The social costs of tropical cyclones

Tropical cyclones (TCs) are among the most harmful extreme weather events. They affect on average 20.4 million people annually and caused mean direct annual economic losses of US$ 51.5 billion over the last decade. There is increasing empirical evidence that TC impacts can reduce economic growth in affected countries for more than a decade, leading to overlapping repercussions where economies lack time to recover between events. Projected increases in the frequency of the most intense TCs under global warming may make such overlaps more likely, amplifying long-term growth losses without additional adaptation, and potentially worsening development prospects, especially for strongly affected low- and middle-income countries. The study groups countries by the World Bank (2024) income classification to assess impacts across development levels and addresses the gap that long-term growth effects of TCs are not adequately incorporated in current SCC estimates.

Prior TC damage functions range from simple estimations linking damages to global mean temperature (GMT) and socioeconomic development to complex event-based approaches that account for storm characteristics such as wind-affected areas and lifetime rainfall. Much of the literature focuses on the USA or applies USA-based functions elsewhere, with fewer studies deriving country- or region-specific functions (e.g., Philippines, China, South Korea, or global regional sets). Except for notable studies (Hsiang et al.; Elliott et al.), most projections statistically link TC predictors to reported direct damages, facilitating calibration to loss databases but failing to capture persistence of damages in the economic system. This omission risks underestimating future damages and limits usefulness for policy evaluation and climate finance decisions. The paper builds on Mendelsohn et al. and Bakkensen et al. by using synthetic TCs to reflect climatology changes, and on Elliott et al. and Hsiang et al. by incorporating persistent growth effects, ultimately expressing damages as functions of GMT for use in integrated assessment models.

The framework comprises three main components: (1) Estimation of historical growth responses to TC strikes accounting for persistence, (2) probabilistic event-based projections of future TC damages using a TC emulator and socioeconomic scenarios, and (3) derivation of country-specific temperature-dependent damage functions and computation of DAD and SCC. - Historical growth response: A three-way fixed-effects panel regression (1981–2015) over 41 TC-affected countries estimates per-capita GDP growth responses to national annual shares of population exposed to ≥34 kt winds, including 0–8 lag years to capture persistence. Population-weighted country-year temperatures (linear and quadratic terms) are included following Burke et al., enabling joint estimation of temperature and TC effects. Maximum entropy bootstrapping (1,200 bootstraps per country) preserves spatio-temporal dependence, propagating uncertainty in historical estimates. - Exposure modeling: Historical exposure uses TCE-DAT; future exposure employs a statistical TC emulator (calibrated to four CMIP5 GCMs: HadGEM2-ES, MIROC5, IPSL-CM5A-LR, GFDL-ESM2M) to generate 100 probabilistic TC landfall time series per basin per GCM for RCP2.6, RCP6.0, RCP8.5 and a “no further climate change” baseline. CLIMADA produces wind footprints; exposed populations are aggregated annually by country using SSP2 and SSP5 population grids (HYDE-based downscaling). Simulated exposed-population series are bias-corrected to match historical country averages (1980–2015). - Discounted annual damage (DAD): For each RCP-SSP-GCM-realization-bootstrap combination, the observed historical growth response is applied to projected exposed shares to perturb SSP GDPpc trajectories (2010–2100), yielding GDP with and without additional warming (baseline vs RCP). DAD is the discounted GDP difference (growth-adjusted discounting with rate r_t = ρ + η g_t, with three calibrations: Stern, Nordhaus DICE, Ricke et al.). Uncertainties propagated include: (UD1) emissions (RCPs), (UD2) socioeconomic development (SSPs), (UD3) discounting, (UD4) historical growth response (bootstraps), (UD5) TC futures (emulator across GCMs and realizations), (UD6) temperature response to emission pulse for SCC. - Temperature-dependent damage functions: For each country, mixed-effects regressions relate TC-induced growth-rate changes to GMT change relative to preindustrial, with RCP-specific random effects and fixed effects shared across RCPs. Structural tests indicate SSP independence and weak RCP dependence (modeled as random effects). Country-level fixed-effect coefficients define temperature damage functions used in SCC calculations. - SCC computation: Following Ricke et al., SCC is the net present value of additional damages from a 1 GtC pulse in 2025, using coupled carbon–climate impulse responses (15 carbon-cycle models x 4 GCMs) to 2200 with constant post-2100 marginal effects. National GDP per-capita trajectories are perturbed by (a) aggregate temperature effects (estimated within the same regression framework) and (b) TC-induced temperature-dependent growth impacts from the derived damage functions. Growth-adjusted discounting is applied; the global SCC sums country-level SCCs. TC-induced SCC (TC-SCC) is the difference between SCC with and without TC contributions.

