## wpiea2023002-print-pdf - Executive Summary

## Source details

**Canonical URL:** [wpiea2023002-print-pdf - Executive Summary](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023002-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023002-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023002-print-pdf.pdf.json)

---

### Introduction and research question
- Core question: When and how do natural disasters worsen within-country income inequality and how persistent are these effects?
- Approach:
  - Develop a conceptual framework outlining channels through which natural disasters may have distributional effects.
  - Empirically analyze when and which types of disasters affect inequality in advanced economies (AEs) and in emerging and developing economies (EMDEs).
  - Run a systematic search for plausible empirical associations using local projections with different disaster definitions and severity cutoffs.

### Empirical strategy and definitions
- Empirical method:
  - Local projections (up to four years ahead) are used to estimate dynamic responses of inequality to disaster events.
  - A systematic grid search over percentile-based disaster definitions and severity cutoffs is implemented to identify plausible associations.
- Key outcome measure:
  - Income inequality proxied by the market Gini (natural logarithm of the market Gini is the dependent variable).
- Severe disaster definitions:
  - Two severity measures: social costs (affected population including deaths, in percent of population) and economic damages (normalized by GDP, in percent of GDP).
  - Selected percentile cutoffs (aggregate, top x percent disasters — AVG Affected Population; AVG Damages/GDP):
    - >= 25: 1.0; 0.5
    - >= 20: 1.8; 0.8
    - >= 15: 3.3; 1.1
    - >= 10: 6.0; 2.0
    - >= 5: 13.6; 4.7
    - >= 1: 39.1; 25.3
  - Event-level, top x percent disasters (AVG Affected Population; AVG Damages/GDP):
    - >= 25: 0.1; 0.1
    - >= 20: 0.2; 0.2
    - >= 15: 0.5; (damages entry not listed)
    - >= 10: 1.0; 0.7
    - >= 5: 3.3; 1.9
    - >= 1: 22.8; 12.1
  - Bottom percent control examples (aggregate; event-level):
    - < 75: 0.01; 0.52 (aggregate); 0.1; 0.1 (event-level)
    - < 50: 0.001; 0.001 (aggregate); 0.01; 0.02 (event-level)
    - < 25: 0.0001; 0.0002 (aggregate); 0.000004; 0.00003 (event-level)

### Main findings on distributional effects
- General result:
  - Within-country income inequality (market Gini) tends to increase significantly following severe and repeated severe natural disasters; effects are persistent, lasting several years.
- Advanced Economies (AEs):
  - Inequality increases after severe disasters.
  - Severe disasters co-occurring with growth slowdowns show larger effects.
  - Multiple severe disasters in a year amplify inequality increases.
  - Earthquakes are identified as a significant disaster type associated with rising inequality.
  - Magnitudes (examples preserving source estimates):
    - Aggregate approach (top 15 percent): increase in market Gini of about 2.3 percent one year after the disaster and 3.3 percent two years after the disaster.
    - Event-based approach (>= 75th percentile): increases of about 1 percent two years after the shock, and 1.4 percent, 1.3 percent and 2 percent in the following three years.
    - Severe earthquakes (event-based, > 75th percentile thresholds): projected increases in the Gini of 2.7 percent after one year; 3.8 percent after two years; 4.3 percent after three years; 3.7 percent after four years.
    - Repeated severe disasters (event-level, above the 75th percentile): contemporaneous increase of 1.2 percent; increases of 2.4 percent, 3.5 percent, 2.5 percent, 2.6 percent in the four consecutive years.
    - Comparative magnitude note: estimated size of the increase in inequality from severe disasters is larger than the average percent increase in market Gini across AEs before and after the GFC (2007-2009) which was around 0.9 percent.
- Emerging Markets and Developing Economies (EMDEs):
  - Inequality increases when there are multiple severe disasters in a year and when severe disasters coincide with growth slowdowns.
  - Disaster types with significant effects: severe epidemics, droughts, and floods.
  - Magnitudes (examples preserving source estimates):
    - Severe disasters co-occurring with growth slowdowns: estimated peak inequality effect slightly above 1 percent three years after the shock.
    - Repeated severe disasters (>= twice in same country-year): increase in the Gini of 1.4 percent three years after the shock and 1.9 percent four years after the shock (EMDE results noted as not robust to varying the control group cutoff).
    - Heterogeneous event-based 90th percentile thresholds (damages > 0.7 percent of GDP or affected > 1 percent of population) yield: severe droughts 1.3 percent after three years; severe epidemics 0.6 percent after three years; severe floods 0.3 percent after one year.
    - Comparative magnitude note: average percent increase in the market Gini across EMDCs before and after the “taper tantrum” episode (2013-2015) was 0.6 percent.

### Severity, frequency, persistence
- Both individually large disasters and a series of smaller disasters can raise inequality.
- Effects of severe and repeated disasters are persistent, lasting several years (responses estimated up to four years ahead).
- Aggregate statistics (1990–2019; EM-DAT-based):
  - EM-DAT records 11,739 natural disasters globally between 1990 and 2019.
  - Annual counts: typically between 300–400 disasters per year, around 500 in 2009, and slightly below 300 per year in the early 1990s.
  - Incidence by country group: EMDEs experienced 8,707 disasters; advanced countries experienced 2,251 disasters.
  - AEs were hit on average by 63 disasters per country compared to 59 disasters per country in EMDEs.
- Top-10-percent most severe disasters (1990–2019 averages):
  - Average affected or dead population: 8.7 percent of a country’s total population, counting 5.9 million people on average.
  - Average economic damages: 2.3 percent of a country’s GDP, approximately 2.1 billion USD in damages.
- For all disasters combined (1990–2019 averages):
  - Average social cost: 1.8 percent of a country’s population, counting 1.7 million people on average.
  - Average economic damages: 0.2 percent of a country’s GDP on average, approximately 0.9 billion USD in damages.
- Frequency of top-10-percent severe disasters: generally between 30 and 40 per year over 1990–2019, with a peak above 60 in 2009.

