operations you perform on lists are
set operations such as addition or deletion of members, union, intersection and so on. Unlike sets,
however, hansl lists are ordered, so individual list members can be accessed via the [] syntax, as
in X[3].
There are several ways to assign values to a list. The most basic sort of expression that works in
this context is a space-separated list of series, given either by name or by ID number. For example,
Chapter 11. Series and lists 47

list xlist = 1 2 3 4
list reglist = income price

An empty list is obtained by using the keyword null, as in

list W = null

or simply by bare declaration. Some more special forms (for example, using wildcards) are de-
scribed in the Gretl User’s Guide.
The main idea is to use lists to group, under one identifier, one or more series that logically belong
together somehow (for example, as explanatory variables in a model). So, for example,

list xlist = x1 x2 x3 x4
ols y 0 xlist

is an idiomatic way of specifying the OLS regression that could also be written as

ols y 0 x1 x2 x3 x4

Note that we used here the convention, mentioned in section 11.1, by which a series can be identi-
fied by its ID number when used as an argument to a command, typing 0 instead of const.
Lists can be concatenated, as in as in list L3 = L1 L2 (where L1 and L2 are names of existing
lists). This will not necessarily do what you want, however, since the resulting list may contain
duplicates. It’s more common to use the following set operations:

Operator Meaning
|| Union
&& Intersection
- Set difference

So for example, if L1 and L2 are existing lists, after running the following code snippet

list UL = L1 || L2
list IL = L1 && L2
list DL = L1 - L2

the list UL will contain all the members of L1, plus any members of L2 that are not already in L1; IL
will contain all the elements that are present in both L1 and L2 and DL will contain all the elements
of L1 that are not present in L2.
To append or prepend variables to an existing list, we can make use of the fact that a named list
stands in for a “longhand” list. For example, assuming that a list xlist is already defined (possibly
as null), we can do

list xlist = xlist 5 6 7

xlist = 9 10 xlist 11 12

Another option for appending terms to, or dropping terms from, an existing list is to use += or -=,
respectively, as in

xlist += cpi
zlist -= cpi

A nice example of the above is provided by a common idiom: you may see in hansl scripts some-
thing like
Chapter 11. Series and lists 48

list C -= const
list C = const C

which ensures that the series const is included (exactly once) in the list C, and comes first.

Converting lists to or from matrices

The idea of converting from a list, as defined above, to a matrix may be taken in either of two
ways. You may want to turn a list into a matrix (vector) by filling the latter with the ID numbers
contained in the former, or rather to create a matrix whose columns contain the series to which the
ID numbers refer. Both interpretations are legitimate (and potentially useful in different contexts)
so hansl lets you go either way.
If you assign a list to a matrix, as in

list L = moo foo boo zoo

matrix A = L

the matrix A will contain the ID numbers of the four series as a row vector. This operation goes
both ways, so the statement

list C = seq(7,10)

is perfectly valid (provided, of course, that you have at least 10 series in the currently open dataset).
If instead you want to create a data matrix from the series which belong to a given list, you have to
enclose the list name in curly brackets, as in

matrix X = {L}

The foreach loop variant with lists

Lists can be used as the “catalogue” in the foreach variant of the loop construct (see section 8.2).
This is especially handy when you have to perform some operation on multiple series. For example,
the following syntax can be used to calculate and print the mean of each of several series:

list X = age income experience

loop foreach i X
printf "mean($i) = %g\n", mean($i)
endloop
 Chapter 12

Estimation methods

You can, of course, estimate econometric models via hansl without having a dataset (in the sense
in which we’re using that term here) in place — just as you might in Matlab, for instance. You’ll
need data, but these can be loaded in matrix form (see the mread function in the Gretl Command
Reference), or generated artificially via functions such as mnormal or muniform. You can roll your
own estimator using hansl’s linear algebra primitives, and you also have access to more specialized
functions such as mols (see section 4.2) and mrls (restricted least squares) if you need them.
However, unless you need to use an estimation method which is not currently supported by gretl,
or have a strong desire to reinvent the wheel, you will probably want to make use of the built-
in estimation commands available in hansl. These commands are series-oriented and therefore
require a dataset. They fall into two main categories: “canned” procedures, and generic tools that
can be used to estimate a wide variety of models based on common principles.

