Building a scalable climate coalition for heavy industry
Data files
Aug 31, 2026 version files 3.21 MB
-
README.md
17.60 KB
-
replication.zip
3.19 MB
Abstract
Since greenhouse gases are a global pollutant, every country benefits from other countries’ mitigation efforts. Recent developments in climate and trade policy create an opportunity to address this coordination problem. This piece describes a climate coalition that builds on that opportunity, initially focused on four emissions-intensive industries: steel, aluminum, cement, and nitrogen fertilizers. Coalition members commit to domestic carbon pricing in covered industries, apply border carbon adjustments to imports from non-members, and deploy complementary incentives to encourage participation by low- and middle-income countries. Using plant-level data and trade modeling, we estimate a potential coalition’s effects on emissions, revenues, output, and prices. The results indicate that a first-wave coalition could achieve meaningful emissions reductions while limiting leakage, preserving competitiveness, respecting fairness, and generating substantial public revenue.
Corresponding author: Catherine Wolfram, cwolfram@mit.edu
Manuscript: Building a Scalable Climate Coalition for Heavy Industry. Science, Policy Forum, 2026. DOI: [10.1126/science.xxxxx]
Replication package DOI: 10.5061/dryad.t1g1jwth4
Package version / date: v1.0, 2026-08-31
1. Overview
This package reproduces all four figures in the main text of the Policy Forum. The analysis models the equilibrium effects of three carbon pricing scenarios — a Current Policy Baseline, a Uniform Coalition Price, and a Graduated Coalition Price — on global production, emissions, tax revenue, and prices in four industrial sectors: aluminum, steel, cement, and ammonia (proxy for nitrogenous fertilizers).
Running bash run_all.sh from the replication/ directory (after unzipping replication.zip) executes the full pipeline end-to-end and writes all figures to figures/. The pipeline has three steps: (1) build model-ready supply and demand inputs from raw plant-level data; (2) solve the partial-equilibrium simulation in Julia; (3) generate the four figures in R. Pre-built intermediate files (data/model/) are included so that figures can be reproduced without re-running the simulation.
2. Data Availability Statement
The simulation draws on several data sources. Raw proprietary sources are not redistributed.
Plant-level emissions and production data (steel, cement, ammonia)
Source: Climate TRACE v4 (2024). Facility-level greenhouse gas monitoring based on satellite imagery, machine learning, and reported data. Freely available at https://climatetrace.org/data. We use the 2023 annual data from the iron and steel, cement, and chemicals (ammonia) sector files. Coverage: 892 steel plants, 2,241 cement plants, 223 ammonia plants globally.
Plant-level production, capacity, cost, and emissions data (primary aluminum)
Source: Wood Mackenzie (proprietary). Plant-level records for all active primary aluminum smelters globally, including total production cost per tonne broken down by component, and Scope 1 and Scope 2 emissions. Licensed; not redistributed. Coverage: 153 smelters globally, year 2023.
Plant-level marginal cost data (steel)
Source: Global Steel Cost Tool (GSCT), maintained by TransitionZero, March 2022 update (2021 data, inflated to 2023 USD using BLS CPI factor 1.12). Freely available at https://www.transitionzero.org/products/global-steel-cost-tracker Unmatched Climate TRACE plants receive imputed costs from country/regional/global means by production mode.
Secondary aluminum production and capacity
Source: World Bureau of Metal Statistics (WBMS), published by the London Stock Exchange Group (LSEG). Monthly production records for 10 major secondary-aluminum-producing countries (Brazil, China, Germany, India, Italy, Japan, South Korea, Norway, Taiwan, United States), 2013–2023. Licensed; not redistributed. Capacity is proxied as maximum annual production in the most recent 10-year window.
Demand data (all sectors)
Source: UN Comtrade (2023; Russia: 2021). Country-level apparent consumption computed as production plus net imports. Trade values and quantities are also used to construct global benchmark prices for steel, cement, and ammonia, and scrap prices for secondary aluminum. Freely available at https://comtradeplus.un.org.
