A Unified Framework for Understanding and Mitigating the Global Polycrisis

The domains are built from public, comparable indicators. Environmental: INFORM's natural hazard and exposure score, the mean recorded disaster damage as a share of GDP and the mean disaster death rate, 2000–2024. Economic: central-government interest as a share of revenue, external debt service as a share of exports, reserves in months of imports and the commodity share of exports. Technological: the diffusion index (connectivity, power, services) and the IMF AI Preparedness Index, both inverted. Geopolitical: INFORM's human hazard (conflict), net energy imports and the food import bill. Capacity: government effectiveness, the Statistical Performance Index and INFORM's institutional coping gap.

Poorer countries score worse on three of the four domains and on capacity. The mean technological score is 18 in high-income, 48 in upper-middle, 72 in lower-middle and 89 in low-income countries; the economic score 40, 51, 58 and 57; the geopolitical 46, 53, 60 and 64; and weak capacity 22, 48, 66 and 78. The environmental score is nearly flat, 48, 58, 50 and 47, because hazard is a property of geography and exposure, and is large for rich countries as well (earthquakes, storms and fires). Upper-middle-income countries, with the largest populations in hazard zones, score highest.

Exposure and weak capacity travel together: the correlation between mean exposure and weak capacity is 0.72, which is mostly the technological and geopolitical domains and the income gradient. Of 145 countries with all scores, 56 (2.1 billion people, 26% of the total) have above-median exposure and weak capacity, 25 (2.5 billion) have high exposure and strong capacity (including India and Indonesia), 48 (3.0 billion) low exposure and strong capacity, and 16 (0.3 billion) low exposure and weak capacity. The first group is where diagnosis and the response gap coincide.

The pairwise correlations are low. Environmental exposure correlates with the economic, technological and geopolitical scores at −0.02, 0.02 and 0.05. The economic and technological scores correlate at 0.44 and the technological and geopolitical at 0.39, mostly because both fall with income; controlling for income, the correlations become 0.23 (economic and technological), −0.29 (economic and geopolitical, because commodity exporters have healthy external positions but high conflict and import exposure) and −0.10 (technological and geopolitical). No pair is strong enough to treat as a single dimension.

The principal components agree: the first carries 39% of the variance, the second 27%, the third 24% and the fourth 10%. A single score, which is the first component in all but name, discards 61% of the information in the four domains; three components are needed to retain 90%. This is not a defect of the data. It reflects the substance of the polycrisis: the places exposed to earthquakes and cyclones are not the places most dependent on imported energy or lagging in digital capacity, and different countries are in trouble for different reasons.

The implication for a unified framework is that the unity lies in the method, not in a number. Putting all four domains on the same 0–100 scale, defined by the same rule, with capacity alongside, allows comparison across countries and domains. Adding them up loses the one thing a policymaker needs to know, which is which domain is the problem, and the sections that follow show what else is lost.

Counting domains in the high band gives a rule that cannot be gamed by weights. Of the 146 countries, 51 have none (33 are high-income), 47 have one and 48 have two or more; 14 have three and none has four. The 20 countries with only environmental risk high hold 4.3 billion people because they include India and China; the 10 with only technological risk high include Nigeria, the Democratic Republic of the Congo and Tanzania; the 10 with only geopolitical risk high are small, import-dependent open economies (the Netherlands, Belgium, Hong Kong, Jordan, Lebanon and Belarus); and the 7 with only economic risk high include Egypt, Kenya and Argentina.

The most common multi-domain pairs are economic and technological (11 countries), technological and geopolitical (10), and the environmental, technological and geopolitical triple (5). Those are the poorest and most fragile: the 14 countries with three high domains are Afghanistan, Benin, Guinea-Bissau, Honduras, Haiti, Jamaica, Cambodia, Mozambique, Malawi, Nicaragua, Pakistan, Senegal, El Salvador and Yemen. The pattern is the polycrisis in miniature: not one cause but several, each in a different channel, and each calling for a different lever.

A clustering of the four scores into groups finds no sharp archetypes: the best k-means solutions have a silhouette of only 0.22–0.28, which means profiles form a continuum, so the report uses the transparent counting rule, not a clustering. The profile is what the explorer in section 7 reports for any country, alongside its rank under the chosen weights and the range of ranks across all weightings.

