Defining and Understanding Polycrisis Resilience

Holling's 1973 paper separated resilience (the amount of disturbance a system can absorb and still persist) from stability (the return to an equilibrium), and his 1996 essay called the second "engineering resilience" and the first "ecological resilience". The distinction matters for a polycrisis. A system that returns quickly to the same state (engineering) may be a system that never changes; one that absorbs large shocks by reorganising (ecological) may look unstable on the way. Development economics tends to use the first when it measures recovery times and the second when it speaks of adaptation and transformation.

The UN's list of verbs, resist, absorb, accommodate, adapt, transform, recover, is already a set of different properties. A system can be good at resisting and poor at recovering, or recover fast by adapting away from what it was. Here the list is turned into five measures: resistance (the depth of the output loss, the verbs resist and absorb), recovery (how far output sits from its old path after three or four years), continuity (whether life expectancy keeps to its trend, preserving an essential function), price stability (the rise in inflation) and adaptation (growth after the shock relative to growth before).

The polycrisis usage adds a fourth idea, compounding. If risks interact so that the whole exceeds the sum of the parts, resilience to one shock may not predict resilience to a combination, and a country's standing may depend on which shocks arrive together. The implication for definition is that "resilient" has no meaning without a shock class, which is the starting point for the relational definition in section 6. Critics who call the term too elastic to operationalise are making the same point: it needs to be tied down.

In 2009 the median country's growth fell 4.3 points below its 2004–08 average and three years later its output was 8.5 points below its old trend; 85% of countries grew more slowly than before. In 2020 the median shortfall was 6.1 points, with 98% of countries growing more slowly than before, and four years later output was 3.3 points below the 2015–19 trend. The two episodes are different tests: 2009 was a financial and trade shock concentrated in open and financial economies, while 2020 was a health and mobility shock that hit nearly everyone and was followed by a global price shock in 2022.

Each dimension is a percentile rank among the countries with data (higher = more resilient). The composite is the mean of the available percentile ranks, with at least four of five required, so it is a relative standing, not an absolute score. By income, the composite is similar across groups in 2020 (about 50 in each group) and varies more in 2009 (44 for high-income, 49, 57 and 55 for the others) because high-income economies lost more relative to their pre-crisis trend.

These choices have consequences, set out in section 6. Measuring against a country's own trend rewards countries with weak pre-crisis growth, because they have less to lose; measuring recovery against the pre-shock trend rewards countries that grew fast before; and measuring inflation as a rise penalises countries that had it low. Every definition embeds a reference path, and the choice of reference path is a choice about what counts as resilient.

In the 2009 episode resistance and recovery correlate at 0.80 and resistance and adaptation at 0.49: countries that lost least tended to rebound. In 2020 the correlation between resistance and recovery falls to 0.35 and resistance and adaptation to 0.07, since the depth of the 2020 loss reflected exposure to the pandemic and to mobility restrictions, plausibly, while recovery depended on what came after. Recovery and adaptation correlate at 0.85 and 0.77, partly by construction, since both use post-shock growth. The mean absolute correlation among the other pairs is 0.21 in 2009 and 0.13 in 2020.

Price stability and continuity are the dimensions that do not fit. The correlation of price stability with the other four dimensions is between −0.10 and 0.25 in both episodes, and of continuity between −0.14 and 0.21. A country that keeps prices steady is not systematically one that keeps output near trend or people alive; countries that preserved life expectancy (whose pandemic mortality was modest, often because of age structure) are not those that preserved output. A definition that emphasised any one of these would rank countries quite differently from one that emphasised another.

Few countries are resilient on every count. 10 of 144 countries in 2020 (7%) and 14 of 140 in 2009 (10%) are in the top third on four or five dimensions, against about 4.5% if the dimensions were independent; 4 countries are in the top third on all five in 2020 and one in 2009. Dimensions agree a little more than chance, not enough to support a single resilience score. Counts at the bottom are similar: 9 countries in 2020 are in the bottom third on four or five dimensions.

