The Winners and Losers of Climate Policies: A Sufficient Statistics Approach

Layer 1: Overview

This paper asks who wins and loses from climate policies — carbon taxes, renewable subsidies, and carbon tariffs — across 193 heterogeneous countries, and by how much. The motivation is that the standard IAM literature aggregates welfare into a global number, obscuring the distributional structure that determines political feasibility. Without knowing which countries gain and lose, and through which channels, it is impossible to understand why international cooperation is so difficult or which club structures can sustain themselves.

The authors build a static Integrated Assessment Model (IAM) with heterogeneous countries, international trade in goods (Armington CES), international trade in fluid fossil (oil and gas), locally traded coal, and locally supplied renewables. Production uses a nested CES combining labour with a composite of three energy types. A reduced-form climate system maps world emissions linearly to global temperature, then to country-specific local temperatures, which damage TFP through a quadratic damage function. The key methodological contribution is a first-order (log-linear) decomposition of welfare around the current equilibrium, which expresses welfare changes analytically as a function of five observable sufficient statistics: (i) direct TFP damage, (ii) export terms-of-trade, (iii) import price index, (iv) energy cost effects (change in energy prices faced by producers), and (v) energy rent effects (change in profits of domestic fossil and renewable producers). This decomposition requires no model simulation; it reads off welfare directly from observables and a small set of elasticities.

Two sets of structural parameters are estimated. First, a structural damage function is estimated using bilateral trade data from the ITPD-E dataset (2000–2016, 169 countries) via a Poisson pseudo-maximum-likelihood gravity regression that instruments temperature shocks against within-trading-partner variation in import penetration, controlling for energy market effects. The preferred specification recovers a global peak temperature of T* = 14.02°C and a damage slope parameter γ = 0.012. This strategy is designed to be robust to the Lucas critique: unlike reduced-form GDP regressions, it nets out general-equilibrium spillovers through trade and energy channels. Second, country-specific energy supply elasticities for oil-gas and coal are estimated from time-series variation in fossil rent shares and international prices (1985–2019 data), using OLS country-by-country and then an empirical Bayes shrinkage procedure with a truncated-normal prior that enforces positive elasticities. Coal is found to be substantially more elastically supplied than oil-gas; OPEC nations (e.g., Saudi Arabia) have near-inelastic oil-gas supply, while the US has relatively elastic supply.

Key quantitative results from the policy experiments follow. (1) Business-as-usual: a 3°C warming by 2100 generates a 17% loss in consumption-equivalent world welfare under utilitarian weights, implying a Social Cost of Carbon of $203/tCO₂ at the current equilibrium point-of-approximation, rising to $302/tCO₂ if computed at 3°C of warming. Under Negishi (income-proportional) weights, the SCC falls to $3.31, reflecting that damages are concentrated in low-income countries with high marginal utility. Winners include Canada and Russia; losers are concentrated in Africa, Latin America, and South-East Asia. (2) Unilateral carbon tax (China, $50/tonne): global emissions rise by less than 0.07% (not fall) because China’s carbon tax shifts its energy mix from coal toward oil-gas (coal is ~1.44× dirtier per unit of energy), raising the international oil-gas price by approximately 5%, which boosts fossil exporters’ rents and induces other countries to substitute back to coal. Global utilitarian welfare falls by 0.2%. China itself gains on net through falling coal prices and improved terms of trade. EU nations lose from higher energy import costs. (3) Unilateral carbon tax (USA, $50/tonne): global emissions fall by 0.8%; US welfare effects are small but positive (energy cost increases largely offset by terms-of-trade gains with Canada and Europe). (4) Renewable subsidies (42.6%, calibrated to produce the same average relative-price shift as a $50 carbon tax): on average substantially less effective than carbon taxation and more harmful to welfare because subsidies push countries up their upward-sloping domestic renewable supply curves, wasting resources on costly domestic generation (especially in countries with high baseline renewable shares such as France). (5) EU climate club ($50 carbon tax + CBAM tariffs): global emissions fall by 3%; global utilitarian welfare rises by around 5% (1% under Negishi weights), but the EU itself is a net loser — only Southern Europe (Spain, Portugal, Italy) gains; Germany and Scandinavian nations lose both from direct policy costs and from cooling that harms countries that benefit from warming. Oil-gas price falls by 4.6% within the club. (6) ASEAN climate club (same structure): global emissions fall by 0.5%; global utilitarian welfare rises by about 0.8% (0.2% Negishi); ASEAN members broadly benefit because they are already losers from climate change and the carbon-reduction benefit outweighs policy costs. Oil-gas price falls by 0.6%. (7) Global $50 carbon tax (all 193 countries): global emissions fall by 3.82%; global oil-gas price rises by 0.96% (substitution from coal toward oil-gas under a global carbon tax); global utilitarian welfare rises by about 6% (1% Negishi). Most of the utilitarian gain reflects reduced international inequality, since benefits concentrate in low-income tropical countries. Fossil exporters such as Saudi Arabia and Nigeria see energy rents rise as coal is substituted for by oil-gas globally.

