Comprehensive evidence implies a higher social cost of CO₂

The study addresses how updated, state-of-the-science components affect estimates of the social cost of CO₂ (SC-CO₂), a central metric used to value the benefits of emissions reductions in policy analysis. Traditional integrated assessment models (IAMs) estimate SC-CO₂ via four steps: projecting socioeconomics (population, GDP) to generate emissions, simulating climate response (concentrations, temperature, sea level), monetizing climate impacts as damages, and discounting future damages to present value. A 2017 NASEM report concluded that government-used SC-CO₂ estimates lag current science and recommended improvements in socioeconomics, climate modeling, damages, and uncertainty/disclosure practices. This paper introduces the Greenhouse Gas Impact Value Estimator (GIVE), an open-source IAM integrating probabilistic socioeconomics, updated climate and sea-level models, modern sectoral damage functions, and a growth-linked stochastic discounting framework to provide transparent, risk-consistent SC-CO₂ estimates.

The paper situates its contribution within critiques that prior SC-CO₂ models used outdated climate modules and damage functions, and poorly characterized compounding uncertainties. It references NASEM (2017) recommendations, IPCC AR6 scenario practices, and recent advances in sectoral damages (agriculture, mortality, energy, sea-level) and discounting. It notes previous influential IAMs (e.g., DICE) and meta-analytic damage functions, and highlights recent literature showing higher SC-CO₂ when incorporating improved science and uncertainty, as well as debates on scenario likelihoods and climate model structural differences (e.g., FaIR vs MAGICC).

The authors develop SC-CO₂ estimates using the GIVE model built on the open-source Mimi.jl platform. Key components: (1) Probabilistic socioeconomics via the Resources for the Future Socioeconomic Projections (RFF-SPs), providing multi-century, country-level distributions for population and GDP per capita, and global CO₂, CH₄, and N₂O emissions based on statistical models and expert elicitation that incorporate policy and technology uncertainty. (2) Climate and sea-level response using FaIR v1.6.2 for atmospheric GHG concentrations and temperature, and BRICK for sea-level rise, with parametric uncertainties (including tail risks) propagated. (3) Sectoral, regionally disaggregated damage functions covering four sectors: temperature-related human mortality, agriculture, building energy expenditures, and sea-level rise (with the Coastal Impact and Adaptation Model (CIAM) enabling optimal adaptation). Damages are monetized and aggregated across 184 countries (agriculture at 16 regions). (4) Discounting through an empirically calibrated, stochastic, growth-linked (Ramsey-like) framework that ties risk-free discount rates to uncertain consumption growth and incorporates risk preferences. The SC-CO₂ is computed as the discounted sum of incremental damages from a 2020 CO₂ emissions pulse along uncertain future trajectories. Uncertainties in socioeconomics, climate, and damages are sampled via 10,000 Monte Carlo draws to produce probabilistic SC-CO₂ distributions. Sensitivity analyses include alternative near-term discount rates (1.5%, 2.0%, 2.5%, 3.0%), comparisons to DICE-2016R, and substitution of aggregate global damage functions from meta-analyses.

- Preferred SC-CO₂: Mean $185 per tCO₂ (5%–95%: $44–$413) at a 2% near-term risk-free discount rate (2020 USD), 3.6× higher than the US government’s current $51 per tCO₂ (3% constant rate). - Sensitivity to discount rate (mean, 5%–95% range): 3.0% → $80 ($12–$197); 2.5% → $118 ($23–$280); 2.0% → $185 ($44–$413); 1.5% → $308 ($94–$626). - Incremental contributions (Table 1): DICE-2016R baseline $44; replacing climate/socioeconomics/discounting while retaining DICE damages (3% near-term) → $59 (+$15, 11% of total change); using GIVE sectoral damages (3% near-term) → $80 (+$21, 15%); lowering near-term discount rate to 2% (preferred) → $185 (+$105, 74%). - Sectoral partial SC-CO₂ under 2% near-term rate: mortality mean $90 (5%–95%: $39–$165); agriculture mean $84 (−$23–$263) with wide uncertainty allowing for potential benefits in some realizations; energy expenditures mean $9 ($4–$15); sea-level rise mean $2 ($0–$4). - RFF-SPs characterize socioeconomics: median world population peaks at ~11 billion ~2130 then declines to 7.3 billion by 2300 (5%–95%: 2.8–21 billion); median global per capita growth averages 0.88% (2020–2300; 5%–95%: 0.17%–2.7%); median net CO₂ emissions ~17 GtCO₂ in 2100 (~40% of today; 5%–95%: −7 to 62 GtCO₂), declining thereafter. - Substituting two global meta-analytic damage functions yields SC-CO₂ means within −18% to +11% of the preferred estimate at 2%.

The results demonstrate that integrating comprehensive, probabilistic socioeconomics, updated climate and sea-level models, modern sectoral damage functions (including non-market mortality risk), and growth-linked stochastic discounting substantially elevates SC-CO₂ compared to values used in current policy analysis. The dominant driver of the increase is the lower near-term discount rate consistent with observed real interest rate trends, followed by updated damage representations. Sectoral analysis shows mortality and agriculture as primary contributors, with sea-level and building energy costs smaller due to physical inertia, adaptation, offsetting heating/cooling effects, and discounting. Probabilistic propagation across components enables a risk-consistent valuation, highlighting substantial uncertainty ranges important for policy. The higher SC-CO₂ implies larger estimated benefits of mitigation and higher expected net benefits of more stringent climate policies. Sensitivity to aggregated damage functions and discount rates suggests robustness of a higher SC-CO₂, though results may still be conservative given omitted sectors and tipping elements.

By implementing NASEM-recommended improvements across the SC-CO₂ estimation chain, the GIVE model provides transparent, probabilistic, and risk-consistent SC-CO₂ estimates. The preferred mean estimate of $185/tCO₂ (2% near-term discount rate) substantially exceeds currently used policy values, implying greater benefits from emissions reductions. The open-source, modular Mimi.jl framework facilitates updates, extension to other gases and years, and incorporation of new scientific evidence. Future work should expand sectoral coverage (e.g., biodiversity, labor productivity, conflict, migration, additional non-market impacts), incorporate additional climate system tipping elements and model structures, and further examine adaptation behaviors and equity weighting. These enhancements are expected to raise SC-CO₂ further and refine uncertainty characterization.

- Damage coverage: Only four sectors (mortality, agriculture, building energy, sea-level rise) are included; many market and non-market damages (e.g., biodiversity, labor productivity, conflict, migration, cultural and ecosystem losses) are omitted and would likely increase SC-CO₂. - Adaptation modeling: Coastal damages assume optimal, forward-looking adaptation via CIAM; real-world adaptation may be constrained, potentially increasing damages. - Physical system representation: While BRICK approximates rapid Antarctic ice-sheet dynamics, broader tipping elements and structural climate-model uncertainty (e.g., differences between FaIR and MAGICC) are not fully integrated. - Temporal discounting: Results are sensitive to discount-rate assumptions, although the chosen framework is empirically calibrated; alternative ethical or empirical choices would shift estimates. - Spatial resolution: Most damages are at the country level (agriculture at 16 regions), which may mask subnational heterogeneity and distributional effects; equity weighting is not applied. - Data and parameter uncertainty: Agriculture and mortality damages incorporate parameter uncertainty, but other sectors have more limited quantified uncertainty.