12.1 Canned estimation procedures

“Canned” maybe doesn’t sound very appetizing but it’s the term that’s commonly used. Basically it
means two things, neither of them in fact unappetizing.

• The user is presented with a fairly simple interface. A few inputs must be specified, and
perhaps a few options selected, then the heavy lifting is done within the gretl library. Full
results are printed (parameter estimates plus numerous auxiliary statistics).

• The algorithm is written in C, by experienced coders. It is therefore faster (possibly much

faster) than an implementation in an interpreted language such as hansl.

Most such procedures share the syntax

commandname parameters options

where parameters usually takes the form of a listing of series: the dependent variable followed by
the regressors.
The line-up of procedures can be crudely categorized as follows:

Linear, single equation: ols, tsls, ar1, mpols

Linear, multi-equation: system, var, vecm
Nonlinear, single equation: logit, probit, poisson, negbin, tobit, intreg,
logistic, duration
Panel: panel, dpanel
Miscellaneous: arima, garch, heckit, quantreg, lad, biprobit

Don’t let names deceive you: for example, the probit command can estimate ordered models,
random-effect panel probit models, . . . The hansl “house style” is to keep to a relatively small
number of command words and to distinguish variants within a class of estimators such as Probit
by means of options, or the character of the data supplied.

49
Chapter 12. Estimation methods 50

Simultaneous systems (SUR, FIML and so on) constitute the main exception to the syntax summary
above; these require a system block— see the chapter on Multivariate models in the Gretl User’s
Guide.

12.2 Generic estimation tools

Hansl offers three main toolkits for defining estimators beyond the canned selection. Here’s a quick
overview:

command estimator User’s Guide

nls nonlinear least squares chapter 20
mle maximum likelihood estimation chapter 21
gmm generalized method of moments chapter 22

Each of these commands takes the form of a block of statements (e.g. nls . . . end nls). The user
must supply a function to compute the fitted dependent variable (nls), the log-likelihood (mle), or
the GMM residuals (gmm). With nls and mle, analytical derivatives of the function in question with
respect to the parameters may (optionally) be supplied.
The most widely used of these tools is probably mle. Hansl offers several canned ML estimators,
but if you come across a model that you want to estimate via maximum likelihood and it is not
supported natively, all you have to do is write down the log-likelihood in hansl’s notation and run
it through the mle apparatus.

12.3 Post-estimation accessors

All of the methods mentioned above are commands, not functions; they therefore do not return
any values. However, after estimating a model — either using a canned procedure or one of the
toolkits—you can grab most of the quantities you might wish to have available for further analysis
via accessors.
Some such accessors are generic, and are available after using just about any estimator. Examples
include $coeff and $stderr (to get the vectors of coefficients and standard errors, respectively),
$uhat and $yhat (residuals and fitted values), and $vcv (the covariance matrix of the coefficients).
Some, on the other hand, are specific to certain estimators. Examples here include $jbeta (the
cointegration matrix, following estimation of a VECM), $h (the estimated conditional variance series
following GARCH estimation), and $mnlprobs (the matrix of per-outcome probabilities following
multinomial logit estimation).
A full listing and description of accessors can be found in the Gretl Command Reference.

12.4 Formatting the results of estimation

The commands mentioned in this chapter produce by default quite verbose (and, hopefully, nicely
formatted) output. However, in some cases you may want to use built-in commands as auxiliary
steps in implementing an estimator that is not itself built in. In that context the standard printed
output may be inappropriate and you may want to take charge of presenting the results yourself.
This can be accomplished quite easily. First, you can suppress the usual output by using the
--quiet option with built-in estimation commands.1 Second, you can use the modprint command
to generate the desired output. As usual, see the Gretl Command Reference for details.
1 For some commands, --quiet reduces but does not eliminate gretl’s usual output. In these cases you can give the

--silent option. Consult the Gretl Command Reference to determine which commands accept this option.
Chapter 12. Estimation methods 51

12.5 Named models

We said above that estimation commands in hansl don’t return anything. This should be qualified
in one respect: it is possible to use a special syntax to push a model onto a stack of named models.
Rather than the usual assignment symbol, the form “<-” is used for this purpose. This is mostly
intended for use in the gretl GUI but it can also be used in hansl scripting.
Once a model is saved in this way, the accessors mentioned above can be used in a special way,
joined by a dot to the name of the target model. A little example follows. (Note that $ess accesses
the error sum of squares, or sum of squared residuals, for models estimated via least squares.)