Auxiliary cost and price data
The following additional sources are used to construct plant-level marginal costs and are not redistributed. They are open-access or licensed as noted:
| Source | Use | Access |
|---|---|---|
| World Bank Commodities Price Data ("Pink Sheet") | Aluminum 2023 LME benchmark price | Open access |
| IEA Energy End-Use Prices (2023) | Electricity, natural gas, coal prices (cement, ammonia) | Proprietary |
| BloombergNEF Climatescope 2025 | Electricity prices for non-IEA countries | Open access |
| Mission Possible Partnership (MPP) Ammonia Roadmap | Feedstock intensities, fixed OPEX, electricity use (ammonia) | Open access |
| Vidovic et al. (2023), JRC Technology Report EUR 31653 EN | Feedstock GJ/t NH₃ by region (ammonia) | Open access |
| Refinitiv Workspace (LSEG) | Cement fixed OPEX; Africa/Russia coal prices | Proprietary |
| World Bank Carbon Pricing Dashboard (2023) | Carbon prices as of April 1, 2023; policy coverage | Open access |
| World Bank Income Classification | Coalition membership tier (HIC/UMIC/LMIC/LIC) | Open access |
Pre-built model outputs
data/model/sims_plantlevel.csv — plant-level simulation output produced by code/02_simulate.jl — is included so that code/03_figures.R can be run standalone.
3. Repository structure
replication/
├── DATA_DICTIONARY.xlsx (variable definitions, types, units, and file locations)
├── README.md (this file)
├── run_all.sh (master script — runs all three steps)
├── code/
│ ├── 01_build_inputs.R (plant level inputs → data/model/supply and demand CSVs)
│ ├── 02_simulate.jl (simulation → data/model/sims_plantlevel.csv)
│ └── 03_figures.R (simulation output → figures/)
├── data/
├── inputs/ (plant-level CSVs)
│ ├── aluminum_supply.csv
│ ├── aluminum_demand.csv
│ ├── steel_supply.csv
│ ├── steel_demand.csv
│ ├── cement_supply.csv
│ ├── cement_demand.csv
│ ├── ammonia_supply.csv
│ └── ammonia_demand.csv
└── model/ (intermediate and pre-built files)
├── supply_cop30_v2.csv (output of step 1 — model-ready supply)
├── demand_cop30_v2.csv (output of step 1 — model-ready demand)
├── sims_plantlevel.csv (output of step 2 — plant-level simulation results)
├── alum_countrylevel.csv (sector-level simulation output)
├── steel_countrylevel.csv
├── cement_countrylevel.csv
└── ammonia_countrylevel.csv
4. Computational requirements
Operating system tested: macOS 15 (Apple Silicon, arm64).
Languages and versions:
- R 4.5.1 — packages:
data.table1.17.8,ggplot24.0.1,scales1.4.0,showtext0.9.7,stringr(tidyverse). Install with:install.packages(c("data.table", "ggplot2", "scales", "showtext", "stringr")). - Julia 1.12.0 — packages:
CSV0.10.15,DataFrames1.8.1,Optim1.13.2,StatsBase0.34.7,Distributions0.25.123,FixedEffectModels1.12.0. Install withusing Pkg; Pkg.add(["CSV", "DataFrames", "Optim", "StatsBase", "Distributions", "FixedEffectModels"]).
Hardware: Reproducible on a standard laptop. Total expected runtime: approximately 15–25 minutes for the full pipeline (dominated by the Julia simulation). If using the pre-built sims_plantlevel.csv, step 3 alone runs in under 1 minute.
Random seed: Set in code/02_simulate.jl via Random.seed!(0). The regression coefficient β used in the simulation (0.0023) is hardcoded from a one-time regression run on aluminum and steel data from a previous verion of the Clausing et al., 2025 NBER working paper "The Global Effects of Carbon Border Adjustment Mechanisms".
Proprietary software: None. R and Julia are free and open-source.
