Persistent inequality in economically optimal climate policies

Traditional neoclassical benefit-cost IAM analyses (e.g., DICE) have often deemed stringent climate targets (well-below 2°C) economically inefficient, suggesting optimal temperatures near or above 3.5–4°C by 2100. This has spurred critiques regarding normative choices (discount rates, intertemporal elasticity), impact functions, climate modules, and mitigation cost structures, with studies showing high sensitivity to these assumptions. Recent recalibrations of DICE that incorporate updated climate and economic science suggest Paris-consistent targets may be optimal, but they largely rely on single-region models, omitting regional heterogeneity and inequality. Empirical evidence indicates substantial cross-country heterogeneity in climate impacts and mitigation opportunities, and aggregated benefit-cost analyses mask inequality. The study’s purpose is to develop and apply a highly disaggregated benefit-cost framework that explicitly represents regional heterogeneity and inequality in both costs and benefits, comparing cooperative and self-interested behaviors under diverse socioeconomic pathways, impact specifications, and preferences over time and equity. The central research questions are: how do optimal climate policies change when accounting for country-level heterogeneity and inequality aversion, and what are the distributional consequences for global inequality under economically optimal pathways?

Prior work has critiqued DICE’s normative parameters and damage specifications, and highlighted sensitivity to modeling choices. Updated DICE variants incorporating empirically derived impact functions suggest tighter climate targets can be optimal, but they lack spatial heterogeneity. Studies have estimated country-level social cost of carbon and analyzed how climate change affects between-country inequality, yet without optimal policy evaluation. Regionalized benefit-cost IAMs (e.g., RICE, AD-RICE, PAGE, FUND, CWS, MICA, C³IAM, STACO) disaggregate into 6–16 macro-regions, limiting the capture of heterogeneity and often not integrating the latest climate-economy evidence. CGE models provide sectoral and regional detail but typically assess prescribed policy scenarios rather than intertemporal optimization. Hence, there is a gap for a high-resolution, welfare-expanded IAM that incorporates contemporary impact science and allows rigorous comparison between cooperative and non-cooperative optimal policies with explicit inequality preferences.

The study develops RICE50+, a regionalized and recalibrated DICE-based IAM with 57 independent regions/countries. Socioeconomic drivers (population, GDP) follow SSP1–SSP5 pathways. The carbon cycle employs a three-box model; radiative forcing follows log-CO2 forcing with exogenous OGHG forcing added via regression on CO2 forcing. The global temperature response uses a two-layer model recalibrated to emulate MAGICC6. Country-level temperatures are obtained via statistical downscaling of CMIP5 (0.5° grid), aggregated with population weights and regressed against global mean temperature to derive country-level temperature responses; subsequently aggregated to the 57 regions. Climate-economic impacts are modeled as empirically estimated impacts on per-capita growth rates using multiple specifications: Burke-Hsiang-Miguel (four variants covering short-run and long-run, with/without income differentiation), and robustness checks with Dell et al. and Kahn et al. These impacts enter a recursive formula linking growth-rate damages to output, consistent with growth impacts; an approximate recursion is used if savings are endogenous. To avoid numerical extremes, cumulative impacts are bounded within ±100% of baseline. Emissions are linked to output via exogenous regional carbon intensity. Calibration uses Enerdata/EnerFuture and POLES for 2025–2040, SSP marker information for the rest of the century, and smooth convergence post-2100 toward DICE-2016R2 global intensity by 2200, with SSP-dependent multipliers to match regional emissions pathways. Mitigation is the fraction of baseline industrial emissions abated, μ∈[0,1.2], allowing negative emissions. Regional marginal abatement cost (MAC) curves are fitted to Enerdata 2025–2040 abatement-price data with a polynomial MAC(t,μ)=a(t)μ+b(t)μ⁴, adjusted by a time-varying global correction ν(t) to align RICE50+ global abatement with SSP ensembles. Post-2100, regional MACs gradually converge to a common backstop technology curve consistent with DICE-2016R2. Mitigation ramp-up is constrained: μ(t+1) ≤ 1.2μ(t)+0.2 every 5 years, capturing inertia. Two behavioral regimes are modeled: (1) non-cooperative Nash equilibrium where each region maximizes its own intertemporal welfare taking others’ strategies as given (solved iteratively), internalizing only own damages; (2) cooperative global social planner maximizing a social welfare function aggregating regional welfare. Welfare expands on RICE by introducing inequality aversion γ that disentangles spatial inequality aversion from intertemporal parameters. The utility discount rate ρ is typically 1.5% (tested 0.1–3%), and the intertemporal elasticity parameter (η=1.45) follows expert elicitation. Inequality aversion spans γ=0 (inequality neutral) to γ=2 (high aversion), default γ=0.5. Population weights are used, and Negishi weights are avoided. Land-use CO2 and other GHGs are exogenous; LU follows PRIMAP-hist initial levels with declining trends per DICE-2016R2 (declines applied only to positive emitters in benefit-cost scenarios), while OGHGs are added using a linear relation to CO2 forcing from SSPs. Savings rates are fixed over time, converging to DICE-2016R2 long-run values by 2200 to reduce complexity; tests indicate inequality results are robust to endogenizing savings. The model evaluates a broad uncertainty space over SSPs, impact functions, ρ, and γ.

