Climate Adaptation and Development Convergence Across the Global South

Executive Summary

Poor countries only began catching up with rich ones in the 2000s, after a century of falling behind. Climate change is the first systematic force in modern development capable of switching that off again — and adaptation is the policy that keeps it on.

The central claim of development economics is that poor countries should grow faster than rich ones, because capital is scarcer where there is less of it. For most of the twentieth century the data flatly contradicted this. Lant Pritchett's survey of the long record, published under the title "Divergence, Big Time", found that the ratio of income in the richest countries to income in the poorest had risen roughly sixfold between 1870 and 1985. Backwardness conferred no advantage; it compounded.

Something changed around the turn of the millennium. Work by Kremer, Willis and You and, separately, by Patel, Sandefur and Subramanian shows that the cross-country relationship between a country's starting income and its subsequent growth, positive for decades, turned negative in the 2000s and has stayed negative. Poorer countries, as a group, began to catch up. This report is about the force most likely to reverse that again, and about the policy that answers it.

Set the first two figures beside each other. They are not commensurable — one is a share of the income gap closed each year, the other a share of output destroyed — and no arithmetic converts one into the other. But the orders of magnitude are close enough to make the point that matters: the climate drag on Global South economies is not a second-order correction to the convergence story. It is the same size as the convergence itself.

The policy conclusion follows directly and is unusual. Adaptation is normally justified as risk management, sold against a probability-weighted disaster that may not arrive. Read against convergence, it is something else: a growth-retention policy, and the cheapest available way to raise the net rate at which developing economies close the distance. That reframing changes which ministry owns it, which budget pays for it, and what counts as its return.

1. Convergence, Interrupted

What the convergence literature actually established, how thin the resulting margin is, and why the growth environment was already deteriorating before the climate drag was priced in.

Three claims need to be kept separate. Divergence is the historical observation that poor countries grew more slowly than rich ones and the distribution of income widened. Conditional convergence is the weaker claim that a poorer country grows faster once you hold constant the things that set its long-run income — savings, schooling, institutions. Absolute convergence is the strong claim that poorer countries grow faster, full stop.

Through the 1990s the consensus was pessimistic: conditional convergence held at a slow and remarkably stable rate, absolute convergence did not. That combination means catch-up is available only to countries that first acquire the institutions and human capital of richer ones; absent that, a country converges toward its own low steady state rather than toward the frontier. The newer finding is that absolute convergence has resumed.

Two caveats travel with the result, and both matter here. The first is speed: the estimated rate of unconditional convergence since 2000 is on the order of one to two percentage points of the gap a year, which means halving the proportional distance to the frontier takes something close to a working lifetime. The second is dispersion. "On average" conceals enormous variation — East and South Asia have converged rapidly, much of Sub-Saharan Africa and parts of Latin America have not. The World Bank's 2024 World Development Report frames the stall as a middle-income trap, noting that at recent growth rates many middle-income countries would need the better part of a century to reach even a quarter of United States income per head.

1.1 The engine was already sputtering

The climate drag is arriving on top of a weakening baseline rather than a booming one. The World Bank's January 2026 Global Economic Prospects projects per capita income growth in developing economies at about three per cent in 2026 — roughly a percentage point below the 2000–2019 average — and describes the decade as on track to be the weakest for global growth since the 1960s. The IMF's January 2026 World Economic Outlook update reaches a similar conclusion, calling the gains uneven and insufficient to reduce poverty at the needed pace.

1.2 Convergence runs on particular escalators

Convergence is an outcome, not a mechanism. The mechanism, in the sectoral view developed at the World Bank and elsewhere, is that certain activities act as growth escalators: once a country gets a foothold in them, productivity rises toward the global frontier almost regardless of the country's overall institutions. Dani Rodrik's finding that formal manufacturing exhibits unconditional convergence — catching up at roughly two to three per cent a year — is the clearest case.

This matters for what follows because the escalators are climate-exposed in specific ways. Manufacturing needs power that is present when the line is scheduled to run, and ports and roads that are open. Urbanisation concentrates people and firms in places that must be cooled, drained and supplied. Traded services need connectivity that does not stop. Human capital is produced in schools and clinics that heat, flood and close. A hazard that interrupts any of these does not merely destroy output in the year it strikes; it slows the escalator that was carrying the country upward.

