Policy Research Working Paper 11129 Global Socio-economic Resilience to Natural Disasters Robin Middelanis Bramka Arga Jafino Ruth Hill Minh Cong Nguyen Stéphane Hallegatte Urban, Disaster Risk Management, Resilience and Land Global Department Climate Change Group & Poverty and Equity Global Department May 2025 Policy Research Working Paper 11129 Abstract Most disaster risk assessments use damages to physical 67% greater well-being losses per dollar of asset losses assets as their central metric, often neglecting distributional and require 56% more time to recover. Socio-economic impacts and the coping and recovery capacity of affected resilience is uncorrelated with exposure or vulnerability to people. To address this shortcoming, the concepts of natural hazards. However, a 10 percent increase in GDP well-being losses and socio-economic resilience—the ability per capita is associated with a 0.9 percentage point gain in to experience asset losses without a decline in well-being— resilience, but this benefit arises indirectly—such as through have been proposed. This paper uses microsimulations to higher rate of formal employment, better financial inclusion, produce a global estimate of well-being losses from, and and broader social protection coverage—rather than from socio-economic resilience to, natural disasters, covering higher income itself. This paper assess ten policy options 132 countries. On average, each $1 in disaster-related asset and finds that socio-economic and financial interventions losses results in well-being losses equivalent to a $2 uniform (such as insurance and social protection) can effectively national drop in consumption, with significant variation complement asset-focused measures (e.g., construction within and across countries. The poorest income quintile standards) and that interventions targeting low-income within each country incurs only 9% of national asset losses populations usually have higher returns in terms of avoided but accounts for 33% of well-being losses. Compared to well-being losses per dollar invested. high-income countries, low-income countries experience This paper is a product of the Urban, Disaster Risk Management, Resilience and Land Global Department, the Climate Change Group, and the Poverty and Equity Global Department. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at shallegatte@worldbank.org and bjafino@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Global Socio-economic Resilience to Natural Disasters Robin Middelanisa , Bramka Arga Jafinoa Ruth Hilla , Minh Cong Nguyena Stéphane Hallegattea aThe World Bank Keywords: natural disasters, risk reduction, well-being, resilience, recovery JEL codes: Q50, Q54, I30, I32, D63 Disaster risk assessments often inform the design of policies and interventions aimed at reducing the harm to people and their well-being. Many of these assessments focus on damages to physical assets (”asset losses” hereafter) as the key metric to measure risk1–3 . This can obscure the disproportionate burden on poorer households, who may lose less in absolute asset terms but suffer more relative to their means, and experience greater well-being impacts for every dollar lost4,5 . Indeed, $1 in asset losses—or even a loss equal to 1% of their total wealth—will have a more severe impact on the well-being of a poor person than on that of a comparatively well-off person. A more comprehensive disaster risk assessment needs to not only account for the distribution of asset losses across the population, but also for people’s capacity to cope with and recover from these losses6 . This capacity depends on socio-economic factors such as poverty7 , sources of income and livelihoods8,9 , income diversification, reliance on natural resources, health, disability, education10 , asset ownership, savings, and access to financial services11–13 . To more accurately capture the impacts of disasters on heterogeneous populations, the concept of well-being losses has been proposed14 . This approach uses a simple economic model to estimate how asset losses reduce household consumption, and then applies a utility function to translate those consumption losses into well-being impacts. The model accounts for the differences in households’ ability to maintain their consumption and replace lost assets after a disaster, depending on factors such as asset vulnerability, sources of income, and savings. The utility function captures the decreasing marginal benefit that individuals gain from each additional dollar consumed. This reflects that the impact on well-being of a one dollar loss in consumption is greater for a poor person than for a wealthier person. While this approach has been widely used in country-specific studies6,15–18 , we apply it here at a global scale, modeling household well-being losses in 132 countries. We start from asset losses estimated by a recent state-of-the-art global catastrophe model developed as part of the Global Resilience Risk Model and Resilience Index (’GIRI’)1 of the Coalition for Disaster Resilient Infrastructure. The next section explains the basic underpinnings of the model, with more details and equations presented in the Methods section. Household-level disaster impacts In each country, we model representative households across five income quintiles19 . Households earn labor income, which relies on a mix of household-owned assets (e.g., livestock or small shops), other non household-owned private assets (e.g., factories and machines), and public assets (e.g., transport infrastructure). Labor income also includes imputed ”housing services” from owner-occupied dwellings, which are considered part of households’ assets. Asset ownership shares are estimated from a variety of data sources20–28 . Additional income comes from diversified sources such as transfers, remittances, social protection, and capital income (as estimated from previous studies29,30 ). Population exposure to disasters—including floods, wind and storm surge from tropical cyclones, tsunamis, earthquakes— and corresponding asset losses are derived from the GIRI data. These asset losses are then distributed across house- holds based on empirically measured "exposure bias," which reflects differences in hazard exposure between poorer and wealthier households31 , as well as on observed and estimated flood protection levels32,33 , and estimated asset vulnerabilities (derived from refs34,35 ). These asset losses lead to consumption losses both because affected households lose income and because they must pay to repair or replace their damaged assets. Then, we use a standard utility function to translate consumption losses into a well-being measure that captures the higher vulnerability of poor people who have to sacrifice basic needs like food, education, and health care when they need to cut consumption. 2 Assets Income Liquidity Vulnerability a [$PPP 1,000] b [$PPP 1,000 / yr] c [$PPP] d [%] hh-owned diversified HTI 80 other 8 labor 200 60 TJK 60 6 150 40 40 4 100 20 20 2 50 0 0 0 0 q1 q2 q3 q4 q5 q1 q2 q3 q4 q5 q1 q2 q3 q4 q5 q1 q2 q3 q4 q5 Asset losses Recovery time Well-being losses, affected Well-being losses, non-affected e [$PPP 1,000] f [yr] g [$PPP 1,000] h [$PPP 1,000] 40 60 2.0 30 hh-owned other 2.0 50 25 30 1.5 20 1.5 40 15 20 30 1.0 1.0 10 20 10 0.5 10 0.5 5 0 0 0.0 0 0.0 q1 q2 q3 q4 q5 q1 q2 q3 q4 q5 q1 q2 q3 q4 q5 q1 q2 q3 q4 q5 Household income quintile i HTI-q5 j TJK-q5 10 Income loss 8 Reconstruction loss [$PPP 1,000 / yr] Liquidity Consumption 6 Consumption 4 2 0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Time [yr] Time [yr] Figure 1: Characteristics of households in Haiti and Tajikistan affected by a 100-year return period earthquake. (a) Total assets used by the household to generate income from labor. Light shading indicates assets directly owned by the household (including owner-occupied dwellings), dark shading other assets (i.e. public or firm-owned capital, including rented housing). (b) Income from labor (dark shading) and diversified sources, such as transfers and social protection (light shading). (c) Liquidity of households, including liquid savings and the ability to borrow money. (d) Household-specific asset vulnerability to earthquakes. (e) Total asset losses of directly affected households. Light shading indicates the share of destroyed assets directly owned by the household, dark shading other asset losses. (f) Recovery duration as the time to reconstruct 95% of destroyed assets (note the secondary axis for Tajikistan). (g) Per capita well-being losses of affected and (h) non-affected households. (i, j) Consumption dynamic of affected households in the top income quintile (q5) in the aftermath of the disaster. Consumption losses result from foregone income (dark shading), and from reconstruction costs of household-owned asset losses (light shading). Hatched area indicates used liquidity to offset consumption losses. All characteristics are shown for households in Haiti (blue) and Tajikistan (orange). While consumption losses can be partially offset by precautionary savings (estimated from ref30 ), in practice affected households often receive additional aid from governments, insurance mechanims, or other humanitarian actors. In our 3 analysis, we first estimate well-being risks and socio-economic resilience in the absence of government support, and then evaluate how various policy options influence outcomes and improve household welfare. Key to the recovery is the fact that each affected household has to balance the speed of reconstruction (or replacement of lost assets) with the need to maintain consumption. At one extreme, households could rapidly rebuild lost assets by drastically cutting consumption, potentially causing severe impacts on short-term well-being36,37 . At the other extreme, delaying or not investing in reconstruction at all might preserve short-term consumption but prolong income loss over time. Assuming rational behavior, our model identifies the optimal recovery rate that minimizes the total impact on household well-being. We then calculate the discounted sum of utility losses across all households and aggregate these for the total population. Finally, we calculate the ”well-being loss” as the uniform drop in national consumption that would cause the same overall impact on people’s well-being as the disaster. This conversion is based on each country’s average sensitivity of well-being to changes in consumption (or marginal utility of consumption) and allows us to compare disaster impacts across countries in terms of an equivalent percentage loss in national income. Details on the model, variables, and datasets are provided in the Methods section and Supplementary Materials (Supplementary Figs. 1 and 2 and Supplementary Tbls. 1–4). Missing data are imputed based on countries with better data availability (see Methods, Supplementary Fig. 3 and Supplementary Tbls. 4 and 5). For illustration, we show the characteristics and recovery dynamics of representative households in Haiti and Tajikistan affected by a 100-year return period earthquake (Fig. 1). Affected households in Tajikistan recover much more quickly—with all household income restored in less than 2.5 years—compared to household in Haiti, where recovery can take up to 40 years for those in the second income quintile. This is despite households in Tajikistan experiencing higher absolute asset losses. Note that these recovery times assume no disaster aid, which can significantly accelerate recovery as we show further below. Consistent with observations38 , the bottom two income quintiles in Haiti are the hardest hit. At the country level, $5.5bn in asset losses in Haiti results in $19.9bn in well-being losses, while $14.2bn in asset losses in Tajikistan lead to only $18.5bn in well-being losses. This contrast illustrates the different capacities of the two countries to cope with disasters. Several factors contribute to these results. While households in Tajikistan have higher income and own more assets— resulting in higher absolute asset losses than households in Haiti—the two countries differ notably in their income structures. First, households in Tajikistan receive a significant share of their income from diversified sources. Second, due to lower self-employment rates and less owner-occupied housing, Tajikistani households own a smaller share of the assets used to generate income (or housing services) compared to Haitians (Supplementary Fig. 4). These factors contribute to more sharing of disaster-related risks in Tajikistan. As a result, well-being losses are mitigated for directly affected populations (even though this also causes higher well-being losses of households that are not directly affected). Socio-economic resilience in 132 countries We compute well-being losses in 132 countries for each hazard and return period, and aggregate these to estimate the national risk to well-being—expressed as the annual average well-being losses as a share of GDP. A country’s socio-economic resilience reflects how well it can protect people’s well-being in the face of disaster-related asset losses. It is defined as the ratio between the economic value of asset losses and the resulting well-being losses. In an idealized scenario—in which risk is perfectly shared, reconstruction is immediate, and inequality is absent—well-being losses would match asset losses, and resilience would be equal to 100%. In reality—when losses are heterogeneous, poorer 4 R2=0.03 (n.s.) a b 2.5 TJK HTI 2.0 risk to assets [% of GDP] 1.5 1.0 IBRD 48825 | APRIL 2025 0.5 0.0 0.000 0.125 0.250 0.500 1.000 3.000 R2=0.08** 8 c d HTI LICs 7 LMICs 6 UMICs risk to well-being HICs [% of GDP] 5 4 3 TJK 2 IBRD 48826 | APRIL 2025 1 0 0.00 0.25 0.50 1.00 2.00 6.00 R2=0.54*** 80 e f TJK 70 socio-economic resilience [%] 60 50 40 IBRD 48827 | APRIL 2025 30 HTI 20 10 20 30 40 50 60 70 80 8 10 12 ln(GDPpc [$PPP]) Figure 2: Risk to assets, simulated risk to well-being, and the resulting socio-economic resilience for 132 countries. (a,b) Risk to assets and (c,d) resulting simulated risk to well-being, both as share of national GDP. (e,f) Socio-economic resilience as the ratio of the risk to assets and the risk to well-being. All data are shown on maps (left) and over the logarithm of GDP per capita (right) with colors indicating country income groups. Level-log regressions with GDP per capita as the independent variable are shown as dashed lines with 95% confidence intervals. R2 and significance levels are shown at the top of the panels. Colored lines show kernel density estimates for the distribution within country income groups. households are hit harder, or reconstruction is delayed by lack of resources or other obstacles—well-being losses exceed asset losses, and resilience declines towards zero. The larger this gap, the lower a country’s resilience. Fig. 2 shows the risk to assets, the corresponding risk to well-being, and socio-economic resilience at the global level. Full results are provided in Supplementary Tbl. 5. Importantly, these results should not be interpreted as a forecast of losses, but rather estimates of risk and resilience based on current socio-economic