_   _   ___
               | | | | /   |
  __ _ _ __ ___| |_| |/ /| |_ __  _   _
 / _` | '__/ _ \ __| / /_| | '_ \| | | |
| (_| | | |  __/ |_| \___  | |_) | |_| |
 \__, |_|  \___|\__|_|   |_/ .__/ \__, |
  __/ |                    | |     __/ |
 |___/                     |_|    |___/

gretl4py: Python Bindings for gretl

Python bindings for the gretl econometrics library.

gretl4py provides a Python interface to libgretl, the numerical and econometric engine underlying gretl. It brings gretl's econometric functionality, numerical routines, and data-handling capabilities into Python while providing Python-oriented interfaces for matrices, datasets, models, bundles, and other gretl objects.

gretl4py is not a separate econometrics library. It is a bridge between Python and libgretl: estimation and many numerical operations are performed by the mature gretl engine, while Python provides the programming environment and ecosystem around it.

Windows users: gretl4py requires the Microsoft Visual C++ 2022 Redistributable (x64). Download it from https://aka.ms/vs/17/release/vc_redist.x64.exe.

---

Highlights

gretl4py provides:

• Econometric estimation, including OLS, weighted and least-absolute-deviation regression, IV/2SLS, maximum likelihood, GMM, quantile regression, regularized regression, and various limited-dependent-variable models.
• Time-series and multivariate methods, including AR/ARIMA, GARCH-type models, VAR and VECM models, structural VAR functionality, and related tests.
• Panel-data methods, including dynamic-panel and instrumental-variable estimators.
• Mixed-frequency modelling, including MIDAS regression.
• Econometric tests, covering unit roots, cointegration, heteroskedasticity, autocorrelation, specification, parameter restrictions, structural stability, and other commonly used procedures.
• Native gretl matrices and datasets, exposed as Python objects with Python-friendly indexing and conversion facilities.
• Bundles and other gretl objects, allowing Python code to interact directly with objects used by libgretl.
• NumPy and pandas interoperability, including conversion to NumPy arrays and pandas DataFrames and, where appropriate, zero-copy views of matrix data.
• gretl's hansl language, allowing Python programs to execute hansl code and exchange user-defined variables with libgretl.
• gretl function packages, making functionality implemented in hansl packages directly usable from Python.
• LIBSVM-based machine-learning functionality available through gretl.

The package is designed to allow users to combine the econometric functionality of gretl with the broader Python ecosystem, including NumPy, pandas, matplotlib, and other scientific and statistical libraries.

---

Installation

The recommended way to install gretl4py is with pip:

pip install gretl4py

Pre-built packages are available for supported Python versions and platforms.

For platform-specific requirements and available distributions, see the gretl4py project page.

---

A quick example

A typical gretl4py workflow consists of loading a dataset, estimating a model, and working with the resulting Python object:

import gretl

data = gretl.get_data("bjg.gdt")

model = gretl.ols(
    [1, 0, 2],
    data=data
).fit()

print(model)

The exact estimator interface depends on the model being estimated. The gretl module exposes both high-level estimator functions and corresponding gretl model objects.

---

Working with datasets

gretl datasets are represented by the gretl.Dataset class. Datasets can be loaded from a number of formats supported by gretl, including:

• .gdt
• .gdtb
• .csv
• .dta
• .wf1
• .xls
• .xlsx
• .ods

For example:

import importlib.resources as resources
import gretl

data_dir = resources.files("gretl").joinpath("data")
data = gretl.get_data(str(data_dir.joinpath("bjg.gdt")))

print(data)

Datasets can be manipulated directly from Python, including adding and transforming series, working with lists, and passing datasets to estimators.

gretl4py also provides conversion to pandas:

df = data.to_pandas()

---

Matrices

The gretl.Matrix class provides a Python interface to gretl's native matrix objects.

Matrices support:

• Python-style indexing and slicing;
• matrix operations;
• row and column names;
• conversion to NumPy;
• conversion to pandas;
• zero-copy NumPy views where appropriate.

For example:

import gretl

m = gretl.Matrix.ones(3, 3)
m[0, 0] = 10

print(m)

NumPy interoperability is available both through copying and through zero-copy views:

a = m.to_numpy()       # independent NumPy array
v = m.numpy_view       # view of the underlying gretl matrix

A pandas representation is also available:

df = m.to_pandas()

The distinction between copies and views is important when modifying data; see the documentation for details.

