Breaking the Climate–Development Trap

The trap rests on a simple asymmetry. A rich economy hit by a flood draws on insurance, reserves, credit and a functioning state. Damaged roads and factories are rebuilt within a year or two, and output catches up. A poor economy hit by the same flood has few of these buffers. Farmers sell livestock and seeds to eat, children leave school, firms close, and governments divert budgets or borrow expensively. Some of the lost capital is never replaced, and the loss of income becomes permanent. Hsiang and Jina's study of 6,700 tropical cyclones found that national incomes do not recover within twenty years. A 90th-percentile event lowered income per person by 7.4% two decades later.

When shocks recur faster than economies recover, the losses compound. They also feed back: a poorer country spends less on drainage, early warning and building codes, so the next hazard does more damage. The divergence chart shows the outcome at the level of income groups. Measured in purchasing-power dollars, the median low-income country was barely richer in 2024 ($1,900 per person) than in 1990 ($1,700), and below its 2019 peak of $2,400. Over the same period the high-income median rose from $31,000 to $49,000.

Climate shocks are not the main cause of that divergence. Conflict, the pandemic, debt crises and commodity busts all weigh on it, and section 2 shows that the average severe climate shock costs a few percent of income, not a third. The trap matters because it is cumulative and growing. Lower-income countries averaged 1.16 severe climate shocks each over 1990–2024, and one in six suffered three or more. Severe shocks in low-income countries were more than twice as frequent in the 2020s as in the 2000s. A loss that never heals turns each shock into a permanent step down.

The event study follows GDP per person from the year before a climate shock to eight years after it, compared with the same country in other years and with all countries in the same year. In low- and lower-middle-income countries, 88 severe climate shocks since 1981 were followed by a 1.7% fall in GDP per person in the shock year. Instead of recovering, the loss deepened: −2.5% after two years, −5.1% after four (statistically significant at the 5% level) and −5.0% after six. After eight years income was still 3.7% below its pre-shock path, a result significant at roughly the 10% level.

Upper-middle-income countries took a larger initial hit, −2.4% across 47 severe shocks, but recovered within four years. High-income countries suffered few severe shocks by these measures (10 in 45 years), and their estimates are too imprecise to interpret. Moderate shocks, those affecting 5% of the population or causing damage of 1% of GDP, cost about 1% of income in the shock year in every income group, and in developing countries the loss was made good within three to four years. It is the large, destructive events in poorer countries that leave lasting scars.

Two cautions apply. In the three years before severe shocks, incomes in poorer countries were growing slightly more slowly than usual. This may reflect smaller shocks just below the threshold, conflict or reporting patterns, and it suggests the estimates include some pre-existing weakness. The second caution is precision: with fewer than a hundred severe events, the confidence bands are wide. Split by income group, the lower-middle-income estimate is the most precise and the most persistent (−7.1% after eight years, standard error 1.9 points). The low-income estimate, from only 28 shocks, is too noisy to interpret on its own.

Splitting shocks by type and definition shows where the scar comes from. Among low- and lower-middle-income countries, the 30 most destructive events, those with recorded damage of 5% of GDP or more, were followed by a 4.4% fall in GDP per person in the shock year, a loss of about 13% after four years and 10% after eight. The confidence interval is wide, but the effect is large. Hurricane Mitch in Honduras and Nicaragua (1998), Cyclone Idai in Mozambique (2019) and Cyclone Nargis in Myanmar (2008) are among these events. Severe floods and storms show a similar but smaller pattern: −2.7% at impact, about −6.5% after three to five years and −3.9% after eight.

Droughts that affected a fifth or more of the population show no measurable effect on GDP. That does not mean they are harmless. Droughts strike food, health, schooling and household assets, much of which national accounts capture poorly, and they are slow-onset, so the year of the shock is hard to date. Shocks defined only by people affected show smaller, imprecise losses. Since 2000, severe shocks have been followed by smaller measured losses (−2.3% after eight years, not statistically significant). That is consistent with better early warning and response, but also with more complete recording of smaller events.

Earthquakes and volcanic eruptions provide a useful comparison. They are not driven by climate, but they destroy capital in similar ways. Geophysical shocks cut GDP per person in the shock year by about 1% across all countries, and output then tended to rise above its previous path as reconstruction spending arrived, significantly so in high-income countries. That climate shocks scar while geophysical ones do not is consistent with the trap mechanism: climate shocks recur and compound, often hit agriculture and the rural poor, and less often attract the reconstruction finance that follows a major earthquake.

