A harmonized monthly panel of macroeconomic, fiscal and monetary indicators with business-cycle labels

A harmonized monthly panel that merges eight official macroeconomic, fiscal and monetary data sources into a single country–month matrix covering 170 economies and 469 indicators from 1970 to 2025, comprising 114,229 unique country–month records. Five World Bank collections and three International Monetary Fund collections are retrieved, of which six contribute indicators meeting the coverage standard. Existing large macroeconomic panels are either annual and cross-country, or monthly and single-country. This resource is monthly and cross-country, and ships per-cell provenance and a business-cycle label as part of the release rather than leaving them to the user. What the release contains The complete panel (panel.parquet) — 473 columns: 239 macroeconomic, 110 fiscal and 120 monetary indicators, plus 4 series derived by the authors from World Bank GDP. Per-cell provenance (native_observation_mask.parquet) — every cell carries a flag distinguishing values reported by the source in that month from values supplied by frequency alignment or produced by imputation, so analyses can be restricted to natively observed data. Business-cycle labels — every country–month observation carries a five-state label (Depression, Recession, Recovery, Expansion, Overheating) derived from a one-sided Hamilton regression-filter output gap with country-adaptive thresholds, together with a confidence flag (label_confidence.parquet) from an independent Bry–Boschan / Harding–Pagan turning-point cross-check. Indicator dictionary (indicator_dictionary.csv) — the correspondence between every released column and its source collection. Column licence manifest (column_licence_manifest.csv) — the release basis of every column. Diagnostic and audit reports — coverage, imputation, economic sanity, collinearity and business-cycle audits. Imputation-cutoff configuration (split_config.json). Method, in brief Annual series are placed on the monthly calendar by period-end anchoring rather than interpolation across the year. Missing values are filled by a five-stage procedure — short-gap interpolation, seasonal-trend decomposition, country-wise and cross-country gradient-boosted regression, then continuity and fallback — whose learned components are fit only on rows preceding a documented cutoff, so no future information enters a filled cell. Countries are retained only where natively observed coverage reaches 60% macroeconomic, 45% monetary and 30% fiscal simultaneously. Licensing and redistribution Of the 473 released columns, 345 derive from World Bank collections and are released under CC BY 4.0, whose terms permit redistribution of modified data with attribution; 4 are the authors' own derived series, released under CC BY 4.0. The remaining 124 columns derive from IMF collections. Written permission was obtained from the International Monetary Fund to deposit data derived from publicly available IMF sources in a public research repository under that repository's standard terms, on a non-exclusive basis, subject to four conditions: the deposit comprises derived, processed or otherwise transformed values rather than raw IMF datasets; the underlying IMF sources are cited in the accompanying documentation; no endorsement by the IMF is implied; and the permission does not extend to third-party material for which the IMF does not hold redistribution rights. The released panel satisfies these conditions, since every value has been harmonized to a monthly grid, imputed where required and relabelled, and the correspondence between released columns and IMF collections is recorded in the indicator dictionary. The complete panel is therefore released under CC BY 4.0, and no reconstruction step is required of the user. The processing code is released separately so that the panel can be rebuilt from its original sources if desired, but running it is not a precondition for using this deposit. Disclaimer This panel is provided as is. The indicator values originate with their providers and are reproduced as retrieved on the vintages recorded in the accompanying documentation; they represent what each provider published at that vintage rather than a definitive estimate of the underlying quantity, and are subject to routine provider revision. No endorsement by the World Bank or the International Monetary Fund is implied. Related Processing code: 10.5281/zenodo.21583081 (MIT License) — retrieval and harmonization scripts, the five-stage imputation pipeline, the business-cycle labelling and turning-point cross-check routines, and the audit suite.