---

Estimation

gretl4py exposes a range of gretl estimators through Python. Depending on the model, these include:

• OLS and weighted least squares
• AR and ARIMA
• GARCH-type models
• IV / 2SLS
• LAD and quantile regression
• logit, probit, tobit and related limited-dependent-variable models
• count and duration models
• sample-selection models
• panel-data estimators
• dynamic-panel estimators
• MIDAS regression
• VAR and VECM
• structural VAR models
• regularized regression, including LASSO, Ridge and Elastic Net
• maximum-likelihood and GMM-based estimation

The resulting objects are instances of gretl4py model classes and provide access to estimation results, coefficients, covariance matrices, residuals, fitted values, forecasts, tests, and other model-specific information.

For example:

import gretl

data = gretl.get_data("bjg.gdt")

model = gretl.ols([1, 0, 2], data=data).fit()

print(model.coeff)
print(model.vcv)

The available methods depend on the particular model class. For example, VAR and VECM models provide functionality specific to multivariate time-series analysis, including impulse-response analysis.

---

Econometric tests

Many of gretl's statistical and econometric tests are available directly from Python. These include tests for, among other things:

• unit roots and stationarity;
• cointegration;
• autocorrelation;
• heteroskedasticity;
• normality;
• structural stability;
• parameter restrictions;
• specification;
• omitted variables;
• functional form;
• multicollinearity;
• panel-data dependence and specification.

Tests may operate either on a dataset or on a fitted model, depending on the particular procedure.

---

Function packages

One of gretl's distinctive features is its function-package system. Function packages extend gretl by providing additional functionality implemented in hansl, gretl's scripting language.

gretl4py provides a bridge to this ecosystem: functions defined in gretl function packages can be loaded and used as Python functions. This means that the Python interface is not limited to functionality implemented directly in the gretl4py bindings. A gretl function package can effectively extend the Python API.

Loading a package

A package is loaded with gretl.include():

import gretl

BMA = gretl.include("BMA")

The returned package object can be inspected, for example:

BMA.show_functions()

After a package has been included, its functions are available through the gretl module. For example:

import gretl

BMA = gretl.include("BMA")

data = gretl.get_data("FLS_41t72.gdt", frompkg="BMA")

opt = gretl.Bundle()
opt["Nrep"] = 10**5
opt["model_prior"] = "binomial"
opt["Nrank"] = 5

result = gretl.BMA(
    Y="GDP_growth",
    X_list="2..13",
    quiet=True,
    Options=opt
)

gretl.BMA_Print(result)

The important point is that BMA() and BMA_Print() are functions supplied by the gretl function package rather than native gretl4py bindings.

Package functions and gretl objects

Package functions can exchange gretl4py objects with the underlying hansl code. For example, a package function can receive a gretl.Matrix, a NumPy view, a gretl.Bundle, a dataset, or other supported gretl objects.

For example, the extra package can be used with both a gretl matrix and its NumPy view:

import gretl

extra = gretl.include("extra")

m = gretl.Matrix(4, 1, "normal")

print(gretl.combinations(m, 1))
print(gretl.combinations(m.numpy_view, 1))

This allows package functions to participate naturally in Python workflows while retaining the functionality implemented in the original hansl package.

Packages demonstrated by gretl4py

The repository currently contains package examples for:

Package	Main functionality
BACE	Bayesian Averaging of Classical Estimates
BayTool	Bayesian regression methods
BMA	Bayesian Model Averaging
BVAR	Bayesian VAR models
criteria	Model-selection criteria
gig	GARCH models
ParMA	Parallel Bayesian Model Averaging
SVAR	Structural VAR models
TVC	Time-varying coefficient models

The corresponding examples are located in:

gretl/examples/packages/

The repository also contains a combined package demonstration in:

demo/packages.py

These examples are intended to show not only how a package is loaded, but also how package functions interact with native gretl4py objects.

Package ecosystem

The gretl function-package ecosystem is considerably larger than the examples included with gretl4py. It contains official gretl add-ons as well as a large collection of contributed packages covering, among other areas:

• Bayesian methods and model averaging;
• unit-root, stationarity and structural-break tests;
• cointegration and long-run analysis;
• univariate time-series models;
• VAR, SVAR and factor models;
• volatility and financial econometrics;
• panel-data models;
• discrete and limited-dependent-variable models;
• hypothesis testing and diagnostics;
• estimation and regression tools;
• forecasting and prediction;
• machine learning and nonparametric methods;
• spatial econometrics;
• graphics and visualization;
• data access and management;
• programming utilities and GUI tools.

