Abstract
The study compiles a global subnational dataset of gross regional product (GRP) for over 1500 regions to estimate historic climate impacts. Using panel, long-difference and cross-sectional regressions, it finds no evidence that temperature changes affect long-run growth, but clear evidence that higher temperatures reduce productivity levels, especially in hotter and poorer regions. A 3.5 °C global warming scenario by 2100 is projected to reduce global output by 7–14%, with higher losses in tropical regions. Updating the DICE-2016 model using these estimates increases the social cost of carbon to 73–142 USD/tCO₂ in 2020.
Introduction
The paper addresses key uncertainties in the climate–economy literature, particularly whether temperature shocks have temporary or persistent economic effects. Earlier research shows strong temperature effects on labour and land productivity but mixed evidence on growth impacts. By using a large subnational dataset at multiple time scales, the study provides improved empirical damage estimates and clarifies the distinction between productivity-level and growth-rate impacts.
Conceptual framework on climate and growth
A Ramsey-type growth model illustrates how weather and climate influence output levels through a productivity factor but do not necessarily change long-run productivity growth. Permanent temperature increases lower consumption and production levels but do not alter long-run growth unless they affect innovation or capital accumulation channels.
Empirical strategy
The authors estimate climate impacts using: (i) annual panel regressions for short-run weather shocks; (ii) long-difference regressions for decadal changes; and (iii) cross-sectional regressions for long-run climate conditions. Non-linearities are captured through quadratic temperature terms and temperature/precipitation bins. Country fixed effects reduce omitted variable bias and allow identification from within-country variation.
Data
Climate data from CRU TS v3.23 are aggregated to regions using an algorithm that accounts for partial grid-cell overlaps. ERA5 data provide robustness checks. GRP data cover high-level administrative regions worldwide. Additional geographical covariates include altitude, distance to coasts and rivers, and oil and gas reserves. Future warming projections follow the RCP8.5 pathway from the ISIMIP dataset.
Results
Panel regressions show significant non-linear temperature effects: warming raises GRP in cold regions but reduces it in hot regions. A 1 °C increase at 25 °C lowers GRP by roughly 3.5%. Precipitation effects are weaker. Long-difference models find no evidence that climate affects growth rates. Cross-sectional regressions identify persistent negative temperature effects on GRP levels (about 2–4.3%). Binned estimates show GRP 28–35% lower in regions with average temperatures between 20–28 °C compared with moderate-temperature regions.
Future climate damages
Projected warming under RCP8.5 reduces global GRP in 2099 by 11–14% using panel estimates and 6–8% using cross-sectional estimates. Losses reach up to 20% in tropical regions. Poorer regions face stronger impacts because they start from hotter baseline climates. Integrated assessment models underestimate damages; incorporating the study’s function into DICE-2016 more than triples the social cost of carbon.
Conclusions and outlook
Temperature increases strongly affect productivity levels but not long-run growth. Damages are larger in hotter and poorer regions. The study recommends updating IAM damage functions to reflect higher empirical sensitivity and emphasises that current models exclude important channels such as extreme weather and non-market damages.