conditions. The global risk to assets amounts to $314 bn, resulting in a well-being risk of $620 bn. This corresponds to an average global resilience of 51%, with high-income countries generally more resilient than poorer countries (Fig. 2 panel e). Risk to well-being is strongly correlated with risk to assets (Fig. 2 and Supplementary Fig. 5), as every additional $1 5 of asset losses leads to additional well-being losses. Although countries usually invest in risk reduction projects after disasters, there is no statistically significant correlation between resilience and risk to assets (Supplementary Fig. 6). This suggests that socio-economic resilience, as defined here, is shaped by broader development pathways more than by specific disaster risk responses. In the sections that follow, we explore the determinants of socio-economic resilience and assess how well-being losses and recovery times can be reduced. We first show that higher levels of economic development and low inequality are associated with greater socio-economic resilience. We also demonstrate that higher-income countries tend to recover more quickly from natural disasters than lower-income countries. We then use our model to simulate the benefits of asset-focused measures (reducing exposure or physical vulnerability), structural socio-economic changes (economic growth, reducing inequality, promoting formal employment, and income diversification), and disaster risk finance options (adaptive social protection and insurance) on disaster risks, resilience, and recovery times. Resilience, development, and inequality We start by investigating how risk to assets, risk to well-being, and socio-economic resilience vary with a country’s economic development level, proxied by GDP per capita. We run level-log regressions with GDP per capita as the independent variable and find that a 10% increase in GDP is associated with a 0.9 percentage point (pp) increase in socio-economic resilience (Fig. 2 panel f, Supplementary Tbl. 6). This suggests that wealthier countries are better equipped to cope with disaster losses and that economic development can play an important role in increasing a country’s resilience. However, this relationship is not necessarily causal, as higher GDP likely enhances resilience indirectly through factors such as financial inclusion and social safety nets (see below for an exploration of the role of different drivers). We distinguish between countries at different levels of economic development using the World Bank’s 2024 country income group classification. We find a median resilience of 60% in high-income countries (HICs), 51% in upper-middle income countries (UMICs), and only 36% in lower-middle-income (LMICs) and low-income countries (LICs). Socio- economic resilience can differ widely within country income groups, particularly among LMICs, where resilience ranges from as low as 18% (Zambia, overall lowest resilience) to as high as 77% (Tajikistan, overall highest resilience). This variation shows that, while economic development positively influences a country’s resilience, the relationship is likely mediated through additional factors. We find no significant correlation between the risk to assets and GDP. Consistent with previous findings39 , the fraction of assets destroyed appears to be dominated by hazard and exposure footprints rather than by a country’s economic level of development—potentially because trends in exposure40 and vulnerability41 offset each other. We do find a relationship between economic development and decreasing risk to well-being, albeit with limited explanatory power. While economic growth appears to increase a country’s capacity to minimize well-being losses, total well-being impacts are mostly driven by geography (Supplementary Fig. 5). The median share of asset losses borne by the bottom income quintile in a country is only 9% of total asset losses, compared to 33% when considering well-being losses (Fig. 3 panel a). Conversely, the median share of asset losses for the top quintile lies at 36%, while their share of well-being losses amounts to only 9%. There is a negative relationship between income inequality within countries—as measured by the Gini index19 —and socio-economic resilience (Fig. 3 panel b). 6 These correlations do not imply a direct causal influence of economic development and inequality on socio-economic resilience. While income growth and inequality directly increase well-being losses—due to the declining marginal utility of consumption at higher income levels—they both also have an indirect effect through other determinants of resilience such as financial inclusion, coverage of social protection programs, and employment rates. We explore these relationships through simulations below, showing that, ceteris paribus, increases in GDP per capita have no effect on socio-economic resilience. a 80 well-being loss asset loss 70 60 Loss share [%] 50 40 30 20 10 0 q1 q2 q3 q4 q5 Household income quintile b 80 LICs LMICs 70 UMICs HICs 60 socio-economic resilience [%] 50 40 30 20 R2=0.19*** 25 30 35 40 45 50 55 60 Gini index [%] Figure 3: Assessing the relationship between inequality and socio-economic resilience. (a) Global statistics over asset and well-being losses from all hazards experienced by income quintiles across all countries. Values represent the relative share of national losses that accrue to households in each income quintile. Box plots show the median (center line), upper and lower quartiles (box limits), whiskers extending to 1.5 times the interquartile range, and outliers (points beyond the whiskers). (b) Socio-economic resilience over income inequality expressed by the Gini index. Dashed black line indicates a linear regression with 95% confidence intervals. Colored lines represent kerned density estimates for the distribution of the Gini index by country income groups. 7 Recovery from natural disasters Socio-economic resilience reflects how quickly households recover from disasters. We define recovery duration as the time needed to restore 95% of lost assets. We find that countries with higher resilience tend to recover more quickly (Supplementary Fig. 7). Consistent with findings from empirical analyses and case studies42,43 , we observe that developing countries exhibit longer recovery times than developed countries (Fig. 4 panel a). Households in LICs and LMICs consistently take longer to recover than those in UMICs and HICs, both for individual hazards and across all hazards (Fig. 4 panel b). Across all hazards, recovery for households in HICs and UMICs is 36% and 28% faster, respectively, than those in LICs, highlighting that economic development helps accelerate recovery. This faster recovery is observed across all income groups within countries (Fig. 4 panel c). Recovery generally improves with household income, which is consistent with both empirical and anecdotal evidence44–47 . However, the median recovery time for households in the bottom income quintile (q1) is often shorter than for households in the second quintile (q2), as seen in the cases of Haiti and Tajikistan (Fig. 1). This is explained by the fact that, as seen in empirical studies48 , access to liquidity accelerates households’ recovery after disasters and mitigates well-being losses (Supplementary Fig. 9). And although the poorest households have fewer resources, their liquidity relative to their asset losses is often higher (Supplementary Fig. 8). In the absence of financial inclusion and access to borrowing, precautionary savings create strong incentives for households to hold resources in liquid, non-productive forms49 . However, such precautionary savings are costly, as they reduce investments in income-generating assets50,51 . 8 a Average recovery duration [yr] IBRD 48828 | APRIL 2025 0.5 1.0 2.0 4.0 8.0 16.0 30.0 b Earthquake Flood 6 Tsunami by country income group [yr] Median recovery duration Storm surge 5 Wind 4 3 2 1 0 LICs LMICs UMICs HICs Country income group c LICs LMICs by household income quintile [yr] 5 UMICs Median recovery duration HICs 4 3 2 1 0 q1 q2 q3 q4 q5 Household income quintile Figure 4: Recovery duration by country and income quintile. (a) National average recovery duration of affected households conditional on hazard occurrence, across all hazard types and weighted by exposed population. (b) Median recovery duration by country income group for individual hazards (bars) and across all hazards (horizontal lines). (c) Median recovery duration by income quintile, for each country income group (bars) and across all countries (horizontal lines). Policy options to increase resilience We investigate ten policy options and simulate their impacts on risk to assets, risk to well-being, socioeconomic resilience, and the recovery duration, comparing each to the ”baseline” case as analyzed so far (Fig. 5). The options cover asset- focused disaster risk reduction measures (options 1–4), structural socio-economic changes (options 5–8), and adaptive social protection measures, such as post-disaster support and insurance (options 9 and 10). 9 Asset-focused disaster risk reduction measures typically aim to reduce either exposure or asset vulnerability. Exposure can be reduced through investments in protective infrastructure, such as dikes, or through risk-informed land- use planning. Vulnerability can be reduced by retrofitting dwellings, enforcing stricter building standards, or improving the quality and accessibility of early warning systems. We model these measures by reducing the country’s total exposure (options 1 and 2) or vulnerability (options 3 and 4) by 5%, and assess the effects when targeting only the bottom 20% of the income distribution (q1; options 1 and 3) or the entire population (options 2 and 4). Asset losses decrease at the same rate as exposure and vulnerability when targeting the entire population, but at a smaller rate when targeting only households in the bottom expenditure quintile (i.e., q1 households), because poorer households tend to own assets with lower economic values. Reducing exposure across the entire population decreases the number of people affected by disasters, but does not directly improve the resilience or recovery time of households that are still impacted. In other words, while fewer people experience losses, the severity of losses for those who are affected remains the same. However, in practice, having fewer people affected could still help resilience and recovery indirectly, by reducing indirect effects through demand or supply chains and facilitating government’s response. Across all country income groups, resilience gains are higher when risk reduction measures target the bottom 20%, even though doing so prevents fewer asset losses overall. Put differently, one dollar of avoided asset losses yields a greater return in terms of avoided well-being losses for the worse-off population (e.g., the bottom 20%). Options 5–8 target socio-economic structural changes. A 5% increase in GDP per capita and household liquidity (option 5) has no direct effect on resilience or recovery, supporting empirical findings that higher income alone does not necessarily build resilience9 . This contrasts with the observed positive correlation between economic growth and socio-economic resilience (Fig. 2) and corroborates the assumption that the relationship is indirect—likely operating through factors such as improvement in financial inclusion, social protection, formal employment, or income distribution as a country develops. To assess the role of income inequality, we redistribute 10% of each household’s income evenly across the population (option 6). This reallocation implies a shift of productive assets toward poorer households, who typically have higher asset vulnerability (Supplementary Fig. 10), leading to higher asset losses and longer recovery times. The resulting increase in resilience substantiates a causal relationship between income inequality and socioeconomic resilience. Next, we investigate the role of formal employment and find that reducing self-employment by 10% (option 7) lowers well-being losses in all countries, with the largest resilience gains observed in LICs. This result underscores the idea that firms (and formal employment) function as risk-sharing institutions52 . When people are formally employed, a larger share of the assets they use to generate output is not owned by households directly, but by their employer (e.g., a firm). This reduces the financial burden on households after a disaster, as the responsibility for the reconstruction of firm-owned assets is distributed more broadly across the economy rather than falling solely on individual households. A similar effect applies to renting, as opposed to owner-occupied housing. Since self-employment rates are particularly high in LICs (Supplementary Fig. 4), formal job creation in these countries results in especially strong gains in socioeconomic resilience . Income diversification can reduce households’ dependency on labor income and thus lower their vulnerability to local shocks. We shift 10% of labor income to diversified sources—such as transfers, social protection, or capital income—while keeping total consumption constant (option 8). In the model, this is implemented by raising labor income taxes by 10% and increasing diversified income accordingly (eq. (3)). While asset ownership, and thus the risk to assets, remains unchanged, the risk of foregone labor income from (locally) destroyed assets is partially shared across 10 the population. The results show that income diversification improves resilience and reduces well-being losses in all countries, especially LICs. Lastly, we examine the impact of financial instruments and social protection programs. We model two mechanisms: (1) a post-disaster support (PDS) program that provides a one-time lump-sum payment equal to 40% of asset losses for households in the lowest income quintile (option 9), and (2) a mandatory national insurance program that covers 20% of asset losses, with costs shared across the entire population based on pre-disaster income (option 10). The 40% and 20% parameters are chosen to target similar total losses across countries, ensuring comparable global costs. Both mechanisms increase the available liquidity of households and can be used to offset consumption losses and accelerate reconstruction. Both PDS and insurance significantly reduce well-being risks and improve resilience across all income groups. In practice, benefit-cost analysis must take into account targeting issues, coverage, and transaction costs. For PDS programs, targeting errors may occur when it is difficult to identify and reach the most affected populations—such as in countries lacking a comprehensive social registry. Coverage also depends on the availability of government resources, including disaster risk funds and contingent credit lines. Transaction costs are typically low, especially when a social protection system or registry is already in place. Insurance programs, on the other hand, face the risk of adverse selection and rely on participation rates. Insurance also tends to have higher administrative and financing costs, particularly when reinsurance is involved. Without accounting for these transaction costs and implementation challenges (such as targeting errors), the PDS and the insurance program both offer high benefit-cost ratios, ranging from $3.26 in HICs to $10.58 in LICs per dollar invested for the PDS and $2.50 to $7.11 for insurance. This means that even with pessimistic estimates for transaction costs, such programs would still deliver high well-being gains per dollar invested. 