The Architecture of Global Polycrisis Mitigation

The framework borrows from systems engineering. A shock propagates through a network: it first hits nodes (households, firms, budgets and services), then travels along links (trade, prices, finance and supply chains), and is either absorbed by buffers or becomes an outcome: deaths, hunger, lost output, debt distress or unrest. Each function acts at a different point. Robustness lowers how much damage a node takes, decoupling reduces how many links a shock can use, and buffers absorb what arrives.

Steering is different in kind: it is a feedback loop. Governments with good information and capacity can see a shock coming, target relief, reroute supply and coordinate the other functions. In engineering terms it is the controller, not a component. The global backstop is the outer layer. Some shocks, such as a pandemic, a global price spike or a sudden stop in capital flows, are larger than any one country's buffers and correlated across countries, so they can only be pooled at the global level.

Two systems ideas shape the analysis. The first is complementarity: a strong buffer is of limited use if steering cannot deploy it, so architecture quality may depend on the weakest function, not the average. The second is coupling: shocks interact, and a country that absorbs one badly enters the next weaker. Sections 2 to 5 measure the architecture and test it against these ideas. Sections 6 and 7 turn to the global layer.

Scored as percentile ranks among 160 countries, the median low-income country scores 13 on robustness, 42 on decoupling, 29 on buffers and 14 on steering. The median high-income country scores 81, 56, 68 and 86. Three of the four functions climb steeply with income, as health systems, fiscal capacity and administrations mature. Decoupling barely moves: low-income countries are dependent on commodity exports, while rich ones import most of their energy.

The functions are also lumpy. Robustness and steering are almost the same thing across countries (r = 0.80). Buffers go with both (r = 0.56 and 0.48). Decoupling is nearly independent of everything (r = 0.08 with buffers and 0.27–0.29 with the others). Building health systems and state capacity tends to happen together, but structural exposure to the world's energy, food and commodity markets is a separate decision, shaped by geology, geography and trade policy.

This matters for design. A strategy that raises general state capacity will lift robustness, buffers and steering together but leave decoupling behind. Decoupling needs its own instruments: domestic clean energy to cut fuel imports, agricultural productivity and storage to reduce food import dependence, and export diversification. These are industrial and energy policies more than social or fiscal ones, so they are often missing from resilience plans.

Because the functions are lumpy, nearly every country has one that lags well behind the others. Among the 134 countries with complete scores, the median gap between the average and the weakest function is about 16 percentile points, and in high-income countries the weakest function scores a median of 51 against an average of 70. Wealth raises the floor, but it does not remove the weak link.

The weak links differ. Decoupling is the weakest function for 49 countries: energy importers in Europe and East Asia such as Germany, France, Korea and Singapore, small import-dependent economies such as Jordan, Jamaica and the Gambia, and commodity exporters such as Australia, Chile, Saudi Arabia and Mongolia. Robustness binds in 29 countries with 3.3 billion people, including India, Nigeria, Pakistan, Bangladesh, Ethiopia and Kenya, where health coverage, sanitation and poverty are the main gaps. Buffers bind in 30, among them the United States, where central-government interest payments exceed a fifth of revenue, the United Kingdom, Egypt, Brazil and Mexico. Steering binds in 26, including China and Vietnam, where statistical performance scores lower than other capacities, and conflict-affected states such as Ukraine, Sudan and Yemen.

If architecture works like a chain, investment should go to the weakest function first. The 2020–2023 data cannot settle whether shock damage tracked the weakest function more than the average: both explain about the same amount. But the weak-link view has a practical advantage. It points each country to a different priority, instead of the same generic package everywhere.

Between 2020 and 2023 the world ran an involuntary stress test: a pandemic, the deepest global recession since the 1930s, then an energy and food price shock and the fastest monetary tightening in decades. If national architecture matters, countries with stronger pre-shock functions should have fared better, after allowing for income. The matrix shows each function's association with each outcome, signed so that positive means protective. Darker cells are statistically stronger.

For the economic outcomes, the answer is close to no. Function scores and income together explain 2–5% of the variation in the 2020 growth hit, scarring by 2023 and the 2022–23 inflation spike, and no function is significant at conventional levels. These shocks were global and correlated. Lockdown timing, tourism and remittance exposure, energy-market structure and world interest rates moved outcomes far more than the slow-moving capacities measured here. Some apparent buffers were also spent defending against the shock itself.