The same dimension measured in the two episodes barely correlates: resistance 0.20, recovery −0.04, price stability −0.07, adaptation 0.02, continuity −0.14, and the composite 0.09. The growth shortfalls of three global downturns tell the same story: 2009 and 2015 correlate at 0.06, 2009 and 2020 at 0.21 and 2015 and 2020 at −0.14. Raw growth in the shock years correlates a little more (0.16 to 0.23), because countries differ in their long-run growth, but that is structure, not resilience. Within income groups the composite's persistence is 0.43 for high-income countries, −0.08 for upper-middle, 0.01 for lower-middle and −0.14 for low-income countries.

The composite ranks move a lot. A quarter of the countries change by 25 percentile points or more between the episodes and 8% by 40 or more; the median change is 15. Lebanon goes from the 68th percentile in 2009 to the 3rd in 2020 (the 2019–20 financial collapse); Iraq from 86 to 28, Lao PDR from 85 to 31, Afghanistan from 88 to 37; Equatorial Guinea rises from 23 to 90 and Venezuela from 11 to 69, on rebounds from earlier collapses. Only four countries (Togo, Gabon, Congo and Turkmenistan) are in the top third in both, which is what chance gives.

The right reading is not that countries have no durable strengths, but that the strengths relevant to one shock are not those relevant to the next. Open economies with deep finance may be hit by a financial crisis and not by a pandemic; tourism-dependent economies the reverse. Measurement noise also lowers correlations, since growth shortfalls are noisy and the episodes are only two. The finding is robust in direction across three shocks and in every income group but high-income, and it argues for shock-specific questions: not "how resilient is this country?" but "how resilient is it to this?".

Resistance in 2009 correlates −0.52 with income and −0.47 with trade openness: richer, more open economies lost more relative to their pre-crisis trend, as one would expect from a crisis that travelled through finance and trade. Interest burden is positively correlated with resistance (0.33), meaning that countries with heavy interest bills held up better, probably because they were less integrated. In 2020 the signs shrink or reverse: income correlates −0.14 with resistance and 0.13 with recovery, and openness −0.19 with resistance; government effectiveness correlates −0.18 with resistance, and with continuity 0.06.

Five features together (income, trade openness, reserves, interest burden, government effectiveness) explain 20% of the variation in composite resilience in 2009 and 7% in 2020 (104 countries): 37% of resistance and 30% of recovery in 2009, but 2% and 10% in 2020. For life-expectancy continuity the figures are 10% and 23%, the one dimension better explained in 2020, though mostly by the age structure that these features proxy. Capacity, in the sense used in earlier reports, is not what these dimensions measure, and report 2026-62's finding that steering mattered for excess deaths and hunger sits alongside this one.

The implication is that "build resilience" cannot be a single list of capacities. What protected countries in the financial crisis (low integration) is not what protected them in the pandemic, and the dimensions that matter to people (survival, prices) are explained by different things from those that matter to output. Resilience policy has to start by naming the shock and the function, as the working definition in the next section does.

The definition. Resilience R(S, H, F, T) is the degree to which function F of system S stays close to its reference path over horizon T after a shock of class H. S is a country, a sector, a firm or a household; H is a financial crisis, a pandemic, a price shock or a combination; F is output, survival, prices or service delivery; T is the impact year or three to four years on; the reference path is a pre-shock trend, a peer benchmark or a policy target. Each of the five measures here is one filling of the four slots: resistance is (country, global shock, output, impact year), recovery (country, global shock, output, +3–4 years), and so on.

Why four slots. The evidence above shows that changing one slot changes the answer. Changing the function (output to survival) leaves almost no correlation; changing the horizon (impact year to three or four years on) leaves resistance and recovery correlated at 0.80 in one shock and 0.35 in the other; changing the shock halves or removes persistence. A statement of resilience that omits a slot is a statement the data do not support. The polycrisis adds a fifth question, whether the shock class is a single event or a combination, which this report could not test with two sequential episodes.