11 a b c d LICs LMICs 1: Reduce total exposure by 5% UMICs HICs 2: Reduce total exposure by 5% targeting the poor 3: Reduce total vulnerability by 5% 4: Reduce total vulnerability by 5% targeting the poor 5: Increase GDP and liquidity by 5% 6: Reduce income inequality by 10% 7: Reduce self-employment rate by 10% 8: Reduce non-diversified income by 10% 9: PDS equal to 40% of asset losses of the poor 10: National insurance covering 20% of all asset losses 5 0 5 0 20 40 0 10 20 0 20 40 60 Avoided risk to Avoided risk to Socioeconomic resilience Recovery time assets [%] well-being [%] change [pp] reduction [%] Figure 5: Assessing the effects of policy options. The impact of asset-focused disaster risk reduction measures (options 1–4), structural socio-economic changes (options 5–8), and adaptive social protection measures (options 9 and 10) on avoided (a) risk to assets and (b) risk to well-being, (c) on socio-economic resilience, and (d) on a reduction of the recovery time. Impacts are in comparison to the baseline scenario. Recovery times are averaged across all hazards, weighted by the affected population. Box plots show the median (center line), upper and lower quartiles (box limits), and whiskers extending to 1.5 times the interquartile range. Outliers are not shown. Discussion We conducted a global analysis of well-being losses from, and socio-economic resilience to, natural disasters. Our findings reveal significant differences both within and between countries. While socio-economic resilience tends to 12 increase with economic development, this relationship is likely mediated by broader structural factors such as income inequality, formal employment, financial inclusion, and income diversification. Our simulation results also highlight the effectiveness of both asset-focused and socio-economic interventions in reducing disaster risks. At the same time, they underscore the importance of aligning risk metrics with the specific objectives of policies. Not all interventions that reduce well-being losses accelerate recovery, and some that enhance resilience do so at the expense of greater asset losses. Our model allows us to investigate the causal effects of risk reduction measures on different risk metrics. Our results highlight the value of targeting the poorest households, who experience the largest well-being losses even though their asset losses remain small in absolute values. These findings reinforce existing evidence53–56 at the global scale, suggesting that prioritizing low-income population groups can yield greater returns, particularly when resources are limited. The role of formal employment has been widely emphasized in discussions on social protection, development, and poverty57 . Empirical evidence also suggests that formal employment can help prevent income losses caused by hazards58 . Our simulations underscore this critical role of employment in strengthening households’ ability to absorb and recover quickly from disaster-related shocks. The metrics of well-being loss and socio-economic resilience can complement the commonly used asset-focused disaster risk assessment approach. However, they remain imperfect measures, as they account only for the material aspects of well-being. Importantly, these indicators do not capture other critical dimensions, such as direct impacts on human lives (e.g., fatalities or psychological impacts), the environment (e.g., pollution, soil degradation, or ecosystem destruction), or cultural losses (e.g., damage to heritage sites). Like any model, ours involves important simplifications. Our analysis of well-being losses uses a standard constant relative risk aversion utility function, which assumes that households can always reduce consumption to enable recovery after a disaster. Alternatively, it would be possible to incorporate a minimum subsistence line, in which case elasticity would no longer be constant. In this case, some households might be unable to cut consumption to replace lost assets, and therefore never recover from the shock. This would create the possibility of poverty traps due to shocks and disasters59–61 and would reinforce our finding that post-disaster cash transfers can significantly accelerate recovery, if well designed58,62,63 . Also, our model considers only the direct economic impacts of natural disasters (i.e., output losses due to asset destruction). Although households can be affected through their diversified income, we do not account for other indirect impact channels, such as the ripple effects through supply chains or local demand64–66 . While this simplification is appropriate for smaller shocks, it likely underestimates the impacts of large-scale disasters, which can have far-reaching consequences67 . Other limitations are linked to data availability. Our results suggest that asset ownership and available household liquidity play crucial roles in approximating people’s capacity to cope with natural disasters. Because such data are scarce, we relied on indirect methods to differentiate between ownership of different asset types and to estimate the available liquidity at the household level. Our liquidity estimates thus likely underestimate the true amount of available savings, particularly for higher-income households. As a result, their recovery could be faster than we estimate here. Other data—such as information on social and private transfers, housing tenure, national accounts details, and disaster preparedness—are not available for all countries and missing data had to be inferred. While this approach is 13 suitable for our global-scale study, more complete data would improve accuracy, enable more localized analyses (e.g., investigating patterns within regions or country income groups), and allow inclusion of more countries in future work. A rigorous validation of the results creates inherent challenges, as the primary outcome of the model—well-being losses—is not directly observable. This difficulty is compounded by the variability of disaster impacts, both across individual households and different hazards types (e.g., flood, earthquakes). While our simulations use aggregated population quintiles, real-world impacts often differ sharply at the household level, complicating direct comparisons with empirical studies. Nevertheless, observed patterns in consumption losses and recovery times provide useful points of reference, even if imperfect. Significant declines in consumption following asset losses—and their severe impacts on well-being—are well-documented68 . This evidence also shows that disasters disproportionately impact the poorest populations9,69,70 , consistent with the patterns demonstrated by our results. Our finding that disasters can have long-lasting impacts on well-being, especially in low-income countries71 and among poorer households60 , is also in line with existing observations. As disaster-related asset losses continue to rise, the need for effective risk reduction strategies becomes increasingly urgent. To evaluate risk, well-being losses offer a more comprehensive and meaningful metric than asset losses alone, as they better reflect the severity of disasters’ impacts on people’s lives. Using multiple metrics, including well-being losses, also allows to compare or combine interventions that reduce asset losses from disasters (such as dikes and construction standards) and interventions that make people better able to cope with these losses (such as financial inclusion, formal employment, or social protection). An expanded risk management toolbox, supported by multiple metrics to target interventions where they matter the most, can contribute to designing more efficient policies and strategies to make people more resilient, protect them against shocks, and improve their lives and well-being. 14 Methods Model description Population counting The model considers representative households by income quintile, derived from household survey data (see data section). Individuals within population groups are counted using a weighting approach. The sum over all five quintile weights nq , representing fq = 20% of the population returns the full population P : ∑ nq = ∑ fq · P = P . (1) q q Additional indices or superscripts are used to further divide income quintiles into sub-groups, for example to count a. individuals within an income quintile affected by a natural disaster nq eff to generate output. These assets can be divided into a share Household capital Each household uses its capital kq owned by the household directly κ h , as well as other private assets, e.g., firm-owned, κ f , and public assets κ p (see data section). At the national level, capital adds up to K = ∑ nq · kq eff . (2) q Income from labor and transfers In the absence of a disaster, consumption cq is equal to a household’s income, lab (generated using the household’s effective capital) and diversified income i div (i.e. which consists of labor income iq q div of labor income is paid as a non-labor income from transfers, social protection, remittances, and capital). A share δtax flat ”tax” to fund the diversified income. Pre-disaster consumption results to div eff Γdiv q div cq = (1 − δtax ) · Π · kq + · δtax · Π · K, (3) nq lab iq div iq eff the effective capital used by the household to where Π is the average productivity of capital (see data section), kq generate income from labor, and Γdiv q the share of all transfers at the national level paid to households in income quintile q. Transfers at the national level are equal to all taxes paid: ∑ nq · iq div div = δtax · Π · K. (4) q div that comes from transfers for each household (see Given household consumption cq and the share of income γq data section), the tax to fund transfers results to div · c div ∑q γq q δtax = (5) ∑q cq 15 and the share of transfers that is paid to households in quintile q to div · c γq q Γdiv q = div · c ′ . (6) ∑q ′ γq ′ q The effective capital used by a household is given by div ) · c (1 − γq eff q kq = div ) · Π . (7) (1 − δtax Exposure, asset losses, and income losses For each considered hazard and return period above the hazard protection threshold (see data section), we derive country-level exposure 0 ≤ f a ≤ 1 from the share of assets that is ∆K destroyed due to the disaster K and national vulnerability V (see data section). Hazard and return period indices are omitted in the following for better readability. To derive the exposure, we use the relationship that asset losses are given by the product of exposed assets and asset vulnerability: ∆K = K · f a · V ∆K 1 (8) ⇒ fa = · . K V National exposure is corrected for exposure biases (see data section) per income quintile to obtain household exposure: a fq = f a · EBq . (9) a can be interpreted as a Since we consider representative households of the population within each quintile, fq probability for any given household in income quintile q to be affected by the disaster. The number of affected and a and nn , respectively) thus result to non-affected households within each quintile (nq q a a nq = nq · fq (10) n a nq = nq · (1 − fq ). eff is destroyed when a Depending on household vulnerability vq (see data section), a fraction of the used capital kq household is affected by the disaster. An affected household thus suffers capital losses of a,eff ,(0) eff ∆kq = vq · kq (11) which sum to losses at the national level ∑ nq a a,eff ,(0) ∆K = · ∆kq . (12) q Thereby, we assume the vulnerability of all assets (household-owned, public, and firm-owned) used by one household to be identical, calibrated on housing quality (see data section). This means that if a household’s dwelling is fragile, we also assume the local public infrastructure and the factory or building in which people work to be fragile. 16 According to eq. (3), the destruction of capital stock reduces labor and diversified income by a,lab div a,eff ∆iq (t ) = (1 − δtax ) · Π · ∆kq (t ) Γdiv (13) div q div ∆i q (t ) = · δtax · Π · ∆K ( t ) , nq which are time-dependent, as destroyed capital is reconstructed over time. Only directly affected households suffer a,lab from labor income losses ∆iq , while all households—also those not directly affected—experience a decrease of their div . diversified income ∆iq Reconstruction loss In addition to income losses, exposed households need to reduce their consumption by the a,reco a ) of the time-dependent amount ∆cq (t ) to pay for the reconstruction of destroyed assets. For the fraction (1 − fq n,reco a of the population. We population that is not affected, ∆cq results to zero. Now, we consider the affected fraction fq assume effective household asset losses (including those owned directly by the household and public or firm-owned) to be reconstructed exponentially at a rate λq 1 , which yields the effective capital loss over time a,eff ,(0) a,eff ∆kq ( t ) = ∆k q · e − λq t . (14) This adds a temporal dimension to all elements of the previously defined income and consumption losses. The reconstruction cost is partly borne by the household (for the share of effective capital owned by the household directly) and partly assumed to be credit-financed (for the share of public and firm-owned assets), with repayments far in the future, widely distributed at the national level, and spread over time (cf. Long-term well-being losses). Note that eq. (14) does not distinguish by asset ownership and uses the effective capital loss, assuming that all assets used by a household are reconstructed at the same recovery rate λq as the household’s privately owned assets. The assets that households have to rebuild at their own expense are equal to a,reco,(0) a,eff ,(0) ∆k q = κ h · ∆kq , (15) where κ h is the share of a country’s total capital owned by private households (see data section). Using eq. (14), the reconstruction cost for the affected household over time is given by d a,reco,(0) −λq t a,reco,(0) a,eff ,(0) a,reco ∆cq (t ) = − ∆k q e = ∆kq λq e −λq t = κ h ∆kq λ q e − λq t , (16) dt where the negative sign is set to obtain positive loss values, in accordance with the convention set above. As eq. (16) shows, a fast recovery (i.e., high recovery rate λq ) causes higher reconstruction costs right after the disaster, because the cost is distributed only over a shorter period of time. In return, a fast recovery also decreases the time during which the household experiences income losses from destroyed capital. Thus, the speed of reconstruction is a trade-off between, on the one hand, short and large consumption losses and, on the other hand, long and smaller consumption losses. 1 The recovery rate λ is obtained for each affected household as the rate that minimizes total well-being losses, see Recovery optimization and q optimal spending. 