Steering is the clear exception. One standard deviation more steering capacity is associated with about 62 fewer excess deaths per 100,000 people in 2020–21 after age structure and income, and with a 5.7-point smaller rise in food insecurity in 2021–23. Steering alone raises the explained share of the food insecurity change from 3% to 22%. Robustness points the same way for deaths but is not significant once income is controlled. The lesson is not that buffers and decoupling are useless. It is that in a global shock, the capacity to see, decide and target counted most.

Polycrisis means shocks interact. The outcomes of 2020–2023 were linked: countries whose economies were furthest below their pre-COVID trend in 2023 had larger inflation spikes (r = −0.35) and larger rises in food insecurity (r = −0.29), and inflation and food insecurity rose together (r = 0.32). The 2022 price shock hit hardest where the pandemic had already left economies weaker, and translated into hunger where incomes could not keep up. COVID mortality, by contrast, was largely unrelated to the economic outcomes.

Counting the outcomes in which a country was among the hardest-hit third, 33 countries with 2.2 billion people were hit in three or more of five, and 10 in four or five. They are mainly middle-income: India, Pakistan, Sri Lanka, Myanmar and Nepal in Asia; Argentina, Peru, Colombia, Bolivia and Ecuador in Latin America; Lebanon, Iraq, Libya and Tunisia; and Ukraine, Moldova, much of Central and Eastern Europe, Spain and Portugal, where excess mortality and energy-driven inflation compounded.

Compounding is the strongest argument for treating mitigation as one architecture. A country that spends its buffers on the first shock enters the second with less, and a country whose steering fails once loses public trust it needs next time. Rebuilding buffers between shocks, and monitoring compound exposure, not single hazards, should be part of the design.

A national buffer is thin when foreign reserves cover less than three months of imports or interest payments take more than a fifth of government revenue. By these thresholds, 43 countries with 3.5 billion people have thin buffers. They include India, Pakistan, Bangladesh and Sri Lanka, Egypt, Ethiopia, Kenya, Ghana, Uganda, Malawi and Zambia, and Brazil and Mexico. The United States crosses the interest threshold, but as the issuer of the reserve currency it faces a different constraint.

For developing countries with thin buffers, the effective buffer in a crisis is external: IMF programmes, regional financing arrangements, central-bank swap lines and multilateral development bank crisis tools. That safety net has grown but is uneven. The IMF's 2025 stocktaking put regional financing capacity at about US$1.3 trillion, concentrated in Europe and Asia. Bilateral swap lines reached about US$517 billion in 2025, according to Boston University's tracker, but most developing countries have no access to the swaps advanced economies extend to each other, and middle-income countries have the least emergency finance available.

Thin buffers also help explain why national scores did not predict 2022 inflation well. Many countries with thin buffers spent reserves and fiscal space defending currencies and subsidising fuel, so the buffer was consumed by the shock. Rebuilding buffers between shocks, through reserve accumulation, debt reprofiling and pre-arranged contingent finance such as climate-resilient debt clauses, is cheaper than borrowing in a crisis.

Since 2022 the global layer has grown. Early Warnings for All aims to cover everyone on Earth by 2027, and 119 countries, about 60%, reported multi-hazard early warning systems in 2025, up 113% in a decade, though only 43% of small island states. The Pandemic Fund's three funding rounds are set to mobilise more than US$11 billion with co-financing across 98 countries. The Fund for responding to Loss and Damage had received about US$790 million in pledges by 2025, and the IMF's Resilience and Sustainability Facility approved five new programmes in its 2026 financial year.

Coverage remains thin. International public adaptation finance runs at about US$26 billion a year, against needs UNEP estimates at US$310 billion or more a year by 2035, under a tenth. The loss-and-damage fund's first call drew US$2–3 billion of requests for US$250 million, and its first disbursements slipped to the end of 2026. The financial safety net is largest where it is least needed. Each mechanism also tends to work in its own silo, financing one hazard through one channel.

A systems view suggests three fixes. Capitalise the layer in proportion to correlated risk, because shocks that hit many countries at once cannot be self-insured. Connect it: pre-arranged finance should be triggered by the same early warnings that move people out of harm's way, as anticipatory-action pilots already show. And fund steering globally, through data, statistics and public-sector capacity, because it was the national function that did most work in 2020–2023 and is the one the backstop relies on to deploy money well.

The architecture layer shows the mean of the four function scores. It is highest in Switzerland, Finland, Germany, Norway, Australia, Israel, Korea and Singapore, and lowest in Liberia, Sierra Leone, Yemen, Guinea, Afghanistan, Haiti and much of Central Africa. The weakest-link layer shows the lowest of the four, which makes many middle- and high-income countries look less secure than their average.

