Skip to content

Changelog

All notable changes to PanelBox are documented here. The format follows Keep a Changelog, and PanelBox adheres to Semantic Versioning.

Sections: Added, Changed, Fixed, Deprecated, Removed, Security, Performance.

[Unreleased]

Added

Interactive Tutorial System

Status: Module 1 Complete — Production Ready

A comprehensive, hands-on learning system for panel data econometrics with PanelBox:

  • Module 1: Fundamentals (3.5–4.5 hours, 4 tutorials)
    • Introduction to Panel Data Structures
    • Model Specification with Formulas
    • Estimation and Results Interpretation
    • Spatial Fundamentals (optional)
  • Structured learning paths (linear, spatial, complete)
  • Real datasets (Grunfeld, simulated spatial)
  • Competency checkpoints and self-assessment

Advanced GMM Estimators

  • CUE-GMM (panelbox.gmm.ContinuousUpdatedGMM) — Hansen-Heaton-Yaron (1996) with continuously updated weighting matrix; HAC and cluster-robust options; bootstrap variance
  • Bias-Corrected GMM (panelbox.gmm.BiasCorrectedGMM) — Hahn-Kuersteiner (2002) analytical bias correction with first- and second-order terms
  • Enhanced GMM Diagnostics (panelbox.gmm.GMMDiagnostics) — Hansen J-test with bootstrap p-values, C-statistic, weak instruments detection

Panel Selection Models

  • Panel Heckman (panelbox.models.selection.PanelHeckman) — Two-step (Wooldridge 1995) and MLE estimation; Murphy-Topel SE correction; IMR diagnostics

Advanced Cointegration Tests

  • Westerlund (2007) ECM-based tests — Gt, Ga, Pt, Pa with bootstrap critical values
  • Pedroni (1999) — 7 residual-based statistics (panel and group)
  • Kao (1999) — DF and ADF statistics

Panel Unit Root Tests

  • Hadri (2000) LM test — Tests H0: all series stationary; heteroskedasticity-robust variant
  • Breitung (2000) — Bias-corrected pooled estimator with detrending
  • Unified interface panel_unit_root_test() — Run multiple tests simultaneously with comparative summary

Specification Tests and Specialized Models

  • Davidson-MacKinnon J-test — Non-nested model comparison with cluster-robust SEs
  • Encompassing tests — Cox test, Wald encompassing, likelihood ratio
  • Multinomial Logit — FE (conditional MLE), RE (GLS), pooled; marginal effects
  • PPML — Poisson Pseudo-Maximum Likelihood for gravity models; handles zeros

Phase 6 — Integration and Documentation (In Progress)

  • Namespace integration: 60+ advanced methods accessible via panelbox.*
  • Comprehensive documentation restructuring (34 pages across 5 deliverables)

[1.0.0] — 2026-02-08

Summary

Panel VAR Module — Major Release

PanelBox v1.0.0 introduces the complete Panel VAR module, establishing PanelBox as the first Python library with full feature parity to R's pvar and Stata's pvar packages.

Key metrics: 150+ tests, 90%+ coverage, validated against R (coefficients within ±1e-6).

Added

  • PanelVAR — OLS with fixed effects and GMM with FOD/FD transformations; automatic lag selection (AIC, BIC, HQIC, MBIC, MAIC, MQIC)
  • PanelVARResult — .params, .std_errors, .pvalues, .is_stable(), .hansen_j, .summary(), .to_latex()
  • IRF Module — Cholesky and Generalized impulse response functions; bootstrap and analytical confidence intervals
  • FEVD Module — Forecast Error Variance Decomposition (Cholesky and Generalized)
  • Granger Causality — Pairwise Wald tests, Dumitrescu-Hurlin (2012) panel Granger causality; bootstrap inference; causality network visualization
  • Forecasting — h-step ahead iterative forecasts; bootstrap and analytical CIs; out-of-sample evaluation (RMSE, MAE, MAPE)
  • PanelVECM — Panel Vector Error Correction Model for I(1) cointegrated systems; Johansen rank selection; long-run (beta) and short-run (gamma) separation