2. Convergence Is a Net Rate

The analytical move on which the rest of the report rests: catch-up is accumulation minus destruction, and climate policy operates on the second term.

Growth is conventionally discussed as a single number — the rate at which output expands. For a country exposed to recurrent hazard, that number is better read as a difference between two processes running at once. An economy accumulates: it invests, it moves workers onto escalators, it educates children, it builds capital stock. It also loses: floods take the road, the cyclone takes the harvest, the heatwave takes the school year and the working day, and the reconstruction budget takes the money that would have built the next thing.

What converges is the residual. Write it plainly:

Two properties of the erosion term make it more damaging than an equivalent shortfall in accumulation, and both are frequently missed.

The first is persistence. A year of slower investment is a year of forgone growth; a destroyed asset is a permanent subtraction from the capital stock that must be rebuilt before anything new is added. A child out of school during a heatwave does not recover the year later. Erosion does not merely lower this year's growth rate — it lowers the base from which every subsequent year compounds.

The second is asymmetry of timing. Accumulation is smooth and erosion is lumpy, which means the political economy treats them differently. A steady one per cent of GDP lost to disasters is invisible in any given year and catastrophic over a generation, and there is rarely a moment at which the case for preventing it is politically legible. This is precisely the profile of a problem that is systematically under-addressed.

3. The Four Erosion Channels

Hazard subtracts from growth through four distinct channels. Only one of them requires a disaster to actually occur — which is why the drag is larger than the disaster record suggests.

Table 1 — How hazard erodes the growth base

The fourth channel is the one absent from most adaptation appraisal, and the only one that operates continuously.

3.1 Capital destruction

The most visible channel is also the best measured. Natural hazards already cost low- and middle-income countries on the order of US$390 billion a year, equivalent to one to two per cent of their combined GDP, and the International Finance Corporation's 2026 assessment projects that unmanaged climate risk could put some 43 million jobs across 49 countries at risk by 2050. Each destroyed asset must be rebuilt before any net addition to the capital stock resumes — reconstruction shows up in GDP as activity while representing, in growth-accounting terms, running to stand still.

3.2 Human-capital interruption

The second channel is less visible and, over a generation, may be larger. It also has the cleanest natural experiment attached to it. Research on millions of secondary-school students found that, absent air conditioning, a school year one degree Fahrenheit hotter reduced that year's learning by about one per cent — with the effect driven by hot days during instruction rather than on weekends or in summer, which is what identifies the mechanism as heat in the classroom rather than heat in general.

Heat takes adult output as well as child learning. The Adrienne Arsht–Rockefeller Foundation Resilience Center's Hot Cities, Chilled Economies analysis estimates the productivity cost already being paid in major Global South cities.

A city losing eight per cent of its output to heat is not experiencing a disaster. It is experiencing a permanently lower productivity level, which in convergence terms is a permanently lower escalator speed.

3.3 Fiscal absorption

The third channel converts a physical shock into a fiscal one. Reconstruction is financed from a budget that was going to build something else, or from borrowing that was already constrained. UNCTAD reports developing-country external debt at a record US$11.4 trillion, with net interest payments of US$921 billion in 2024. A government in that position absorbs a disaster by cancelling development spending, which is to say by lowering the accumulation term to pay for the erosion term.

3.4 The risk premium — the channel that never stops

The fourth channel is the one this report wants to put at the centre, because it operates continuously and appears in no disaster ledger. Capital is priced against perceived risk, and a visibly exposed economy pays more for it in every year, including the quiet ones.

Because capital-intensive, front-loaded assets are exactly the ones whose economics are dominated by the price of money, a high cost of capital does not reduce investment uniformly. It selectively suppresses the long-lived infrastructure that convergence depends on — power, transport, water, and adaptation itself — while leaving short-payback activity relatively unaffected. The premium is, in effect, a tax levied on the future.

Worse, the four channels compose into a loop.

4. Why the Drag Falls on the Converging

If climate losses were distributed evenly, they would lower world growth without touching convergence. They are not distributed evenly, and the pattern runs precisely against catch-up.