17 Disaster aid As an optional policy measure, we model post-disaster support (PDS) as a one-time payout aid a,reco Sq = σ · ∆k q =1 , (17) equal for all helped households. We assume that countries provide each affected household with the same lump sum, equal to σ = 40% of the reconstruction cost faced by an affected household in the poorest income quintile. Alternatively, we model an insurance scenario where aid reco Sq = σ · ∆kq , (18) with σ = 20%, where payouts are proportional to the experienced reconstruction cost of affected households. Then, total aid costs at the country level are C aid = ∑ nq a aid Sq . (19) q Similar to the reconstruction cost for destroyed public and firm-owned capital, we assume these costs to be credit- financed, and that repayment is far in the future and widely distributed at the national level and over time (cf. Long-term well-being losses). Dynamic consumption loss sav from own savings or through Households can have access to short-term liquidity Sq borrowing (simply ’liquidity’ hereafter, see data section). Together with aid payouts, households can use these funds to sav +aid (partly) offset consumption losses over time by ∆cq (t ). The overall time-dependent consumption loss is given by the sum of foregone diversified and labor income, as well as reconstruction costs, and is mitigated by the use of liquidity and PDS: lab div reco sav +aid ∆cq (t ) = ∆iq (t ) + ∆iq ( t ) + ∆c q ( t ) − ∆c q (t ). (20) Short-term well-being losses We model the trade-off between the goal to minimize the total consumption loss and the desire to keep consumption at a minimum level by using a constant relative risk aversion utility function with diminishing marginal utility c 1− η − 1 w (c ) = , (21) 1−η short −term with η = 1.5 the elasticity of the marginal utility of consumption. The total short-term well-being loss ∆Wq after the disaster is obtained by aggregating the (discounted) loss of utility during recovery: ∞ short −term ∆W q = w cq − w cq − ∆cq (t ) e −ρt dt (22) 0 with discount rate ρ = 6%. Evaluating eq. (22) requires the (unknown) recovery rate and the spending of aid payouts and liquidity over time. We set both such that well-being losses are minimized. 18 opt Recovery optimization and optimal spending In the following, we derive the optimal recovery rate λq such that the well-being loss as per eq. (22) is minimized. For this, we simplify eq. (20) by omitting losses from diversified income (which depend on the recovery rates of all other quintiles) and taxes: lab reco sav +aid ∆cq (t ) ≈ ∆iq (t ) + ∆cq (t ) − ∆cq (t ) eff −λq t sav +aid (23) ≈ Π + λq κ h ∆kq e − cq (t ). α When optimally used, liquidity and aid payouts reduce consumption losses to a constant level ξ · α (because the ˆ. If ξ = 0, the utility function penalizes deviations from pre-disaster consumption cq ) until they are used up after time t ˆ = 0 means no liquidity is available to offset destroyed capital is instantly repaired from liquidity and aid payouts, while t consumption losses. Thus, the optimal spending is given by   α e − λq t − ξ  ˆ 0≤t ≤t sav +aid cq (t ) = (24) 0  else. ˆ, we can set the constraints As all liquidity and aid payouts should be spent at time t sav +aid ˆ cq (t ) = 0 ˆ t (25) sav +aid sav aid cq (t ) = Sq + Sq , 0 which yields ˆ = − 1 ln(ξ ) t (26) λq and λq sav aid ξ 1 − ln(ξ ) = 1 − (Sq + Sq ), (27) α which cannot be solved analytically and we therefore solve numerically for ξ . Note that a solution for eq. (27) with ξ ∈]0, 1[ exists only if λq sav aid 0 < 1− Sq + Sq < 1. (28) α If the term is less than or equal to 0, this implies that consumption losses can be fully offset with liquidity and post-disaster support. If the term is greater than or equal to 1, this implies no (or negative) liquidity. In the following, we assume that liquidity and post-disaster support are non-negative and not enough to fully offset all consumption losses. Plugging eq. (24) into eq. (23), (simplified, i.e. without diversified income and tax) consumption losses over time can be expressed as   αξ ˆ 0≤t ≤t ∆cq (t ) = (29)  α e − λq t  else. opt To obtain for each household a reconstruction rate λq that minimizes well-being losses, we plug eq. (29) into eq. (22) and minimize with respect to λq : opt short −term λq = arg min (∆Wq ). (30) λq 19 short −term opt Since ∆Wq cannot be solved analytically, we find the optimal recovery rate λq numerically2 . Long-term well-being losses aid and for the reconstruction of We previously assumed the costs for aid payouts Cq reco,shared public and firm-owned assets Cq to be credit-financed, and that repayment is far in the future (after full recovery), widely distributed among the entire population, and spread over time. We distribute these costs proportionally to the recovered household consumption level, which yields the costs born by each household: aid cq Cq = · C aid ∑q ′ nq ′ cq ′ cq (31) reco,shared Cq = · ∆K · ( 1 − κ h ) , ∑q ′ nq ′ cq ′ which all households (independently of exposure or receiving of aid) have to pay. Together with the amount of liquidity sav (which differs by exposure and aid received), the long-term consumption loss amounts to used ∆Sq long −term aid reco,shared sav ∆c q = Cq + Cq + ∆Sq . (32) As the costs are assumed far in the future and distributed over time, we use the marginal utility of a household’s consumption to compute the long-term well-being losses as long −term ∂w long −term −η long −term ∆W q = ∆c q = cq · ∆ch , (33) ∂c c =cq assuming for simplicity that borrowing was done at the discount rate. Of course, the actual cost would include an additional risk premium, which we do not consider here because it would require assumptions on the time and duration of the repayment. We note that such a risk premium depends on a country’s creditworthiness and would, in reality, further increase well-being losses, especially in lower income countries. Total well-being losses Total well-being losses at the household level result to short −term long −term ∆W q = ∆W q + ∆W q . (34) and aggregation to the country level yields ∆W = ∑ nq ∆Wq . (35) q As the calculated loss expresses a decrease in utility, it cannot be compared directly to asset losses. Therefore, we calculate a reference well-being loss that the disaster losses would cause in an idealized scenario of perfect risk risk sharing at the national level, and without inequality as ∂w −η ∆W ref = ∆K · = ∆K · cavg , (36) ∂c c =cavg with the national average consumption cavg . 2 For the numerical optimization, we use the Python package scipy.optimize. For computational reasons, we do not evaluate the integral to infinity, but over 50 years, which is longer than the full recovery time after almost all disasters. 20 Conversely, we compute the equivalent (country-level) consumption loss ∆C eq that would, in an ideal scenario, cause the experienced well-being losses ∆W : ∆W ∆C eq = −η . (37) cavg Note that using the country’s average consumption (rather than the global average) makes the above an equivalent country-level consumption loss (and not an equivalent global loss). The measure therefore accounts for consumption differences within countries, but not across countries, and the results are in country units (which allows to express well-being losses as percent of GDP). Event aggregation The previous sections have described the approach to calculate well-being losses for a single hazard-return period event in one country and indices for different hazards and return periods were omitted for better readability. With the set of return periods R in ascending order and the set of hazards H , the approach yields well-being losses {∆Wj ,h,rp }h∈H ,rp ∈RP for each country j . We average over return periods to obtain annual average well-being losses: ∆ W j ,h , − = ∑ ∆Wj ,h,rpi · (Prpi − Prpi +1 ) · 1rpi >HPj ,h , (38) rpi ∈R 1 with Prpi = rpi the occurrence probability, HPj ,h the protection level against hazard h (see data section), and 1(·) an indicator function that equals 1 if the return period exceeds the hazard protection, and 0 otherwise. The resulting annual average losses can be summed over hazards: ∆W j , − , − = ∑ ∆ W j ,h , − . (39) h ∈H Note that the same aggregation applies to other hazard and return-period specific metrics, e.g., asset losses, or the consumption equivalent of well-being losses. Then, the socio-economic resilience Ψj is given by the ratio of the well-being losses experienced in an ideal scenario to the actual well-being losses or, equivalently, the asset losses and the consumption equivalent of well-being losses: ∆Wjref ,−,− ∆Kj ,−,− Ψj = = eq . (40) ∆W j , − , − ∆ Cj , − , − Data GDP and population. Gross domestic product (GDP) per capita (in constant 2021 international dollars) and population data are obtained from the World Bank’s International Comparison Program72 and World Development Indicators73 . Income shares. We compute per-quintile income cq from GDP per capita using the share of income held by each income quintile, obtained from the World Bank Poverty and Inequality Platform19 . At the household level, we assume that consumption is equal to income. Financial inclusion. fin of the population with savings at a financial For each income quintile, we obtain the fraction γq institution from the most recent Global Financial Inclusion Database (Findex)30 . 21 Diversified income. For each income quintile, the share of income that is received from diversified sources (i.e., income that is not generated from labor but that comes from e.g., social protection, transfers, and remittances) is obtained from the ASPIRE data set29 . This data set provides coverage (Cov , the share of the population receiving transfers) and adequacy (Ade, the amount received as a share of total welfare) for private transfers (pt) and social protection and labor programs (sp) each income quintile. We compute the overall share of income from diversified sources as div pt pt sp sp fin γq = Covq · Adeq + Covq · Adeq +0.1 · γq , (41) sp,pt γq where we assume that the fraction of income from diversified sources increases by 10% for people with bank accounts (see previous paragraph). The ASPIRE data set does not cover all countries. To increase the country coverage, for each sp,pt income quintile we fit a regression model for the share of income from social protection and transfers γq based on the available data. Then, we use this model to estimate shares of diversified income for missing countries. As regressors, we use (i) the social protection expenditure as share of GDP (SOC )74 (ii) the unemployment rate21 (UNE ), (iii) the amount of personal remittances as a share of GDP73 (REM ), and (iii) binary indicators for HICs (IHICs ), UMICs (IUMICs ), former Soviet and Yugoslav countries (IFSY ), and for World Bank regions Middle East and North Africa (IMNA ), Europe and Central Asia (IECA ), East Asia and Pacific (IEAP ), and Latin America and Caribbean (ILAC ). The regressions read sp,pt γq =1 ∼ SOC + UNE + IHICs + IUMICs + IMNA + IFSY sp,pt γq =2 ∼ SOC + UNE + IHICs + IUMICs + IMNA + IECA + IEAP sp,pt γq =3 ∼ SOC + UNE + IHICs + IUMICs + IMNA + IECA + IEAP (42) sp,pt γq =4 ∼ SOC + REM + IHICs + IUMICs + IMNA + ILAC + IEAP sp,pt γq =5 ∼ SOC + REM + IUMICs + IEAP . Regression results are shown in table Supplementary Tbl. 7. For country coverage see Supplementary Fig. 3 and Sup- plementary Tbls. 4 and 5. Household liquidity. sav (simply ”liquidity”) are estimated for each Liquid savings and the ability to borrow money Sq income quintile from Findex30 . For this, we calculate the share of respondents who reported that they could come up with the sum of 1/20 of the gross national income per capita (GNIpc) for an emergency through savings, family and relatives or friends, money from working, borrowing, or other sources as the primary source (excluding respondents who would need to sell assets). For each income quintile, we multiply the share of positive respondents with the sum of 1/20 of the GNIpc (from ref72 ) to obtain average liquidity within the income group. Note that this imperfect estimate does not capture savings below the threshold of 1/20 of the GNIpc, nor does it capture that households can have liquid savings beyond this threshold. Because of these limitations, our estimate should be considered a lower estimate on liquid savings. Capital productivity. We use the Penn World Table version 10.0175 to derive the average productivity of capital Π per country as the ratio of output-side GDP to capital stock. Fraction of destroyed assets. We calculate the share of assets that is destroyed by natural hazards using modelled loss estimates from the Global Infrastructure Risk Model and Resilience Index1 (”GIRI”). The data contains modelled 22 asset losses for floods, tropical cyclones (accounting for storm surge and wind separately), tsunamis, and earthquakes at various return periods from 1-in-10 to 1-in-5,000 years. To maintain data consistency in our model, we do not use absolute losses from GIRI, but compute the share of assets that are destroyed for each hazard-return period combination, ∆Kh,rp K , as the ratio of modelled asset losses to the total capital stock as per the GIRI data. Hazard protection. We consider hazard protection levels HPh against a hazard. The protection level is expressed in terms of return period events against which assets are protected. We obtain protection levels for riverine floods from the merged protection layer of the FLOPROS dataset33 , based on reported and modelled protection, as well as established protection standards. We merge data from refs33,76 to create a similar protection layer for storm surges. Protection levels are aggregated to the national level weighted by population density (ref77 ) using country shapefiles from ref78 . For all other hazards, we assume no protection, i.e., HPh = 0. Country-level vulnerability. We estimate average vulnerability of the national capital stock VGEM from the Global Exposure Model (GEM)34 . GEM provides a global inventory of buildings with a classification of the building structure and materials used, as well as replacement costs (including building contents). Thus, we assume that on average, building contents (e.g., household property, or machinery) and other assets (e.g., public infrastructure) have the same vulnerability as the building stock. For each of the considered hazards, we map 119 identifiers of lateral load system materials, types, and building heights to three categories with associated vulnerabilities (fragile: 0.7, median: 0.3, and robust: 0.1; mapping in Supplementary Tbl. 8). We then average vulnerability at the country level, weighted by the total replacement cost. We further account for the availability of early warning systems, which can significantly reduce vulnerability79 . Based on analyses of past disasters80,81 , we reduce vulnerability by up to 20% in case of a perfectly functioning early warning system for all hazards except tsunamis and earthquakes. Thus, country-level vulnerability results to V = VGEM · (1 − 0.2 · qEW ), where qEW ∈ [0, 1] is an indicator for the availability of early warning systems (see data section on early warning). Household-level vulnerability We use a representative household’s relative (to the national average) vulnerability vq ,rel to derive household-level vulnerability. We obtain this relative vulnerability from the harmonized household survey microdata of the World Bank’s Global Monitoring Database (GMD)35 , which includes information on dwellings of individual households. We use data on the types of materials used for roofs, walls, and floors and classify these into the categories fragile (0.7), median (0.3), and robust (0.1), which results in up to nine housing vulnerability variables. We then apply a principal component analysis at the country level and retain the first principal component, which we normalize such that it is bounded between 0 and 1. This yields a vulnerability of individual households in the survey, which we aggregate to the national and income quintile level using a weighted average. Finally, we obtain vq ,rel as the income quintile average relative to the national average. If vq ,rel is greater (lower) than 1, this means that households in this quintile are more (less) vulnerable than the national average. The vulnerability of representative households per income quintile is then given by the product of the national to the relative vulnerability: vq = V · vq ,rel . While this vulnerability is calibrated on the national building stock and household dwellings, we assume that it applies to all assets used by one household (including other household-owned assets, as well as public and firm-owned assets). Since GMD does not cover all countries in our simulation, we apply the heuristic that the poorest people live in the most vulnerable buildings as identified by the GEM data set for those countries without GMD coverage. For country coverage see Supplementary Fig. 3 and Supplementary Tbls. 4 and 5. 