To test the stability of a unified ranking, the four domain weights were drawn at random 10,000 times (a flat distribution over all weightings that sum to one) and three aggregators were applied: the weighted arithmetic mean, the weighted geometric mean (which punishes a country that is low in one domain less than the arithmetic mean punishes it), and the worst domain (each score multiplied by its weight, take the maximum). Across the weightings, the median country's 5th–95th percentile rank interval is 62 places wide out of 146. Even the riskiest countries move: Afghanistan ranks between first and 87th, Pakistan between third and 40th.

Only three countries, Haiti, Mozambique and Yemen, are in the worst twenty in at least 90% of arithmetic and geometric weightings. 44 more are in it sometimes (between 10% and 90% of the time), and 75 of the 146 (51%) reach the worst twenty under at least one weighting. At equal weights the arithmetic and geometric lists share 16 of 20 countries, the arithmetic and worst-domain lists 13 and the geometric and worst-domain lists only 10. The rank correlations between aggregators fall from 0.98 (arithmetic against geometric) to 0.72 (geometric against worst-domain).

The practical rule is to publish the range. A country's rank is a statement about a weighting as much as about the country, and a hotspot list should state how many weightings it holds under. Three countries are hotspots under nearly every reasonable choice; the rest of any equal-weight list is there because of the domains it happens to weight. The explorer in section 7 lets the reader set the weights and see this directly.

The test uses the five outcomes that report 2026-62 built for the 2020–23 polycrisis: cumulative excess deaths to end-2021, the 2020 growth shortfall, 2023 GDP-per-head scarring, the 2022–23 inflation rise and the rise in food insecurity. The harm index is the mean percentile of those available (at least three). The environmental score has the strongest rank correlation with harm, 0.31 (0.37 with excess deaths and 0.30 with the 2020 shortfall); the economic score is −0.16, the technological 0.00 and the geopolitical 0.12. The combined scores do worse than the environmental score alone: 0.09 for the arithmetic mean, 0.15 for the geometric, −0.05 for the worst-domain and 0.02 for the breadth count.

Regressing harm on the four domains gives R² = 0.14 (0.17 with log income); adding the six pairwise interactions raises it to 0.26, and the interactions are jointly significant (F = 2.7, p = 0.017). But the signs are mixed: the technological-geopolitical product is negative (−7.5, standard error 2.8), the economic-technological product negative (−5.0, 2.3) and the environmental-economic product positive (4.5, 2.9). Counting the domains in the high band is non-monotonic: mean harm of 52.8 with none, 49.2 with one, 46.6 with two and 52.5 with three (13 countries). The data do not show compounding in general. A weak-capacity score added to the model contributes little.

These results do not say the framework is wrong, but they bound what it can claim. The 2020–23 episode was one global shock whose damage tracked the pandemic and the price shock more than national exposure to any one domain; exposures are measured at different dates; the environmental correlation likely reflects regional factors; and four scores are not enough to forecast harm. As a diagnostic, the framework describes where each country is exposed and why; as a predictor of harm it is untested beyond this episode. The multi-episode shock register of report 2026-65, which finds that economic shocks raise the next year's risk of displacement or mortality shocks 2.5 times, is the better evidence on transmission.

The framework's last step is the match. A country whose profile is led by the environmental domain needs early warning and building standards; one led by the economic domain needs cost-of-capital relief and revenue; one led by the technological domain needs the foundations (power, connectivity, skills) before applications; one led by the geopolitical domain needs to reduce import and commodity dependence and invest in prevention; and any country with weak capacity needs steering capacity (statistics, government effectiveness) first, because the evidence of report 2026-62 is that steering is the function associated with the smallest harm in 2020–23.

Multi-domain countries need a sequence, not a menu. Fourteen countries are high in three domains and 48 in two or more; for them the levers interact, since early warning needs connectivity, relief needs a revenue base, and conflict prevention needs the fiscal room that debt service consumes. The transmission evidence suggests starting with the economic and technological foundations that the other channels run through: the observed network of report 2026-65 has economic shocks and shutdowns raising the next year's risk of displacement or mortality shocks 2.5 and 2.0 times.

The resource question is scale. The reference early-warning investment of report 2026-80 would cost $4 billion a year for all 8 billion people; the interest bill of the 34 squeezed countries is $495 billion (2026-79); the adaptation finance gap is $310–365 billion a year against $26 billion of international public flows. The framework does not make resilience cheaper, but it can make it better targeted: robust hotspots first, then the countries whose dominant domain has a proven lever, and the weights stated.

