Global CBAM Exposure, Green Compliance Capability, and Latent Trade-Readiness Regimes: Replication Data and Code, 2007–2024

Published: 25 July 2026| Version 1 | DOI: 10.17632/72c5rjkj3g.1
Contributor:

Description

This dataset and replication package supports a global longitudinal study of carbon-border exposure, green-compliance capability, and trade-readiness mobility under the European Union Carbon Border Adjustment Mechanism (CBAM). The construction panel contains 3,566 country-year observations for 201 non-EU exporting economies and territories over 2007–2024. EU-facing HS07 exports are classified into direct CBAM products, near-CBAM value-chain products, and broader green-compliance products. The baseline weighted exposure measure assigns regulatory-proximity weights of 0.60, 0.25, and 0.15 to these three channels, respectively. The package also contains the Green Compliance Capability Core, constructed from renewable-energy, electricity-access, and green-transition-capacity indicators, together with macroeconomic and structural variables used to examine regime mobility. The primary latent-state sample contains 3,010 observations for 178 economies over 2007–2023. The covariate-dependent transition analysis contains 1,831 observations and 1,695 consecutive one-year transitions for 135 economies. The state analysis ends in 2023 because the capability components required for the baseline index are not sufficiently populated for 2024. The replication workflow estimates two- to six-state hidden Markov models, identifies latent carbon-compliance regimes, calculates posterior state probabilities, transition matrices, expected regime durations, and one-, three-, and five-year mobility probabilities. It also evaluates state-dependent transition effects and performs robustness analyses using alternative exposure weights, capability definitions, state structures, and sample exclusions. The deposited materials include processed and analysis-ready data, product-exposure variables, posterior country-year classifications, model-selection results, transition estimates, robustness outputs, a data dictionary, source-provenance documentation, checksums, Python estimation scripts, R verification and figure-generation scripts, and publication-ready monochrome figures. The package supports numerical verification, figure reproduction, and full model re-estimation. The data should be interpreted as evidence on historical exposure-readiness structures and mobility preceding the definitive financial implementation of CBAM in 2026. They do not constitute a causal estimate of realized post-2026 CBAM effects.

Files

Steps to reproduce

Download all deposited files and extract the replication-package ZIP while preserving the original folder structure. Install Python 3.11 and create the computational environment using 02_CODE/Python/environment.yml, or install the packages listed in 02_CODE/Python/requirements.txt. Open R in the extracted package directory and run source("02_CODE/R/00_INSTALL_R_PACKAGES.R") to install the required R packages. For numerical verification using the frozen reference outputs, run source("02_CODE/R/VERIFY_REPLICATION.R"). This verifies the principal sample sizes, the preferred six-state hidden Markov model, model-selection statistics, the carbon-compliance-trap profile, its self-transition probability, and expected duration. To reproduce the publication figures, run source("02_CODE/R/MAKE_PUBLICATION_FIGURES_PORTABLE.R"). The recreated figures will be saved in 03_OUTPUTS/generated/figures. For complete model re-estimation, run source("02_CODE/R/99_RUN_FULL_REPLICATION.R"). This executes the two- to six-state hidden Markov model selection, covariate-dependent transition estimation, posterior classification, duration and mobility calculations, alternative exposure-weight tests, capability-index robustness checks, and sample-exclusion analyses. Newly generated tables, posterior classifications, verification results, and model outputs will be saved in 03_OUTPUTS/generated. The numerical outputs reported in the associated manuscript are retained in 03_OUTPUTS/frozen for comparison. The full estimation procedure is computationally intensive because the candidate hidden Markov models use multiple random starting values and country-cluster bootstrap replications. Small last-decimal differences may occur across operating systems, Python versions, numerical libraries, and BLAS implementations. The complete construction panel covers 201 economies and territories over 2007–2024. The principal latent-state analysis covers 178 economies over 2007–2023 because the capability components required for the Green Compliance Capability Core are not sufficiently available for 2024. Further variable definitions, transformations, source information, and file descriptions are provided in 04_DOCUMENTATION/DATA_DICTIONARY.csv, SOURCE_PROVENANCE.csv, FILE_MANIFEST.csv, and README.md

Institutions

Categories

Economics, International Trade, Environmentalism

Licence