Between 2000 and 2024, climate shocks occurred in 17% of country-years in low-income countries, 14% in lower-middle-income and 12% in upper-middle-income countries, against 2% in high-income countries. Severe shocks followed the same gradient, at 3.6, 3.2, 2.7 and 0.5 per 100 country-years. On average, 2.9% of the population of low-income countries was affected by a climate disaster each year, against 0.4% in high-income countries. Rich countries are not free of hazards. But their floods, storms and heatwaves rarely affect a large share of the population, because housing, infrastructure and warning systems absorb more of the blow.

For low-income countries the trend is upward. Climate shocks rose from 14 per 100 country-years in the 2000s to 15 in the 2010s and 26 in the 2020s, and severe shocks from 2.7 to 6.4. Some of this rise reflects better recording of disasters in EM-DAT, which improved markedly after 2000. Some reflects larger and more exposed populations. Some is the climate signal that the IPCC has documented for heavy rainfall, droughts and tropical cyclone intensity. Lower-middle- and upper-middle-income countries show no clear trend in shock frequency over the same period.

Recorded deaths tell a different story. They fell sharply in developing countries as early warning and evacuation improved, which shows that preparedness works. Economic exposure, however, has grown with population, urban assets and agricultural intensification in hazard zones. The trap operates through income and assets more than through lives, and income losses are much harder to see.

Recovery can be measured by whether GDP per person returns to its pre-shock level. After severe climate shocks, low-income countries were back above their pre-shock income within three years only 56% of the time, against 75% in lower-middle-income, 78% in upper-middle-income and 75% in high-income countries. In the remaining cases, income was still below where it started three years later, before allowing for the growth the country would otherwise have enjoyed. This simple count does not control for other events, but it matches the event study.

A second gap concerns visibility. Of the climate shocks that affected at least 5% of a country's population, only 22% in low-income countries have any economic damage recorded in EM-DAT, against 45% in lower-middle-income, 73% in upper-middle-income and 71% in high-income countries. Losses go unrecorded where insurance is rare, assets are informal and statistical capacity is weak, which is where they hurt most. Global loss tallies therefore understate the burden on the poorest economies, and finance allocated by recorded losses flows away from them.

The visibility gap also matters for policy. Insurance, catastrophe bonds and parametric risk pools need reliable loss data to be priced. Loss-and-damage funds need it to allocate grants, and finance ministries need it to plan. Investments in hazard monitoring, damage assessment and household surveys after disasters are therefore part of breaking the trap. They make losses visible and so financeable.

Insurance is the first buffer. Swiss Re estimates that natural catastrophes caused $318 billion of economic losses in 2024, of which 57% ($181 billion) was uninsured. In emerging economies 80–90% of catastrophe losses are typically uninsured. Without insurance, rebuilding depends on household savings, informal networks, government budgets and aid, and these arrive slowly, partially or not at all. Speed matters. The Centre for Disaster Protection estimated that only about 2.7% of crisis financing in 2021 was arranged in advance, so most money arrives months after the shock, when households have already sold assets and taken children out of school.

Warning and preparation form the second buffer. The Global Commission on Adaptation found that 24 hours' warning of a coming storm or heatwave can cut the resulting damage by 30%. It also estimated that $800 million invested in early warning in developing countries would avoid $3–16 billion of losses a year. The UN's 2024 review of multi-hazard early warning systems found that countries with limited coverage have disaster mortality about six times higher than those with comprehensive coverage, and that fewer than half of the least developed countries have such systems.

Prevention is the third buffer, and it is the most underfunded. The UNEP Adaptation Gap Report 2025 puts developing countries' adaptation finance needs at $310–365 billion a year by 2035, against $26 billion of international public adaptation finance in 2023. That gap is 12 to 14 times, and about 58% of the finance that does flow comes as loans. The World Bank's Lifelines report estimated that every dollar invested in more resilient infrastructure in developing countries yields about four dollars in benefits. The returns are high, but the money is scarce in exactly the countries where the scar is deepest.

The severe-shock layer counts the years since 1990 in which a climate shock affected a fifth of the population or caused damage of 5% of GDP. Such shocks cluster in southern and eastern Africa, the Sahel and Horn, Central America and the Caribbean, South and South-East Asia, and Central Asia. Eswatini (7), Somalia (5), Bangladesh (4) and Tajikistan (4) have the most among lower-income countries. The frequency layer counts all climate shocks since 1990, and the affected layer shows the average share of people hit each year since 2000.

The missing-data layer shows the share of each country's large shocks with no recorded economic damage, which is highest across sub-Saharan Africa. The divergence layer shows each country's income as a share of the high-income median in 2024, and the change layer shows how that share has moved since 1990. It identifies the countries that converged (much of East and South Asia, parts of Central Europe and Latin America) and those that fell behind (much of sub-Saharan Africa and several fragile states). The coping layer shows INFORM's lack-of-coping-capacity score, a proxy for the buffers that set the scar.