• Baselines and temperatures: BAU with impacts (no mitigation) yields 2100 global mean temperature (GMT) of +3.65°C (10th–90th percentile: 2.99–4.49°C). A self-interested non-cooperative optimum yields relatively flat emissions and a lower GMT of about +3.0°C (2.10–4.19°C). Cooperative global optimization leads to rapid emissions reductions approaching carbon neutrality by mid-century and temperatures mostly below +2°C, with a mean around 1.80°C (1.53–2.31°C) in 2100. The 1.5°C target is not cost-optimal in this framework.
• Country mitigation and NDCs: In non-cooperative equilibria, large emitters with high impacts and relatively low mitigation costs (India, China, USA) optimally reduce emissions substantially out of self-interest, achieving 20–75% CO2 reductions vs BAU by 2050. Under cooperation, nearly all regions decarbonize close to maximum potential except very low emitters (e.g., Sub-Saharan Africa). Current 2030 NDCs are generally closer to non-cooperative levels; cooperation implies higher ambition across all regions. Mitigation costs are small in 2030 but can reach up to about 10% of GDP by 2050 for the most exposed regions (e.g., Russia), consistent with heterogeneity in abatement opportunities and carbon intensity.
• Inequality and distributional impacts: Even under cooperative, Paris-consistent temperature outcomes, climate impacts are highly regressive across countries. Poorer countries face large GDP losses often exceeding 20% of GDP, with impacts increasing by roughly 11 percentage points per $10,000 decrease in baseline per-capita GDP. Climate change widens the global income distribution: in cooperative optimal pathways, the 90:10 income ratio increases by about 117% and the 80:20 ratio by about 63% relative to no-climate-change baselines. Approximately 2.3 billion people in Africa and 1.6 billion in India experience income reductions up to 60% by 2100 due to impacts under cooperative outcomes. Delaying full cooperation to 2030 further elevates inequality (90:10 ratio increase from ~+117% to ~+148%).
• Role of preferences and impacts: Inequality aversion (γ) strongly affects regional burden-sharing and per-capita emissions but has limited influence on global aggregate emissions and temperatures. Persistent, nonlinear temperature-growth impacts (e.g., Burke et al.) drive long-lasting reductions in growth for warmer regions, making inequality increases robust across SSPs, impact specifications, and discounting choices. Only under the BHM long-run specification (with higher losses across all countries) do cooperative policies slightly improve inequality compared to no-climate-change BAU; otherwise, inequality worsens markedly.
• Baselines matter: Accounting for climate-economy feedbacks produces lower non-cooperative emissions and forcing than original SSP baselines, addressing concerns about implausibly high counterfactuals.

The analysis demonstrates that integrating regional heterogeneity and explicit inequality aversion into a benefit-cost IAM materially changes both optimal policies and their distributional outcomes. While cooperative optimization makes Paris-consistent temperature stabilization economically optimal, this requires immediate and full global cooperation and substantial near-term mitigation ramps. Even under such favorable conditions, climate damages remain regressive and significantly increase between-country inequality due to the nonlinearity of temperature-growth impacts and the geographic distribution of baseline temperatures. Preferences for equity redistribute mitigation efforts toward richer nations but cannot eliminate the regressive pattern of impacts. These findings underscore that mitigation alone, even when globally optimal, is insufficient to prevent climate-induced increases in inequality. Policy implications include accelerating mitigation, exploring earlier deployment of CO2 removal, considering geoengineering assessments, and prioritizing adaptation planning and financing in highly vulnerable countries alongside mechanisms for compensation and inclusive development. The work links welfare-based optimization to burden-sharing principles, clarifying the limited capacity of equity-weighted objectives to offset structurally regressive climate damages.

This study introduces RICE50+, a high-resolution, welfare-expanded benefit-cost IAM that incorporates country-level climate damages, empirically based growth impacts, detailed abatement costs, and explicit inequality aversion. The model shows that globally cooperative, welfare-optimal policies can stabilize temperatures below 2°C by 2100, yet climate change still markedly increases between-country inequality, with severe losses in poorer, warmer regions. Contributions include: (1) demonstrating the economic optimality of near-2°C targets in a regional framework with updated impact science; (2) quantifying persistent inequality under both cooperative and non-cooperative optima; and (3) linking inequality preferences to burden-sharing while showing limited effect on aggregate temperature outcomes. Future research should integrate within-country inequality, sectoral and socioeconomic impact heterogeneity (energy systems, land use, agriculture, biodiversity), adaptation dynamics and capacity, interactions with health and other risks, and refined damage functions over long horizons to better capture distributional consequences and inform policies for mitigation, adaptation, and compensation.

Empirical impact functions are estimated from historical data and may misrepresent future adaptation and economic responses over long timescales. The model abstracts from many dimensions of heterogeneity, including within-country inequality, sectoral detail in energy and land-use systems, agriculture, biodiversity, health and other environmental risks, and heterogeneous adaptive capacity and consumption patterns. Land-use and other GHGs are treated exogenously. Immediate full cooperation is an optimistic assumption; delays increase inequality beyond reported cooperative levels. While savings are fixed for tractability (with robustness checks), simplifying choices may still influence some dynamic paths. Overall, omitted dimensions likely imply that real-world inequality effects could be larger, especially when within-country disparities are considered.