Replication Package for Beyond Gravity: Machine Learning Forecasts of Nigeria's Cocoa Exports

This replication package supports the article “Beyond Gravity: Machine Learning Forecasts of Nigeria’s Cocoa Exports.” It contains the audited 163-observation annual analytical dataset, a public relative-path Python pipeline, higher-frequency trade and price inputs, frozen model results, partner-level predictions, source data for tables and figures, feature documentation, uncertainty diagnostics, software-environment files and integrity manifests. The package reproduces the leakage-controlled expanding-window model-selection exercise, the selected annual Random Forest forecasts, fold-wise PPML benchmarks, the frozen 2024 external evaluation, SHAP-based predictive interpretation, multiplicative conformal prediction intervals and the selected Q1 mirror-trade hybrid nowcast. The private raw-to-cleaned reconciliation workbook is excluded because it contains local filesystem paths and internal workflow information. It is replaced by a path-free public audit summary, the analytical dataset checksum, explicit sample definitions and public reproducibility documentation. Third-party source data remain subject to the access, use and redistribution terms of their originating databases.