Who emits, what it costs, who bears the income impact, and where emissions may go: the United States and Bangladesh against 190+ economies, 1960–2024 actuals and scenarios to 2100
Open the interactive dashboard: world maps of 13 indicators, an animated emissions map, a scenario explorer for every economy, and an income-vs-emissions view.
Stack: Python · pandas · statsmodels · scikit-learn · SciPy · matplotlib · Plotly.js · Jupyter · pytest
Situation. Emissions are set by a small number of large economies, but the income impact of the resulting warming is expected to fall unevenly. Lenders, investors and corporates need country-level numbers they can defend.
Complication. The available evidence is dated (much public analysis stops at 2016–2018), comes from sources that disagree, and is usually reported without uncertainty. The two economies studied here sit at opposite ends: the United States (advanced, high-income, high-emitting) and Bangladesh (emerging, lower-middle-income, low-emitting, highly climate-exposed).
Question. What drives national emissions, how far can they be forecast, what do they cost, who bears the risk, and what does development need to look like?
Audience. A country-risk, sustainable-finance or strategy team that needs defensible numbers with their ranges for framing the emissions/exposure gap.
| # | Finding | So what |
|---|---|---|
| 1 | The per-person gap is 23× (US 14.2 t vs Bangladesh 0.62 t in 2024), down from 83× in 2004. | The gap is closing, but through US decline (−32% per person) and Bangladesh growth (×2.5), not convergence at a common level. |
| 2 | Emissions still follow income, less tightly. Elasticity 1.06 in 2024 vs 1.29 in 2004. Bangladesh emits 0.54× what its income predicts; the US 1.21×. | Income is a weakening proxy for emissions: efficiency and energy mix matter more each year. |
| 3 | Country emissions are only moderately forecastable. Error is ~4% at 1 year, ~11% at 5, ~18% at 9. Over the 2018→2024 test, trend models over-forecast every year (world +4.5%, US +8%, Bangladesh +15% in 2024). | Use ranges and scenarios, not point forecasts; and treat model bias as a risk in its own right. |
| 4 | Even fast decarbonisation overshoots a 2 °C budget. Cumulative fossil CO₂ 2025–2100: Stress 6,987 Gt, Base 2,934 Gt, Fast 1,556 Gt vs about 940 Gt remaining for 2 °C (AR6, 67% chance, net of 2020–24 use). Staying within the 2 °C budget would need carbon intensity to fall about 6.2% a year in every economy from 2025. | The gap is structural; the question is how large a transition and physical risk to plan for, not whether one exists. |
| 5 | Carbon liability is large and not highest where income is highest. At $206/t, 2024 emissions carry $1.0 tn (4.1% of GDP) for the US and $22 bn (5.9%) for Bangladesh. China is 16%, India 21% of GDP. World: $7.6 tn (8.2% of GDP). | Relative to output, the exposure sits with carbon-intensive middle-income economies. |
| 6 | A fast transition is worth about $204 tn in present value versus the Base case (range $109–381 tn across EPA discount rates), about 2.2 years of world output. | This is avoided damage only; mitigation cost is not modelled. |
| 7 | Income at risk is far from where emissions are. Low- and lower-middle-income economies hold 44% of people and about 40% of income at risk but only 6.5% of cumulative emissions. | The physical-risk burden is borne mostly by the economies with the smallest balance-sheet capacity to absorb it. |
| 8 | Bangladesh's 2100 income impact (Base case) depends on one assumption: −1.1% if a temperature penalty is a one-off level change, −38% if it compounds (90% interval −51% to −21%). | Persistence is the dominant model risk; the data cannot yet resolve it, so both readings are shown. |
| Stakeholder | Use | Caution |
|---|---|---|
| Sovereign / country-risk analysts | Carbon liability as a share of GDP (fig. 21) and income-at-risk maps as screening indicators for transition and physical risk. | Treat the growth reading as a stress test, not a central case; the sign of the impact for cool, high-income economies (e.g. the US, +20%) is not robust. |
| ESG and climate-risk investors | Per-person emissions vs income (fig. 11), decoupling chart (fig. 13) and forecast ranges to judge whether an economy's trajectory is credible. | Territorial emissions omit those embodied in trade; check consumption-based figures (fig. 8). |
| Development lenders | Where adaptation finance protects the most income: decarbonising avoids a median 9.6 percentage points of 2100 income loss in low-income economies (Bangladesh 7.7). | Income effects only: sea-level rise, cyclones and mortality are excluded. |
| Corporate strategy / carbon pricing | The shadow carbon price range ($120–$340/t in 2020, EPA) and its present-value effect under three scenarios. | A modelled global value; it says nothing about who bears the damage. |
The project is a six-stage pipeline. Each notebook states its question, method, result, and limitations, and hands a cleaned table to the next.