diff y x
ADL <- ols y const y(-1) x(0 to -1)
ECM <- ols d_y const d_x y(-1) x(-1)
# the following two values should be equal
ssr_a = ADL.$ess
ssr_e = ECM.$ess
 Part III

Reference

52
 Chapter 13

Rules regarding white space

Programming languages differ in their rules regarding the use of white space in a program. Here
we set out the rules in hansl. The rules differ somewhat between commands on the one hand and
function calls plus assignment on the other.

13.1 White space in commands

Hansl commands are structured as follows: first comes a command word (e.g. ols, summary); then
come zero or more arguments (often the names of series); then come zero or more options (some
of which may take parameters). The relevant rules are:

1. The individual elements just mentioned must always be separated by at least one space, and
where one space is required you are free to insert as many as you like.

2. Whenever a parameter is supplied with an option flag, the parameter must be attached to the
flag with an equals sign, with no intervening space:
ols y 0 x --cluster=clustvar # correct
ols y 0 x --cluster =clustvar # broken!

13.2 Spaces in function calls and assignment

For the most part, white space in function calls and assignment is not significant; it can be inserted
or not at will. For example, in the following sets of statements each member is equally acceptable
syntactically (though some are ugly!):

# set 1
y = sqrt(x)
y=sqrt(x)
# set 2
c = cov(y1, y2)
c=cov(y1,y2)
c = cov(y1 , y2)

However, please note these exceptions:

1. When an assignment starts with a type keyword such as series or matrix, this must be
separated from what follows by at least one space, as in

series y = normal() # or: series y=normal()

2. In a function call, the opening parenthesis that marks the start of the argument list must be
attached to the name of the function without intervening space:
c = cov(y1, y2) # correct
c = cov (y1, y2) # broken!

53
 Chapter 14

Operators

14.1 Precedence
Table 14.1 lists the operators available in gretl in order of decreasing precedence: the operators on
the first row have the highest precedence, those on the second row have the second highest, and so
on. Operators on any given row have equal precedence. Where successive operators have the same
precedence the order of evaluation is in general left to right. The exceptions are exponentiation
and matrix transpose-multiply. The expression a^b^c is equivalent to a^(b^c), not (a^b)^c, and
similarly A’B’C’ is equivalent to A’(B’(C’)).

Table 14.1: Operator precedence

() [] . {}
! ++ -- ^ ’
* / % \ **
+ - ~ |
> < >= <= ..
== !=
&&
||
?:

In addition to the basic forms shown in the Table, several operators also have a “dot form” (as in
“.+” which is read as “dot plus”). These are element-wise versions of the basic operators, for use
with matrices exclusively; they have the same precedence as their basic counterparts. The available
dot operators are as follows.

.^ .* ./ .+ .- .> .< .>= .<= .=

Each basic operator is shown once again in the following list along with a brief account of its
meaning. Apart from the first three sets of grouping symbols, all operators are binary except
where otherwise noted.

() Function call
[] Subscripting
. Bundle membership (see below)
{} Matrix definition
! Unary logical NOT
++ Increment (unary)
-- Decrement (unary)
^ Exponentiation
’ Matrix transpose (unary) or transpose-multiply (binary)

54
Chapter 14. Operators 55

* Multiplication
/ Division, matrix “right division”
% Modulus
\ Matrix “left division”
** Kronecker product
+ Addition
- Subtraction
~ Matrix horizontal concatenation
| Matrix vertical concatenation
> Boolean greater than
< Boolean less than
>= Greater than or equal
<= Less than or equal
.. Range from–to (in constructing lists)
== Boolean equality test
!= Boolean inequality test
&& Logical AND
|| Logical OR
?: Conditional expression

The interpretation of “.” as the bundle membership operator is confined to the case where it is
immediately preceded by the identifier for a bundle, and immediately followed by a valid identifier
(key).
Details on the use of the matrix-related operators (including the dot operators) can be found in the
chapter on matrices in the Gretl User’s Guide.