5. Instructions to replicate
Full pipeline (requires licensed input data in data/inputs/)
From the replication/ directory:
bash run_all.sh
This runs in sequence:
Rscript code/01_build_inputs.R— reads the 8 raw CSVs fromdata/inputs/, constructs COP30 coalition membership flags, and writesdata/model/supply_cop30_v2.csvanddata/model/demand_cop30_v2.csv.julia code/02_simulate.jl— solves the partial-equilibrium model for three scenarios × four sectors and writesdata/model/sims_plantlevel.csv.Rscript code/03_figures.R— readssims_plantlevel.csvand writes all four figures tofigures/.
Figures only (using pre-built simulation output)
If the licensed plant-level data are unavailable, data/model/sims_plantlevel.csv is already included. To regenerate the figures only:
cd replication
Rscript code/03_figures.R
Figures are written to figures/ as both PNG (300 dpi) and SVG.
6. Map of scripts to outputs
| Figure in paper | Script | Key inputs | Output files |
|---|---|---|---|
| Figure 1a — GHG change by coalition membership | code/03_figures.R |
data/model/sims_plantlevel.csv |
figures/fig_1a.{png,svg}, figures/data_fig_1a.csv |
| Figure 1b — Carbon revenue by scenario | code/03_figures.R |
data/model/sims_plantlevel.csv |
figures/fig_1b.{png,svg}, figures/data_fig_1b.csv |
| Figure 1c — Output change by income group | code/03_figures.R |
data/model/sims_plantlevel.csv |
figures/fig_1c.{png,svg}, figures/data_fig_1c.csv |
| Figure 1d — Price change by sector | code/03_figures.R |
data/model/sims_plantlevel.csv |
figures/fig_1d.{png,svg}, figures/data_fig_1d.csv |
sims_plantlevel.csv |
code/02_simulate.jl |
data/model/supply_cop30_v2.csv, data/model/demand_cop30_v2.csv |
data/model/sims_plantlevel.csv, data/model/{sector}_countrylevel.csv |
supply_cop30_v2.csv, demand_cop30_v2.csv |
code/01_build_inputs.R |
data/inputs/{sector}_{supply,demand}.csv (×8) |
data/model/supply_cop30_v2.csv, data/model/demand_cop30_v2.csv |
8. Data citations
- Climate TRACE. (2024). Climate TRACE greenhouse gas emissions data, v4 [Dataset]. Climate TRACE Coalition. https://climatetrace.org/data (Accessed 2024. Steel, cement, ammonia sectors.)
- Wood Mackenzie. (2023). Aluminum smelter cost and emissions data [Dataset]. Wood Mackenzie Ltd. Proprietary; not redistributed. (Primary aluminum supply data.)
- TransitionZero. (2022). Global Steel Cost Tool (GSCT), March 2022 update [Dataset]. TransitionZero. Proprietary; not redistributed. (Steel marginal cost data, 2021, inflated to 2023 USD.)
- World Bureau of Metal Statistics (WBMS) / London Stock Exchange Group. (2023). Secondary aluminum production and capacity [Dataset]. LSEG. Proprietary; not redistributed. (10 countries, 2013–2023.)
- United Nations Statistics Division. (2024). UN Comtrade database [Dataset]. United Nations. https://comtradeplus.un.org (Trade values and quantities for all sectors, 2023; Russia: 2021. Used for demand construction and global price benchmarks.)
- World Bank. (2024). Commodity markets: Pink Sheet [Dataset]. World Bank Group. https://www.worldbank.org/en/research/commodity-markets (Aluminum 2023 LME cash price benchmark.)
- International Energy Agency. (2025). Energy prices (June 2025 ed.) [Database]. IEA, Paris. https://www.iea.org/data-and-statistics/data-product/energy-prices (Electricity, natural gas, coal prices for cement and ammonia cost models, 2023.)