The package mechanism is therefore an important part of the overall gretl4py architecture: it allows the Python interface to benefit from the existing and continuously growing gretl function-package ecosystem without requiring every package to be reimplemented as a separate Python extension.

---

NumPy and pandas interoperability

gretl4py is intended to work naturally with the Python scientific-computing ecosystem.

For matrices, the following interfaces are provided:

numpy_array = matrix.to_numpy()
numpy_view = matrix.numpy_view

dataframe = matrix.to_pandas()

to_numpy() and to_pandas() create independent objects. In contrast, numpy_view exposes the underlying matrix storage directly and therefore avoids a data copy.

Datasets can be converted to pandas DataFrames with:

dataframe = dataset.to_dataframe()

This makes it possible to use gretl for estimation while using NumPy, pandas, matplotlib, or other Python packages for subsequent processing and visualization.

---

Interoperability with hansl

gretl4py also provides an additional interoperability layer with hansl, gretl's scripting language.

Python code can execute hansl code directly:

import gretl

gretl.run_hansl("""
function void foo (void)
    print "Hello from hansl"
end function

foo()
""")

An existing hansl script can be executed with:

gretl.run_script("my_script.inp")

This functionality is particularly useful for testing existing hansl code, accessing functionality implemented in gretl function packages, and using parts of gretl that are naturally expressed in hansl.

User-defined variables

gretl4py can also exchange user-defined variables with libgretl.

For example, a matrix created in Python can be placed in the gretl namespace:

import gretl

m = gretl.Matrix.ones(2, 2)

gretl.genr(name="mat", value=m)
gretl.run_hansl("mat = mat .* 2")

m = gretl.get_uservar("mat")

These variables live in the libgretl namespace rather than in Python's namespace. Consequently, they can be accessed by subsequently executed hansl code and can also be retrieved from Python.

Important: hansl execution and user-defined-variable interoperability are supplementary features of gretl4py. They are provided primarily to facilitate interoperability with existing gretl/hansl code and should not be regarded as the primary programming interface of the package. For new Python applications, the native gretl4py API is generally preferable.

---

Examples and demonstrations

The source distribution contains a growing collection of examples and demonstrations.

Estimator examples

Estimator examples are located in:

gretl/examples/estimators/

They provide small, self-contained examples of individual estimators. For example:

import gretl.examples.estimators.ols

gretl.examples.estimators.ols.run_example()

The source of an example can also be inspected directly from Python:

import inspect
import gretl.examples.estimators.ols

print(inspect.getsource(
    gretl.examples.estimators.ols.run_example
))

Package examples

Examples demonstrating gretl function packages are located in:

gretl/examples/packages/

They provide focused examples of loading and using individual packages.

Feature demonstrations

More extensive demonstrations are located in:

demo/

These cover not only estimation, but also data handling, matrices, bundles, user-defined variables, hansl interoperability, forecasting, VAR/VECM functionality, function packages, filters, nonlinear models, and other gretl4py features.

The examples and demonstrations are often the best starting point for seeing how a particular feature is intended to be used.

---

Package contents

A gretl4py installation contains:

1. libgretl and its plugins together with the required runtime dependencies;
2. the compiled _gretl Python extension module providing the binding to libgretl;
3. Python-level helper modules and utilities;
4. example modules and scripts;
5. bundled datasets and other supporting resources.

The Python extension is implemented using C++ and pybind11, while the underlying econometric functionality is provided by libgretl.

---

Documentation

The main documentation is available at:

https://gretl.sourceforge.net/gretl4py.html

The current PDF documentation is available from the SourceForge files area.

The documentation is actively evolving alongside the Python API.

The source code, examples, demonstrations, and development history are available in the gretl4py repository.

---

License

gretl4py is distributed under the GNU General Public License, version 3 or later (GPL-3.0-or-later).

gretl and its components are distributed under their respective free-software licenses. See the accompanying license files and the gretl documentation for details.

---

Development status

gretl4py is under active development. The Python interface continues to evolve as more of the functionality provided by libgretl is exposed through a native Python API.

The project aims to keep the Python interface consistent and Pythonic while retaining close correspondence with gretl's established econometric functionality.

For current functionality, examples, and API details, consult the documentation and source tree rather than relying solely on this README.