The central mechanism finding is that leakage operates primarily through energy trade, not goods trade: energy market effects are consistently larger than goods-market terms-of-trade effects across all policy experiments. This quantifies why unilateral climate policy is so limited in effectiveness. International coordination through climate clubs overcomes leakage but creates winners and losers within member coalitions depending on each member’s energy mix, trade exposure, and baseline climate damage.

Layer 2: Deep Dive

What is the identification strategy for the structural damage function and what are the main threats to it?

The authors estimate the damage function using a Poisson pseudo-maximum-likelihood gravity regression on bilateral import penetration ratios (Xij/Xii) as a function of temperature differences between exporters and importers (and their squares), with country-pair fixed effects and year fixed effects. Controls for GDP/capita (polynomial), oil rent share, and renewable energy share proxy for the time-varying component of factory-gate prices driven by energy prices and wages. The key identifying assumption is that conditional on these controls and fixed effects, temperature shocks are uncorrelated with time-varying bilateral preference or cost shifters. Threats include: (1) confounding time-varying bilateral shocks correlated with temperature, such as ENSO events or specific geopolitical shocks; (2) the possibility that global (rather than local) temperature drives damages, which the paper cannot address given limited time-series variation and potential spurious correlation concerns (following Goulet Coulombe and Klieber, 2025); (3) the treatment of θ = 5 as a known parameter in computing γ from the regression coefficient, which propagates calibration error. The authors argue their strategy is robust to the Lucas critique because it nets out general-equilibrium effects on GDP that would contaminate GDP-based damage regressions.

How does the paper’s welfare decomposition work and what are its five channels?

The welfare decomposition is a first-order log-linearisation of the indirect utility around the current equilibrium. Changes in consumption-equivalent welfare for country i decompose into: (i) direct climate TFP damage (change in Dy_i); (ii) export terms-of-trade effect (change in domestic good price p_i); (iii) import price-index effect (change in price index P_i); (iv) energy cost effects (changes in oil-gas price q^f, coal price q^c_i, and renewable price q^r_i weighted by their shares in production); and (v) energy rent effects (changes in profits from fossil, coal, and renewable extraction weighted by their shares in household income). The key insight is that none of these five terms requires solving the full model; each can be computed from observable data moments (energy mix, energy rent shares, trade shares) and a small number of estimated or calibrated elasticities.

What heterogeneity in climate damages is documented and what drives it?

Winners from climate change (3°C warming) are primarily cold countries: Canada, Russia, Scandinavian nations. Losers are concentrated in Africa (Djibouti, Niger, Burkina Faso, Sudan), Latin America, and South-East Asia. The heterogeneity arises from: (1) differences in baseline temperature relative to the estimated global peak productivity temperature T* = 14.02°C; countries hotter than T* lose productivity with further warming, while colder countries gain; (2) partial local adaptation (αT = 0.5) so each country’s effective peak temperature is halfway between T* and its current local temperature; (3) indirect effects through trade networks — cold, open economies can lose if major trading partners are damaged; (4) energy rent effects — fossil exporters lose energy rents as warming reduces global energy demand, partially offsetting their direct productivity gains.

Why does China’s unilateral carbon tax at $50/tonne raise global emissions rather than lower them?

China relies heavily on coal, which has a carbon concentration ratio of approximately ξc/ξf ≈ 1.44 (coal is ~44% dirtier per unit energy than oil-gas). A carbon tax on both fuels raises the effective cost of coal more than oil-gas, inducing China to substitute toward oil-gas imports. This raises the international oil-gas price by approximately 5%, which: (1) increases energy rents for fossil exporters (Gulf states, Russia) and (2) makes oil-gas costlier for other countries, incentivising them to substitute back toward coal. The net effect on global emissions is a slight increase of less than 0.07%, rather than a decline. This is the carbon leakage effect operating through energy trade.