Performance

Benchmark vs R vs Stata
OLS estimation ~1.5x faster —
GMM estimation ~1.3x faster ~1.5x faster
IRF Bootstrap ~1.5x faster —
Memory usage ~1.5–2x less —

Validation

Validated against R (plm, pvar, panelvar, urca):

Metric Tolerance Result
OLS Coefficients ± 1e-6 max diff = 3.2e-7
GMM Coefficients ± 1e-4 max diff = 8.4e-5
Hansen J statistic ± 1e-3 diff = 0.003
IRFs ± 1e-6 max diff = 8.4e-7
FEVD ± 1e-3 diff < 0.001
Granger p-values ± 1e-3 diff < 0.001

[0.8.0] — 2026-02-08

Summary

Test Runners and Master Report System

Added

  • ValidationTest Runner — Configurable test runner with quick, basic, full presets; integrates with model .validate() methods
  • ComparisonTest Runner — Multi-model comparison with automatic metric extraction (R², AIC, BIC)
  • Master Report System — Comprehensive HTML report with experiment overview, model summary grid, navigation to sub-reports
  • PanelExperiment.save_master_report() — One-liner master report generation

Fixed

  • Fixed KeyError when accessing timestamp in model metadata (now uses .get() with fallback)

[0.7.0] — 2026-02-08

Summary

ResidualResult Container and Chart Registration Fix

Added

  • ResidualResult — Diagnostic container with Shapiro-Wilk, Jarque-Bera, Durbin-Watson, Ljung-Box tests; standardized residuals for outlier detection; summary(), save_html(), save_json() methods
  • PanelExperiment.analyze_residuals() — One-liner residual analysis

Fixed

  • Chart registration: Root cause was plotly not installed; all 35 charts now register correctly
  • Fixed jarque_bera() return value unpacking
  • Fixed ljung_box() DataFrame indexing (statsmodels return type change)
  • HTML reports now include embedded interactive charts (102.9 KB vs 77.5 KB)

[0.6.0] — 2026-02-08

Summary

Experiment Pattern and Result Containers

Added

  • PanelExperiment — Unified workflow for fitting, validating, and comparing panel models
  • BaseResult — Abstract base class for result containers
  • ValidationResult — Model specification test results with save_html() and save_json()
  • ComparisonResult — Multi-model comparison results with automatic best-model identification

[0.5.0] — 2026-02-08

Summary

Comprehensive Visualization System — 28+ Interactive Charts

Added

  • 28+ interactive chart types using Plotly
  • 3 professional themes: Professional (corporate blue), Academic (journal-ready), Presentation (high-contrast)
  • Validation charts (5): test overview, p-value distribution, test statistics, comparison heatmap, validation dashboard
  • Residual diagnostics (7): QQ plot, residual vs. fitted, scale-location, residual vs. leverage, time series, distribution, partial regression
  • Model comparison (4): coefficient comparison, forest plot, model fit comparison, information criteria
  • Panel-specific (4): entity effects, time effects, between-within decomposition, panel structure heatmap
  • Econometric test visualizations (4): ACF/PACF, unit root test, cointegration heatmap, cross-sectional dependence
  • Export system: HTML, JSON, PNG, SVG, PDF; batch export across formats
  • Report generation: HTML reports for validation, residuals, comparison, and panel diagnostics
  • Registry/Factory architecture: @register_chart decorator, ChartFactory.create()

[1.0.0] — 2026-02-05

Summary

Production Release — Complete Panel Data Econometrics Suite

600+ unit tests, 93% passing, validated against Stata xtabond2 and R plm.