Reference paths and artefacts. Every measure here is relative to a pre-shock trend and shows the artefacts that implies. Zimbabwe is the top country in the 2009 composite because its prices had just stopped collapsing, and Equatorial Guinea shares first place in 2020 because its pre-shock path was already falling, which left it little to lose; Lebanon scores in the top third in 2009 and bottom in 2020. A country ranked on how little it lost can be one that had little to lose. The corrective is to report levels and profiles with the ranking, to use peer benchmarks as well as own-trend ones, and to state the reference path.

The explorer recomputes a composite from the five dimension percentiles under the weights and episode chosen (weights are renormalised over the dimensions a country has). For "both" it averages the two episodes' composites. It reports how many of the equal-weight top twenty for the same episode remain, how many of your top twenty are also in the top twenty under the other episode, the rank correlation between your composite and the other episode's, and the people-weighted share.

Put all the weight on continuity and the list is a list of countries that kept life expectancy near trend; put it on price stability and it becomes a list of countries whose inflation fell or stayed low; put it on resistance and it favours countries that lost little output. The list changes completely, which is the point: there is no definition-free answer to "who was resilient?". The lists contain countries that are not what the word suggests (see section 6), because they are measured against their own past.

The explorer is a way of seeing the definition's consequences, not a ranking to cite. It uses two episodes and crude proxies, and a reader who prefers another definition (peer-relative, level-based, or including health and social outcomes) should read the explorer as showing how much such choices matter.

State the slots. Any claim of resilience should name the system, the shock class, the function and the horizon. A country that is resilient to a financial crisis in output terms may be fragile to a pandemic in survival terms; the evidence here is that the correlation between the two is about zero. Programmes, indices and rankings should carry the four-part statement with them.

Report profiles, not ranks. The five dimensions agree only partly and none persists across shocks, so a single score is a measure of a weighting. Reporting the five dimensions, with levels as well as relative standing, lets users choose the function they care about, and shows when two countries with the same composite are resilient in different ways.

Test across episodes and name the reference path. A resilience claim made on one shock is a claim about that shock. Multi-episode panels, combination shocks and peer-relative as well as own-trend benchmarks are the minimum for a definition used to allocate money, and the reference path should be stated, since the rankings here show how much it matters. Investment should follow the shock and function in question, not an abstract resilience score.

Series: World Bank WDI NY.GDP.PCAP.KD (GDP per person, constant 2015 US$) and NY.GDP.PCAP.KD.ZG (growth), FP.CPI.TOTL.ZG (inflation), SP.DYN.LE00.IN (life expectancy), SP.POP.TOTL (population, at least 1 million in 2023), NE.TRD.GNFS.ZS (trade openness), FI.RES.TOTL.MO (reserves, months of imports), GC.XPN.INTP.RV.ZS (interest, % of revenue) and government effectiveness from report 2026-62; 160 countries with a World Bank income group. Episodes: 2009 (pre-period 2004–08; post-period 2010–12) and 2020 (pre-period 2015–19; post-period 2021–24).

Measures (higher = more resilient). Resistance = growth in the shock year minus mean growth over the five pre-shock years. Recovery = 100 × (ln GDP per person at +3 years (2012) or +4 years (2023) minus ln at the last pre-shock year) minus the pre-shock mean growth times the number of years. Price stability = minus the rise in mean inflation (2008–09 or 2022–23) over its pre-shock mean (2004–08 or 2017–19), excluding countries with pre-shock inflation above 50%. Adaptation = mean growth in the post-period (2010–12 or 2022–24) minus the pre-shock mean. Continuity = life expectancy two or three years after the shock minus its pre-shock level plus the pre-shock trend. Dimensions are converted to percentile ranks within the sample; the composite is the mean of at least four. Spearman correlations are used throughout; the predictor R² values are from OLS of each percentile on the five standardised features (104 countries).

Caveats: two episodes of different kinds; own-trend measures reward weak pre-shock growth and collapsing prices; growth and life expectancy are crude proxies for adaptation and continuity; recovery and adaptation overlap by construction; composite ranks are relative; correlations are attenuated by measurement noise; countries have unequal data coverage; and nothing here tests combination shocks. Definitions are paraphrased from UNDRR's Sendai terminology (2017), Holling (1973, 1996) and secondary reports of the WEF Global Risks Report 2023, which was not itself consulted.