<p dir="ltr"><b>Partial-Pooling Gradient Boosting for Heterogeneous Panels: An Adaptive Shrinkage Estimator Validated Across Two Independent Panels</b></p>

<p dir="ltr">This repository contains the data, code, and reproducibility materials supporting the study <b>“Partial-Pooling Gradient Boosting for Heterogeneous Panels: An Adaptive Shrinkage Estimator Validated Across Two Independent Panels.”</b> The study introduces Partial-Pooling Gradient Boosting (PPGB), a nonlinear panel-data estimator that combines pooled gradient boosting with unit-specific warm-start corrections and empirical-Bayes shrinkage.

The empirical validation uses two independent datasets: an unbalanced Ghana National Fire Service panel comprising <b>904 region-month observations across 11 reporting units</b>, and a larger cross-country macroeconomic panel containing <b>1,532 country-year observations for 37 countries</b>, used to classify inflation regimes. The repository is intended to support transparent replication of the data-processing, panel-aware cross-validation, PPGB estimation, benchmark-model comparisons, unit-blocked bootstrap inference, shrinkage-weight analysis, and robustness exercises reported in the manuscript.