23 Early warning. To estimate the availability of early warning systems, we employ the Hyogo Framework for Action82 (HFA) with country-level indicators on an ordinal 1 to 5 scale, which we normalize to 1. We complement this top-down government-reported indicator with bottom-up household survey-based information from the 2021 World Risk Poll83 (WRP). We calculate a score for the availability of early warning systems as the mean of two individual data set indicators: EWHFA + EWWRP qEW = . (43) 2 As the government-reported score EWHFA , we use HFA indicator P2-C3 ("Early warning systems are in place for all major hazards, with outreach to communities"). For the bottom-up early warning indicator EWWRP , we compute the share of WRP respondents that reported to have received a warning about a previously experienced disaster via internet or social media, the local government or police, broadcasting or newspapers, or local communities or organizations. In cases where only one indicator of the two indicators is available, we use this indicator directly. For countries where both indicators are missing, we use the average indicator value of countries in the same income group and world region, assuming . For country coverage see Supplementary Fig. 3 and Supplementary Tbls. 4 and 5. Exposure bias. We calculate the ratio of each income quintile’s exposure to national exposure, also referred to as the exposure bias EBq . It measures to what extent people in one income group are more or less exposed to a given hazard than people on average in the entire country: if the value is greater (lower) than 1, a higher (lower) share of people within the income group are affected than in the average population. To compute the bias, we use estimates of exposed people by poverty line and at the national level for tropical cyclones (which we apply to wind and storm surge) and floods as described in ref31 . We first calculate the exposure for population shares segmented by poverty line. Then, we divide by national exposure to obtain exposure biases. For each income quintile, we calculate the population-weighted average EBq over the exposure biases of all segments that fall into this income quintile. Since estimates of affected people are not available for all countries, we fill missing data with population-weighted exposure biases of countries where data is available. For country coverage see Supplementary Fig. 3 and Supplementary Tbls. 4 and 5. Asset shares. Households use productive assets with different ownership to generate income from labor. This includes assets that they own directly (in case of self-employed activities), other private assets (e.g., machines and factories), and public assets (e.g., infrastructure). To capture the full impact of a disaster, we further account for housing services. We here assume that these services are recorded in household consumption for rented dwellings, and housing capital needs to be imputed for households who own their dwelling (and thus receive a return from housing capital that is not recorded as a transaction). In case of destroyed assets due to a disaster, owners need to pay for the reconstruction of their home (while tenants don’t) and self-employed people need to pay for the reconstruction of their productive assets (as opposed to employees). We split capital into shares of public κ p , private or firm-owned κ f , and household-owned capital κ h . Because data is not available to perform estimates per income quintile, we assume these shares to be equal for all income quintiles. We obtain the public capital share κ p as the sum of general government and public-private partnership capital shares from the IMF investment and capital stock dataset20 . Next, we derive the capital share of the real-estate sector κ r from the value added share of this sector, using national accounts data from the United Nations Statistics Division27 . For this, we use a ratio of 3.7, derived from Eurostat data23,24 for 24 countries, where 24 both shares are available. Using national home ownership rates γh (refs22,25,26,28 ) and self-employment rates γs (ref21 ), household-owned and private capital shares result to κ h = γh · κ r + γs (1 − κ p − γh · κ r ) (44) κf = 1 − κp − κh where we assume that the capital share of owner-occupied dwellings is equal to the country’s home ownership fraction of real-estate capital. We find this assumption to work well on six European countries for which Eurostat provides capital data of imputed owner-occupied rents. We split the remaining capital between self-employed (i.e., household-owned) and other private capital using a country’s self-employment rate. As real-estate shares of value added and home ownership rates are not available for all countries, we fill values of missing countries with the median of the country’s income group. For country coverage see Supplementary Fig. 3 and Supplementary Tbls. 4 and 5. 25 References 1. CDRI. Global Infrastructure Resilience: Capturing the Resilience Dividend - A Biennial Report from the Coalition for Disaster Resilient Infrastructure 2023. https://giri.unepgrid.ch. 2. United Nations Office for Disaster Risk Reduction. Global Assessment Report on Disaster Risk Reduction 2022: Our World at Risk: Transforming Governance for a Resilient Future (2022). 3. UNISDR. Technical Guidance for Monitoring and Reporting on Progress in Achieving the Global Targets of the Sendai Framework for Disaster Risk Reduction. Collection of Technical Notes on Data and Methodology (2017). 4. Hallegatte, S., Vogt-Schilb, A., Rozenberg, J., Bangalore, M. & Beaudet, C. From poverty to disaster and back: A review of the literature. Economics of Disasters and Climate Change 4, 223–247 (2020). 5. Kawasaki, A. & Shimomura, N. Accelerated widening of economic disparity due to recurrent floods. International Journal of Disaster Risk Reduction 102, 104273 (2024). 6. Markhvida, M., Walsh, B., Hallegatte, S. & Baker, J. Quantification of disaster impacts through household well-being losses. Nature Sustainability 3, 538–547 (2020). 7. Skoufias, E., Kawasoe, Y., Strobl, E. & Acosta, P. Identifying the vulnerable to poverty from natural disasters: The case of typhoons in the Philippines. Economics of Disasters and Climate Change 4, 45–82 (2020). 8. De Silva, M. & Kawasaki, A. Socioeconomic vulnerability to disaster risk: a case study of flood and drought impact in a rural Sri Lankan community. Ecological Economics 152, 131–140 (2018). 9. Erman, A. et al. The road to recovery the role of poverty in the exposure, vulnerability and resilience to floods in Accra. Economics of Disasters and Climate Change 4, 171–193 (2020). 10. Reimann, L., Koks, E., de Moel, H., Ton, M. J. & Aerts, J. C. An Empirical Social Vulnerability Map for Flood Risk Assessment at Global Scale (“GlobE-SoVI”). Earth’s Future 12, e2023EF003895 (2024). 11. Moore, D., Niazi, Z., Rouse, R. & Kramer, B. Building resilience through financial inclusion: A review of existing evidence and knowledge gaps (2019). 12. Bakkensen, L. A., Fox-Lent, C., Read, L. K. & Linkov, I. Validating resilience and vulnerability indices in the context of natural disasters. Risk analysis 37, 982–1004 (2017). 13. Rufat, S., Tate, E., Burton, C. G. & Maroof, A. S. Social vulnerability to floods: Review of case studies and implications for measurement. International journal of disaster risk reduction 14, 470–486 (2015). 14. Hallegatte, S., Bangalore, M. & Vogt-Schilb, A. C. Socioeconomic resilience: multi-hazard estimates in 117 countries. World Bank Policy Research Working Paper (2016). 15. Walsh, B. & Hallegatte, S. Measuring natural risks in the Philippines: socioeconomic resilience and wellbeing losses. Economics of Disasters and Climate Change 4, 249–293 (2020). 16. Verschuur, J., Koks, E., Haque, A. & Hall, J. Prioritising resilience policies to reduce welfare losses from natural disasters: a case study for coastal Bangladesh. Global Environmental Change 65, 102179 (2020). 17. Kerblat, Y., Walsh, B., Simpson, A. & Hallegatte, S. Overlooked: examining the impact of disasters and climate shocks on poverty in the Europe and Central Asia region. World Bank (2021). 18. Sauer, I. J., Walsh, B., Frieler, K., Bresch, D. N. & Otto, C. Understanding the distributional effects of recurrent floods in the Philippines. iScience 28 (2025). 26 19. World Bank. Poverty and Inequality Platform (version 20240326_2017) 2023. www.pip.worldbank.org. [Dataset] Accessed 01 Apr 2025. 20. IMF. Investment and Capital Stock Dataset (ICSD) 2022. https://data.imf.org/?sk=1ce8a55f-cfa7-4bc0- bce2-256ee65ac0e4. [Dataset] Accessed 13 Feb 2025. 21. International Labour Organization. ILO modelled estimates database, ILOSTAT 2025. https://ilostat.ilo.org/ data/. [Dataset] Accessed 01 Apr 2025. 22. OECD. OECD Affordable Housing Database 2025. https://www.oecd.org/content/oecd/en/data/datasets/ oecd-affordable-housing-database.html. [Dataset] Accessed 13 Feb 2025. 23. Eurostat. Capital stocks by industry (NACE Rev.2) and detailed asset type (nama_10_nfa_st) 2025. https : //ec.europa.eu/eurostat/databrowser/view/nama_10_nfa_st/table. [Dataset] Accessed 10 Feb 2025. 24. Eurostat. Gross value added and income by detailed industry (NACE Rev.2) (nama_10_a64) 2025. https : //ec.europa.eu/eurostat/databrowser/view/nama_10_a64/default/table. [Dataset] Accessed 03 Feb 2025. 25. Eurostat. Distribution of population by tenure status, type of household and income group (ilc_lvho02) 2025. https://ec.europa.eu/eurostat/databrowser/view/ilc_lvho02/default/table. [Dataset] Accessed 13 Feb 2025. 26. United Nations Statistics Division. Households in housing units by type of housing unit, tenure of household and urban/rural residence 2025. http://data.un.org/Data.aspx?d=POP&f=tableCode%3a304. Demographic Statistics Database [Dataset]. Accessed 18 Feb 2025. 27. United Nations Statistics Division. Table 2.4 Value added by industries at constant prices (ISIC Rev. 4) 2025. http://data.un.org/Data.aspx?d=SNA&f=group_code%3a205. National Accounts Official Country Data [Dataset]. Accessed 13 Feb 2025. 28. Centre for Affordable Housing Finance in Africa. 2024 Yearbook: Housing Finance in Africa http://housingfinanceafrica. org (Centre for Affordable Housing Finance in Africa, Johannesburg, South Africa, 2024). 29. World Bank. The atlas of social protection indicators of resilience and equity 2025. https://www.worldbank.org/ en/data/datatopics/aspire. [Dataset] Accessed 01 Apr 2025. 30. Demirgüç-Kunt, A., Klapper, L., Singer, D. & Ansar, S. The Global Findex Database 2021: Financial inclusion, digital payments, and resilience in the age of COVID-19 (World Bank Publications, 2022). [Dataset] Accessed 18 Feb 2025. 31. Doan, M. K. et al. Counting People Exposed to, Vulnerable to, or at High Risk From Climate Shocks—A Methodology. World Bank Policy Research Working Paper (2023). 32. Tiggeloven, T. et al. Global-scale benefit–cost analysis of coastal flood adaptation to different flood risk drivers using structural measures. Natural Hazards and Earth System Sciences 20, 1025–1044 (2020). 33. Scussolini, P. et al. FLOPROS: an evolving global database of flood protection standards. Natural Hazards and Earth System Sciences 16, 1049–1061 (2016). [Dataset] Accessed 19 Jan 2024. 34. Yepes-Estrada, C. et al. Global building exposure model for earthquake risk assessment. Earthquake Spectra 39, 2212–2235. https://github.com/gem/global_exposure_model (2023). [Dataset] Accessed 06 Jan 2025. 35. World Bank. Global Monitoring Database (GMD) 2024. [Dataset] Accessed 28 Feb 2024. 27 36. Behrman, J. R. The impact of health and nutrition on education. The World Bank Research Observer 11, 23–37 (1996). 37. Glewwe, P., Jacoby, H. G. & King, E. M. Early childhood nutrition and academic achievement: a longitudinal analysis. Journal of public economics 81, 345–368 (2001). 38. Canavire-Bacarreza, G., Carrasco, N., Aliaga, G. A. G., Nsababera, O. & Botero, M. C. Household vulnerability to and preparedness for disasters in Haiti. International Journal of Disaster Risk Reduction 111, 104752 (2024). 39. Schumacher, I. & Strobl, E. Economic development and losses due to natural disasters: The role of hazard exposure. Ecological Economics 72, 97–105 (2011). 40. Rentschler, J. et al. Global evidence of rapid urban growth in flood zones since 1985. Nature 622, 87–92 (2023). 41. Estrada, F., Botzen, W. W. & Tol, R. S. Economic losses from US hurricanes consistent with an influence from climate change. Nature Geoscience 8, 880–884 (2015). 42. Naoaj, M. S. From catastrophe to recovery: the impact of natural disasters on economic growth in developed and developing countries. European Journal of Development Studies 3, 17–22 (2023). 43. Platt, S., Brown, D. & Hughes, M. Measuring resilience and recovery. International Journal of Disaster Risk Reduction 19, 447–460 (2016). 44. Barton-Henry, K. & Wenz, L. Nighttime light data reveal lack of full recovery after hurricanes in Southern US. Environmental Research Letters 17, 114015 (2022). 45. Sou, G., Shaw, D. & Aponte-Gonzalez, F. A multidimensional framework for disaster recovery: Longitudinal qualitative evidence from Puerto Rican households. World Development 144, 105489 (2021). 46. De Alwis, D. Distributional impacts of disaster recovery: Sri Lankan households a decade after the 2004 Indian Ocean tsunami. Economics of Disasters and Climate Change 4, 195–222 (2020). 