A uniform shock is not a convergence problem. If every economy lost the same share of output to hazard, the income gap would close at the same rate as before, on a slightly lower path. What makes climate a convergence issue is that the losses fall hardest on the countries that were catching up.

The geography is the first reason. Hazard exposure is concentrated in the tropics and subtropics, where most of the developing world sits; heat effects are non-linear, so a given increment of warming does far more damage to an economy already close to the threshold at which outdoor work and classroom learning degrade. Economic structure is the second: a larger share of output and a much larger share of employment sits in agriculture and outdoor work, which are the activities most directly weather-exposed — and, as Figure 2 showed, agriculture is also the sector with no measured catch-up, so damage there is not recovered by convergence elsewhere.

The modelled estimates bear this out. The Swiss Re Institute's climate stress test, summarised by the World Economic Forum, found global GDP some 11 to 14 per cent smaller by mid-century than in a no-warming baseline even on a path consistent with well below 2°C — with the emerging economies of South and Southeast Asia facing losses two to three times the global average. The same body of work finds that the countries least responsible for emissions face income losses roughly 60 per cent larger than higher-income countries and 40 per cent larger than higher-emitting ones.

The third reason is the one developed in Section 3.4: the risk premium. Exposure is priced, and it is priced against the same countries. A frontier economy and a vulnerable one facing the same physical hazard do not face the same cost of protecting themselves against it, because the vulnerable one borrows at three times the rate. Vulnerability raises the price of the cure.

5. Adaptation as Retention Policy

Most of what adaptation buys is not insurance against a disaster. It is output that would otherwise have been lost in ordinary years — which places it in the growth budget, not the contingency budget.

The standard case for adaptation is actuarial: multiply the damage by the probability, discount it, compare with the cost. That framing has done adaptation considerable harm, because it presents a protective asset as a bet on a bad outcome and invites a finance ministry to weigh it against a school.

The evidence says the framing is simply wrong about where the return comes from. The World Resources Institute's 2025 study of 320 real adaptation investments across twelve countries found benefits above ten dollars per dollar committed and an average internal rate of return of 27 per cent — and, critically, that more than half of those benefits accrued whether or not a climate disaster actually occurred, through higher yields, more reliable services, lower insurance costs and reduced business interruption. An asset with that profile is not insurance. It is infrastructure with an unusually good risk-adjusted return.

The clearest single demonstration comes from the heat-and-learning literature, because the same research design that measures the damage also measures the remedy.

Note what this figure is and is not. It is not a claim that air conditioning is the right adaptation everywhere — in most of the Global South passive cooling, building design, shading and school-calendar adjustment will do more per dollar, and air conditioning on a fossil grid carries its own costs. It is a demonstration that a specific, measurable erosion of human capital was substantially reversible with a known intervention. That is the general shape of the adaptation case: the losses are real, they are measured, and they are largely recoverable.

5.1 Retention is cheapest when specified early

The cost of retention depends almost entirely on when it is decided. Across the four essential systems the World Bank examined in Lifelines, building resilience in from the design stage adds only about 3 to 5 per cent to upfront capital costs, while the damage and service disruption avoided over an asset's life routinely runs to 50 to 100 per cent of the asset's value.

For a converging economy this is a closing window rather than a standing option. Most of the infrastructure that will carry the Global South through the 2040s and 2050s is being built now, or will be built in the next two decades. Specified with resilience, it retains its output for fifty years at a three-to-five per cent premium. Specified without, it becomes a permanent liability that must be protected later at many times the cost, or written off in the first serious event.

6. What Retention Is Worth

The arithmetic that makes the case: what a difference in the erosion rate does to income over a development horizon.

Take two economies with identical fundamentals. Both accumulate at five per cent a year — the same investment, the same escalators, the same human capital. They differ in one respect only: how much of that growth survives. One loses half a point a year to shocks; the other loses two, which is within the range of what hazard already costs low- and middle-income countries.

Thirty years is roughly one development generation. Over it, a one-and-a-half point difference in the erosion rate produces an income gap of more than half — larger than most successful reform programmes deliver, and achieved without changing the savings rate, the institutions, the education system or anything else the growth literature normally argues about. The two economies did everything else identically.