Hover over or tap a country for its card. It shows a strip of every year from 1980 to 2024 coloured by climate shock, the largest recorded shock, the share of the population affected, damage and its recording, recovery after severe shocks, and income relative to the high-income median. The sortable table below lists every country.

The simulator draws random sequences of severe climate shocks for a typical country in the chosen income group, at the annual frequency observed in 2000–2024 multiplied by a climate scenario factor. Each shock follows the estimated income path from section 2 for eight years. After that, a chosen share of the loss is permanent: 100% if scars never heal, 0% if economies eventually catch up. Income is shown relative to the same country's path with no shocks, so 100 means no loss.

Three levers reduce the scar. Early warning and preparedness cut the initial loss, up to the 30% damage reduction the Global Commission on Adaptation associates with a day's warning. Fast finance and adaptive social protection, through pre-arranged insurance, contingent credit and cash transfers that scale up after shocks, cut the deepening of losses after the first year, which is the part of the scar that comes from missing buffers. Resilient infrastructure and land-use planning reduce the chance that a hazard becomes a severe shock at all.

The outputs show the median income loss by 2050, the bad-luck case (the 10th percentile of the 1,000 futures), the chance of at least one severe shock, and the share of futures in which the country ends up poorer relative to the high-income median than it started, given its underlying catch-up growth. The model is illustrative. It treats shocks as independent, applies average effects and ignores the feedback from lower income to higher vulnerability, so it probably understates the trap.

Breaking the trap starts with prevention: keeping hazards from becoming disasters. Drainage and flood defences, building codes that are enforced, resilient roads and power, protected mangroves and wetlands, and land-use plans that steer new settlement away from the worst flood zones reduce the share of hazards that become severe shocks. The returns are high, at about four dollars for every dollar invested in resilient infrastructure according to the World Bank. They are highest where exposure is growing fastest, in the expanding cities of Africa and South Asia.

Warning and fast money come next. Early warning systems are among the best-documented investments in development. Bangladesh's cyclone-preparedness programme is the classic case, cutting cyclone deaths from hundreds of thousands to dozens in a generation. The economic scar, however, is set by what happens in the months after the shock. Pre-arranged finance (sovereign catastrophe insurance, parametric risk pools, catastrophe bonds, contingent credit lines and climate-resilient debt clauses that pause debt service after a disaster) puts money in place before the shock. Adaptive social protection, which expands cash transfers automatically when a flood or drought is forecast or declared, keeps households from selling productive assets and taking children out of school. That is how a temporary loss becomes permanent.

Finally, rebuild better and make losses visible. Reconstruction that restores the same exposure resets the trap, while rebuilding to higher standards and in safer places lowers it. Damage assessments, loss databases and household surveys after disasters turn the 78% of low-income shocks with no recorded damage into losses that insurers, budgets and donors can see and finance. None of this is cheap. But against an adaptation finance gap of $284–339 billion a year and losses that accumulate over decades, it is the cheapest way to stop climate shocks from widening the gap between rich and poor countries.

Coverage: 155 countries with at least 1 million people and at least 25 years of GDP data, 1980–2024. Shocks: EM-DAT (CRED, UCLouvain) via Our World in Data, total people affected and economic damage as a share of GDP by hazard. Climate hazards are drought, flood, landslide, extreme weather (storms), wildfire and extreme temperature. Geophysical hazards are earthquakes and volcanic activity. A climate shock is a country-year with people affected ≥ 5% of the population or damage ≥ 1% of GDP; a severe shock uses ≥ 20% or ≥ 5%. Income: GDP per person in constant 2015 US$ (WDI) for the event study, and PPP 2021 $ for divergence. Income groups are the World Bank's current classification.

Local projections: for horizons h = 0…8, the dependent variable is ln y(t+h) − ln y(t−1). It is regressed on the shock in year t with controls for shocks in t−1 to t−3 and GDP growth in t−1 and t−2, and with country and year fixed effects, estimated separately by income group. Standard errors are clustered by country. Pre-shock horizons (h = −3, −2) omit the growth controls. Recovery counts compare GDP per person three years after a severe shock with the year before it. The simulator draws Bernoulli shocks at the group's 2000–2024 annual rate, applies the estimated impulse path for eight years and then a chosen permanent share.

Caveats: EM-DAT under-records smaller events and economic damage, especially before 2000 and in poorer countries, and recording has improved over time. Affected counts are imprecise. Shocks are not randomly assigned, and slightly weaker growth before severe shocks suggests some confounding. Estimates for high- and low-income countries alone rest on few events, and confidence intervals are wide throughout. Across countries, the number of severe shocks since 1990 is not significantly related to average growth over 1990–2024. Shocks explain a small part of long-run differences between countries, and the trap appears in the paths after events. The simulator is illustrative. Literature figures are quoted as published.