Why are renewable subsidies substantially less effective than carbon taxes?

Several mechanisms distinguish the two policies. First, a carbon tax directly raises the relative price of all fossil fuels versus renewables and pushes production up the upward-sloping renewable supply curve only modestly. A renewable subsidy instead directly subsidises a reduction in the cost of renewables, which expands renewable supply — but this requires moving up the domestic renewable supply curve, wasting real resources in countries where the marginal renewable site is expensive (e.g., France with over 40% baseline renewable share). Second, a carbon tax creates a reallocation from coal to oil-gas (since the tax raises the coal price more per unit of energy), which can inadvertently raise oil-gas prices and redistribute income to exporters. A renewable subsidy does not have this feature in the same way. Third, the lump-sum financing of subsidies has a direct income cost, while carbon tax revenues are rebated, so only general equilibrium price effects matter for welfare. On average across countries, renewable subsidies cause more harm and generate smaller emission reductions per dollar.

What is the distinction between the EU and ASEAN climate clubs, and why do outcomes differ so substantially?

The EU club ($50 carbon tax + CBAM on imports from non-members) reduces global emissions by 3%, raises global utilitarian welfare by about 5%, but makes EU members net losers on average. The reason is that EU countries include many cold nations (Germany, Scandinavia) that benefit from warming; by cooling the climate, the policy harms them. Additionally, energy cost effects within the EU are heterogeneous — energy costs rise in France but fall in Poland and Germany — and Ireland is harmed through goods trade with Great Britain. The ASEAN club reduces global emissions by only 0.5% (ASEAN is smaller and less fossil-intensive in global terms), raises global utilitarian welfare by 0.8%, and ASEAN members broadly benefit because: (1) all ASEAN members are in the tropical/sub-tropical zone and thus lose from warming; (2) reducing global temperature yields direct productivity gains for members; (3) the energy rent loss for fossil exporters within ASEAN (Brunei, Indonesia) is outweighed by the climate benefit for others. The key structural difference is that the ASEAN club’s members are already losers from warming and hence have aligned incentives for carbon reduction.

What is the Social Cost of Carbon computed in this framework and how does it vary with assumptions?

Under utilitarian Pareto weights (ωi = 1, equal weight per person) and a 3°C warming by 2100, the global consumption-equivalent welfare loss is 17%, implying SCC = $203/tCO₂ at the current baseline temperature. Changing the point of linearisation to the 3°C warmer world raises the SCC to $302/tCO₂, indicating that damages accelerate as warming progresses and that the baseline approximation understates future costs. Under Negishi weights (proportional to income, ωi ∝ 1/u’(ci)), the SCC falls dramatically to $3.31/tCO₂, because damages are concentrated in low-income countries which receive little weight under income-proportional welfare aggregation. The authors note their static, log-linearised model provides a lower bound: fully dynamic IAMs with nonlinearities, uncertainty, or catastrophic-tail risks would further raise the SCC.

How does the paper estimate energy supply elasticities and what are the key findings?

The authors regress changes in the oil-gas rent share of GDP on changes in the international oil-gas price (and changes in GDP as a control) country-by-country using first differences, recovering country-specific supply elasticities. Because some OLS estimates are noisy, negative, or below 1 (implying negative supply elasticity, inconsistent with theory), the authors apply an empirical Bayes shrinkage procedure: they impose a truncated-normal prior (truncated below 1) whose hyperparameters come from a pooled regression, and compute the posterior mean for each country. Key findings: oil-gas supply is nearly inelastic in OPEC nations (Saudi Arabia) and Russia and China, consistent with market power compressing effective supply elasticity; the US has relatively elastic oil-gas supply. Coal supply is substantially more elastic on average than oil-gas; the US and India have relatively inelastic coal supply; Russia and China have more elastic coal supply. Coal rents never exceed 1% of GDP even in the largest producers, consistent with near-competitive flat supply curves. These spatial patterns matter significantly for which countries gain or lose from energy price changes induced by climate policy.

What is the main mechanism through which leakage operates — energy trade or goods trade — and how is this established?