47. Francisco, J. P. Determinants of property damage recovery time among households affected by an extreme flood event in Metro Manila, Philippines. Philippines (September 15, 2013) (2013). 48. Ivaschenko, O., Doyle, J., Kim, J., Sibley, J. & Majoka, Z. Does ‘Manna from Heaven’help? The role of cash transfers in disaster recovery—lessons from Fiji after Tropical Cyclone Winston. Disasters 44, 455–476 (2020). 49. Ndirangu, L., Burger, K., Moll, H. A. & Kuyvenhoven, A. Seasonality, precautionary savings and health uncertainty: Evidence from farm households in central Kenya (2009). 50. Banerjee, A. V. & Duflo, E. Poor economics: A radical rethinking of the way to fight global poverty (Public Affairs, 2011). 51. World Bank. World development report 2014: Risk and opportunity–managing risk for development http://hdl. handle.net/10986/16092 (The World Bank, 2013). 52. Shavell, S. Risk sharing and incentives in the principal and agent relationship. The Bell Journal of Economics, 55–73 (1979). 53. Soden, R. et al. The importance of accounting for equity in disaster risk models. Communications Earth & Environment 4, 386 (2023). 54. De Bruijn, K. M. et al. Flood risk management through a resilience lens. Communications Earth & Environment 3, 285 (2022). 28 55. Kim, J. H. & Sutley, E. J. Implementation of social equity metrics in an engineering-based framework for distributing disaster resources. International Journal of Disaster Risk Reduction 64, 102485 (2021). 56. Pelling, M. & Garschagen, M. Put equity first in climate adaptation. Nature 569, 327–329 (2019). 57. World Bank. State of Social Protection Report 2025: The 2-Billion-Person Challenge. (2025). 58. Noy, I., Nguyen, C. & Patel, P. Floods and spillovers: Households after the 2011 great flood in Thailand. Economic Development and Cultural Change 69, 829–868 (2021). 59. Carter, M. R. & Barrett, C. B. The economics of poverty traps and persistent poverty: An asset-based approach. The Journal of development studies 42, 178–199 (2006). 60. Carter, M. R., Little, P. D., Mogues, T. & Negatu, W. Poverty traps and natural disasters in Ethiopia and Honduras. World development 35, 835–856 (2007). 61. Barrett, C. B., Garg, T. & McBride, L. Well-being dynamics and poverty traps. Annual Review of Resource Economics 8, 303–327 (2016). 62. Gilligan, D. O., Hoddinott, J. & Taffesse, A. S. The impact of Ethiopia’s Productive Safety Net Programme and its linkages. The journal of development studies 45, 1684–1706 (2009). 63. Francken, N., Minten, B. & Swinnen, J. F. The political economy of relief aid allocation: evidence from Madagascar. World Development 40, 486–500 (2012). 64. Van den Hurk, B. J. et al. Climate impact storylines for assessing socio-economic responses to remote events. Climate Risk Management 40, 100500 (2023). 65. Verschuur, J., Koks, E. E. & Hall, J. W. Systemic risks from climate-related disruptions at ports. Nature Climate Change 13, 804–806 (2023). 66. Hallegatte, S. An adaptive regional input-output model and its application to the assessment of the economic cost of Katrina. Risk Analysis: An International Journal 28, 779–799 (2008). 67. Middelanis, R. et al. Wave-like global economic ripple response to Hurricane Sandy. Environmental Research Letters 16, 124049 (2021). 68. Baez, J. E., Caruso, G. & Niu, C. Extreme weather and poverty risk: evidence from multiple shocks in Mozambique. Economics of Disasters and Climate Change 4, 103–127 (2020). 69. Erman, A. E. et al. Wading out the storm: The role of poverty in exposure, vulnerability and resilience to floods in Dar Es Salaam tech. rep. (The World Bank, 2019). 70. Bangalore, M., Smith, A. & Veldkamp, T. Exposure to floods, climate change, and poverty in Vietnam. Economics of Disasters and Climate Change 3, 79–99 (2019). 71. Dercon, S. in Macroeconomic Policies and Poverty 308–329 (Routledge, 2004). 72. World Bank. International Comparison Program 2025. https://www.worldbank.org/en/programs/icp. [Dataset] Accessed 01 Apr 2025. 73. World Bank. World Development Indicators 2025. https://doi.org/10.57966/6rwy-0b07. [Dataset] Accessed 01 Apr 2025. 74. International Labour Organization. World Social Protection Report 2024-26: Universal social protection for climate action and a just transition (2024). [Dataset] Accessed 25 Feb 2024. 29 75. Feenstra, R. C., Inklaar, R. & Timmer, M. P. The next generation of the Penn World Table. American economic review 105, 3150–3182 (2015). [Dataset] Accessed 12 Dec 2023. 76. Tiggeloven, T. Benefit-cost analysis of adaptation objectives to coastal flooding at the global scale (Version 2) (Zenodo, 2020). https://doi.org/10.5281/zenodo.4275517. [Dataset] Accessed 19 Jan 2024. 77. Center For International Earth Science Information Network-CIESIN-Columbia University. Gridded Population of the World, Version 4 (GPWv4): Population Density Adjusted to Match 2015 Revision UN WPP Country Totals, Revision 11 (Version 4.11) Palisades, NY, 2018. https://doi.org/10.7927/H4F47M65. [Dataset] Accessed 15 Jan 2024. 78. GADM. GADM Admin 0 shapefiles (version 4.1) 2022. https://gadm.org. 79. Hallegatte, S. A cost effective solution to reduce disaster losses in developing countries: hydro-meteorological services, early warning, and evacuation. World Bank policy research working paper (2012). 80. Thieken, A. H., Müller, M., Kreibich, H. & Merz, B. Flood damage and influencing factors: New insights from the August 2002 flood in Germany. Water resources research 41 (2005). 81. Kreibich, H., Thieken, A. H., Petrow, T., Müller, M. & Merz, B. Flood loss reduction of private households due to building precautionary measures–lessons learned from the Elbe flood in August 2002. Natural hazards and earth system sciences 5, 117–126 (2005). 82. UN-ISDR. Hyogo framework for action 2005-2015: building the resilience of nations and communities to disasters in (2007). 83. Lloyd’s Register Foundation. 2021 World Risk Poll 2022. https://wrp.lrfoundation.org.uk/data-resources. [Dataset] Accessed 30 Jan 2024. 84. Middelanis, R., Jafino, B. A., Hill, R., Nguyen, M. C. & Hallegatte, S. Data and code for "Global socio-economic resilience to natural disasters" 2025. https://doi.org/10.5281/zenodo.14889021. [Dataset]. 30 Declarations Data availability Preprocessed data to run the model is provided alongside with the model code in ref84 . The preprocessed data were generated using the following datasets. Where World Bank indicator codes are provided, datasets are openly available at the World Bank Open Data portal (https://data.worldbank.org). GDP and GNI per capita in constant 2021 international dollars were obtained from the International Comparison Program72 and are accessible with World Bank indicator codes NY.GDP.PCAP.PP.KD and NY.GNP.PCAP.PP.KD. Population and personal remittances as a share of GDP were obtained from the World Development Indicators database73 and are accessible with World Bank indicator codes SP.POP.TOTL and BX.TRF.PWKR.DT.GD.ZS. Per quintile income shares were obtained from the Poverty and Inequality Platform (PIP)19 and are accessible with World Bank indicator codes SI.DST.FRST.20, SI.DST.02nd.20, SI.DST.03rd.20, SI.DST.04th.20, and SI.DST.05th.20. Gini index data was obtained from PIP with World Bank indicator code SI.POV.GINI. Coverage and adequacy of private remittances and private transfers by income quintile were obtained from the ASPIRE dataset29 and are accessible with World Bank indicator codes per_pr_allpr.cov_q{i}_tot, per_allsp.cov_q{i}_tot, per_pr_allpr.adq_q{i}_tot, and per_allsp.adq_q{i}_tot (where {i} is to be replaced by the income quintile index 1, 2, 3, 4, or 5). Unemployment rates and self-employment rates were obtained from the International Labor Organization’s ILOSTAT database21 and are accessible with World Bank indicator codes SL.UEM.TOTL.ZS and SL.EMP.SELF.ZS. Findex30 microdata is accessible at https://microdata.worldbank.org/index.php/catalog/global-findex. ILO Public health and social protection expenditure data74 is available at https://www.social-protection.org/gimi/ ShowWiki.action?id=52. The Penn World Table v10.0175 is available at https://www.rug.nl/ggdc/productivity/pwt/. Modeled asset losses from the Global Infrastructure and Resilience Index1 are available at https://giri.unepgrid.ch. The FLOPROS33 dataset is available as supplementary material from https://nhess.copernicus.org/articles/16/ 1049/2016/ and the modeled coastal protection layer (ref76 ) is available at https://doi.org/10.5281/zenodo.4275517. GADM78 shapefiles used for flood protection aggregation are freely available at https://gadm.org. Population density data from the Gridded Population of the World data set77 , Version 4 (GPWv4), used for flood protection aggregation, was obtained from https://doi.org/10.7927/H4F47M65. The Global Exposure Model dataset34 is available at https://github.com/gem/global_exposure_model. The Global Monitoring Database dataset35 cannot be made publicly available, but the derived relative per-quintile vulnerabilities can be requested from the authors. Performance indicator scores from the Hyogo Framework For Action82 can be obtained from country reports at https://www.preventionweb.net/sendai-framework/Hyogo-Framework-for-Action/reports. The World Risk Poll dataset83 can be obtained from https://www.lrfoundation.org.uk/wrp. 31 Estimates of the number of exposed people by poverty line and hazard (ref31 ) are available from the authors upon request. The IMF investment and capital stock dataset20 is available at https://data.imf.org/?sk=1ce8a55f-cfa7-4bc0-bce2-256ee65ac0e4. United Nations Statistics Division datasets on global value added by industry (ref27 ) and household tenure (ref26 ) are publicly available at http://data.un.org/. Eurostat capital stock (ref23 ), value added by industry (ref24 ), and housing tenure (ref25 ) data are available at https: //ec.europa.eu/eurostat/ with dataset identifiers nama_10_nfa_st, nama_10_a64, and ilc_lvho02. The OECD Affordable Housing Database22 is available at https://www.oecd.org/content/oecd/en/data/datasets/ oecd-affordable-housing-database.html. Code availability The data and code to reproduce the results of this study will be made available on Zenodo upon publication at https://doi.org/10.5281/zenodo.14889021 (ref84 ). Acknowledgments We thank Antonios Pomonis for his help in mapping building materials to hazard-specific vulnerability classes and David P Allen for editorial support. We are also grateful to Rashmin Gunasekera for his support of this study and to Jamele Rigolini, Samuel Freije-Rodriguez, and Yohannes Yemane Kesete for their valuable feedback and suggestions. This work has received support from the Whole of Economy: Social and Just Transition Trust Fund, part of the Climate Support Facility Trust Fund, and from the Global Facility for Disaster Reduction and Recovery. Author information R.M., B.A.J., R.H., and S.H. designed the study. R.M. conducted the simulations and analysis. M.C.N. provided the relative household vulnerability data. R.M. wrote the manuscript with the help of B.A.J. and S.H. All authors discussed the results. Competing interests The authors declare that they have no competing interests. Additional information Supplementary information is available for this paper. 32 Supplementary information for: Global Socio-economic Resilience to Natural Disasters Robin Middelanis, Bramka Arga Jafino, Ruth Hill, Minh Cong Nguyen, Stéphane Hallegatte Table of contents Supplementary figures Supplementary Fig. 1: Modeling approach. Supplementary Fig. 2: Illustration of loss and income flows in the model. Supplementary Fig. 3: Data coverage. Supplementary Fig. 4: Estimated capital shares. Supplementary Fig. 5: Relationship between risk to well-being and risk to assets. Supplementary Fig. 6: Relationship between socio-economic resilience and risk to assets. Supplementary Fig. 7: Relationship between average recovery duration and socio-economic resilience. Supplementary Fig. 8: Liquidity by household income quintile across countries. Supplementary Fig. 9: Recovery durations increase without liquidity. Supplementary Fig. 10: Household vulnerability by income quintile and hazard. Supplementary tables Supplementary Tbl. 1: Model parameters used in the simulations. Supplementary Tbl. 2: Exogenous variables in the model and data sources. Supplementary Tbl. 3: Endogenous variables in the model. Supplementary Tbl. 4: Datasets used. Supplementary Tbl. 5: Simulation results for all countries in the baseline scenario. Supplementary Tbl. 6: Regression results for disaster risks, resilience, and recovery time. Supplementary Tbl. 7: Regression results for imputed income share from social protection and transfers. Supplementary Tbl. 8: Mapping of GEM taxonomy codes to vulnerability classes.. 1 Supplementary figures Asset losses • At the country level Asset-focused • Tropical cyclone (storm surge, wind), tsunami, Risk assessment flood, earthquake • Taken from CDRI (2023) Consumption utility Asset losses losses function Household-level • Of representative productivity Well-being • Income losses wellbeing impact disaggregate households Reconstruction losses losses • 5 income quintiles • Post-disaster support • Liquidity recovery Well-being minimization Recovery duration Optimal dynamics recovery rate Supplementary Figure 1: Modeling approach. The approach expands on asset-focused risk assessments by ac- counting for socio-economic disparities and the temporal component of recovery dynamics. Asset losses at the national level are disaggregated into household asset losses per income quintile. From these asset losses, we derive decreases in consumption and well-being. Assuming rational households, we optimize for the recovery duration that minimizes well-being losses of each household. 2 Assets Output Diversification Consumption Well-being losses (short-term / long-term) qa 1 qa 2 qa 3 Consumption qa 4 Income loss Reconstruction loss Liquidity Time qa 5 qna avg Supplementary Figure 2: Illustration of loss and income flows in the model. Example assets and aggregated output, consumption from labor and diversified income and their losses are shown, as well as the resulting well-being losses. Quantities are shown for individual affected households in each income quintile, as well as an average non-affected household. Flows (i.e. output, diversified income, and consumption) are aggregated over the time until all households have recovered. Total bar heights denote values in the absence of a disaster, grey bars are losses that occur due to the disaster. Diagonally hatched areas denote losses of assets owned directly by the households (as experienced right after the disaster) and the resulting aggregated consumption losses to pay for their reconstruction. Cross-hatched areas denotes used liquidity to offset consumption losses. Note that bar heights can be compared only within the same column. For example, pre-disaster assets of a q1 household are worth about half the pre-disasters assets of a q2 household, but not equal to the pre-disaster output of a q1 household. The flow of the model is as follows. When a share of each affected household’s assets gets destroyed (according to the households vulnerability) by a disaster, this causes output losses which reduce both the income from labor (for affected households only) and diversified income (all households). The consumption of affected households is further decreased by reconstruction payments, and consumption losses persist until recovery is completed. Over time, dynamic (i.e., short-term) well-being losses aggregate. In addition to these short-term well-being losses, the reconstruction of public and firm-owned assets, the replenishing of used liquidity as well as potential disaster aid payments cause long-term well-being losses. 