### Heterogeneity and mechanisms (summary)
- Channels identified in the conceptual framework:
  A. Macroeconomic channels:
    - Negative supply shocks (damage to physical and human capital), negative demand shocks (wealth losses), worsened external balances, fiscal positions, and financial sector stress.
  B. Policy responses as adaptation channels:
    - Emergency responses, reconstruction, fiscal buffers, disaster insurance, contingency financing can mitigate distributional impacts; reconstruction can temporarily raise demand and employment.
  C. Macroeconomic vulnerabilities as intensification channels:
    - Limited fiscal space, high public debt, weak financial systems, lack of diversification, poor medical care, high informality —70 percent of the labor force in many EMDEs (World Bank, 2020)— widespread poverty, and food insecurity.
  D. Socioeconomic vulnerabilities as intensification channels:
    - Low-income, non-male, non-white, non-prime-age, and less-educated groups face higher exposure and lower recovery capacity, producing a “double hit.”
  E. Complex interactions:
    - Interactions (e.g., lack of access to credit, prolonged unemployment, psychological distress) can produce lasting inequality effects.
- Heterogeneous effects by disaster type and country group:
  - EMDEs: severe epidemics, droughts, and floods show significant distributional impacts.
  - AEs: earthquakes are highlighted as significant.
- Repeated events and growth interactions:
  - Repeated severe disasters and severe disasters that coincide with economic growth slowdowns are particularly associated with rising inequality.

### U.S. descriptive evidence on distributional channels
- Event analysis using U.S. Current Population Survey (CPS) data for 1990–2019 (state-month panels matched to EM-DAT events) finds:
  - Disproportionate adverse labor market effects on:
    - the less educated
    - women
    - racial minorities (Hispanic highlighted in illustrative results)
    - the young
  - Illustrative findings:
    - Larger adverse employment effects on less-educated workers (less than college) relative to college-educated workers.
    - Larger adverse employment effects on less-educated females compared to less-educated males.
    - Larger employment effects on less-educated Hispanic workers.
    - Larger adverse unemployment effects on youth compared to prime-age workers.
- U.S. event-list of major disasters (selected entries and source figures preserved):
  - Northridge earthquake, 1994 — CA — Damages as Percent of GDP: 20.58 — Percent of Affected population: 0.09
  - Tropical cyclone "Andrew", 1992 — FL, LA — Damages as Percent of GDP: 10.16 — Percent of Affected population: 1.40
  - Tropical cyclone "Katrina", 2005 — AL, LA, MS, GA, FL — Damages as Percent of GDP: 8.23 — Percent of Affected population: 1.32
  - Hurricane Harvey, 2017 — TX, LA — Damages as Percent of GDP: 5.00 — Percent of Affected population: 8.04
  - Storm Jonas (Snowzilla), 2016 — DC, NY, NJ, PA, MD, VA, TN, KY, DL, WV, GA, NC — Damages as Percent of GDP: 0.01 — Percent of Affected population: 94.81

### Robustness, sensitivity, and limitations
- Results sensitivity:
  - Findings are highly sensitive to the precise definition of a “severe” disaster (treatment group cutoff), the control group cutoff, and specification choices (treatment alone; coinciding with growth slowdowns; repeated disasters; by disaster type).
  - Grid search visualizations indicate many specifications are not robust; only subsets show significant positive associations within a four-year horizon.
- Data and measurement caveats:
  - SWIID Gini data have known data quality and comparability issues (e.g., imputations), likely more problematic for EMDEs.
  - EM-DAT records realized destruction (dead/affected, damages) not destructive potential; realized measures may be endogenous to institutions, policies, and income status.
  - Event-based approach better captures individually severe disasters that could have aggregate effects; highly localized disasters are less likely to show macroeconomic or inequality effects in this analysis.
- Causal inference:
  - The empirical approach is exploratory/descriptive and not a tool for causal identification.
- Estimation constraints:
  - Some equations could not be reliably estimated for AEs due to fewer severe disasters; restrictive cutoffs reduced observations and sometimes prevented estimation.

### Policy-relevant implications and recommendations
- Overarching concern:
  - The increasing incidence of natural disasters poses a palpable challenge to social cohesion given documented effects on within-country inequality and concentration of adverse labor market outcomes among vulnerable groups.
- Three broad policy areas:
  - Address pre-existing macro- and socioeconomic vulnerabilities:
    - Deepen insurance and financial markets.
    - Enhance social safety nets.
    - Invest in resilient infrastructure.
    - Use macro policies to build buffers and/or ensure access to external credit lines.
    - Design policies attentive to disadvantaged population groups who rely on them.
  - Provide prompt, adequate and effective emergency support:
    - Use policy space to provide economic relief.
    - Ensure disaster relief funds reach less affluent communities.
    - Coordinate disaster relief with donors for low-income and vulnerable countries with limited fiscal space.
  - Implement smart distributional policies to reduce poverty and inequality while supporting recovery:
    - Active labor market policies to re-skill workers in affected sectors and regions.
    - Targeted measures to support vulnerable groups likely to experience more prolonged effects.

*Executive Summary, IMF Working Paper "Why Some Don’t Belong - The Distributional Effects of Natural Disasters"*

### Executive Summary ......................................................................................................