The paper establishes that energy market effects are consistently larger in magnitude than goods-market terms-of-trade effects across all policy experiments (see Appendix Table A3). Leakage through energy trade operates because: (1) a domestic carbon tax reduces domestic demand for fossil fuels, lowering the international price of oil-gas (for small countries) or shifting demand between fuels; (2) lower oil-gas prices benefit importing countries and encourage them to use more fossil fuels, partially offsetting the original emission reduction. Goods-market leakage (productivity and competitiveness effects through the trade network) exists but is secondary. This finding has implications for policy: carbon border adjustment mechanisms (CBAMs) target goods trade leakage, but the model suggests the larger channel — energy trade leakage — is not addressed by CBAM alone.

What robustness checks or sensitivity analyses does the paper report?

The paper reports several robustness exercises: (1) The damage function estimation reports results under OLS (Columns 1-2) and Poisson (Columns 3-4), with separate or restricted coefficients on importer and exporter temperatures; the preferred Poisson specification with restricted coefficients yields T* = 14.02 and γ = 0.012, and the separate-coefficient specification yields statistically indistinguishable estimates. (2) The SCC is computed at two points of approximation — the current baseline and a 3°C warmer world — yielding $203 and $302/tCO₂ respectively, giving a sense of nonlinearity bias from log-linearisation. (3) Welfare is reported under both utilitarian (ωi = 1) and Negishi (ωi ∝ 1/u’(ci)) weights throughout, and the results differ sharply, highlighting how inequality weighting matters. (4) The partial local adaptation parameter αT = 0.5 nests pure global peak (αT = 1) and pure local baseline (αT = 0) damage specifications. (5) Appendix Table A3 provides a comprehensive decomposition of welfare into climate, energy, and trade effects for all six policy scenarios (BAU, global carbon tax, China tax, US tax, EU club, ASEAN club), enabling consistency checks across experiments.

How does this paper relate to the broader literature on IAMs and sufficient statistics?

The paper makes three connections. First, it is related to the large IAM literature (Nordhaus and Yang 1996; Barrage and Nordhaus 2024; Cruz and Rossi-Hansberg 2024) but differs by explicitly decomposing welfare into observable sufficient statistics, avoiding the need to solve a large dynamic system. Second, it is related to the sufficient statistics literature in trade (Lashkaripour 2021 on trade wars; Baqaee and Farhi 2024 on trade barriers; Kleinman, Liu, and Redding 2024 on productivity shocks in trade models) — the paper extends this approach to a broad set of climate instruments in a model with detailed energy markets. Third, it differs from Bourany (2025) — a companion paper by one author — which solves for optimal climate agreement design; the present paper instead uses sufficient statistics to evaluate many given policies, trading optimality for analytical tractability and decomposability. The paper also distinguishes from Krusell and Smith (2022), which does not allow cross-border energy trade, and from Cruz and Rossi-Hansberg (2024), which does not model heterogeneous energy rents across space.

What are the scope conditions and limitations of the approach?

Scope conditions and limitations are significant. (1) The model is static, so it cannot capture dynamic considerations: optimal intertemporal extraction paths, green paradox effects (whether carbon taxes accelerate fossil extraction), directed innovation toward renewables, adaptation capital accumulation, or dynamic leakage in energy markets. (2) The first-order log-linearisation abstracts from nonlinearities in the climate system, making the results most relevant as marginal effects near the current equilibrium rather than for large climate-policy changes or for evaluating policies at future, warmer states of the world. (3) The paper does not model market power in international energy markets (OPEC behaviour), abstracting from strategic behaviour by fossil exporters. (4) Labour is internationally immobile, so migration as a margin of adaptation is excluded. (5) Utility damages from climate change (mortality, amenity loss) are excluded — only productivity (TFP) damages are modelled; including utility damages would amplify gains and losses proportionally. (6) The framework cannot evaluate dynamic policy environments such as climate coordination with commitment problems or intergenerational redistribution from carbon taxation.

What are the policy implications of the paper’s findings?

Several policy implications follow from the paper’s results, with important scope conditions. (1) Unilateral climate policy is largely ineffective for reducing global emissions and can even increase them (as in China’s carbon tax case); the standard free-rider analysis understates the problem because energy-market leakage can reverse the direction of emissions. (2) Renewable energy subsidies are generally a worse policy instrument than carbon taxes, because they push countries up costly domestic supply curves rather than reallocating away from fossil fuels through price signals; policy prescriptions that favour subsidies (such as the US Inflation Reduction Act) should account for this comparative inefficiency. (3) Climate clubs with both a domestic carbon tax and carbon tariffs (CBAMs) can overcome leakage effects and yield positive global welfare gains, but impose net costs on members whose composition makes them net losers from cooling (cold, energy-exporting member nations). This suggests club membership incentives are heterogeneous even within a bloc and require side payments or complementary redistribution to be stable. (4) ASEAN-style clubs where all members are hot-country losers from warming can achieve a Pareto-improvement for members while also improving global welfare, making them potentially more robust to free-riding than clubs like the EU where some members prefer a warmer climate. (5) The SCC estimated under utilitarian weights ($203/tCO₂) is substantially higher than under Negishi weights ($3.31/tCO₂), implying that the appropriate SCC for policy depends critically on how inequality across countries is weighted in the social welfare function.

