PanelBox

Panel Data Econometrics in Python

PanelBox is a comprehensive Python library for panel data econometrics, with 70+ models across 13 families, 50+ diagnostic tests, and 35+ interactive charts. It brings the capabilities of Stata's xtabond2, xtreg, xtfrontier, and R's plm, splm, frontier to Python with a modern, unified API.

Installation

pip install panelbox

Quick Start

import panelbox as pb

# Load bundled dataset (103 datasets available)
data = pb.datasets.load_grunfeld()

# Fixed Effects model
fe = pb.FixedEffects(
    formula="invest ~ value + capital",
    data=data,
    entity_col="firm",
    time_col="year"
)
results = fe.fit(cov_type='clustered')
print(results.summary())

Model Families

Static Panel Models

Dynamic Panel GMM

Full diagnostic suite: Hansen J, Sargan, AR(1)/AR(2), Windmeijer correction, instrument ratio monitoring, overfit diagnostics.

Panel VAR

Includes IRF, FEVD, Granger causality network visualization, lag selection (AIC/BIC/HQIC), and Johansen cointegration rank test.

Spatial Models

Stochastic Frontier Analysis

JLMS, BC, and Mode efficiency estimators. TFP decomposition and frontier visualization.

Count Data Models

Discrete Choice Models

Quantile Regression

Selection & Censored Models

Diagnostic Tests (50+)

Robust Standard Errors (8 types)

• HC0-HC3: Heteroskedasticity-consistent (White, leverage-adjusted)
• Clustered: One-way (entity/time) and two-way (Cameron-Gelbach-Miller 2011)
• Driscoll-Kraay: Spatial and temporal dependence
• Newey-West: HAC for serial correlation
• PCSE: Panel-corrected (Beck-Katz 1995)
• Spatial HAC: For spatial panel models

Visualization (35+ interactive charts)

from panelbox.visualization import (
    create_residual_diagnostics,
    create_validation_charts,
    create_comparison_charts,
    create_panel_charts,
    export_charts,
)

# Residual diagnostics (Q-Q, fitted vs residual, scale-location, etc.)
charts = create_residual_diagnostics(results)
export_charts(charts, "diagnostics.html")

# Entity/time effects, between-within decomposition, panel structure
charts = create_panel_charts(results)

Three professional themes: professional, academic, presentation. Export to HTML, JSON, PNG, SVG, PDF.

Experiment Pattern

The PanelExperiment class provides a factory-based workflow for comparing models:

import panelbox as pb

data = pb.datasets.load_grunfeld()

experiment = pb.PanelExperiment(
    data=data,
    formula="invest ~ value + capital",
    entity_col="firm",
    time_col="year"
)

# Fit and compare models
experiment.fit_all_models(names=['pooled', 'fe', 're'])
comparison = experiment.compare_models(['pooled', 'fe', 're'])
print(f"Best model: {comparison.best_model}")

# Validate specification
validation = experiment.validate_model('fe')
validation.save_html('validation.html', test_type='validation')

# Residual diagnostics
residuals = experiment.analyze_residuals('fe')
print(residuals.summary())

# Master report linking all sub-reports
experiment.save_master_report('report.html', theme='professional', reports=[...])

AutoExperiment (AutoML for Panel Data)

AutoExperiment automates the entire panel data modeling pipeline — from variable transformation and selection to multi-model estimation, econometric validation, and ranking:

from panelbox.autoexperiment import AutoExperiment

auto = AutoExperiment(
    data=data,
    depvar="invest",
    entity_col="firm",
    time_col="year",
    sign_constraints={"value": "+", "capital": "+"},
)
results = auto.run()
print(results.summary())
results.report("report.html")

• Variable transformations: lags, diffs, logs, growth rates, rolling means, squares
• Forward stepwise selection by BIC/AIC with sign constraints from economic theory
• Multi-model comparison: Pooled OLS, Fixed Effects, Random Effects, First Difference
• Automatic validation: Hausman, Pesaran CD, RESET, Modified Wald, and more
• Auto-selected standard errors based on diagnostic test results
• Composite ranking combining fit, diagnostics, theory compliance, and parsimony
• Data mining warnings when too many combinations are tested

Bundled Datasets (103 datasets)

from panelbox.datasets import load_dataset, list_datasets, list_categories

# Browse categories
print(list_categories())
# ['censored', 'count', 'diagnostics', 'discrete', 'frontier', 'gmm',
#  'marginal_effects', 'production', 'quantile', 'spatial', 'standard_errors',
#  'validation', 'var']

data = load_dataset("healthcare_visits")
grunfeld = load_dataset("grunfeld")

All 80+ example notebooks use load_dataset() and work directly in Google Colab.

Comparison with Other Packages

Requirements

• Python >= 3.9
• NumPy, Pandas, SciPy, statsmodels, scikit-learn
• Plotly, Matplotlib, Seaborn (visualization)
• Numba, Joblib (performance)

See pyproject.toml for full dependency list.

Documentation

• User Guide - Comprehensive guides for all model families
• API Reference - Full API documentation
• Tutorials - Interactive Jupyter notebooks
• Examples - 80+ example notebooks across all model families
• Theory - Econometric theory guides
• Benchmarks - Validation against R/Stata

Citation

@software{panelbox2026,
  author = {Haase, Gustavo and Dourado, Paulo},
  title = {PanelBox: Panel Data Econometrics in Python},
  year = {2026},
  version = {1.0.0},
  url = {https://github.com/PanelBox-Econometrics-Model/panelbox}
}

Contributing

Contributions are welcome! See CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE.

Support

• GitHub Issues
• Documentation
• Discussions
• Changelog

Made with care for econometricians and researchers