### wpiea2023002-print-pdf - Executive Summary

### Introduction and research question
- Core question: When and how do natural disasters worsen within-country income inequality and how persistent are these effects?
- Approach:
  - Develop a conceptual framework outlining channels through which natural disasters may have distributional effects.
  - Empirically analyze when and which types of disasters affect inequality in advanced economies (AEs) and in emerging and developing economies (EMDEs).
  - Run a systematic search for plausible empirical associations using local projections with different disaster definitions and severity cutoffs.

### Empirical strategy and definitions
- Empirical method:
  - Local projections are used to estimate dynamic responses of inequality to disaster events.
  - A systematic grid search over disaster definitions and severity cutoffs is implemented to identify plausible associations.
- Key outcome measure:
  - Income inequality proxied by the market Gini.

### Main findings on distributional effects
- Severe and repeated severe disasters tend to increase income inequality, with effects lasting several years.
- In advanced economies (AEs):
  - Inequality increases after severe disasters.
  - Inequality increases when severe disasters are associated with growth slowdowns.
  - Inequality increases when there are multiple severe disasters in a year.
  - Earthquakes are identified as a significant disaster type associated with rising inequality.
- In emerging and developing economies (EMDEs):
  - Inequality increases when there are multiple severe disasters in a year.
  - Severe epidemics, droughts, and floods are significant events associated with increases in inequality.
- Disaster severity and frequency:
  - Both individually large disasters and a series of smaller disasters can raise inequality.
- Persistence:
  - The inequality-increasing effects of severe and repeated disasters are persistent, lasting several years.

### Heterogeneity and mechanisms (summary)
- Heterogeneous effects by disaster type and country group:
  - EMDEs: severe epidemics, droughts, and floods show significant distributional impacts.
  - AEs: earthquakes are highlighted as significant.
- Repeated events and growth interactions:
  - Repeated severe disasters and severe disasters that coincide with economic growth slowdowns are particularly associated with rising inequality.

### U.S. descriptive evidence on distributional channels
- Labor market impacts in the United States suggest adverse effects disproportionately fall on vulnerable groups:
  - The less educated
  - Women
  - Minorities
  - The young
- This descriptive evidence supports the view that adverse labor market channels are important pathways for post-disaster increases in inequality.

### Policy-relevant implication
- The increasing incidence of natural disasters poses a palpable challenge to social cohesion for policy makers going forward, given documented effects on within-country inequality and the concentration of adverse labor market outcomes among vulnerable groups.

*Executive Summary, IMF Working Paper "Why Some Don’t Belong - The Distributional Effects of Natural Disasters"*

### Introduction

### Introduction

### Background and motivation
- Within-country income inequality has been rising over several decades in many countries; the Covid-19 pandemic exacerbated these trends by hitting individuals employed in more contact-intensive sectors, the lower-skilled, women, and other vulnerable groups particularly hard.
- Other major epidemics in this century have had adverse macroeconomic effects and led to a significant and persistent increase in income inequality.
- Natural disasters and inequality can create a vicious cycle: pre-existing social and economic vulnerabilities amplify immediate effects of shocks because poor households tend to be more exposed to natural disasters and often have little means to prepare for or react to them.
- Macroeconomic vulnerabilities (e.g., high public debt, low fiscal buffers) can limit the ability to respond to severe disasters in many emerging markets and developing economies (EMDEs).
- Structural features that amplify shocks in EMDEs include poor health care, high informality —70 percent of the labor force in many EMDEs (World Bank, 2020)— widespread poverty, and food insecurity.

### Research question and approach
- Core question: When and how do natural disasters worsen within-country income inequality and how persistent are these effects?
- Steps taken:
  - Develop a conceptual framework outlining macroeconomic and socioeconomic channels through which natural disasters may have distributional effects and survey the literature.
  - Build a comprehensive dataset of natural disasters from the past four decades for 180 advanced and emerging/developing economies, expanding the EM-DAT database with major pandemic/epidemic events.
  - Empirically analyze when natural disasters affect within-country income inequality using local projections with different disaster definitions and severity cutoffs.
  - Complement macro analysis with a descriptive event analysis of heterogeneous labor market effects of major natural disasters across affected versus non-affected U.S. states using U.S. Current Population Survey and EM-DAT data.

### Main empirical findings (local projections)
- General result: Within-country income inequality, proxied by the market Gini, tends to increase significantly following severe and repeated severe natural disasters.
- Advanced Economies (AEs):
  - Effects are detectable for severe disasters alone.
  - Estimated peak inequality effect of severe disasters co-occurring with growth slowdowns is around 4 percent three years after the shock.
  - The estimated natural disaster effects on inequality are consistently larger in AEs than in EMs.
- Emerging Markets and Developing Economies (EMDEs):
  - Income inequality increases if the severe disaster coincides with a growth slowdown or is repeated at least twice within the year.
  - Severe epidemics, droughts, and floods tend to increase income inequality in EMDEs.
  - Estimated peak inequality effect of severe disasters co-occurring with growth slowdowns is slightly above 1 percent three years after the shock.
- Heterogeneity by disaster type and country income status:
  - Severe epidemics, droughts, and floods are more relevant for EMDEs; severe earthquakes are relevant in AEs.
  - Results are sensitive to the definition of a severe disaster (percentile/severity cutoff).
- Comparative magnitudes:
  - The estimated size of the increase in inequality from severe disasters is larger than the increase in the Gini observed before and after the GFC on average for AEs, and before and after the “taper tantrum” for EMDCs.
  - The average percent increase in the market Gini across AEs before and after GFC (2007-2009) was around 0.9 percent, while that across EMDCs before and after the “taper tantrum” episode (2013-2015) was 0.6 percent.