Key Concepts

Sufficient statistics (for climate policy): In this paper’s sense, a set of observable data moments and estimable elasticities — specifically nations’ energy mix (shares of oil-gas, coal, renewables), energy rent shares of GDP, bilateral trade shares, energy supply and demand elasticities, and damage parameters — that fully characterise, to the first order, the welfare impact of a climate policy change without requiring the full model to be solved. The approach follows Chetty (2009) and extends it from tax incidence to climate policy in an IAM with trade.

Carbon leakage: In this paper’s framework, the phenomenon by which a unilateral domestic carbon tax reduces domestic fossil demand and lowers the international price of oil-gas, inducing countries outside the policy to increase their fossil fuel consumption, partly or fully offsetting the original emission reduction. The paper shows leakage operates primarily through energy trade (oil-gas price channel) rather than through goods trade competitiveness effects, with energy effects consistently dominating in magnitude across all policy experiments.

Local Cost of Carbon (LCC): The country-specific welfare cost of an additional unit of global carbon emissions, measured in monetary units as the negative of the partial derivative of country i’s welfare with respect to aggregate emissions, divided by the marginal utility of consumption. Distinct from the global Social Cost of Carbon (SCC), which aggregates LCCs across countries with Pareto weights. Countries whose productivity is harmed more by warming have a higher LCC; cold countries may have a negative LCC (they benefit from marginal warming).

Structural damage function: The function Dy_i(E) mapping world cumulative emissions E to country i’s TFP via a quadratic temperature-productivity relationship with peak temperature T* and slope parameter γ, estimated in this paper from bilateral trade data (import penetration ratios and temperature differences) rather than from GDP-temperature regressions. The estimation is designed to be robust to the Lucas critique by netting out general-equilibrium propagation through trade and energy markets that would bias GDP-based estimates.

Climate club: In this paper’s usage (following Nordhaus 2015), a coalition of countries that jointly impose a domestic carbon tax on their own emissions and levy carbon tariffs (carbon border adjustment mechanism, CBAM) on imports from non-member countries scaled by the carbon intensity of those imports. The paper studies EU and ASEAN climate clubs and finds they differ sharply in welfare distribution: the EU club creates net losers among members (because some EU countries benefit from warming), while the ASEAN club delivers welfare gains for all members because all are hot-country losers from climate change.

Energy rent effect: The component of the welfare decomposition arising from changes in profits of domestic energy producers (fossil extractors, coal producers, renewable firms) due to changes in energy prices. Captured in the sufficient statistics formula as the profit share of GDP weighted by the relevant price change. Fossil-fuel-exporting countries have large positive exposure to oil-gas price increases (gains from price rises) and are harmed when global carbon policy reduces the fossil price — this is a key redistribution channel distinct from both climate damages and goods trade.

Empirical Bayes shrinkage (energy supply elasticities): In this paper, a procedure that estimates country-specific fossil and coal supply elasticities by first running OLS regressions of rent share changes on price changes country-by-country, then shrinking noisy or negative estimates toward a pooled mean by imposing a truncated-normal prior (truncated below 1 to enforce positive elasticities) and computing posterior means. Used because country-level time series are short and noisy, while the prior encodes the theoretical constraint that supply must be upward-sloping.

Negishi weights vs. utilitarian weights: Two distinct social welfare aggregation methods used throughout the paper to aggregate country-level welfare changes into global welfare. Utilitarian weights (ωi = 1 per person) put equal importance on each person globally, so welfare gains in low-income tropical countries count fully; this yields high SCCs ($203/tCO₂) and large global welfare gains from carbon taxation. Negishi weights (ωi ∝ 1/u’(ci), proportional to income) downweight poor countries and upweight rich ones, yielding dramatically lower SCCs ($3.31/tCO₂) and smaller measured global welfare gains because damages concentrate in low-income countries that receive little weight.