| Stage | Notebook | Question | Method |
|---|---|---|---|
| Data | 01_data_foundation |
Is the data current, complete and consistent? | Multi-source reconciliation, coverage and missing-data audit, outlier review, master panel |
| Diagnostic | 02_case_study_us_vs_bangladesh |
What drove each economy's emissions? | Kaya decomposition, year-on-year correlation with detrending, consumption vs territorial |
| Diagnostic | 03_global_analysis |
Do the two economies generalise? | Log-log elasticities, fixed-effects panel, LOWESS, decoupling, Lorenz/Gini, attribution |
| Predictive | 04_emissions_forecasting |
How far can emissions be forecast? | Rolling-origin backtest of six models, empirical intervals, 2018→2024 out-of-sample test, scenarios, carbon budgets |
| Financial | 05_financial_and_socioeconomic_impact |
What do emissions and decarbonisation cost, and who is exposed? | Social cost of carbon valuation, present values, per-capita parity, exposure indicators |
| Risk | 06_climate_risk_by_country |
Which economies lose income from warming? | Replication of Burke–Hsiang–Miguel (2015), pattern scaling, coefficient-draw uncertainty |
Supporting material: an interactive dashboard built by code/build_dashboard.py, 35 unit and data-integrity tests, and a downloader that records source, timestamp and checksum for every file.
| Scenario | Carbon-intensity trend | Purpose |
|---|---|---|
| Stress case | Each economy's carbon intensity of GDP stops improving | Stress test, not a forecast |
| Base case | Each economy's 2010–19 trend continues | Reference path |
| Fast transition | Every economy reaches, within 10 years, the pace of the fastest tenth in 2010–19 (about −4.2% a year) | What rapid action would achieve |
Each economy's income growth starts at its 2010–19 rate and fades to 1.5% a year over 30 years in all three; population follows UN WPP 2024 (medium variant).
How the two economies moved (2004–2024). US emissions fell because carbon per dollar of GDP fell 47%, more than offsetting population (+18%) and income (+28%). Bangladesh's rose ×3.0 on income (×2.76) and population (×1.21).
Decoupling is real but not at scale. 64 of 188 economies (34%) grew income while cutting per-person emissions, but they hold 26% of world population, and world emissions still rose 36% between 2004 and 2024.
Forecast error grows with the horizon, and the 2018→2024 test shows the bias.
Scenarios to 2100 and the carbon budget.
Carbon liability relative to output.
The carbon cost of a long life has fallen about 3.5× since 1990 (3.4 t per person for a life expectancy of 70 in 1990, 1.0 t in 2023).
Income at risk versus share of cumulative emissions.
Bangladesh and the United States to 2100, under both readings of the temperature effect.
All 31 figures are in figures/, numbered in the order the notebooks produce them.
Recency was the first requirement: the previous version of this project ended in 2018. Every headline number now uses 2024 data.
| Control | Result |
|---|---|
| Recency | Emissions to 2024 (Global Carbon Project via Our World in Data); economic and health data to 2024 (World Bank, UN WPP 2024) |
| Coverage | 192 of 218 economies have complete 2024 data, holding 97.9% of world population and 98.6% of national CO₂ |
| Independent CO₂ check | Global Carbon Project vs EDGAR (World Bank): log-correlation 0.995, median gap 5.2%. Bangladesh is 14.7% higher in EDGAR; the US 5.5% lower |
| Second source for health | World Bank vs UN WPP life expectancy: 95% within 1 year |
| Second source for income growth | World Bank vs Maddison Project: correlation 0.95 |
| Known break | Central African Republic life expectancy from 2009 is an upstream UN error, flagged and excluded |
| Missing data | No imputation; sample sizes reported next to each result; sparse indicators used as latest value with the year observed |
| Provenance | manifest.json records URL, retrieval time and SHA-256 for every downloaded file |
| Tests | 35 tests: country-name joins, Kaya identity, forecast models, damage function, master-panel integrity |
Details, the data dictionary and the source list are in data/README.md.
