The economic commitment of climate change

The study addresses how much macroeconomic damage from climate change is already unavoidable (committed) due to past emissions and socio-economic inertia and how these near-term damages compare with mitigation costs. It responds to challenges in long-horizon projections—uncertainty, human biases, and changing socio-technical conditions—by focusing on the near term to enhance credibility and communicability. It aims to produce robust global and sub-national projections by incorporating multiple climate variables (average temperature, daily temperature variability, precipitation metrics and extremes) and by constraining the persistence of climate impacts on economic growth using an empirical framework that provides a conservative lower bound on persistence. The research is important for informing adaptation planning, macroeconomic risk management, and climate justice debates, as well as for contextualizing mitigation cost-benefit discussions.

Prior macroeconomic damage projections often rely on national annual mean temperature and long horizons, with substantial divergence depending on whether climate impacts affect growth rates persistently (growth effects) or only current output levels (level effects). Foundational work shows non-linear temperature impacts on economic output (Burke et al., 2015) and stresses the importance of persistence assumptions (Newell et al., 2021; Dell et al., 2012). Recent sub-national studies identify impacts from daily temperature variability and multiple precipitation dimensions (Kotz et al., 2021; 2022), suggesting broader climatic drivers beyond mean temperature. Cost-benefit analyses and IAM-based studies indicate mitigation’s net benefits typically emerge after mid-century (e.g., Drouet et al., 2022), potentially underplaying near-term damages. Literature also highlights unequal climate impacts, with larger burdens on countries with lower historical emissions and lower income (Diffenbaugh & Burke, 2019; Callahan & Mankin, 2022). The present study builds on these by integrating sub-national econometrics with a conservative treatment of persistence and multiple climatic components.

• Data and scope: Panel of >1,600 sub-national regions worldwide over ~40 years, combining climate and income data. Climate projections from an ensemble of 21 CMIP6 models, bias-adjusted to accurately reproduce observed climatological patterns.
• Econometric framework: Fixed-effects panel regressions estimate impacts of changes in climate variables on economic growth, using first-differenced climate variables to relate growth rates to changes in climate rather than levels. This specification makes the no-lag baseline a level-effects prior (instantaneous impact only). Distributed lag models (DLMs) are used to test and estimate persistence by including lags for each climate variable, providing a conservative lower bound on persistence (avoiding assumptions of infinite persistence inherent in level-based specifications).
• Climate variables: Annual mean temperature, daily temperature variability, total annual precipitation, annual number of wet days (>1 mm), and extreme daily precipitation. Interaction terms capture heterogeneity by prevailing local climatic conditions. Moderating variables are capped at the 95th percentile of their historical distributions to limit out-of-sample extrapolation.
• Lag selection and robustness: Initial models include up to 10 lags for temperature and up to 4 for precipitation; significant lag terms are retained. Information criteria guide the trade-off between fit and overfitting. Robustness checks include: restricted DLMs to limit oscillations; Monte Carlo simulations showing robustness to autocorrelation in lagged climate variables; tests for multicollinearity; and the necessity of including multiple climate variables to isolate effects. Final uncertainty sampling includes models with 8–10 lags for temperature and 4 for precipitation.
• Projections: The empirical response functions are combined with bias-adjusted CMIP6 projections under SSP2-RCP2.6 (low emissions) and SSP5-RCP8.5 (high emissions). A Monte Carlo procedure samples uncertainty from: (1) climate model selection; (2) empirical model choice (lag structure); and (3) regression parameter uncertainty via 1,000 block-bootstraps. Projections quantify the exogenous effect of future climate on income relative to socio-economic baselines (ceteris paribus), not forecasting overall growth. Population-weighted aggregation yields global impacts.
• Currency and comparability: Damages are reported in 2005 international dollars. Structural choices (e.g., timescale for moderating variable changes; order of intertemporal and international currency comparison) are tested and found not to alter conclusions.
• Spatial analysis: Main models do not include spatial spillovers. A supplementary spatial-lag model (contemporaneous, excluding temporal lags) explores spillovers among neighboring regions, indicating amplification of magnitude and heterogeneity of impacts.