3 a Fraction of income from social protection and transfers b Availability of early warning systems IBRD 48829 | IBRD 48830 | APRIL 2025 APRIL 2025 c Real estate share of value added d Home ownership rate IBRD 48831 | IBRD 48832 | APRIL 2025 APRIL 2025 e Dwelling materials microdata f Number of exposed people by hazard type and poverty line IBRD 48833 | IBRD 48834 | APRIL 2025 APRIL 2025 Supplementary Figure 3: Data coverage. Data is available for countries in green and was imputed for countries in sp,pt purple. Only data with incomplete coverage is shown. (a) Fraction of income from social protection and transfers γq from ref1 , (b) availability of early warning systems qEW from refs2,3 , (c) real-estate share of value added from ref4 , (d) home ownership rate γh from refs5–8 , (e) household microdata including dwelling materials used to derive relative household vulnerability vq ,rel from ref9 (for countries in dark grey, vq ,rel is solely determined from ref10 ), (f) number of people exposed to floods and tropical cyclones according to ref11 . 4 a public capital share p b firm-owned capital share f c household-owned capital share h COD 70 NPL LICs IRQ TCD UGA NER LMICs 60 CAF TJK BOL BLR DEU CHE IRL GHA ZMB COG IND TZA MDA ZWE CIV BEN ARM UMICs RWA TGOLAO EGY AZE ZAF TUR LTUCYP AUT NOR BEL SWE NGA CMR COM MRT SEN MLI LBR MDG HTIBGD MMR GIN LBN BTN HICs 50 LSO MOZ ETH CHN UKRCHL SVN KOR ISR AUS MLTDNK LUX SLE KEN BFA AGO ETH IDN IRN ALB MWI SLE GMB SWZDZA ECU MYS BGR CRI POL CZE ISL NLD PAK MOZ GTM GEO MKD SRB COL GRC PAN SVK HUNITA MDG KAZ SDN VNM PER DOM share [%] GIN MLI BFA ARG MUS SRB BIH EST ROU RUS ESP GBR FRA FIN USA TGO MWI HND GMBMAR JOR LKA LAO SLV PHL PRY MEX ROU LVA CAN 40 BEN NGA HTI AGO LBRUZB MNG GEO TUN LKA PHL MAR SLV BWA COL IDNDOM MNE MEX LVA MYSPRT JPN CAN NIC CAFUZB BRA GAB ECU CRIKAZ URY CHL PRT GBR ESP ISRFIN FRA KEN TZA SEN PAKNAM TUNBIH BWA MUS BRA URY JPN PRY GTM PAN GABURY HRV SVK ITA NAM BOL AZE CHN MNG ARG MNE EST HRV RUS CZE AUS BEL MLTISL 30 NIC TCD MMR HND NER UGA SDN JOR IND CIV CMR COG BTN VNM BGD MKD GAB ALB PER MNEGRC HRV RUS DNK NLD USA NOR SDN TJKLBN NAM NIC HND MRT COM JOR PER MNG VNMBRA HUN IRN DZA RWA LSO TUN SWZDZA UKR BWA MUS BIH BGR TUR BLR JPN POLKOR SVN LTUCYP USA NLD AUT PHL ZWE NPLSLV ARM MEX IRN PRY ZAF GTM ARG BGR POL PRT EST HUNSVNFRA FIN CYP SWE ISL LUX LSO PAKSWZ UZB BGD MKD IRQ ZAF EGY SWE DNK IRL LUX MAR ZMB MRT COM UKR LKA MDA COL DOM LVA SVK PAN KAZ LTU ROU GBR ESP CZE KORCAN ITAMLT AUT GHA ZWE ZMB ALB EGY CHN GRC MDA ECU DEU CHE NOR 20 COD GHA CRITUR DEU MMR CMR RWA BTN ARM MYS SRB IDN BLR LBN CHL ISR AUS BEL CHE IRL AGO KEN GMB CIV COG SEN MWI HTI LBRIND GEO 10 BFA UGA NPL COD TZA IRQ BOL AZE TJK NGA GIN MOZ MLI NERLAO BEN ETH MDG TCD TGO SLE CAF 0 0 25 50 75 100 125 0 25 50 75 100 125 0 25 50 75 100 125 GDP per capita [$PPP 1,000] GDP per capita [$PPP 1,000] GDP per capita [$PPP 1,000] d e f COD 70 IRQ NPL TCD CAF DEU CHE IRL GHA ZMB UGA NER 60 TJK BOL BLR NOR AUT MDA ARM COG IND NGA ZWE CIV TZA BEN EGY RWA AZE LAOTGO ZAF TUR LBN COM CMR MRT SEN MLI LBR MDG 50 LSO MOZ BEL LTU SWE CYP AUS DNK LUX UKRCHL KOR IRN BGD IDN MMR ALB BTN KEN HTI BFA GINSLE SWZCHN DZA MYS ECU MWI GMB ETH SLE ISR MLT SVN BGR POL NLD CZE MKD SVK HUN GRC PAN GEO AGO ETH MDG ISL CRI KAZ ITASRB GTM COL SDN DOM VNM PERPAK MOZ share [%] GIN MLI USA ESP EST GBR FRAMUSARG ROU JOR CAN PRY MAR LKAHND MWIGMB TGO LAO CAF 40 BIH MNG GEO AGOBFA UZB HTI LBR BEN NGA RUSFIN LVA JPNPRT CAN BIH MNE SRB TUNBWA MYS MEX LKA PHLDOM MAR SLV COL IDN LVA GBR PRT ESP FIN ISR FRA ROU MEX CRI KAZBRA URY CHL SLV PHL GAB NIC UZB ECU AZE MUS BWA KEN HRV ITA PAN PRY EST RUS CZE HRV AUS BEL ARG NAM MNGCHN 30 JPN HRV JORMKD MNE URY TUNBRA GRC NAM GABNIC HNDVNM ALB PER SEN PAKMMR BTNIND TZA CIV CMR TCD NER SVK HUN JOR GAB LBN URY TJK BRA GTM NAM SDN NIC MNG HNDPERMRT COM USAISL MLT BGR JPN LTU MNE NLD POL SVN UKR AUT KORTUR TUNDZA SWZRWA BOL RUS DNK USA LUX NORSWE FRA PRT ESTNLD POL ZAF ARGMEXSLV SDN IRN PRY BGD ARM COGUGA MKDEGY DZA SWZ IRN LSO UZBVNM BLRCYP IRL LUX SWE DNK ZAFBIH BWA MUS IRQ LSO BGR SVN FIN HUN ISL CYP GBR LVA CAN LTU UKR SVK ESP CZEKOR ITA ROU PHL GTM MAR LKACOL COM ZWE MDA ZMB NPL GRC BGD ALBPAK MDA GHA DEU CHEEGY 20 MLT AUT TUR PAN KAZ DOM MRT CHNECU MMRZWE ZMB BTN NOR MYS DEU BLRISRCHE AUS BEL IRL CRI SRB CHL LBN IDN GHA COD RWA GEO ARM AGO KEN MWI CMR GMB SEN AZECIV COG HTI LBR IND TJK 10 IRQ BOLBFA UGA NPL LAO COD TZA NGA GIN MOZ MLINER BEN ETH TGOMDG TCD SLE CAF 0 20 40 60 80 20 40 60 80 20 40 60 80 self employment rate [%] self employment rate [%] self employment rate [%] g h i COD 70 IRQ NPL CAF CHE DEU NER GHA ZMB TCD UGA 60 TJK BOL IRL BLR TZA ZWE COG IND MDA ARM AZE EGY RWA LAOTGO NOR AUT NGA CIV BEN CMR COM MRT SEN LBN MOZ LSO ETH TUR ZAF SWE UKR LTUMLT CYP CHL DNK BEL AUSISR BTNLBR MLI MDG HTI SLEMMR GIN BGD 50 CHN KOR LUXSVNBGR IDN BFA AGO ETH KENALB IRNGEOGRC GMB SWZ MYS SLE DZA ECU MWI POL NLD CRI KAZCZE ISL MOZCOLPAK VNM PAN GTM ITA SRB CAN MKD SVK HUN MDG GMB PER SDN DOM share [%] GIN MLI BFA BIH MUS ARG ROU USA ESP EST GBR FRA FIN RUS SRB LKA PRY MAR HND LAO SLV PHL TGO MWI MEX ROU LVA JOR 40 LBRNGA MNG UZB BEN HTIAGO GEO MYS BWA IDN LKA COLDOMTUN PHL MAR SLV MNE MEX JPN PRTLVA CAN NIC CAF UZB ECU GAB CHL CRI BRA KAZ URY GBR PRT ESP FIN ISR FRA BWA TZA BIH NAM PAK MUS TUN KEN BRA SEN URYJPN PRY GAB PAN GTM HRV ITA SVK HUN AZENAM MNG BOL ARG CHN CZE AUS BEL ISL MNE MLT EST HRV RUS 30 BTN NER VNM PERCMR NIC TCD MMR HND CIV COG IND GAB ALB MNE HRV GRC JOR MKD TJK NAM PER VNM SDN NIC HND MRT COM MNG URY BRA IRN JOR LBN TUR KOR SWZTUN RWAPOL DZA UKR AUT NLD SVN LTU JPNUSA BGR SDNUGA DNK POL NORSLV SWE NLD LUX ARGMEX BGD RUS HUN USA ARM FRA PRT SWZ DZA UZB BWA LSOBIHMUS IRL CYP LUX IRQ SWE DNK BLR 20 ZWE ZMB ZAF LKA COL KOR DOM MRT PRY PHL UKR MAR COM SVN CYP LTU AUT CRI GTM PAN KAZ BGR ISL NPL CZE MLT IRN EST ROUFIN GBR MDA LVA CAN ESP ITA SVK GHA ZWE CMR ZMB LSO PAKEGY ECUCHN BGD ALB MDA GRC MKD CHEZAFEGY DEU NOR TUR GHACHE COD DEU AUSSRB BTN MMR RWA AGO ARM MYS IDN IRL CHL BEL BLR ISR LBN GMBLBR AZE HTICIVKEN COG BFA SEN MWI IND GEO TJK 10 TZA COD NGABOL UGA GIN LAO IRQ NPL MOZ BEN NER MLI ETH MDG TCDTGO SLE 0 CAF 0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10 Owner-occupied housing share Owner-occupied housing share Owner-occupied housing share of value added [%] of value added [%] of value added [%] Supplementary Figure 4: Estimated capital shares. National capital shares of (a, d, g) public capital, (b, e, h) private (firm-owned) capital, and (c, f, i) household capital, over (a-c) GDP per capita, (d-f) self-employment rate, and (g-i) the owner-occupied share of value added, given by the product of the home ownership rate and the real-estate share of value added. κ p is obtained from ref12 , κ f and κ h are derived according to Methods. 5 a 8 b LICs 1011 7 LMICs UMICs HICs 6 1010 risk to well-being [$PPP] risk to well-being 5 [% of GDP] 4 109 3 2 108 1 107 0 0.0 0.5 1.0 1.5 2.0 2.5 107 108 109 1010 1011 risk to assets risk to assets [$PPP] [% of GDP] Supplementary Figure 5: Relationships between risk to well-being and risk to assets. (a) Risks expressed in percent of GDP. (b) Absolute risks on a log-log scale. Colors indicate low income (LICs), lower middle income (LMICs), upper middle income (UMICs), and high income (HICs) countries according to the World Bank country income group classification 2024. 6 R2=0.00 (n.s.) 80 LICs LMICs 70 UMICs HICs 60 socio-economic resilience [%] 50 40 30 20 0.0 0.5 1.0 1.5 2.0 2.5 risk to assets [% of GDP] Supplementary Figure 6: Relationship between socio-economic resilience and risk to assets. The dashed line indicates a linear least squares fit with 95% confidence intervals in grey. Marker colors indicate country income groups. 7 25 R2=0.28*** LICs LMICs UMICs HICs 20 15 Average recovery duration [yr] 10 5 0 20 30 40 50 60 70 80 Socio-economic resilience [%] Supplementary Figure 7: Relationship between average recovery duration and socio-economic resilience. Country-level recovery durations correspond to the data shown in Fig. 3 panel a, calculated as the average population- weighted recovery duration of affected households across all hazards. The dashed line indicates a linear least squares fit with 95% confidence intervals in grey. Asterisks *** indicate significance level p < 0.001. Marker colors indicate country income groups. 8 a 1.2 Earthquake b Flood Tsunami 2.00 1.0 Storm surge Wind Liquidity relative to reconstruction cost Liquidity relative to average liquidity 1.75 0.8 1.50 0.6 1.25 0.4 1.00 0.75 0.2 0.50 0.0 q1 q2 q3 q4 q5 q1 q2 q3 q4 q5 Household income quintile Household income quintile Supplementary Figure 8: Liquidity by household income quintile across all countries. (a) Household liquidity relative to the reconstruction cost of affected households by hazard. (b) Household liquidity relative to average national liquidity. 9 LICs LMICs UMICs 8 HICs Recovery duration [yr] 6 4 2 0 q1 q2 q3 q4 q5 Household income quintile Supplementary Figure 9: Recovery durations increase without liquidity. Median recovery duration by income quintile if households had no liquidity, by country income group (bar height) and across all countries (horizontal lines). Colors indicate country income group. 10 a Earthquake b Flood c Tsunami 0.7 LICs LMICs 0.6 UMICs HICs Household vulnerability 0.5 0.4 0.3 0.2 0.1 q1 q2 q3 q4 q5 q1 q2 q3 q4 q5 q1 q2 q3 q4 q5 Household income quintile Household income quintile Household income quintile d Storm surge e Wind 0.7 0.6 Household vulnerability 0.5 0.4 0.3 0.2 0.1 q1 q2 q3 q4 q5 q1 q2 q3 q4 q5 Household income quintile Household income quintile Supplementary Figure 10: Household vulnerability by income quintile and hazard. Vulnerability as the fraction of an affected household’s assets that is destroyed when affected by (a) earthquakes, (b) floods, (c) tsunamis, (d) storm surges, and (e) wind. Colors indicate the country income group. 11 Supplementary tables Supplementary Table 1: Model parameters used in the simulations. Variable Description Scope Eq. Value σ disaster aid payout factor national (17),(18) 0.4, 0.2 η elasticity of marginal consumption national (21) 1.5 ρ discount rate national (22) 0.06 12 Supplementary Table 2: Exogenous variables in the model and data sources. Variable Description Scope Eq. Source P population national (1) ref13 cq average household consumption by income quintile household (3) refs14,15 Π average productivity of capital national (3) ref16 div γq income diversification household (5) ref1,17 ∆K K fraction of assets destroyed national (8) ref18 V asset vulnerability national (8) ref2,3,10 vq asset vulnerability household (8) ref2,3,9,10 EBq exposure bias household (9) ref11 κh household share of capital national (15) refs4–8,12,19,20 Sqsav liquidity household (25) ref17 HPj ,h hazard protection national (38) refs21,22 13 Supplementary Table 3: Endogenous variables in the model. Variable Description Scope Eq. n population headcount national (1) f fraction of the population national (1) K capital national (2) lab /div iq labor / diversified income household (3) Γdiv q share of total diversified income that accrues to quintile q household (6) kq eff effective capital used to generate income household (2) div δtax tax rate on labor income national (7) ∆K capital loss national (8) ∆kq eff effective capital loss household (10) a,lab /div ∆iq labor / diversified income loss household (13) div /reco ∆cq diversified / reconstruction consumption loss household (13), (16) λq recovery rate household (14), (30) Sq aid disaster aid payout household (19) C aid disaster aid cost national (19) w utility household (21) (short −term /long −term ) ∆Wq (short-term / long-term) well-being loss household (21), (33), (34) sav +aid ∆cq spending of disaster aid and liquidity household (24) aid /reco,shared Cq disaster aid / public and firm asset reconstruction cost household (31) long −term cq long-term consumption loss household (32) ∆W / ∆W ref total / reference well-being loss national (35), (36) ∆C eq equivalent consumption loss national (36) Ψj socio-economic resilience national (40) 14 Supplementary Table 4: Datasets used. Country coverage indicates how many of the 132 countries in the simulations are contained in the original dataset. Missing data points are indicated in Supplementary Fig. 3 and Supplementary Tbl. 5. Dataset Source Country coverage GDP per capita ref15 full GNI per capita ref15 full Population ref13 full Income shares ref14 full Findex ref17 full ASPIRE ref1 85a Self-employment rate ref23 full Penn World Table ref16 full GIRI ref18 full FLOPROS (spatially aggregated using refs24,25 ) refs21,22 full Global Exposure Model ref10 full Global Monitoring Database ref9 77b Hyogo Framework for Action ref2 110c World Risk Poll ref3 106c Exposed people by hazard and poverty line ref11 116c IMF Investment and Capital Stock ref12 full Real-estate share of value added ref4 109c Real-estate share of capital to real-estate share of value added ratio refs19,20 24d Home ownership rate refs5–8 89c a: missing values imputed with regressions using refs13,23,26 b: per-quintile vulnerability for missing countries derived from ref10 only, assuming the poorest people inhabit the most vulnerable buildings c: missing data points imputed based on median or average data of available countries in the same region and / or country income group d: best-fit ratio is applied to all countries 15 Supplementary Table 5: Simulation results for all countries in the baseline scenario. ISO code WB Country GDP per capita risk to assets risk to well-being socio-economic Income group (2021 $PPP) (% GDP) loss (% GDP) resilience (%) AGOb,c,d LMICs 7245 0.04 0.15 25.24 ALBe UMICs 17976 0.53 1.47 35.87 ARGc,d UMICs 27105 0.15 0.26 58.53 ARMe UMICs 19230 0.29 0.71 40.99 AUSa,e HICs 60409 0.08 0.15 52.15 AUTa,e HICs 64336 0.16 0.24 64.46 AZEb,e UMICs 21262 0.18 0.42 43.05 BELa,e HICs 62876 0.09 0.16 55.79 BENd LMICs 3721 0.07 0.23 31.47 BFA LICs 2482 0.02 0.05 35.04 BGD LMICs 8242 0.45 1.24 36.43 BGRe UMICs 33112 0.15 0.27 55.70 BIHd,e UMICs 19829 0.34 0.59 57.31 BLRb,e UMICs 27718 0.25 0.36 69.52 BOLc LMICs 9844 0.21 0.43 49.81 BRA UMICs 19018 0.18 0.35 50.95 BTNb LMICs 14061 1.12 2.13 52.79 BWAa,b UMICs 18846 0.10 0.22 45.37 CAFa,b,c,d LICs 1135 0.40 1.10 36.58 CANa,e HICs 56687 0.05 0.09 56.99 CHEa,e HICs 81684 0.12 0.17 68.86 CHLe HICs 29491 0.71 1.23 57.68 CHNb,e UMICs 22138 0.25 0.42 59.25 CIVc,d LMICs 6485 0.04 0.14 29.51 CMRd LMICs 4871 0.09 0.36 26.29 CODb,c LICs 1456 0.33 1.27 25.73 COGa,c,d LMICs 6172 0.47 2.49 18.72 COL UMICs 18325 0.32 1.03 30.96 COMb,c,d,f LMICs 3499 0.33 1.25 26.73 CRIf UMICs 25990 0.18 0.30 60.93 CYPa,e HICs 52148 0.33 0.53 