The Economics of Resilience in an Age of Global Polycrisis

Write each country's annual growth as its own average plus a deviation. Most deviations are small and, if resilience were only about smoothing, symmetric; their effect on compound growth is the volatility drag, half the variance. The identity used here concentrates on the left tail: the wedge W = φ × p × s, where p is the share of years that are shortfalls, s the mean shortfall in points and φ the share of a shortfall that is never recovered. If φ is zero, shortfalls are fully recovered and resilience adds nothing to long-run income; if it is one, every avoided point is kept.

The four channels act on different terms. Development changes p and s: richer countries have fewer and smaller shortfalls. Adaptation lowers s where the shock is a hazard. Productivity changes φ: if shocks scar productivity, they are permanent, and if they do not, output recovers. Systemic risk reduction, pooling and buffering, lowers the effective s that a country has to bear by moving the idiosyncratic part of shortfalls onto a larger balance sheet. The calculator in section 7 takes each channel as a lever on a slice of the wedge.

Averaged over the 153 countries, shortfalls occur in 14% of years with a mean size of 6 points, so p × s = 0.9 points of growth a year (median 0.6; the mean is pulled up by a few countries with many crises). With φ near 0.85 the wedge is about 0.8 points a year, which compounds to an income shortfall of about 18% over 25 years relative to a path without shortfalls. That is the size of the prize the rest of the report decomposes. It is gross: some of it is the cost of fast-growth strategies that are also riskier, and not all of it can be avoided.

Poorer countries are more volatile: the median standard deviation of growth is 3.6 points in low-income countries and about 2.4–2.8 in the others. They also have more crashes: growth falls below −5% in 6.7% of low-income country-years, against 3.6–3.9% elsewhere, and growth is negative in 27% of years against 17–21%. The mean growth lost to shortfalls is 0.71 points a year in high-income countries, 1.11 in upper-middle, 0.87 in lower-middle and 1.16 in low-income countries.

The gradient is real but modest. A high-income country loses about two thirds of what a low-income country loses, because shortfalls measured against a country's own average occur in every group: the 2009 financial crisis in the rich world, commodity-price swings in the middle, conflict and harvests in the poorest. The scatter shows the range within groups is larger than the gap between them, and that countries with the largest wedges are a mix of oil exporters, small open economies and fragile states.

For development policy this has an implication that is easy to miss. Growth strategies that rely on raising the average (investment, schooling, trade) are widely studied; the wedge, 0.7–1.2 points a year against average growth of 1.6–3.4, is of comparable size and changes with different policies. A country that halves it gains something of the size of the growth gap between income groups. The next sections show why the usual argument for resilience, through smoothing volatility, misses this.

If growth has mean g and variance σ², compound growth is lower than the average by about σ²/2. For a country with a standard deviation of 2.8 points that is 0.04 points a year, and the median across countries is 0.04 (0.03 for high-income and 0.07 for low-income countries). Two decades of this compounds to under 1% of income. The usual argument that smoother growth is better growth therefore needs something more than arithmetic.

The cross-country evidence, which asked whether more volatile countries grow more slowly (the question Ramey and Ramey posed in 1995), does not find it in 2000–2019. Regressing mean growth on the standard deviation alone gives 0.046 (standard error 0.060), controlling for initial income 0.003 (0.056) and for investment and schooling too 0.025 (0.045); with compound growth as the outcome, −0.038 (0.044). Labour-productivity growth is the outcome with a negative sign, −0.08 (0.043), at the margin of significance. The cross-section cannot show that volatility matters for growth.

That leaves two readings. Either shortfalls do not matter for long-run growth, or what matters is not the second moment but the left tail and its permanence, which the standard deviation mixes with ordinary upswings. Sections 4 and 5 test the second: whether shortfalls leave scars, and where they come from. The reframing is important for policy: resilience investment cannot be justified by a general preference for stability, it has to be argued from the specific shortfalls it prevents.

To measure scarring, each country's growth in a given year is compared with its average growth in the other years (excluding the window under study, to avoid a mechanical rebound), and the cumulative later deviation over the next three, five or eight years is regressed on the initial deviation. For shortfalls of three points or more the slope is 0.02, 0.06 and 0.37 at the three horizons (standard errors 0.13, 0.19 and 0.35): no sign of a rebound, so the permanent share is about 1. For shortfalls of six points or more it is −0.24, −0.28 and 0.02 (standard errors 0.21, 0.30 and 0.52), a permanent share of 0.72 to 1.02.