The explorer recomputes the unified score for 146 countries under the weights and aggregator chosen. The weighted mean and geometric mean use the weights directly; the worst domain takes the highest of the four weighted scores (each multiplied by four times its weight); the breadth rule counts the domains in the high band and breaks ties with the weighted mean. The output reports how many of the equal-weight worst twenty remain in the list, how many countries move ten places or more against equal-weight arithmetic scoring, and the flag for each country in the list.

Try putting all the weight on one domain: the ranking becomes that domain's ranking, and a country that was a robust hotspot may leave the list. Try the worst-domain rule at equal weights: countries with one very high score rise, and countries that score moderately in all four domains drop out of the list. The profile panel shows the country's four scores and weak capacity against the income-group median, so that the reader can see which domain leads.

The explorer is a way of seeing the framework's choices, not a recommendation of weights. Any weighting expresses a judgement about which domain matters most for a purpose: a humanitarian planner, a central bank and a climate funder would choose differently, and the report's point is that they should say so, and that the resulting hotspot list should be accompanied by its range.

Report a profile, not a rank. Four domain scores and a capacity score carry information that a single index destroys: the correlations among the domains are 0.05 or less for environmental risk and below 0.45 otherwise, and a single score keeps 39% of the variance. Risk assessments should show the four scores, the dominant domain and the number of domains in the high band, and treat a composite as a convenience.

Publish the range and the robust list. A hotspot is a statement about weights. A list should say how many weightings it survives: three countries are worst-twenty hotspots under nearly every choice, 44 sometimes, and a median country's rank spans 62 places. Funders can start with the robust hotspots, and explain the rest by the domain that put them there.

Match the lever and check the capacity. The levers in section 6 differ by domain, and each requires capacity to deliver. Multi-domain countries need sequencing, starting with the foundations that other channels depend on; capacity needs to be built first where it is weak. The evidence that domains compound in general is not there; the evidence that transmission exists and that steering matters is. A framework that is honest about this will be harder to misuse.

Sample: 146 countries with at least one million people and all four domain scores (10 more have three and are not ranked); population from World Bank WDI/WPP in report 2026-62. Indicators: environmental = INFORM Risk Index Mid-2026 natural hazard and exposure (0–10), mean recorded disaster damage as % of GDP and mean disaster death rate 2000–2024 (EM-DAT via Our World in Data); economic = central-government interest % of revenue, total external debt service % of exports (latest 2019–24, WDI), reserves in months of imports (inverted; not scored for reserve-currency economies) and commodity share of exports (report 2026-62); technological = the H Heuristics diffusion index and the IMF AI Preparedness Index 2023 (both inverted); geopolitical = INFORM human hazard, net energy imports % of use and food import bill % of GDP (report 2026-62); capacity = WGI government effectiveness and the Statistical Performance Index (inverted) and INFORM institutional coping (lack of). Each indicator becomes a percentile rank (0–100, higher = riskier); a domain score is the mean of its available indicators, at least half required.

Robustness: 10,000 weight vectors drawn from a flat Dirichlet distribution on the four domains; for each, countries are ranked under the weighted arithmetic mean, the weighted geometric mean (scores floored at 1) and the worst domain (maximum of four times the weight times the score). Frequencies are the share of draws in which a country ranks in the worst twenty; rank intervals are the 5th and 95th percentiles. Overlaps and Spearman correlations use equal weights. Profiles: a domain is high at 67 or more. Validation: the harm index is the mean percentile of excess deaths per 100,000 to end-2021, the 2020 growth shortfall, 2023 GDP-per-head scarring, the 2022–23 inflation rise and the 2021–23 food-insecurity rise (report 2026-62); OLS of the harm index on standardised domain scores with log GDP per person; the interaction test is an F test of the six pairwise products. K-means with 3–8 clusters (30 initialisations) gave silhouettes of 0.22–0.28.

Caveats: percentile ranks are relative and compress the distribution towards the middle; indicators are measured in different years and with uneven coverage; the technological domain measures the capacity gap and omits the exposure of advanced economies to cyber, platform or AI risks; the geopolitical domain omits sanctions, alliances and trade concentration beyond energy and food; INFORM's hazard scores are model-based; the harm index describes one global episode; the interaction test has 145 observations and ten regressors; and the framework describes cross-sectional exposure and does not forecast. Lever evidence is taken from the cited reports in this series, which carry their own caveats.