- Persistent growth impacts: TC strikes reduce economic growth in the short-, mid-, and long-term. Cumulative median growth losses increase with lag length and plateau around 6–13 years; an 8-lag model balances persistence capture and robustness. - Heterogeneity: Historical average annual growth losses are much higher in strongly exposed countries (e.g., Japan 1.63%, Philippines 2.62%) than in partially exposed ones (USA 0.17%). - DAD (2010–2100): Median global DAD is positive across all scenarios. For the main specification (RCP6.0–SSP2, Ricke discounting), the median global DAD equals 0.18% of 2021 global GDP; it rises up to 1.13% under RCP8.5–SSP5 with Stern discounting. Discounting is the dominant uncertainty driver; lower discount rates notably raise damages. - Per-capita burden: Under the main specification, median average DAD per capita is highest in strongly exposed high-income countries (e.g., Taiwan US$ 2,046; USA US$ 170). When scaled by 2019 average household income, burdens for Mauritius (0.71% AHI), Jamaica (0.33% AHI), the Philippines (0.93% AHI), and Vietnam (0.29% AHI) can exceed that of the USA (0.21% AHI). For the Philippines, average household income losses increase from 3.39 to 12.21 days/year under Stern’s discount rate. - Temperature damage functions: For 37 of 41 countries, TC-induced growth losses increase significantly with GMT. The relationship is SSP-independent and only weakly RCP-dependent. - SCC impacts: Without TC impacts, the median global SCC (main specification) is US$ 173/tCO2 (66% CI: 108–266). Including TCs raises median SCC by 22% to US$ 212/tCO2 (138–318). SCC is highest under RCP8.5 for both SSPs (e.g., US$ 361/tCO2 (222–588) vs US$ 212 (138–318) for main spec). Under Stern discounting, SCC with TCs reaches US$ 1654/tCO2 (1122–2361), over seven times the high-discount case. - Country TC-SCC contributions: Largest absolute TC-induced SCC increases occur for the USA (from US$ 13 to 20/tCO2; 17.7% of global median TC-SCC), Japan (US$ 1 to 8; 17.6%), Taiwan (US$ 2 to 8; 14.4%), China (US$ 8 to 14; 14.4%), and India (US$ 43 to 47; 11.3%). For the USA, the 66% CI for TC-SCC is US$ 4.34–9.79/tCO2 (median 6.64). The study indicates substantial percentage increases in country-level SCC due to TCs, e.g., China +68%, USA +53%, India +9%.

The study demonstrates that TCs impose persistent, compounding growth losses that materially elevate both country-level and global climate damage estimates. By explicitly capturing long-term growth effects and expressing impacts as temperature-dependent damage functions, the work addresses a key omission in SCC assessments and shows that excluding extreme-event damages can significantly underestimate mitigation benefits. DAD results highlight that relative burdens can be larger for lower- and middle-income countries when normalized by household income, informing equitable adaptation planning and international support. The SCC findings underscore that integrating extreme-event damages, notably TCs, materially raises the SCC, thereby strengthening the economic case for stringent mitigation and targeted adaptation. The approach provides IAM-ready damage functions to support broader policy evaluation, including carbon pricing and national adaptation planning.

This paper introduces a transparent, tractable framework that couples event-based TC exposure modeling with empirically estimated, persistent growth effects to produce country-level temperature damage functions and quantify DAD and SCC contributions. Accounting for TCs increases the median global SCC by about 22% and reveals substantial, persistent economic burdens across 41 exposed countries. The results suggest that current policy evaluations likely underestimate the benefits of mitigation and the need for adaptation when extreme-event damages are omitted. Future research should extend this framework to other extremes (e.g., floods, droughts), incorporate additional TC impact drivers (rainfall, storm surge), and embed the derived damage functions in fully integrated assessment models that endogenize mitigation and adaptation responses.

- The model assumes that future growth responses to TCs mirror historical responses; adaptation could reduce (or maladaptation increase) future vulnerability, biasing damages. - Physical impact modeling uses wind fields only; rainfall and storm surge are excluded, likely yielding lower-bound damage estimates, especially as sea-level rise and TC rainfall intensify with warming. - The framework does not endogenously model mitigation/adaptation behaviors or feedbacks on socioeconomics and climate, limiting policy response dynamics. - Results depend on discounting choices; lower rates substantially elevate DAD and SCC. - Temperature-damage relationships are treated as SSP-independent and only weakly RCP-dependent (as random effects); unmodeled structural dependencies may remain. - The TC track emulator is proprietary (limits transparency/access), though other components and code are openly available. - The analysis focuses on TCs and excludes other extremes (floods, droughts, heat); total SCC impacts from extremes may be higher.