63.58 CZEa,e HICs 47452 0.08 0.13 61.50 DEUa,e HICs 63098 0.08 0.11 73.48 DNKa,e HICs 71390 0.02 0.03 70.45 DOM UMICs 23088 0.37 0.70 52.58 DZAa,c,d,e LMICs 15159 0.17 0.38 43.71 ECU UMICs 14472 0.89 1.92 46.45 EGYc LMICs 16691 0.04 0.06 64.11 ESPa,e HICs 47298 0.06 0.11 55.09 ESTa,e HICs 41669 0.03 0.05 63.74 ETHb,c LICs 2755 0.08 0.20 38.15 FINa,e HICs 56455 0.02 0.03 64.19 FRAa,e,f HICs 53969 0.06 0.09 61.23 GABc,d UMICs 18703 0.13 0.32 39.51 GBRa,e,f HICs 52589 0.06 0.13 50.45 GEO UMICs 22591 0.81 1.44 56.27 GHA LMICs 6796 0.03 0.10 35.71 16 ISO code WB Country GDP per capita risk to assets risk to well-being socio-economic Income group (2021 $PPP) (% GDP) loss (% GDP) resilience (%) GINc,d LMICs 3949 0.08 0.25 33.65 GMBb,d LICs 2932 0.03 0.07 38.98 GRCa,e HICs 36821 1.28 2.06 61.82 GTMb,d UMICs 12389 0.39 0.91 42.25 HNDc,d LMICs 6468 1.01 2.90 34.91 HRVe HICs 41100 0.27 0.52 51.26 HTIa,b,d,e LMICs 2956 2.36 7.63 30.97 HUNe HICs 40168 0.15 0.25 60.22 IDNa,f UMICs 13890 0.31 0.74 41.99 INDa,d,e LMICs 9160 0.10 0.35 30.18 IRLa,e HICs 115401 0.02 0.04 60.04 IRNd,f LMICs 15912 0.83 1.51 55.02 IRQc,d UMICs 12711 0.17 0.29 61.13 ISLa,e HICs 66880 0.28 0.38 72.09 ISRa,e HICs 48356 0.13 0.20 62.47 ITAa,e HICs 52589 0.39 0.77 51.07 JORd LMICs 9363 0.12 0.26 44.96 JPNa,e HICs 45949 0.29 0.45 64.79 KAZd,e UMICs 34703 0.17 0.38 44.30 KEN LMICs 5683 0.09 0.26 34.29 KORa,e HICs 49995 0.19 0.31 60.32 LAOa,d,f LMICs 8372 1.26 3.45 36.37 LBNa,d LMICs 11475 0.45 1.22 37.04 LBRb LICs 1617 0.27 0.88 31.02 LKAf LMICs 13030 0.28 0.64 43.76 LSOb LMICs 2596 0.32 0.61 51.68 LTUe HICs 46118 0.09 0.12 71.27 LUXa,b,e HICs 130373 0.04 0.07 57.75 LVAe HICs 38333 0.18 0.29 61.01 MARa LMICs 8869 0.23 0.77 29.69 MDA UMICs 15855 0.46 0.89 51.23 MDGa,b,c,d LICs 1643 0.16 0.48 32.30 MEXe UMICs 21874 0.16 0.39 41.12 MKDa,e UMICs 23324 0.34 0.63 53.90 MLIc,d LICs 2395 0.08 0.21 37.06 MLTa,e HICs 59548 0.06 0.15 40.79 MMRa,c LMICs 5364 1.21 2.92 41.33 MNEb,e UMICs 27343 0.27 0.51 53.17 MNG LMICs 16223 0.24 0.41 58.32 MOZa,d,f LICs 1512 0.13 0.43 29.61 MRTa,b,d LMICs 6259 0.14 0.38 35.80 MUS UMICs 26590 0.32 0.52 60.94 MWIb LICs 1648 0.11 0.26 41.52 MYSd,e,f UMICs 32812 0.12 0.18 64.71 NAM UMICs 10106 0.06 0.17 33.21 NERb,d LICs 1703 0.25 0.80 30.65 NGA LMICs 5593 0.09 0.28 32.36 NICc,d LMICs 7487 0.44 0.97 44.89 NLDa,e,f HICs 70610 0.05 0.09 55.95 17 ISO code WB Country GDP per capita risk to assets risk to well-being socio-economic Income group (2021 $PPP) (% GDP) loss (% GDP) resilience (%) NORa,e HICs 90160 0.02 0.03 76.31 NPL LMICs 4860 0.59 1.63 36.33 PAK LMICs 5439 0.29 0.73 39.72 PANc,d,f HICs 35864 0.22 0.47 46.23 PER UMICs 15294 0.28 0.76 37.05 PHL LMICs 9901 0.65 1.18 54.69 POLe HICs 43585 0.03 0.05 68.00 PRTa,e HICs 41498 0.06 0.10 58.67 PRYd UMICs 15783 0.19 0.51 36.49 ROU HICs 40267 0.21 0.41 51.54 RUSd,e,f UMICs 39753 0.15 0.33 45.50 RWAb LICs 3060 0.24 0.62 38.71 SDNa,b,c,d LICs 2469 0.59 1.63 36.23 SENd LMICs 4317 0.03 0.10 27.05 SLEc,d LICs 3034 0.04 0.12 34.91 SLVd,f UMICs 11404 0.39 0.71 55.27 SRBe UMICs 25718 0.65 1.34 48.68 SVKe HICs 39172 0.07 0.17 42.49 SVNa,e HICs 47825 0.41 0.57 71.37 SWEa,e HICs 62665 0.02 0.03 70.29 SWZb LMICs 10132 0.18 0.37 47.83 TCDb,d LICs 1679 0.10 0.28 37.08 TGOd LICs 2768 0.05 0.13 37.60 TJKa,d LMICs 4472 2.42 3.13 77.34 TUNa,c,d LMICs 12553 0.12 0.28 42.36 TURe UMICs 34252 0.57 1.03 55.28 TZAd,f LMICs 3621 0.20 0.81 24.24 UGA LICs 2791 0.11 0.48 23.86 UKRd,e LMICs 15885 1.66 2.63 62.98 URY HICs 31019 0.06 0.13 44.86 USAa,e HICs 74578 0.09 0.17 54.67 UZBa,d,e LMICs 10008 0.56 1.01 55.19 VNMf LMICs 13492 0.51 1.25 40.87 ZAFc UMICs 13690 0.17 0.31 52.66 ZMB LMICs 3673 0.22 1.23 17.79 ZWE LMICs 3442 0.06 0.18 35.84 Missing and inferred data points (cf. Supplementary Fig. 3): a: Fraction of income from social protection and transfers b: Availability of early warning systems c: Real estate share of value added d: Home ownership rate e: Dwelling materials microdata f: Number of exposed people by hazard type and poverty line 18 Supplementary Table 6: Regression results for disaster risks, resilience, and recovery time. Coefficients for six regression models are shown. Models (1–3), 4, 5, and 6 correspond to regressions shown in Fig. 2, Fig. 3, Supplementary Fig. 5, Supplementary Fig. 6. 1 2 3 4 5 6 Risk to assets Risk to well- Socio-economic Socio-economic Socio-economic Recovery [%GDP] being [%GDP] resilience [%] resilience [%] resilience [%] time [y] ln(GDPpc [PPP]) -0.0548 -0.2189** 8.7089*** (0.029) (0.067) (0.699) Gini index [%] -0.7680*** (0.141) Risk to assets 1.8746 [% GDP] (3.090) Socio-economic -0.1389*** resilience [%] (0.020) const. 0.8249** 2.7774*** -35.8634*** 75.7284*** 47.0135*** 11.0839*** (0.279) (0.643) (6.747) (5.266) (1.510) (0.979) R2 0.027 0.077 0.544 0.187 0.003 0.275 N 132 132 132 132 132 132 Note: ∗ p < 0.05; ∗∗ p < 0.01; ∗∗∗ p < 0.001 19 Supplementary Table 7: Regression results for imputed income share from social protection and transfers. Values are in percent. sp,pt sp,pt sp,pt sp,pt sp,pt γq =1 γq =2 γq =3 γq =4 γq =5 SOC 1.4817*** 1.0243** 0.8312* 1.0608*** 0.9639*** (0.406) (0.349) (0.335) (0.258) (0.177) UNE 0.7630** 0.7032** 0.5194* (0.258) (0.207) (0.199) REM 0.3741* 0.2400* (0.148) (0.120) IHICs 17.1771** 10.4837* 11.7009* 13.0351** (6.216) (5.061) (4.869) (4.657) IUMICs 8.4143* 8.1317** 7.2858* 10.1749*** 3.8510* (3.672) (2.949) (2.837) (2.734) (1.847) IFSY 11.5620* (4.866) IMNA -15.0030* -13.7580** -10.4954* -7.3198 (6.460) (5.155) (4.959) (4.225) IEAP 7.6582 9.8520** 6.7344* 7.3744** (3.863) (3.600) (3.044) (2.473) IECA 9.5125* 8.0773* (4.047) (3.885) ILAC -6.5397* (2.980) const. 5.5110* 3.3008 3.8654 4.3360* 5.1751** (2.700) (2.307) (2.216) (2.017) (1.644) R2 0.557 0.558 0.503 0.468 0.328 N 101 103 104 107 108 Note: ∗ p < 0.05; ∗∗ p < 0.01; ∗∗∗ p < 0.001 20 Supplementary Table 8: Mapping of GEM taxonomy codes to vulnerability classes. Each taxonomy string in the GEM database is mapped to a hazard-specific vulnerability class. The mapping is primarily done on the material of the lateral load-resisting system. Some mappings account in addition for the type of the lateral load-resisting system, as well as on the building height. The GEM taxonomy is described in ref27 . GEM taxonomy Earthquake Flood, storm surge, tsunami Wind lat. load sys. material lat. load sys. type height MUR fragile fragile fragile MUR+MO fragile fragile fragile MUR+ADO fragile fragile fragile MUR+ADO+MOC fragile fragile fragile MUR+ADO+MOM fragile fragile fragile MUR+STRUB fragile fragile fragile MUR+STRUB+MON fragile fragile fragile MUR+STRUB+MOM fragile fragile fragile MUR+STRUB+MOL fragile fragile fragile MUR+STRUB+MOC fragile fragile median MUR+STDRE fragile fragile fragile MUR+STDRE+MOM fragile fragile fragile MUR+STDRE+MOL fragile fragile fragile MUR+STDRE+MOC median fragile median MUR+CL fragile fragile fragile MUR+CLBRS fragile fragile fragile MUR+CLBRS+MOM fragile fragile fragile MUR+CLBRS+MOL fragile fragile fragile MUR+CLBRS+MOC median fragile median MUR+CLBRH fragile fragile fragile MUR+CB99+MOC median fragile fragile MUR+ST fragile fragile fragile MUR+ST+MOM fragile fragile fragile MUR+ST+MOL fragile fragile fragile MUR+ST+MOC median fragile median MUR+CLBLH fragile fragile fragile MUR+CB fragile fragile fragile MUR+CBS fragile fragile fragile MUR+CBH fragile fragile fragile MUR+CBR fragile fragile fragile E+ETO fragile fragile fragile EU fragile fragile fragile EU+ETC fragile fragile fragile EU+ETR fragile fragile fragile ER+ETR fragile fragile fragile MCF median fragile median MCF+CB median fragile median MCF+CBH median fragile median MCF+CBS median fragile median MCF+CBR median fragile median MCF+CL median fragile median MCF+CF median fragile median MCF+CLBRS median fragile median 21 GEM taxonomy Earthquake Flood, storm surge, tsunami Wind lat. load sys. material lat. load sys. type height MCF+CLBRH median fragile median MCF+CLBLH median fragile median MCF+S median fragile median MR median fragile median MR+CB median fragile median MR+CBR median fragile median MR+CBH median fragile median MR+CL median fragile median MR+STRUB+RCB+MOC median fragile median CR median median median CR+CIP LDUAL HBET:3,1 robust fragile robust CR+CIP LDUAL default robust median median CR+CIP LFINF HBET:3,1 median fragile robust CR+CIP LFINF default median median median CR+CIP LFM HBET:3,1 robust fragile robust CR+CIP LFM default robust median median CR+CIP default median median median CR+PCPS LFM HBET:3,1 robust fragile robust CR+PCPS LFM default robust median median CR+PCPS default robust median robust CR+PC LDUAL HBET:3,1 robust fragile robust CR+PC LDUAL default robust median median CR+PC LFINF HBET:3,1 robust fragile robust CR+PC LFINF default robust median median CR+PC LFM HBET:3,1 robust fragile robust CR+PC LFM default robust median median CR+PC default robust median median S robust median median S+SL robust median median SL robust median median S+SR robust median median SR robust median median S+SO robust median median SRC robust median median W median fragile fragile W+WWD fragile fragile fragile W+WBB median fragile fragile W+WO median fragile fragile W+WS median fragile fragile W+WLI LPB median fragile fragile W+WLI LWAL median fragile fragile W+WLI LFBR robust fragile fragile W+WLI LFM median fragile fragile W+WLI default median fragile fragile W+WHE median fragile fragile ME fragile fragile fragile ME+MEO fragile fragile fragile ME+MEIR fragile fragile fragile M+ADO fragile fragile fragile 22 GEM taxonomy Earthquake Flood, storm surge, tsunami Wind lat. load sys. material lat. load sys. type height M+ST fragile fragile fragile M+CB median median median W+S median median median MIX(MUR-STRUB-W) fragile fragile fragile MIX(MUR-STDRE-W) fragile fragile fragile MIX(MUR-W) fragile fragile fragile MIX(MUR-CR) median median fragile MIX(S-CR-PC) robust median robust MIX(S-W) median fragile median MIX(S-CR) robust median robust MIX(C-S) robust median robust MIX(C-W) median fragile median MIX(M-W) fragile fragile fragile MIX(M-ST) fragile fragile fragile MIX(W-M) fragile fragile fragile MIX(W-EU) fragile fragile fragile MIX(CR-W) median fragile median MIX(MR-W) median fragile median MIX median fragile median MATO median fragile median INF fragile fragile fragile UNK median fragile median S+S99+SC99 robust fragile median S+SL+SC99 robust fragile median MR+MUN99+MR99+MO99 median fragile median MUR+MUN99+MO99 fragile fragile fragile 23 References 1. World Bank. The atlas of social protection indicators of resilience and equity 2025. https://www.worldbank.org/ en/data/datatopics/aspire. [Dataset] Accessed 01 Apr 2025. 2. UN-ISDR. Hyogo framework for action 2005-2015: building the resilience of nations and communities to disasters in (2007). 3. Lloyd’s Register Foundation. 2021 World Risk Poll 2022. https://wrp.lrfoundation.org.uk/data-resources. [Dataset] Accessed 30 Jan 2024. 4. United Nations Statistics Division. Table 2.4 Value added by industries at constant prices (ISIC Rev. 4) 2025. http://data.un.org/Data.aspx?d=SNA&f=group_code%3a205. National Accounts Official Country Data [Dataset]. Accessed 13 Feb 2025. 5. United Nations Statistics Division. Households in housing units by type of housing unit, tenure of household and urban/rural residence 2025. http://data.un.org/Data.aspx?d=POP&f=tableCode%3a304. Demographic Statistics Database [Dataset]. Accessed 18 Feb 2025. 6. Eurostat. Distribution of population by tenure status, type of household and income group (ilc_lvho02) 2025. https://ec.europa.eu/eurostat/databrowser/view/ilc_lvho02/default/table. [Dataset] Accessed 13 Feb 2025. 7. OECD. OECD Affordable Housing Database 2025. https://www.oecd.org/content/oecd/en/data/datasets/ oecd-affordable-housing-database.html. [Dataset] Accessed 13 Feb 2025. 8. Centre for Affordable Housing Finance in Africa. 2024 Yearbook: Housing Finance in Africa http://housingfinanceafrica. org (Centre for Affordable Housing Finance in Africa, Johannesburg, South Africa, 2024). 9. World Bank. Global Monitoring Database (GMD) 2024. [Dataset] Accessed 28 Feb 2024. 10. Yepes-Estrada, C. et al. Global building exposure model for earthquake risk assessment. Earthquake Spectra 39, 2212–2235. https://github.com/gem/global_exposure_model (2023). [Dataset] Accessed 06 Jan 2025. 11. Doan, M. K. et al. Counting People Exposed to, Vulnerable to, or at High Risk From Climate Shocks—A Methodology. World Bank Policy Research Working Paper (2023). 12. IMF. Investment and Capital Stock Dataset (ICSD) 2022. https://data.imf.org/?sk=1ce8a55f-cfa7-4bc0- bce2-256ee65ac0e4. [Dataset] Accessed 13 Feb 2025. 13. World Bank. World Development Indicators 2025. https://doi.org/10.57966/6rwy-0b07. [Dataset] Accessed 01 Apr 2025. 14. World Bank. Poverty and Inequality Platform (version 20240326_2017) 2023. www.pip.worldbank.org. [Dataset] Accessed 01 Apr 2025. 15. World Bank. International Comparison Program 2025. https://www.worldbank.org/en/programs/icp. [Dataset] Accessed 01 Apr 2025. 16. Feenstra, R. C., Inklaar, R. & Timmer, M. P. The next generation of the Penn World Table. American economic review 105, 3150–3182 (2015). [Dataset] Accessed 12 Dec 2023. 17. Demirgüç-Kunt, A., Klapper, L., Singer, D. & Ansar, S. The Global Findex Database 2021: Financial inclusion, digital payments, and resilience in the age of COVID-19 (World Bank Publications, 2022). [Dataset] Accessed 18 Feb 2025. 24 18. CDRI. Global Infrastructure Resilience: Capturing the Resilience Dividend - A Biennial Report from the Coalition for Disaster Resilient Infrastructure 2023. https://giri.unepgrid.ch. 19. Eurostat. Capital stocks by industry (NACE Rev.2) and detailed asset type (nama_10_nfa_st) 2025. https : //ec.europa.eu/eurostat/databrowser/view/nama_10_nfa_st/table. [Dataset] Accessed 10 Feb 2025. 20. Eurostat. Gross value added and income by detailed industry (NACE Rev.2) (nama_10_a64) 2025. https : //ec.europa.eu/eurostat/databrowser/view/nama_10_a64/default/table. [Dataset] Accessed 03 Feb 2025. 21. Scussolini, P. et al. FLOPROS: an evolving global database of flood protection standards. Natural Hazards and Earth System Sciences 16, 1049–1061 (2016). [Dataset] Accessed 19 Jan 2024. 22. Tiggeloven, T. Benefit-cost analysis of adaptation objectives to coastal flooding at the global scale (Version 2) (Zenodo, 2020). https://doi.org/10.5281/zenodo.4275517. [Dataset] Accessed 19 Jan 2024. 23. International Labour Organization. ILO modelled estimates database, ILOSTAT 2025. https://ilostat.ilo.org/ data/. [Dataset] Accessed 01 Apr 2025. 24. Center For International Earth Science Information Network-CIESIN-Columbia University. Gridded Population of the World, Version 4 (GPWv4): Population Density Adjusted to Match 2015 Revision UN WPP Country Totals, Revision 11 (Version 4.11) Palisades, NY, 2018. https://doi.org/10.7927/H4F47M65. [Dataset] Accessed 15 Jan 2024. 25. GADM. GADM Admin 0 shapefiles (version 4.1) 2022. https://gadm.org. 26. International Labour Organization. World Social Protection Report 2024-26: Universal social protection for climate action and a just transition (2024). [Dataset] Accessed 25 Feb 2024. 27. Brzev, S. et al. GEM Building Taxonomy Version 2.0, GEM Technical Report 2013-02 V1.0.0 2013. [Dataset] Accessed 24 Jan 2025. 25