### U.S. states event analysis (descriptive)
- Results corroborate that natural disasters disproportionately affect disadvantaged groups:
  - Employment of less educated individuals, women, racial minorities, and young individuals tends to suffer in the aftermath of disasters.
- These findings align with evidence of gender and racial inequality in labor market impacts from the COVID-19 pandemic and distributional effects of major disasters on marginalized groups.

### Stylized facts on disasters (1990–2019; EM-DAT-based)
- EM-DAT records 11,739 natural disasters globally between 1990 and 2019.
- Annual counts: typically between 300–400 disasters per year, around 500 in 2009, and slightly below 300 per year in the early 1990s.
- Incidence by country group:
  - EMDEs experienced 8,707 disasters; advanced countries experienced 2,251 disasters.
  - AEs were hit on average by 63 disasters per country compared to 59 disasters per country in EMDEs.
- Most frequent disaster types globally: storms and floods.
- Geographical concentration: East Asia and Pacific, Sub-Saharan Africa, and Latin America and Caribbean show average incidence well above the global average (ranging between 837 and 3833 disasters per region).
- Country-level incidence: United States and China exhibited the highest incidence of disasters (above 750) over 1990–2019; India experienced 473 disasters, Mexico 208, Democratic Republic of Congo 129.

### Severity measures and top-10-percent severe disasters
- Severity proxied by: number of affected population (including death toll) normalized by national population and direct economic damages normalized by GDP.
- Top 10 percent most severe disasters (1990–2019):
  - Average affected or dead population: 8.7 percent of a country’s total population, counting 5.9 million people on average.
  - Average economic damages: 2.3 percent of a country’s GDP, approximately 2.1 billion USD in damages.
- For all disasters combined (1990–2019 averages):
  - Average social cost: 1.8 percent of a country’s population, counting 1.7 million people on average.
  - Average economic damages: 0.2 percent of a country’s GDP on average, approximately 0.9 billion USD in damages.
- Frequency of top-10-percent severe disasters: generally between 30 and 40 per year over 1990–2019, with a peak above 60 in 2009.
- Severe disasters included many storms, floods, and epidemics; many countries experienced one severe disaster per year on average, while some EMDEs (e.g., China, the Philippines, several Caribbean economies) experienced several severe disasters per year.

### Stylized macro outcomes around severe disasters
- Near-term macroeconomic effects of natural disasters are mixed and depend on severity, frequency, type, and policy responses.
- Some severe disasters (droughts, earthquakes, epidemics) were accompanied on average by a growth slowdown relative to a pre-disaster year; others (e.g., floods) were not.
- Comparing affected versus unaffected countries, poverty and the market Gini increased relatively more in countries experiencing top-10-percent severe natural disasters; increases were more pronounced in emerging market and low-income countries than in advanced economies.

### Conceptual framework and channels for distributional effects
- Identified channels:
  A. Macroeconomic channels
    - Severe disasters induce negative supply shocks (damage to physical and human capital) and can reduce production and labor supply.
    - Disasters can induce negative demand shocks via wealth losses and reduced household spending, lowering firm investment.
    - External balances, fiscal positions, and financial sector health can worsen, constraining recovery and investment.
    - Effects vary by severity, frequency, disaster type, economic structure, and country resilience.
  B. Policy responses as adaptation channels
    - Timely and effective emergency responses, reconstruction, fiscal buffers, disaster insurance, and contingency financing can mitigate distributional impacts; in some cases reconstruction can temporarily raise demand and employment.
  C. Macroeconomic vulnerabilities as intensification channels
    - Limited fiscal space, high public debt, weak financial systems, lack of diversification, poor medical care, high informality —70 percent of the labor force in EMDEs (World Bank, 2020)— widespread poverty, and food insecurity can intensify effects.
  D. Socioeconomic vulnerabilities as intensification channels
    - Low-income, non-male, non-white, non-prime-age, and less-educated groups face higher ex-ante exposure and lower ex-post capacity to recover—resulting in a “double hit.”
    - Women and racial minorities are more likely to be poor, more likely to die from disasters, and more likely to live in high-hazard areas.
  E. Complex interactions
    - Interactions between macro and socioeconomic vulnerabilities can produce lasting effects: lack of access to credit limits insurance, prolonged unemployment can depreciate human capital, and psychological distress can further limit labor market participation.

### Empirical strategy summary
- Main hypothesis: Occurrence of severe disasters bears a positive empirical association with income inequality.
- Methodology:
  - Grid search over percentile-based cutoffs of disaster severity (social costs and economic damages normalized by population and GDP) to capture potential non-linearities and measurement noise in defining “severe.”
  - Local projections (following Jorda, 2005; Furceri et al., 2021) to estimate impulse response functions up to four years ahead.
  - Dependent variable: natural logarithm of the market Gini for country i in year t.
  - Investigation of intensity factors: co-occurrence with growth slowdowns, repeated disasters within the same year, and heterogeneity by disaster type.

### Caveats and limitations
- Measurement error concerns:
  - Gini coefficients from the SWIID database have known data quality issues (e.g., data imputations), likely more widespread for EMDEs.
- Comparability issues:
  - Results for AEs and EMDEs are not directly comparable because severity metrics are based on realized destruction indicators; AEs may reflect disasters with higher realized destructive potential.
- Gini limitation:
  - Using Gini coefficients may obscure heterogeneities along the income distribution; strata such as gender, race, and socioeconomic status may offer additional insights (partly addressed via U.S. states analysis).
- Causal inference:
  - The empirical approach is exploratory and not a tool for causal inference.
- Estimation constraints:
  - Some equations could not be estimated reliably for AEs because fewer severe disasters occurred there; restrictive treatment/control cutoffs reduced observations and sometimes prevented estimation.