- BloombergNEF. (2025). Climatescope 2025: Emerging markets power transition factbook (14th ed.). Bloomberg L.P. https://www.global-climatescope.org/downloads/CS2025_20251103.pdf (Electricity prices for non-IEA countries including China, Russia, Vietnam, Egypt, Saudi Arabia, Mexico, Nigeria, UAE, Thailand, Philippines, Iraq, Mozambique, Togo, Cameroon, Zambia.)
- Mission Possible Partnership. (2022). Making net-zero ammonia possible: An industry-backed, 1.5°C-aligned transition strategy. Mission Possible Partnership. https://www.energy-transitions.org/wp-content/uploads/2022/09/Making-1.5-Aligned-Ammonia-possible.pdf (Ammonia feedstock intensities, fixed OPEX, electricity use by region.)
- Vidovic, D., Marmier, A., Zore, L., & Moya, J. (2023). Greenhouse gas emission intensities of the steel, fertilisers, aluminium and cement industries in the EU and its main trading partners (EUR 31653 EN, JRC134682). Publications Office of the European Union. https://doi.org/10.2760/359533 (Ammonia feedstock GJ/t NH₃ by region.)
- Refinitiv / London Stock Exchange Group. (2025). Refinitiv Workspace [Database]. LSEG. https://workspace.refinitiv.com (Cement industry fixed OPEX; Africa and Russia coal prices, 2023 annual average. Accessed June 2025.)
- World Bank. (2025). State and trends of carbon pricing dashboard [Database]. World Bank Group. https://carbonpricingdashboard.worldbank.org (Carbon prices and policy coverage as of April 1, 2023.)
- World Bank. (2024). World Bank country and lending groups [Dataset]. World Bank Group. https://datahelpdesk.worldbank.org/knowledgebase/articles/906519 (Income classification used to assign coalition membership tiers HIC/UMIC/LMIC/LIC.)
- Clausing, K. A., Colmer, J. M., Hsiao, A., & Wolfram, C. (2025). The global effects of carbon border adjustment mechanisms (NBER Working Paper No. 33723). National Bureau of Economic Research. https://doi.org/10.3386/w33723 (Source of the β = 0.00234 coefficient used in the logit supply model, estimated on aluminum and steel data.)
- Global Climate Policy Project Working Group on Climate Coalitions. (2025). Building a climate coalition: Aligning carbon pricing, trade, and development. Global Climate Policy Project at Harvard and MIT. https://salatainstitute.harvard.edu/building-a-climate-coalition-gcpp-flagship-report/ (Flagship report on which this Policy Forum is based.)
9. Software citations
- R Core Team (2025). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Version 4.5.1. https://www.R-project.org/.
- Bezanson, J., Edelman, A., Karpinski, S., & Shah, V. B. (2017). Julia: A fresh approach to numerical computing. SIAM Review, 59(1), 65–98. Version 1.12.0. https://julialang.org.
- Dowle, M., & Srinivasan, A. (2024). data.table: Extension of
data.frame. R package version 1.17.8. https://CRAN.R-project.org/package=data.table. - Wickham, H. (2016). ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag. Version 4.0.1.
- Gomez, D. (2024). showtext: Using Fonts More Easily in R Graphs. R package version 0.9.7.
- Kwok, T., & Magnusson, A. et al. (2024). FixedEffectModels.jl. Version 1.12.0. https://github.com/FixedEffects/FixedEffectModels.jl.
10. Restrictions, deviations, and notes
- The figures produced by this package are numerically identical to those in the manuscript but may differ slightly in appearance. The published versions were post-processed in Adobe Illustrator for final typesetting (e.g., font adjustments, label positioning, whitespace). All data, colors, and values are unchanged.
- The FuturaPT font files in
fonts/are required to reproduce the exact figure typography. If unavailable,showtextwill fall back to a system font; figures will be numerically identical but typographically different. - The figures included in this replication package are the authors’ initial versions, prior to copy editing and graphic design. The final published figures may therefore differ in formatting, typography, labeling, or layout, but the underlying data and numerical results are unchanged.
11. Contact
Questions or replication issues: pereboom@mit.edu.