The estimates are imprecise, which is the honest message: shortfalls are mostly permanent, and the rebound is somewhere between nothing and a quarter. Labour productivity, the best measure of scarring, tells the same story: a permanent share of 1.05 (standard error 0.16, 336 shortfalls). By income group the point estimates are 1.63 for high-income, 0.78 for upper-middle, 1.61 for lower-middle and 1.20 for low-income countries, with intervals that overlap, so no group is distinguishable. The calculator uses 0.85 as a central permanence and allows 0.5 to 1.

The finding connects the growth literature, which finds that output after a recession often fails to return to its old path, to the resilience question. A shortfall that is permanent has a present value of its full size; avoiding it is worth the same. A shock that is recovered in two years is worth only its transitory cost. Productivity shocks, the loss of firms, skills and investment, are the mechanism, and the policy implication is that resilience investments that protect capital, jobs and firms through a shock (buffers, insurance, social protection) do more for long-run income than those that restore output quickly after it.

The decomposition classifies each of 429 shortfall events. 19 (4%) coincide with a recorded disaster doing damage of 1% of GDP or more (1% of the loss in high-income countries, 3–6% in the other groups); 134 (32% of the loss) fall in a year when global growth was at least a point below its mean, which in 2000–2019 is 2009 (−3.7), 2016 and 2019; 276 (65%) are neither. The idiosyncratic share rises as income falls, from 48% for high-income countries to 63%, 71% and 81%, while the global-year share falls from 51% to 15%.

Disasters nevertheless matter for growth where countries are poor. With country and year fixed effects, one point of GDP in recorded damage lowers growth by 0.106 points in low and lower-middle-income countries in the same year (standard error 0.028), and later years do not offset it (cumulative −0.08 over three years). In upper-middle and high-income countries the same-year effect is −0.015 and later growth is higher, so the cumulative effect is +0.10, which is reconstruction. Pooled, the effect is −0.055 and then a rebound. Poor countries do not get the reconstruction offset: their disaster losses are permanent.

For adaptation economics the implication is twofold. Disasters are a small share of shortfall events, so adaptation alone cannot close the wedge, and a growth argument for adaptation applies to the countries with a record of large events (Haiti, Myanmar, Pacific and Caribbean islands, Pakistan). And since recorded damage is incomplete, the disaster share is a lower bound. The dominant shortfalls are macro-financial, political and terms-of-trade shocks that adaptation does not touch, which is why the next section looks at how much of them can be pooled.

The global business cycle explains only a quarter of national growth variance: 38% in high-income countries, 28% in upper-middle, 13% in lower-middle and 6% in low-income countries. The mean pairwise correlation between national growth rates is 0.16. Poor countries' growth is therefore mostly their own: harvests, commodity prices, conflict, politics. The first principal component of the 153 growth series carries 27% of the variance, the next two 13% and 12%.

That is what makes pooling valuable. An equally weighted pool of all 153 countries has a growth variance 7.6% of the average national variance. A pool of the 42 sub-Saharan countries cuts it to 4.1% and Middle East and North Africa to 8.0%, East Asia to 9.7%, Latin America to 27%, South Asia (4 countries) to 35% and Europe and Central Asia to only 37%, because European growth is correlated (pairwise 0.47). By income group, lower-middle- and low-income pools cut variance to 4–5%, upper-middle to 10% and high-income to 32%. The poorest countries gain the most from sharing risk, and regional pools work in most developing regions.

This is a stylised calculation: it pools growth rates, not insured losses, and ignores moral hazard, basis risk and the fact that insurance pays for specific events. But it shows what mechanisms such as regional risk pools, contingent credit lines, a global financial safety net and sovereign risk pooling are for: they work because the risk is idiosyncratic. It also shows what they cannot do: the 16% of shortfall loss that is global-year (2009) can only be covered by global backstops, central-bank swaps and countercyclical lending, not by mutual insurance among countries that fall together.

For each country the shortfall loss (p × s, in points a year) is split by the cause rule into disaster-linked, global-year and other slices. The levers reduce their slices (adaptation the first, the global backstop the second, pooling and buffers the third); convergence scales all three towards the high-income median exposure; the permanence applies to what remains, with the productivity lever lowering it. The dividend is the baseline loss (at permanence φ) minus the new loss, in points of growth a year. Income compounds the dividend over the horizon; the cost is an annual percentage of GDP, and the payback year is when cumulative extra output first exceeds cumulative cost.