*IMF Working Paper — Introduction (wpiea2023002-print-pdf)*

### 1. Occurrence of

### 1. Occurrence of disasters and inequality - sign and significance

### By country income level: AEs vs EMDEs
- Advanced economies (AEs) and emerging market and developing economies (EMDEs) are analyzed separately to account for differing resilience, early warning systems, information structures, and fiscal capacity to rebuild.
- Country fixed effects and lags of the Gini and the treatment dummy are included to control for differences in income inequality levels and disaster history.
- The treatment dummy, D_{i,t}, equals one when a severe natural disaster occurs in country i in year t; if a severe disaster lasted longer than one year, the dummy is equal to one during the period it took place.
- The panel covers 180 countries over 1990-2019 and is estimated for horizons k=0,...,4 using local projections; robust standard errors are clustered at the country level.

### Constructing the Severe Disaster Dummy
- Two severity measures used: social costs (affected population including deaths, in percent of population) and economic damages (normalized by GDP, in percent of GDP).
- Averages:
  - Average social costs: 0.7 percent of the population.
  - Average economic damages: 0.2 percent of GDP.
  - For the top 10 percent most severe disasters: 2.1 percent of population and 12.7 percent of GDP.
- Two selection approaches to identify severe disasters:
  - Aggregate approach: sum losses from all disasters in a country-year and compare aggregate to severity thresholds.
  - Event-based approach: identify individual disasters exceeding a severity threshold, then sum losses of those individually severe events within a country-year.
- Rationale:
  - Aggregate approach captures cumulative losses from multiple moderate events in the same year.
  - Event-based approach captures individual disasters that are individually severe relative to country size.

- Severity cutoffs (selected entries from Table 1; values preserved exactly):
  - Aggregate, Top x Percent Disasters:
    - >= 25: AVG Affected Population (In percent) 1.0; AVG Damages/GDP (In percent) 0.5
    - >= 20: AVG Affected Population (In percent) 1.8; AVG Damages/GDP (In percent) 0.8
    - >= 15: AVG Affected Population (In percent) 3.3; AVG Damages/GDP (In percent) 1.1
    - >= 10: AVG Affected Population (In percent) 6.0; AVG Damages/GDP (In percent) 2.0
    - >= 5: AVG Affected Population (In percent) 13.6; AVG Damages/GDP (In percent) 4.7
    - >= 1: AVG Affected Population (In percent) 39.1; AVG Damages/GDP (In percent) 25.3
  - Event-level, Top x Percent Disasters:
    - >= 25: AVG Affected Population (In percent) 0.1; AVG Damages/GDP (In percent) 0.1
    - >= 20: AVG Affected Population (In percent) 0.2; AVG Damages/GDP (In percent) 0.2
    - >= 15: AVG Affected Population (In percent) 0.5
    - >= 10: AVG Affected Population (In percent) 1.0; AVG Damages/GDP (In Percent) 0.7
    - >= 5: AVG Affected Population (In percent) 3.3; AVG Damages/GDP (In Percent) 1.9
    - >= 1: AVG Affected Population (In percent) 22.8; AVG Damages/GDP (In Percent) 12.1
  - Bottom y Percent Disasters (control group cutoffs; selected entries):
    - < 75: AVG Affected Population (In percent) 0.01; AVG Damages/GDP (In percent) 0.52
    - < 50: AVG Affected Population (In percent) 0.001; AVG Damages/GDP (In percent) 0.001
    - < 25: AVG Affected Population (In percent) 0.0001; AVG Damages/GDP (In percent) 0.0002
  - Event-level bottom percent examples:
    - < 75: AVG Affected Population (In percent) 0.1; AVG Damages/GDP (In percent) 0.1
    - < 50: AVG Affected Population (In percent) 0.01; AVG Damages/GDP (In percent) 0.02
    - < 25: AVG Affected Population (In percent) 0.000004; AVG Damages/GDP (In percent) 0.00003

- Illustrative differences across approaches (Box 1):
  - Overlap of top-10 countries hit most often by severe disasters across the two approaches is about ½.
  - Example top-10 (Aggregate approach): Cambodia 8 years; China 7; Djibouti 7; Tajikistan 7; Bangladesh 6; Mongolia 6; Namibia 6; Niger 6; El Salvador 5; Haiti 5.
  - Example top-10 (Event-level approach): Cambodia 13 years; Lao P.D.R. 12; China 10; Djibouti 9; Malawi 9; Thailand 9; Ethiopia 8; Mongolia 8; Niger 8; Fiji 7.
  - Country-years with highest frequency (Aggregate approach examples):
    - China 2013: Severe Disasters 6; Disasters 42; Damages (Percent of GDP) 0.4; Affected (Percent of population) 2.0
    - China 2014: Severe Disasters 5; Disasters 41; Damages 0.3; Affected 4.8
    - China 2006: Severe Disasters 12; Disasters 36; Damages 0.5; Affected 6.8
    - China 2015: Severe Disasters 2; Disasters 36; Damages 0.2; Affected 0.3
    - Philippines 2011: Severe Disasters 14; Disasters 36; Damages 0.4; Affected 12.6
  - Country-years with highest frequency (Event-based approach examples):
    - Philippines 2011: Severe Disasters 14; Disasters 36; Damages 0.4; Affected 12.6
    - Philippines 2009: Severe Disasters 13; Disasters 26; Damages 0.5; Affected 14.8
    - China 2006: Severe Disasters 12; Disasters 36; Damages 0.5; Affected 6.8
    - Philippines 2006: Severe Disasters 11; Disasters 20; Damages 0.3; Affected 10.1
    - Philippines 2013: Severe Disasters 10; Disasters 14; Damages 4.7; Affected 26.7