14.2 Assignment
The operators mentioned above are all intended for use on the right-hand side of an expression
which assigns a value to a variable (or which just computes and displays a value — see the eval
command). In addition we have the assignment operator itself, “=”. In effect this has the lowest
precedence of all: the entire right-hand side is evaluated before assignment takes place.
Besides plain “=” several “inflected” versions of assignment are available. These may be used only
when the left-hand side variable is already defined. The inflected assignment yields a value that is
a function of the prior value on the left and the computed value on the right. Such operators are
formed by prepending a regular operator symbol to the equals sign. For example,

y += x

The new value assigned to y by the statement above is the prior value of y plus x. The other
available inflected operators, which work in an exactly analogous fashion, are as follows.

-= *= /= %= ^= ~= |=

In addition, a special form of inflected assignment is provided for matrices. Say matrix M is 2 × 2. If
you execute M = 5 this has the effect of replacing M with a 1 × 1 matrix with single element 5. But
if you do M .= 5 this assigns the value 5 to all elements of M without changing its dimensions.
Chapter 14. Operators 56

14.3 Increment and decrement

The unary operators ++ and -- follow their operand,1 which must be a variable of scalar type. Their
simplest use is in stand-alone expressions, such as

j++ # shorthand for j = j + 1

k-- # shorthand for k = k - 1

However, they can also be embedded in more complex expressions, in which case they first yield the
original value of the variable in question, then have the side-effect of incrementing or decrementing
the variable’s value. For example:

scalar i = 3
k = i++
matrix M = zeros(10, 1)
M[i++] = 1

After the second line, k has the value 3 and i has value 4. The last line assigns the value 1 to
element 4 of matrix M and sets i = 5.
Warning: as in the C programming language, the unary increment or decrement operator should be
not be applied to a variable in conjunction with regular reference to the same variable in a single
statement. This is because the order of evaluation is not guaranteed, giving rise to ambiguity.
Consider the following:

M[i++] = i # don’t do this!

This is supposed to assign the value of i to M[i], but is it the original or the incremented value?
This is not actually defined.

1 The C programming language also supports prefix versions of ++ and --, which increment or decrement their operand

before yielding its value. Only the postfix form is supported by gretl.
 Chapter 15

Greek-letter identifiers

As mentioned in chapter 3, the sole exception to the requirement that hansl identifiers must be
plain ASCII is that they may take the form of a single Greek letter. Here are the details.

• This exception applies only to names of variables other than series; the names of series must
always be ASCII.

• The supported Greek characters are the 24 (unaccented) letters in the basic Greek alphabet,
minus omicron, which is indistinguishable from the Latin ‘o’.

• These letters may be used in lower or upper case (constituting distinct identifiers, as usual in
hansl), except for the several upper-case letters which are indistinguishable in the Latin and
Greek alphabets (‘A’, ‘B’, ‘E’, ‘K’, ‘M’, ‘N’, . . . ).

• The Greek letters must be encoded in UTF-8.

In gretl’s graphical interface (script editor and GUI “console”), the acceptable Greek letters can be
entered by typing a Latin letter while the Alt key is depressed. The mappings from Latin to Greek
are shown below: lower case first, then upper case.

Latin Greek Latin Greek

a α alpha n ν nu
b β beta p π pi
c χ chi q θ theta
d δ delta r ρ rho
e  epsilon s σ sigma
f φ phi t τ tau
g γ gamma u υ upsilon
h η eta v ν nu
i ι iota w ω omega
j ψ psi x ξ xi
k κ kappa y υ upsilon
l λ lambda z ζ zeta
m µ mu

57
Chapter 15. Greek-letter identifiers 58

Latin Greek
D ∆ Delta
F Φ Phi
G Γ Gamma
J Ψ Psi
L Λ Lambda
P Π Pi
Q Θ Theta
S Σ Sigma
U Υ Upsilon
W Ω Omega
X Ξ Xi
Y Υ Upsilon

A word of advice: it’s probably not a good idea to employ Greek-letter identifiers in hansl scripts
that you intend to share via the internet, since one cannot assume that text encodings are preserved
unchanged. This warning applies in particular if you, or any of the intended recipients of your
scripts, work on MS Windows, since Windows does not natively support UTF-8, the mandatory
encoding of such identifiers for use with gretl.