The function layers show where each is strong or weak. Decoupling is weakest in small import-dependent economies, from the Gambia and Liberia to Jamaica, Lebanon and Moldova, and lower than average in energy-importing Europe. Buffers are weakest in countries with low reserves and heavy interest burdens. Steering mirrors robustness, though not exactly. The shocks layer counts how many of the five 2020–2023 outcomes hit a country hard.

Hover over or tap a country for its card: its four function scores and the indicators behind them, its weakest link and its 2020–2023 outcomes. Clicking a country loads it into the architecture builder in section 9. The sortable table lists all 160 countries.

Pick a country and raise any of its four function scores with the sliders, which add percentile points (capped at 100). The radar chart compares the current and strengthened profiles, and the outputs show the new average, the new weakest-link score and which function now binds. Spreading the same points evenly raises the average but often leaves the weakest link where it was. Putting them into the weakest function lifts the floor.

The builder also translates steering gains into the two outcomes where steering was associated with less damage in 2020–2023, using the estimated effects per standard deviation. These figures are illustrative. They come from one episode, they are associations, not causal effects, and the next shock may reward different functions. Gains in the other functions are not translated because their 2020–2023 associations were weak.

Try the United States, whose weakest link is buffers, against Germany (decoupling), India (robustness) and China (steering). Then try a low-income country such as Ethiopia, where every function is low and the architecture has to be built across the board, starting with robustness and steering.

Invest in steering. Government effectiveness and statistical capacity were the national functions most clearly associated with fewer deaths and less hunger in 2020–2023. They are cheap relative to the losses they avert: digital public infrastructure for targeting, timely statistics, disease and food-security surveillance, and a civil service able to act on them. Steering is also what lets the other functions work, from deploying reserves to triggering social protection.

Treat decoupling as resilience policy. It is the most common weak link and the least connected to general development. Domestic clean energy cuts the fuel import link, agricultural productivity and strategic storage cut the food link, and export diversification cuts the commodity-revenue link. These should sit in national resilience plans, not only in industrial or climate strategies. Rebuild buffers between shocks. Accumulate reserves, reprofile expensive debt, pre-arrange contingent finance and keep social protection scalable, so the next shock does not arrive on top of the last.

Design for compounding and capitalise the global layer. Monitor compound exposure, not single hazards. Size global mechanisms to correlated risk, connect them to early warnings with pre-agreed triggers, and close the gaps where they are widest: middle-income access to emergency liquidity, small island early warning, adaptation finance and loss and damage. The architecture exists in outline. The task is to finish it, fund it and wire it together.

Coverage: 160 countries with at least one million people and a World Bank income group; 134 have all four function scores. Indicators (latest 2014–2024; pre-shock latest to 2019): robustness = WHO UHC service coverage index, basic sanitation (SH.STA.BASS.ZS), undernourishment (SN.ITK.DEFC.ZS, inverted), extreme poverty (SI.POV.DDAY, inverted); decoupling = net energy imports % of use (EG.IMP.CONS.ZS), food import bill % GDP (TM.VAL.FOOD.ZS.UN × imports ÷ GDP), commodity share of exports (food, agricultural raw materials, ores and metals, fuels), all inverted; buffers = reserves in months of imports (FI.RES.TOTL.MO, excluded for the US, UK, Japan, Switzerland, Canada, Australia and euro-area members), interest % of revenue (GC.XPN.INTP.RV.ZS, inverted), social protection coverage (ASPIRE); steering = WGI government effectiveness and the Statistical Performance Index.

Each indicator is converted to a percentile rank; a function score is the mean of its available indicators (at least half required). The architecture score is the mean of the four functions, and the weakest-link score is the minimum. Stress-test outcomes: Economist cumulative excess deaths per 100,000 to end-2021; GDP-per-head growth in 2020 minus its 2015–19 mean; log GDP per head 2023 minus 2019 minus four years of 2015–19 trend growth; mean CPI inflation 2022–23 minus 2017–19 (countries with pre-shock inflation above 50% excluded); FIES moderate-or-severe food insecurity 2021–23 minus 2017–19. Outcomes are winsorised at the 2.5th and 97.5th percentiles; each function enters separately, standardised, with log GDP per head PPP (2019) and, for deaths, the over-65 share; HC1 standard errors.

Hardest-hit = worst third of each outcome; thin buffers = reserves under three months of imports or interest above 20% of revenue. Caveats: percentile scores are relative and mask absolute levels; indicators have uneven coverage and years; reserve adequacy depends on exchange-rate regime; one global episode cannot identify causal effects, and the weakest-link and mean models fit equally well; global-layer figures are from institutional reports and press coverage, cited below.