### Econometric results: severe disasters and inequality
- Estimation approach:
  - Local projections with two lags of the dependent variable and the treatment dummy; country and year fixed effects; horizons k=0,...,4.
  - Grid search over percentile-based severity cutoffs to identify threshold above which disasters affect income inequality (market Gini).
  - Exercises include: (i) severe disasters only; (ii) severe disasters co-occurring with a growth slowdown; (iii) severe disasters repeated within a country-year (>= twice); (iv) severe disasters by disaster type.
- Caveats:
  - Grid search is not a tool for causal identification.
  - EM-DAT measures realized destruction (dead/affected, damages) not destructive potential; measurement error may be endogenous to institutions, policies, and income status.
  - Separate regressions for AEs and EMDEs and country fixed effects mitigate—but do not eliminate—concerns.

- Severe disasters (aggregate and event-based findings):
  - Aggregate approach (top 15 percent, i.e., damages > 1.1 percent of GDP or population affected > 3.3 percent): in AEs, a severe disaster leads to an increase in the market Gini of about 2.3 percent one year after the disaster and 3.3 percent two years after the disaster.
  - Event-based approach (>= 75th percentile, i.e., damages > 0.1 percent of GDP or population affected > 0.1 percent): in AEs, significant and persistent increases in the market Gini by about 1 percent two years after the shock, with increases of 1.4 percent, 1.3 percent and 2 percent in the following three years.
  - Interpretation: both individually severe events (event-based) and extremely large aggregate damages (aggregate) affect income inequality in AEs.
  - No significant increase in the market Gini following severe disasters is found in EMDEs for either approach when not conditioning on concurrent macroeconomic conditions.

### Severe disasters accompanied by growth slowdowns
- Definition: a growth slowdown is a year where the change in the growth rate of per capita GDP is less than or equal to 0.
- Treatment dummy equals one when a severe disaster coincides with a non-positive change in GDP per capita growth.
- Results:
  - For AEs, severe disasters co-occurring with a growth slowdown (event-based >= 75th percentile) lead to a stronger and more persistent increase in the market Gini than severe disasters alone:
    - Rise of 1.2 percent on impact, and rises of 2.7 percent, 3.8 percent, 4.3 percent and 3.7 percent in the four years thereafter.
  - For EMDEs, significant effects are found above the 95th percentile according to the aggregate approach (i.e., damages > 4.7 percent or humans affected > 13.6 percent). The text reports finding "peak effects in the range of" but the specific range is not provided in the excerpt.

### Repeated disasters and disaster types (overview)
- The analysis also examines severe disasters that occur repeatedly within the same country-year (>= twice) and heterogeneities by disaster type (earthquakes, droughts, floods, epidemics, storms).
- Conceptual expectation: droughts likely have large effects on harvests and households; wildfires or storms likely affect physical capital and households in affected regions.
- Main summary: severe disasters are more likely to affect within-country income inequality when accompanied by growth slowdowns or when they repeatedly hit affected countries, thereby complicating recovery. Substantial heterogeneity exists across disaster types (detailed type-specific results are discussed elsewhere in the chapter).

*Source: IMF Working Papers — Why Some Don’t Belong - The Distributional Effects of Natural Disasters.*

### 0.4 percent -  1.3 percent three years after the shock, depending on the severity of the event under

### wpiea2023002-print-pdf - 0.4 percent -  1.3 percent three years after the shock, depending on the severity of the event under

### Key findings on severe disasters and inequality
- Severe disasters can raise within-country income inequality, with estimated increases in the market Gini varying by country group, disaster definition, and specification.
- Effects differ between Advanced Economies (AEs) and Emerging Market and Developing Economies (EMDEs), and are often amplified when severe disasters coincide with growth slowdowns or occur repeatedly.

### Magnitude and timing of effects
- Aggregate finding range noted near the top of the excerpt: "0.4 percent -  1.3 percent three years after the shock, depending on the severity of the event under consideration."
- Repeated severe disasters (event-level approach, above the 75th percentile):
  - In AEs, a severe disaster repeated at least twice in the same country-year leads to:
    - contemporaneous increase in the Gini of 1.2 percent
    - increases of 2.4 percent, 3.5 percent, 2.5 percent, 2.6 percent in the four consecutive years (order preserved as in source)
  - In EMDEs, a severe disaster repeated at least twice in the same country-year leads to:
    - increase in the Gini of 1.4 percent three years after the shock
    - increase in the Gini of 1.9 percent four years after the shock
    - (note: EMDE results "are not robust to varying the control group cutoff")
- Heterogeneity by disaster type (event-based approach, severity cutoffs described):
  - AEs: Severe earthquakes (above the 75th percentile; i.e., with damages > 0.1 percent of GDP or affected 0.1 percent of population) are associated with projected increases in the Gini of:
    - 2.7 percent after one year
    - 3.8 percent after two years
    - 4.3 percent after three years
    - 3.7 percent after four years
  - EMDEs: Significant Gini increases found for:
    - severe droughts: 1.3 percent after three years
    - severe epidemics: 0.6 percent after three years
    - severe floods: 0.3 percent after one year
    - (these use 90th percentile severity cutoffs in the event-based approach; i.e., damages > 0.7 percent of GDP or affected > 1 percent of the population)

### Definitions and measurement
- Two severity approaches:
  - Aggregate approach: counts country-year disasters whose aggregate direct costs are above the 75th percentile.
  - Event-based approach: counts individually severe disasters with direct costs above the 75th percentile in that country-year.
- Rationale for 75th percentile cutoff: balance between selecting sufficiently severe disasters and retaining variation in the disaster frequency variable.
- Higher severity cutoffs (e.g., 90th percentile) used for some event-type analyses (droughts, epidemics, floods) with explicit thresholds: damages > 0.7 percent of GDP or affected > 1 percent of the population; for earthquakes example threshold cited as damages > 0.1 percent of GDP or affected 0.1 percent of population.

