The Architecture of Global Convergence Under Systemic Risk

Between 2005 and 2010, 82% of low- and middle-income economies grew faster per person than the high-income average, the broadest convergence on record. Rich economies were stalled by the financial crisis, commodity prices were high and capital flowed freely. The share fell to 68% in 2010–15 and 59% in 2015–19, and in 2019–24, spanning the pandemic, rising interest rates and the commodity and food shocks of 2022, only 49% converged. The median gap between developing and high-income growth, which had been more than two percentage points a year, fell to roughly zero. The World Bank's June 2026 Global Economic Prospects reached a similar conclusion, describing a lost decade for convergence.

The stall shows up in levels too. The median developing economy's income per person, measured at purchasing-power parity, rose from 12% of the high-income average in 2000 to 15% in 2010, and has hovered between 15% and 16% since. China and India continued to converge rapidly, which keeps population-weighted aggregates rising, but the typical developing economy has been treading water for a decade and a half.

Why should a report on systemic risk start with convergence? Because convergence is both a goal and a source of resilience: richer, faster-growing economies have more fiscal space, better services and more diversified production with which to absorb shocks. And because, as the following sections show, the crises that stalled convergence after 2010 — debt distress, inflation and conflict — are precisely the ones that development pathways have been least able to reduce.

The seven crisis domains fall into two families. Crises of deprivation — hunger, child mortality, air pollution and the greenhouse-gas intensity of output — are slow-moving burdens that accumulate in poverty and fade with development. Between the early 2000s and the early 2020s, child mortality fell in 98% of developing economies, emissions per dollar of output in 88%, undernourishment in 76% and exposure to fine-particle pollution in 70%. Each of these improvements is strongly associated with income growth: comparing economies with similar starting points, one standard deviation of faster growth goes with about half a standard deviation more improvement in hunger, child mortality and emissions intensity.

Systemic crises are different. External debt service, inflation and organised violence are fast-moving, prone to sudden spikes and transmitted across borders. They improved in far fewer places: debt service as a share of exports fell in only 31% of the economies with data, inflation in 46%, and deaths from organised violence in 39%. And faster growth does not predict their improvement at all. Rising incomes did not stop debt service rising as borrowing costs and dollar debts grew, did not stop inflation returning after 2021, and did not prevent conflict from spreading across the Sahel and the Horn of Africa.

This split matters for the idea of pathways that reduce many crises at once. Growth-led development has been a powerful multi-crisis reducer — but for the crises of deprivation. Systemic crises need different tools: fiscal and monetary institutions, debt management and restructuring, security and political settlements. The United Nations trade and development agency estimates that developing countries paid a record $921 billion in interest on public debt in 2024 and that 3.4 billion people live in countries that spend more on interest than on health or education — a systemic burden that eats directly into the investments that reduce deprivation.

The matrix shows, for each of 13 development levers, how its improvement between the early 2000s and early 2020s relates to improvement in each crisis domain and to convergence, comparing economies with similar starting incomes, starting crisis levels and starting lever levels. Values are standardised: a value of 0.4 means that economies whose lever improved by one standard deviation more saw the crisis improve by 0.4 standard deviations more. Teal cells are co-benefits, coral cells trade-offs, and bold cells are statistically distinguishable from zero.

Read with growth included, the pattern is broad. Gains in electricity, clean cooking, drinking water, sanitation, internet use, investment, crop yields and government effectiveness all go with faster progress against hunger and child mortality, and most with faster convergence; electricity, clean cooking and government effectiveness also go with falling emissions intensity. Government effectiveness is the only lever with significant associations in four of the seven domains and the strongest association with convergence. Investment and women's labour participation are linked to fewer conflict deaths. The renewable share of energy appears as a trade-off only because, in poor economies, it mostly measures firewood and charcoal, which decline as households move to modern fuels.

Switch to "beyond growth" and most of the matrix fades. Once each economy's income growth is held constant, energy access, internet use and sanitation no longer add much: their association with crisis reduction is largely the growth they came with. What remains is narrower and more interesting. Government effectiveness and women's labour participation are still tied to fewer conflict deaths; gains in lower-secondary completion and in drinking water go with falling debt service; health spending goes with cleaner air. Click a lever to see its profile.

To capture breadth in one number, each economy receives a multi-crisis reduction index: the average of its standardised improvements across the seven domains. An economy that improved more than average on most domains scores above zero. The chart ranks levers by their association with this index, with and without holding income growth constant. With growth included, gains in drinking water, government effectiveness, schooling, electricity, sanitation, crop yields and clean cooking all go with broader crisis reduction.

Beyond growth, only two remain clearly distinguishable from zero. Economies whose government effectiveness improved by one standard deviation more reduced crises by about 0.23 standard deviations more across the board, and those whose access to basic drinking water improved by one standard deviation more by about 0.22. Lower-secondary completion, cereal yields and women's labour participation come close, at the edge of significance. Energy access, sanitation and internet use, so prominent in the first specification, add nothing measurable once growth is accounted for.

The ranking does not say that energy or connectivity are unimportant: they are among the strongest correlates of growth itself, and growth is the biggest single reducer of deprivation. It says that their crisis-reducing power runs through growth. State capacity and basic water services, by contrast, appear to reduce crises in ways growth alone does not — consistent with their role in public health, in managing scarce resources and in preventing and resolving conflict.