This is the sense in which retention is worth as much as acceleration. It is also, on the evidence, substantially cheaper. The Global Commission on Adaptation estimated that US$1.8 trillion invested across five areas — early-warning systems, climate-resilient infrastructure, improved dryland agriculture, mangrove protection and water management — could generate US$7.1 trillion in total net benefits. Early warning alone returns roughly nine to one. There is no comparable menu of interventions with that evidence base on the accumulation side of the ledger.

7. Evidence from the Record

Three economies that retained, and what the pattern has in common.

7.1 Bangladesh: retention at the extreme

Bangladesh is the clearest case in the record because the counterfactual is its own history. Cyclone Bhola killed on the order of 300,000 people in 1970, and a comparably powerful storm around 138,000 in 1991. Cyclone Amphan in 2020 struck a more populous coast and killed a number in the double digits, after 2.4 million people were moved into more than 12,000 shelters (The Conversation).

The mechanism is exactly the composition this report describes: hard capital — embankments and raised concrete shelters — combined with a forecasting service and tens of thousands of trained community volunteers who carry warnings the last mile. And the result is not only lives. An economy that does not lose its coastal labour force, its housing stock and its harvest every decade retains the base from which it compounds. Bangladesh's manufacturing-led convergence over the same period is not separable from the protection that let it accumulate.

7.2 Ahmedabad: retention at almost no capital cost

Ahmedabad's 2013 Heat Action Plan — the first in South Asia — combines an early-warning trigger, public cooling infrastructure and reflective cool roofs on public buildings and low-income housing. It has been credited with avoiding roughly 1,190 heat deaths a year, and the model has since been adopted, with mixed implementation, by more than a dozen Indian states. Against the Dhaka figure in Figure 3, the productivity retained by keeping a city's workforce functional through a heat season is a growth intervention that happens to be administered by a municipal health department.

7.3 The counter-case

The pattern is equally visible in its absence. An economy that rebuilds the same road after every flood, at full cost, from a budget that would otherwise have extended the network, is running the accumulation term and the erosion term against each other indefinitely. It shows the activity in its GDP and none of the convergence — because reconstruction restores the base rather than raising it. This is what a stalled catch-up trajectory frequently is on closer inspection: not a failure to invest, but a failure to keep what was invested.

What the successful cases share is not wealth. Bangladesh built its cyclone system while poor, and Ahmedabad's plan cost almost nothing. What they share is that retention was treated as a standing function of the state with a named owner and a permanent budget, rather than as a response mobilised after each event.

8. An Agenda for Retention

Organised by erosion channel, because an instrument aimed at the wrong channel does not raise the net rate.

8.1 Against capital destruction

• Make design-stage resilience mandatory for publicly financed infrastructure, capturing it at the 3–5 per cent premium Lifelines identifies rather than as a later retrofit — and do it while the Global South's build-out is still ahead of it.
• Treat reconstruction budgets as evidence of an appraisal failure rather than as disaster response, and require a build-back standard that raises the base rather than restoring it.

8.2 Against human-capital interruption

• Specify schools, clinics and workplaces for heat — passive cooling, shading, ventilation, building orientation — on the evidence that the learning and productivity losses are large, chronic and substantially recoverable.
• Complete Early Warnings for All, the highest-return intervention in the repertoire at roughly nine to one and still only about half covered.
• Build shock-responsive social protection so that a household hit by a hazard does not liquidate its children's schooling to absorb it — the mechanism by which a one-year shock becomes a one-generation loss.

8.3 Against fiscal absorption

• Pre-arrange the money: reserves, contingent credit, sovereign and parametric risk transfer, so reconstruction does not cannibalise the development budget.
• Make climate-resilient debt clauses standard, and address the debt overhang UNCTAD documents, which converts every hazard into a fiscal crisis.

8.4 Against the risk premium

• Deploy concessional capital and guarantees explicitly to compress the cost-of-capital gap, and measure the result in basis points achieved rather than dollars pledged.
• Make adaptation legible to investors: published, verifiable resilience standards lower measured exposure, which is the input the premium is priced from. Adaptation that no one can see does not lower the price of capital.
• Recognise the two instruments as complements. As Figure 5 shows, cutting the loop at one point while leaving the other turning does not break it.