### Robustness and sensitivity
- Results are sensitive to:
  - the treatment group cutoff (severity definition)
  - the control group cutoff
  - specification choices (treatment dummy alone; coinciding with growth slowdowns; repeated disasters; by disaster type)
- The grid search (Figures 14–15) shows many specifications are not robust; purple fields indicate significant positive associations within a four-year forecast horizon, white fields indicate no evidence, gray fields not estimated.
- Most robust evidence found for:
  - severe disasters coinciding with growth slowdowns (EMDEs)
  - droughts (EMDEs)
  - repeated severe disasters (AEs)
  - earthquakes (AEs)

### Data caveats and limitations
- SWIID Gini data: known data quality and cross-country comparability issues (Jenkins, 2015; UN, 2018) that may introduce measurement error.
- EM-DAT lacks measures of destructive potential (e.g., earthquake strength); it records realized destruction, which may be endogenous to country institutions, policies, and income status and could bias comparisons between AEs and EMDEs.
- Event-based approach better captures individually severe disasters likely to have aggregate effects; highly localized disasters are less likely to show macroeconomic or inequality effects in this analysis.
- The analysis and grid search are descriptive/exploratory and "not a tool for causal inference"; results should be appropriately caveated.

### Summary conclusions (as stated in the source)
- (i) Severe natural disasters affect income inequality in AEs, with important heterogeneities across disaster types and impacts in EMDEs.
- (ii) Co-occurrence with a growth slowdown and repeated severe disasters amplify inequality effects.
- (iii) Severe earthquakes have significant adverse effects on inequality in AEs; severe droughts, floods, and epidemics are more important drivers of inequality effects in EMDEs.
- Further work with improved measures of destructive potential and more granular data is recommended to address comparability and causal identification issues.

### The United States: local, more granular evidence
- Event analysis using U.S. Current Population Survey (CPS) data for 1990–2020 shows disproportionate adverse labor market effects of major natural disasters on:
  - the less educated
  - women
  - racial minorities
  - the young
- CPS-based state-month panels match labor outcomes (employment, unemployment, average wage) by gender-race-age-education groups to major disaster occurrences to reveal within-state heterogeneities.
- The source notes these granular socioeconomic vulnerabilities are likely more pronounced in low-income and developing economies due to higher poverty rates, limited access to credit and insurance, higher informality, and residence in disaster-prone areas.

*Source: EM-DAT and Furceri and others, 2021, and IMF staff calculations.*

### Box 2. Heterogeneous Labor Market Effects of Natural Disasters in the United States (Continued)

### Box 2. Heterogeneous Labor Market Effects of Natural Disasters in the United States (Concluded)

### Event analysis: heterogeneous labor market effects (U.S. states)
- Method: plots the average change in labor market variables (before and after major disasters) across socioeconomic groups in affected states and compares them to the respective average changes in the same groups in unaffected states (Figure 2.1).
- Main illustrative results:
  - Disproportional adverse effects on employment of less-educated workers:
    - Larger adverse employment effects on less-educated workers (with less than college degree) compared to educated workers (college and above), after controlling for other changes in the employment gap between less-educated and educated workers in unaffected states (top-left chart).
  - Disproportional effects on less-educated female employment:
    - Larger adverse employment effects on less-educated females compared to less-educated males, after controlling for other changes in gender employment gap in unaffected states (top-right chart).
    - Consistent with evidence that women find it more difficult to recover because of lower wages and family care responsibilities (Cutter, 2001).
  - Disproportional effects on less-educated Hispanic employment:
    - Larger employment effects on less-educated Hispanic workers, after controlling for other changes in respective employment gaps in unaffected states (bottom-left chart).
  - Disproportional effects on youth unemployment:
    - Larger adverse unemployment effects on the young, compared to prime age workers, after controlling for changes in the unemployment gaps in unaffected states (bottom-right chart).
- Caveat: Event study results are purely illustrative and further empirical work is necessary to better control for other factors.

### Conclusions (summary of paper’s broader empirical strategy and findings)
- Empirical strategy:
  - Develops a conceptual framework on macroeconomic and socioeconomic channels through which severe natural disasters affect income inequality.
  - Empirically analyzes effects on within-country income inequality proxied by the market Gini for a sample of 180 countries over 1990–2019.
  - Uses a grid-search exercise iterating over the cutoff beyond which a disaster is deemed severe; local projections are used to assess effects for horizons up to t+4.
  - Database: expands EM-DAT with major pandemic/epidemic events from Furceri et al. (2021) and COVID-19 data from Our World in Data.
- Main findings from local projections:
  - Inequality (market Gini) tends to increase significantly after severe disasters in advanced economies (AEs).
  - Inequality increases if severe disasters are associated with growth slowdowns (non-positive change in the growth rate of GDP per capita) or if there are multiple severe natural disasters in a year in AEs and in emerging and developing economies (EMDEs).
  - Heterogeneous effects by disaster type:
    - In EMDEs, severe epidemics, droughts, and floods particularly increase income inequality.
    - In AEs, earthquakes have severe impacts on income inequality.
  - Effects vary based on severity definition, association with growth slowdowns, frequency within the year, disaster type, and country income status.
  - Results are highly sensitive to the precise definition of severe disaster; potential data quality issues (including SWIID Gini issues), obscuring factors (e.g., informality), policy response offsets, and localization/high-frequency visibility of effects are possible explanations.