Reading the output: for the median country with default settings (30% of disaster loss, 20% of idiosyncratic and 15% of global-year loss avoided, with a ten-point fall in permanence) the dividend is about 0.15 points of growth, worth about 4% of income after 25 years; at a cost of 1% of GDP a year the median benefit-cost ratio is 1.7 and the programme pays back within the horizon in 69% of countries. Whether it repays depends on the country's wedge: for a country with a wedge of 1.2 points the dividend exceeds the cost within two decades; for one with a wedge of 0.3 it does not. That is the practical use of the accounting: resilience spending pays where the wedge is large, permanent and addressable.

The calculator is arithmetic on a stylised model. It treats the wedge as stationary, applies the same permanence to every shock, assumes levers act on slices independently and ignores interactions, costs of adjustment and the fact that the 2000–2019 window excludes the pandemic. The slice shares are from a classification rule and the permanence estimates are imprecise. Treat it as showing what would have to be true for a given programme to pay, not as a forecast.

Argue from shortfalls, not volatility. Volatility as such costs 0.04 points of growth a year, and the cross-country evidence for a volatility-growth link is nil. What costs growth is the loss from shortfalls (0.9 points a year) and its permanence. Resilience programmes should be appraised on the shortfalls they prevent and the permanent income they protect, with the dividend calculated country by country: it is large where the wedge is large and small where it is not.

Protect productive capital through a shock. Because shortfalls scar productivity, instruments that keep firms, jobs, skills and investment intact (social protection that scales, credit guarantees, contingent finance, maintained infrastructure) are worth more than those that rebuild after a shock. The permanence estimate of 0.7–1.0 is the single most important unknown; better evidence on scarring is the first research priority, and a lower permanence lowers the dividend of every other lever.

Pool the idiosyncratic and back-stop the global. Two thirds of shortfall loss is idiosyncratic and only 6–15% of low-income growth risk is global, so regional and global risk pools, contingent credit and sovereign insurance should deliver most of their value in developing regions, especially sub-Saharan Africa. The global-year third (2009) needs central-bank swap lines and multilateral countercyclical lending. Adaptation matters most where disasters leave permanent scars, in poor countries with large recorded losses, and is a small part of the wedge elsewhere.

Data: World Bank WDI NY.GDP.PCAP.KD.ZG (growth of GDP per person, constant 2015 US$), NY.GDP.PCAP.KD, NY.GDP.PCAP.PP.KD, SL.GDP.PCAP.EM.KD (GDP per person employed), NE.GDI.TOTL.ZS, SE.SEC.ENRR, NE.TRD.GNFS.ZS and SP.POP.TOTL; EM-DAT damage as % of GDP via Our World in Data; 160 countries with at least one million people, 155 with at least 18 growth observations in 2000–2019 and 153 with all 20. Shortfall: growth at least 3 points below the country's 2000–2019 mean; p = share of years, s = mean size, p × s = growth lost to shortfalls; volatility drag = arithmetic mean minus compound growth. Ramey regressions: mean growth 2000–2019 on the standard deviation, log initial income, mean investment and 2000 secondary enrolment (116 countries).

Permanence: for each country-year with a shortfall below a threshold, the cumulative deviation of growth in the following 3, 5 or 8 years from the country's mean in years outside the window, regressed on the initial deviation; permanent share = 1 + slope. Disasters: growth on recorded damage (% of GDP) in the same year and two lags, with country and year fixed effects, standard errors clustered by country, 2000–2024. Cause rule for each shortfall event: disaster-linked if damage ≥ 1% of GDP in the year or the year before; otherwise global-year if the cross-country mean growth is at least 1 point below its mean; otherwise other. Global factor: R² of each country's growth on the annual cross-country mean (153 balanced countries); pool variance = variance of the equal-weighted mean growth of the group divided by the mean national variance.

Calculator: loss slices (disaster, global-year, other; points a year) are reduced by the levers, scaled by convergence (closing the gap to the high-income median of p × s), and multiplied by permanence (reduced by the productivity lever); the dividend is baseline permanence × baseline loss minus the new loss; income compounds at the country's mean growth with the dividend added, and the payback year compares cumulative extra output with cumulative cost; the benefit-cost ratio discounts at 5%. Caveats: the window ends before the pandemic; shortfalls are relative to each country's own mean; permanence estimates are imprecise; the cause rule cannot separate financial, political and conflict shocks; recorded damage is incomplete; the dividend is gross of costs of adjustment and of any growth sacrificed by safer strategies; pooling is stylised.