8.5 Across all four: fix the appraisal

• Count non-contingent benefits — yields, reliability, avoided interruption — which the WRI evidence puts at more than half the total return, and which a probability-weighted disaster framing discards entirely.
• Move adaptation out of the environment budget and into the growth budget, where its returns actually land, and report it in retained output rather than in avoided disaster.
• Add an erosion term to national growth accounting. A country that does not measure what it loses each year cannot manage the variable this report argues is decisive.

8.6 Conclusion

Convergence is a young and fragile fact. For most of modern history the poorest countries fell further behind the richest, and the reversal that began around 2000 is neither large nor guaranteed — one to two percentage points of a gap that takes a lifetime to halve, distributed very unevenly, arriving in a decade the World Bank already expects to be the weakest for global growth since the 1960s.

Into that thin margin arrives a drag of comparable size, aimed disproportionately at the economies doing the converging, operating through four channels of which only one requires a disaster to occur. The honest reading is that climate change is the most serious threat to development convergence now visible, and that it works less by preventing poor countries from growing than by preventing them from keeping what they grow.

That framing is also the encouraging one, because the erosion term is the tractable half of the problem. The interventions are known, the returns are among the best-evidenced in development, and the cheapest of them are cheap in absolute terms. Bangladesh cut cyclone mortality by four orders of magnitude while poor. Ahmedabad protects a city's working population for the cost of paint and a protocol. What is missing is not capability but accounting — the recognition that an investment which protects growth belongs in the same ledger, and the same budget, as an investment that produces it.

References

Every quantitative claim above is attributed inline. The principal sources are collected here.

• Pritchett (1997)Divergence, Big Time, Journal of Economic Perspectives — the sixfold widening of the rich–poor income ratio, 1870–1985.
• Kremer, Willis & YouConverging to Convergence, NBER Working Paper 29484.
• Patel, Sandefur & SubramanianA New Era of Unconditional Convergence, Center for Global Development — the sign change in the convergence coefficient.
• RodrikUnconditional Convergence in Manufacturing — formal manufacturing productivity catching up at 2–3 per cent a year.
• World BankGrowth Escalators, Convergence and Divergence; and WEF on the absence of catch-up in African agricultural productivity.
• World BankWorld Development Report 2024: The Middle-Income Trap.
• World BankGlobal Economic Prospects, January 2026 — the weakest decade for global growth since the 1960s.
• IMFWorld Economic Outlook Update, January 2026.
• IFCLow Cost, High Yield: The Adaptation and Resilience Investment Opportunity (2026) — US$390bn a year of hazard losses, 1–2 per cent of combined GDP; 43 million jobs at risk by 2050.
• IEACost of Capital Observatory; and How a High Cost of Capital Is Holding Back Energy Development in Kenya and Senegal.
• Goodman et al.Heat and Learning, AEJ: Economic Policy — roughly one per cent less learning per one degree Fahrenheit of school-year heat, absent air conditioning. See Harvard Kennedy School on the air-conditioning offset.
• Arsht-RockHot Cities, Chilled Economies — urban heat productivity losses in Dhaka, Bangkok and New Delhi.
• World Resources InstituteClimate Adaptation Investments Yield Massive Returns (2025) — 320 investments, benefits above $10 per $1, IRR 27 per cent, more than half of benefits non-contingent. Full study.
• World Bank / GFDRRLifelines: The Resilient Infrastructure Opportunity — a 3–5 per cent design premium against avoided losses of 50–100 per cent of asset value.
• Global Commission on AdaptationAdapt Now (2019) — US$1.8tn for US$7.1tn in net benefits.
• UNEPAdaptation Gap Report 2025 — US$310–365bn of assessed need against US$26bn of flows.
• UNCTADA World of Debt — US$11.4tn external debt; US$921bn of net interest in 2024.
• World Economic Forum / Swiss ReThis Is How Climate Change Could Impact the Global Economy — the climate stress test and the distribution of losses.
• WMOEarly Warnings for All; and The Triple Dividends of Early Warning Systems.
• Notre DameND-GAIN Country Index — vulnerability paired with readiness.
• The ConversationBangladesh Has Saved Thousands of Lives from a Devastating Cyclone — Here's How.
• Down To EarthAhmedabad Uses Cool Roofs, Misting Bus Stops and Heat Insurance for Climate Resilience.