### Policy considerations and recommendations
- Overarching point: Timely and effective policy responses can alter how natural disasters affect within-country inequality and can help avert second-round effects such as recessions.
- Three broad areas of action (design specifics require further policy research):
  - Address pre-existing macro- and socioeconomic vulnerabilities:
    - Deepen insurance and financial markets.
    - Enhance social safety nets.
    - Invest in resilient infrastructure.
    - Use macro policies to build buffers and/or ensure access to external credit lines.
    - Design these policies with attention to adequacy for disadvantaged population groups who crucially rely on them.
  - Provide prompt, adequate and effective emergency support:
    - Use policy space to provide economic relief.
    - Ensure disaster relief funds reach less affluent communities that often have harder time accessing information and resources.
    - For low-income and vulnerable countries with limited space for emergency relief, coordinating disaster relief with donors can help (as in the Covid-19 pandemic).
  - Implement smart distributional policies to reduce poverty and inequality while supporting recovery:
    - Active labor market policies to re-skill workers in affected sectors and regions and support necessary reallocation of labor.
    - Targeted measures to support vulnerable groups likely to experience more prolonged effects of natural disasters.

### Appendix III: U.S. states event analysis — data, aggregation, and sample selection
- Data sources:
  - Harmonized IPUMS CPS data and EM-DAT for events in the U.S. from 1990–2019.
- CPS details:
  - Individual-level, monthly frequency; includes demographic information, employment data, program participation.
  - Outgoing rotation group/earners study provides additional labor data on current work and income.
- Aggregation of variables into groups for event analysis (examples):
  - Gender: Female = 1, Male = 2.
  - Race: White = 1, Black = 2, Hispanic = 3, Other = 4.
  - Education: Less than high school = 1, high school = 2, no college degree = 3, college degree = 4, advanced degree = 5.
  - Age groups: Youth (15-24), Prime age (25-54), Older (55-64); coding: Older = 1, Prime age = 2, Youth = 3.
- Aggregation and panel construction:
  - Labor market variables aggregated at state-month-year level for gender-race-age-education subgroups: employment number, unemployment number, labor force number (sums using CPS weights), and weekly wage (simple average using earn weights).
  - Notation details: state-month-year combinations T = 17,748 (50 states + 1 district × 12 months × 29 years, 1990–2019); subgroup combinations S = 120 (2 gender × 4 races × 3 age groups × 5 education groups).
- Focus on most severe disasters:
  - Given many smaller disasters have mild human/social costs, analysis focuses on the top most severe disasters (top 5 severe disasters in terms of human costs and top 5 severe disasters in terms of economic costs).

### Table: United States major disasters over 1990–2019 (top events listed in source)
- Northridge earthquake, 1994 — States affected: CA — Damages as Percent of GDP: 20.58 — Percent of Affected population: 0.09
- Tropical cyclone "Andrew", 1992 — States affected: FL, LA — Damages as Percent of GDP: 10.16 — Percent of Affected population: 1.40
- Tropical cyclone "Katrina", 2005 — States affected: AL, LA, MS, GA, FL — Damages as Percent of GDP: 8.23 — Percent of Affected population: 1.32
- Parasitic disease, 1993 — States affected: ND, MN ... — Damages as Percent of GDP: 8.01 — Percent of Affected population: (not listed)
- Tornado outbreak, 2011 — States affected: MS — Damages as Percent of GDP: 5.39 — Percent of Affected population: 0.02
- Hurricane Harvey, 2017 — States affected: TX, LA — Damages as Percent of GDP: 5.00 — Percent of Affected population: 8.04
- Riverine flood, 2008 — States affected: OK, MN, WI, IA, IL, MS, SD, ND, NE, KS — Damages as Percent of GDP: 0.47 — Percent of Affected population: 23.59
- Tropical cyclone "Frances", 2004 — States affected: FL, NC, SC, OH — Damages as Percent of GDP: 0.71 — Percent of Affected population: 12.30
- Storm Jonas (Snowzilla), 2016 — States affected: DC, NY, NJ, PA, MD, VA, TN, KY, DL, WV, GA, NC — Damages as Percent of GDP: 0.01 — Percent of Affected population: 94.81

*Source: wpiea2023002-print-pdf - Box 2. Heterogeneous Labor Market Effects of Natural Disasters in the United States (Concluded)*

### References

### wpiea2023002-print-pdf - References

### Content summary
- This content unit is a references section comprising bibliographic citations to works on natural disasters, economic resilience, inequality, and the socioeconomic impacts of pandemics, including COVID-19.
- The references cite academic articles, IMF Working Papers, NBER Working Papers, Policy Research Working Papers, books, and reports from institutions such as the International Monetary Fund, World Bank, OECD, and research institutes.
- Themes reflected across the cited works include:
  - Natural disasters: economic impacts, distributional effects, resilience, insurance, and mitigation strategies.
  - Pandemics and COVID-19: labor market impacts, inequality, long-term scars, firm-level effects, and macroeconomic consequences.
  - Income and consumption dynamics: inequality measures, household responses to shocks, and long-run growth effects.
  - Regional and sectoral analyses: Caribbean, Pacific, Sub-Saharan Africa, Latin America, and firm- and county-level studies.
- The unit does not contain primary findings, projections, or policy recommendations itself; it lists sources referenced elsewhere in the parent document.

*Source: wpiea2023002-print-pdf - References*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023002-print-pdf.pdf_