Added

Static Panel Models:

  • Pooled OLS, Fixed Effects (Within), Random Effects (GLS), Between Estimator, First Differences
  • Hausman test for FE vs. RE specification
  • Formula interface (R-style with patsy)

Dynamic Panel GMM:

  • Difference GMM (Arellano-Bond 1991) — one-step, two-step, iterative
  • System GMM (Blundell-Bond 1998) — combined differenced and level equations
  • Automatic instrument generation (GMM-style and IV-style)
  • Instrument collapse (Roodman 2009)
  • Windmeijer (2005) finite-sample SE correction

Robust Standard Errors (8 types):

  • HC0–HC3 (White, MacKinnon-White)
  • One-way and two-way clustering (Cameron, Gelbach & Miller 2011)
  • Driscoll-Kraay (1998)
  • Newey-West HAC (1987)
  • PCSE (Beck & Katz 1995)

Bootstrap Inference:

  • 4 methods: Pairs, Wild, Block, Residual
  • 3 CI methods: percentile, basic, studentized

Diagnostic Tests (50+):

  • Heteroskedasticity: White, Breusch-Pagan, Modified Wald
  • Serial correlation: Wooldridge, Breusch-Godfrey, Baltagi-Wu
  • Cross-sectional dependence: Pesaran CD, Frees, BP-LM
  • Unit root: LLC, IPS, Fisher
  • Cointegration: Pedroni, Kao, Westerlund

Robustness Checks:

  • Influence diagnostics (DFBETA, Cook's D, leverage)
  • Outlier detection (standardized residuals, IQR)
  • Jackknife resampling
  • Panel cross-validation

Report Generation: HTML, Markdown, and LaTeX exporters with customizable templates.

Datasets: load_grunfeld(), load_abdata(), list_datasets().

Performance: Numba JIT compilation for critical paths (up to 348x speedup).

[0.4.0] — 2026-02-05

Added

Robust Standard Errors:

  • HC0–HC3 variants with leverage adjustment and caching
  • One-way and two-way clustering with finite-sample corrections
  • Driscoll-Kraay with 3 kernel options (Bartlett, Parzen, Quadratic Spectral)
  • Newey-West HAC with automatic lag selection
  • PCSE (Beck & Katz 1995) for contemporaneous cross-sectional correlation
  • Integration: Fixed Effects supports 8 covariance types; Random Effects supports 7

[0.3.0] — 2026-01-22

Added

Bootstrap Inference:

  • PanelBootstrap with 4 methods (pairs, wild, block, residual)
  • Performance: ~95–110 iterations/second
  • Confidence intervals: percentile, basic, studentized

Sensitivity Analysis:

  • SensitivityAnalysis with leave-one-out entities, leave-one-out periods, and subset sensitivity
  • Influential unit detection
  • Optional matplotlib visualization

[0.2.0] — 2026-01-21

Added

Dynamic Panel GMM:

  • Difference GMM (Arellano-Bond 1991) with one-step, two-step, and iterative estimation
  • System GMM (Blundell-Bond 1998) with combined differenced and level equations
  • Automatic instrument generation with collapse option
  • Windmeijer (2005) finite-sample correction
  • Smart instrument selection for unbalanced panels (72% retention vs. 0%)
  • Specification tests: Hansen J, Sargan, AR(1), AR(2)

Fixed

  • Arellano-Bond validation now works with unbalanced panels
  • Time dummies no longer cause 0% observation retention
  • System GMM error handling improved

[0.1.0] — 2025-12

Added

Core Framework:

  • PanelData — Panel data container with validation
  • FormulaParser — R-style formula parsing (patsy integration)
  • PanelResults — Base results class

Static Models:

  • PooledOLS, FixedEffects, RandomEffects
  • HausmanTest and HausmanTestResult

Standard Errors: Homoskedastic, heteroskedasticity-robust, clustered.

Validation Framework: Autocorrelation, heteroskedasticity, cross-sectional dependence, and unit root tests.

Reporting: HTML (Plotly), static (Matplotlib), LaTeX table export.

Versioning Policy

PanelBox uses Semantic Versioning:

Component When incremented
Major (X.0.0) Incompatible API changes
Minor (0.X.0) New features, backward compatible
Patch (0.0.X) Bug fixes, backward compatible

Migration Notes

v0.x → v1.0.0

No breaking changes. All pre-1.0 code continues to work. New features are additive:

# v0.1.0 code still works
from panelbox import FixedEffects, RandomEffects

# v1.0.0 adds Panel VAR
from panelbox.var import PanelVAR, PanelVECM

See Also