• Committed global damages: A permanent 19% reduction in global income per capita by 2049 relative to a baseline without climate impacts (population-weighted; likely range 11–29%). Damages under low- and high-emission scenarios are statistically indistinguishable until 2049.
• Regional impacts: Median income reductions around 11% for North America and Europe; about 22% for South Asia and Africa by 2049. All regions experience losses except very high latitudes, where reduced temperature variability brings benefits.
• Monetary magnitude: Under SSP2 income baseline, global annual damages in 2049 are about 38 trillion (2005 international dollars), with a likely range of 19–59 trillion.
• Mitigation costs comparison: In 2050, the median committed damages exceed median mitigation costs for achieving SSP2-RCP2.6 by about sixfold (mitigation costs ~6 trillion 2005 international dollars in IAM estimates). The delayed emergence of net mitigation benefits in formal analyses primarily reflects that damages are indistinguishable across emission paths until mid-century, not that near-term damages are small.
• Drivers of damages: Changes in annual mean temperature account for the largest share. Considering only mean temperature implies ~13% global income reduction in 2049 (likely 5–21%); adding daily temperature variability and precipitation components raises net damages by ~50%.
• Incremental contributions: Including daily temperature variability increases damages by 4.9 percentage points (likely 2.4–8.7 pp; ≈10 trillion international dollars). Precipitation components add smaller but non-negligible amounts: total annual precipitation ≈0.01 pp (−0.37–0.33 pp), number of wet days ≈0.34 pp (0.07–0.90 pp), extreme daily precipitation ≈0.36 pp (0.13–0.65 pp). Climate models may underestimate future changes in variability and extreme precipitation, suggesting true impacts could be larger.
• Inequity in impacts: Damages are larger in countries with lower historical cumulative emissions and lower current income. Lower-income quartile countries face losses 8.9 percentage points (61%) greater than the upper-income quartile (likely 3.8–14.7 pp). Countries in the lowest quartile of historical cumulative emissions face losses 6.9 percentage points (40%) greater than the highest quartile (likely 0.27–12 pp).

By focusing on near-term, already committed damages and using a conservative empirical approach to bound persistence, the study demonstrates that substantial macroeconomic losses are unavoidable by mid-century regardless of emissions path. This reframes policy debates: although formal cost-benefit analyses often indicate net mitigation benefits after mid-century, the magnitudes show that damages before then are already much larger than mitigation costs; the lack of divergence across emission scenarios until 2049 explains the perceived delay in benefits. Inclusion of multiple climate variables reveals larger and more heterogeneous damages than mean temperature alone, highlighting the importance of daily temperature variability and precipitation characteristics. The strong inequity—higher losses in low-income, low-emission countries—underscores climate justice concerns and the need for targeted adaptation and support. Post-2049, damages diverge strongly across scenarios, reinforcing the substantial economic value of mitigation in the long run. The analysis likely understates true damages due to excluded impact channels and spillovers, and conservatism in persistence and extrapolation, but it provides a credible lower-bound estimate to inform policy and planning.

The paper provides global and sub-national projections showing that climate change has already committed the world to substantial macroeconomic damages by mid-century: about a 19% permanent reduction in income per capita globally, largely driven by changes in mean temperature but significantly amplified by temperature variability and precipitation dynamics. These near-term damages already exceed estimated mitigation costs to meet the 2 °C target by a factor of about six, while damages diverge sharply across emission scenarios after mid-century, highlighting the long-term economic benefits of mitigation. The results also reveal profound inequities, with higher relative damages in lower-income and lower-emitting countries. Future research should incorporate additional impact channels (heatwaves, sea-level rise, tropical cyclones, tipping points, non-market damages), better characterize spatial and trade-network spillovers with spatio-temporal models and sub-national trade data, and refine climate model representations of daily variability and extremes to improve damage estimates.

• Impact coverage: Excludes key channels such as heatwaves, sea-level rise, tropical cyclones, climate tipping points, and non-market damages (ecosystems, health), likely making estimates conservative.
• Spillovers: Main analysis omits spatial and trade-network spillovers; a simplified neighboring-region spatial-lag model suggests spillovers can amplify magnitude and heterogeneity of damages.
• Model persistence and extrapolation: The first-difference DLM approach provides a conservative lower bound on persistence. Despite capping moderating variables at the 95th percentile, projections necessarily extrapolate beyond historical climate variability. The empirical models explain <5% of historical variance, but projected climate changes are much larger than historical fluctuations.
• Climate model biases: CMIP6 models may underestimate future changes in daily temperature variability and extreme precipitation, potentially understating damages.
• Affiliation of impacts over time: Currency comparisons and socio-economic baselines introduce structural choices (tested and robust), but remaining structural uncertainty persists.