| Area | Assumption | Sensitivity or mitigation |
|---|---|---|
| Forecasting | Country trends from 2000–2019 data are informative | Backtest error reported by horizon; 80% intervals under-cover (65–76%) and are labelled as such |
| Scenarios | Smooth country trends (2010–19); long-run income growth of 1.5% | Scenarios are not forecasts; ±1 point on long-run growth is tested |
| Valuation | EPA social cost of carbon, central 2.0% discount rate ($206/t in 2024) | Range $120–$340/t in 2020; present values shown at all three discount rates |
| Currency | 2024 GDP restated to 2020 dollars with the GDP deflator (119.0 for the US) | Ratio to GDP uses market exchange rates, which inflates ratios for lower-income economies |
| Damage function | Temperature-growth parabola estimated from repo data (peak 13.1 °C, published 13 °C) | Extended sample moves the peak to 6.1 °C; two persistence readings and coefficient draws shown; net global totals not used |
| Warming | 0.45 °C per 1,000 Gt CO₂ (IPCC AR6), fossil CO₂ plus declining land-use CO₂ | Non-CO₂ gases excluded, so warming is understated |
| Exposure | Elevation below 5 m as a sea-level proxy | Last measured in 2015; ignores defences |
| Grouping | Current World Bank income group applied to all years | Stated wherever it matters |
- Association, not causation in the cross-country results; energy mix, geography and trade are not modelled.
- Territorial emissions attribute offshored production to the producer.
- Scenarios are not forecasts, and fossil CO₂ only (no methane, no other gases), so budget comparisons are lenient.
- Income effects only in the risk model: no sea-level rise, cyclones, mortality, displacement or adaptation. Country data cannot see Bangladesh's coastal districts.
- Mitigation cost is not modelled, so the value of a fast transition is a benefit, not a cost-benefit result.
- Next: consumption-based accounting for all economies, energy-mix data to explain why carbon intensity falls, sea-level exposure by country, mitigation-cost curves, and a slide deck for stakeholders.
| Term | Meaning |
|---|---|
| Carbon liability | 2024 emissions valued at the social cost of carbon; a measure of global damage caused, not damage borne |
| Shadow carbon price / SC-CO₂ | The modelled dollar value of the damage from one extra tonne of CO₂ (EPA 2023) |
| Transition risk | Exposure to the cost of moving to a low-carbon economy |
| Physical risk | Exposure to climate impacts such as heat, floods and sea-level rise |
| Income at risk | Modelled percentage change in income per person by 2100 relative to today's climate (negative = loss) |
| Kaya decomposition | Emissions = population × GDP per person × CO₂ per unit of GDP |
| Decoupling | Income per person rising while per-person emissions fall |
| Elasticity | Percentage change in emissions for a 1% change in income |
| PPP | Purchasing-power-parity adjusted income, comparable across economies |
Climate-Change/
├── README.md
├── requirements.txt
├── docs/index.html interactive dashboard (GitHub Pages)
├── code/
│ ├── 01_data_foundation.ipynb … 06_climate_risk_by_country.ipynb
│ ├── climate_utils.py loaders, statistics, plot style
│ ├── forecast_utils.py forecast models and scenario engine
│ ├── damage_utils.py temperature-growth response and income-impact functions
│ ├── download_data.py fetches external data, writes manifest
│ ├── build_dashboard.py builds docs/index.html
│ └── dashboard_template.html
├── data/
│ ├── raw/external/ downloaded sources + manifest.json
│ ├── raw/reference/ EPA social cost of carbon, carbon budgets, UN population projections
│ ├── processed/ tables written by the notebooks
│ └── README.md provenance and data dictionary
├── figures/ 31 figures + dashboard preview
└── tests/ 35 tests
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cd code
python download_data.py # optional: the downloaded files are already in data/raw/external/
jupyter nbconvert --to notebook --execute --inplace 01_data_foundation.ipynb 02_case_study_us_vs_bangladesh.ipynb \
03_global_analysis.ipynb 04_emissions_forecasting.ipynb 05_financial_and_socioeconomic_impact.ipynb 06_climate_risk_by_country.ipynb
python build_dashboard.py # rebuilds docs/index.html
cd .. && pytest -q # 35 testsRun the notebooks in order: 01 writes the master panel, 04 writes the scenario paths, 06 writes the damage paths. Re-running download_data.py refreshes the sources, so results can shift slightly as they are revised.
- Our World in Data, CO₂ dataset (Global Carbon Project) and country temperature (Copernicus ERA5); UN World Population Prospects 2024 via Our World in Data.
- World Bank, World Development Indicators via the public API (GDP, GDP deflator, population, life expectancy, mortality, poverty, Gini, electricity, energy, agriculture, elevation exposure, EDGAR-based CO₂, income classifications).
- World Bank, Country Climate and Development Report for Bangladesh (2022).
- U.S. EPA, Report on the Social Cost of Greenhouse Gases (Dec 2023), Table ES.1.
- IPCC, AR6 Working Group I (remaining carbon budgets, TCRE); Forster et al., Indicators of Global Climate Change (2025).
- Burke, Hsiang & Miguel (2015), Global non-linear effect of temperature on economic production, Nature 527.
- Gapminder, UN population projections to 2100 (UN WPP 2024, medium variant); Maddison Project via Our World in Data (income cross-check).