Placing each economy by its convergence and its multi-crisis reduction since the early 2000s gives four pathways. Forty-six economies, with two-thirds of the developing world's people, did both: they grew faster than rich economies and reduced crises more than average. They include China, India, Indonesia, Nepal, Myanmar, Ethiopia and Rwanda, much of Central Asia and the Andes, and several West African economies. The two dimensions are clearly related — the correlation is 0.51 — which supports the idea that convergence and resilience can be built together.

Twenty-four economies, with 19% of developing-world people, converged without reducing crises more than average, among them Bangladesh, Pakistan, Nigeria, Kenya, Egypt and Malaysia: growth outpaced rich economies, but debt, inflation, pollution or violence rose or fell too little to keep pace. Only five economies reduced crises without converging, including Brazil and Angola. And 26, with 10% of developing-world people, fell behind on both — a group dominated by conflict-affected states such as Syria, Sudan, Libya and the Central African Republic, by commodity exporters hit after 2014, and by economies such as Argentina, Lebanon and South Africa caught in debt, inflation or stagnation.

The pathways show that the binding constraint differs. For the converging-only group, the missing pieces are mostly systemic: debt burdens and inflation that growth did not tame. For economies falling behind on both, conflict and institutional breakdown sit at the root, and no amount of ordinary development spending substitutes for a political settlement and a functioning state. Hover over a bubble to see each economy.

The map shows each economy's pathway, its convergence, its multi-crisis reduction and its crisis burden today — the average of its standardised levels of hunger, child mortality, air pollution, debt service, inflation and conflict deaths in 2019–23. The converging-and-reducing pathway covers most of South and East Asia, much of Central Asia and the Andes and parts of East and West Africa. Falling behind on both concentrates in the Middle East and North Africa's conflict zones, the Sahel and Central Africa, and in a few middle-income economies with chronic macroeconomic instability.

The burden layer shows that progress has not erased risk. Several economies that converged and reduced crises still carry heavy burdens today, because they started from very high levels of hunger or child mortality; others with modest burdens are deteriorating. The country card lists each domain's change in its own units, from undernourishment and child deaths per 1,000 births to debt service and conflict deaths per 100,000, together with the economy's growth gap with rich economies.

Economies are included if they had at least 1 million people and are classified as low or middle income; high-income economies, the benchmark for convergence, are left blank. Small and data-poor economies, including several small island states, are missing — a significant gap, because their exposure to systemic and climate shocks is among the highest in the world.

The evidence suggests an architecture for development pathways that reduce several crises at once. Its foundation is state capacity: government effectiveness is the strongest correlate of convergence, predicts broad crisis reduction beyond growth and is tied to fewer conflict deaths. Without it, the rest does not stand — the economies that fell behind on both convergence and crisis reduction are disproportionately those where the state broke down. Building capable, accountable institutions is slow and cannot be bought, but it is where development finance and policy reform pay off across the most crises.

The ground floor is growth-linked development: energy, water, sanitation, schooling, connectivity, agricultural productivity and investment. It has been extraordinarily effective at dissolving the crises of deprivation, mostly by raising incomes and partly through direct effects on health and nutrition, and it should be designed for co-benefits — clean rather than dirty energy, water systems that serve health and agriculture together. The upper floor is the one most pathways have not built: buffers against systemic shocks. Fiscal rules and revenue that hold up in downturns, central banks that keep inflation anchored, debt management that limits currency and rollover risk, social protection that scales up automatically, and security and political institutions that settle disputes without violence.

The roof is international. No developing economy can insure itself alone against global interest-rate cycles, commodity shocks or pandemics. A functioning architecture for debt restructuring, contingent and countercyclical finance, liquidity in crises and conflict prevention is what allows the upper floor to hold. When that roof leaks — as rising interest bills, net outflows to creditors and stalled restructurings have shown since 2022 — even well-built national pathways stall, which is part of why convergence has faded.

The sample is the 102 low- and middle-income economies (current World Bank classification) with at least 1 million people and complete income data. All changes are long differences between five-year averages for 2000–04 and 2019–23. Crisis measures, signed so that positive values are improvements, are: prevalence of undernourishment (FAO, from 2001); log under-five mortality (UN IGME); population-weighted PM2.5 exposure; external debt service as a share of exports of goods, services and primary income; log(1 + inflation); log greenhouse-gas emissions per dollar of PPP GDP (EDGAR via WDI); and log(1 + deaths from organised violence per 100,000 people), from UCDP GED 26.1 located by event.

Levers are changes in electricity access, clean cooking access, renewable share of final energy, lower-secondary completion, current health spending as % of GDP, basic drinking water, basic sanitation, internet use, gross capital formation as % of GDP, tax revenue as % of GDP, women's labour-force participation, WGI government effectiveness (1996–2004 baseline) and log cereal yield. Each lever–crisis pair is estimated by OLS with heteroskedasticity-robust standard errors, controlling for initial log GDP per person (PPP), the initial crisis level and the initial lever level; the "beyond growth" specification adds annual growth of GDP per person. Coefficients are standardised. Convergence is annual GDP-per-person growth minus that of the population-weighted high-income aggregate.

The multi-crisis index averages the seven standardised improvements for economies with at least five domains observed; the burden index averages standardised 2019–23 levels of six domains (excluding emissions intensity). Caveats: associations, not causal effects; many tests, so isolated significant cells may be chance; the renewable share includes traditional biomass; debt service and tax data cover fewer economies; levers move together, so their separate associations are not additive. Related reports: infrastructure and economic convergence, convergence through crisis, digital infrastructure as a resilience multiplier and resource scarcity and armed conflict.