## wpiea2023250-print-pdf

## Source details

**Canonical URL:** [wpiea2023250-print-pdf](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023250-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023250-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023250-print-pdf.pdf.json)

---

### Literature context
- Contributes to three strands:
  - Decomposition of cross-border bilateral migration out-flows into origin-country, destination-country, and global shocks using Amiti and Weinstein (2018).
  - Climate and cross-border migration: estimates impact of different climate events/risks on overall migration flows and on origin- and destination-induced portions.
  - Climate impacts on macroeconomic outcomes, extending the DIGNAD general equilibrium model to allow endogenous climate-driven migration and to quantify adaptation via public investment in resilient infrastructure.

### Data and sample
- Primary datasets:
  - United Nations Global Migration database: bilateral migration stocks in 5-year windows, 1990–2020; 238 countries and territories; gender breakdown; analysis focuses on aggregate origin flows using bilateral variation for decomposition.
  - EM-DAT (CRED): counts of climatological disasters (droughts, extreme temperature, floods, landslides, storms, wildfires); aggregated in 5-year windows.
- Complementary datasets:
  - Sautner et al. (2023): firm-level exposures (33 countries, 2000–2022); country aggregates via asset-weighted averages.
  - FAOSTAT: average temperature and land-use.
  - World Bank WDI: other economic outcomes and population.
- Sample composition and outcomes:
  - Maximum sample of 135 countries with additional controls: 44 LICs, 63 EMEs, 28 Advanced Economies.
  - Estimated impacts on: agricultural GDP and remittances as percent of GDP, employment as percent of total labor force, and the Gini index.
  - Controls include agricultural GDP growth, GDP per capita, and initial population.

### Decomposition method (Amiti and Weinstein (2018) applied to migration)
- Bilateral growth formulation:
  - g_{o,d,t} = O_{o,t} + D_{d,t} + G_{t} + ε_{o,d,t}
- Aggregate origin decomposition:
  - gOrigin_{o,t} ≡ (Σ_d L_{o,d,t} − L_{o,d,t−1}) / (Σ_d L_{o,d,t−1}) = O_{o,t} + ˜D_{o,t} + G_{t}
  - ˜D_{o,t} ≡ Σ_d φ_{o,d,t−1} D_{d,t}, where φ_{o,d,t−1} is initial share in destination d.
- Aggregate destination decomposition:
  - gDestination_{d,t} ≡ (Σ_o L_{o,d,t} − L_{o,d,t−1}) / (Σ_o L_{o,d,t−1}) = Σ_o θ_{o,d,t−1} O_{o,t} + D_{d,t} + G_{t}
- Identification notes:
  - Bilateral idiosyncratic shocks average out in aggregate origin/destination growth.
  - Origin and destination factors can be consistently estimated without modeling interaction terms; Appendix B.1 discusses refinements.

### Empirical specification (5-year panel)
- Main regression:
  - ∆M_o,t = β C_o,t + x_o,t Θ′ + δ_o + u_o,t
  - Dependent variables: change in migration outflow and origin-level factor as percent of initial population.
  - C_o,t: number of climatological disasters in EM-DAT in the 5-year window.
  - Controls: country fixed effects, initial population, previous population change, initial migrant stock (% of pop), previous migration outflows, initial origin factor, first and second lags of natural disasters, lagged average temperature difference w.r.t. 1950, initial altering land; robustness checks include GDP per capita and GDP growth.
- Heterogeneity by size:
  - ∆M_o,t = β_0 C_o,t + β_1 Size_o,t × C_o,t + x_o,t Θ′ + δ_o + u_o,t
  - Size measured by land size and initial population.

### Regional and aggregate decomposition findings
- Advanced economies:
  - Migration outflows mainly driven by global factors; origin domestic factors contributed to lower outward migration; net result: very limited outward migration.
- EMDEs:
  - Global, origin-country, and destination-country shocks all worked in the same direction, contributing to migration outflows.
  - Origin-country shocks gained importance over 2005–2015.

### Impact of climate disasters on migration — key regression estimates (Table 1, country-time 5-year windows, sample: 135 countries, observations ~657)
- Origin Factor (percent of initial population):
  - Column (1) Total Disasters coefficient 0.0313*** (0.0115); Observations 657; Number of Countries 135; Country FE YES; Controls YES; R2 0.42; R2 without Disasters 0.41.
  - Column (2) Total Disasters 0.0318*** (0.0115); Time Trend YES; R2 0.42; R2 without Disasters 0.41.
  - Column (3) Total Disasters 0.0264** (0.0123); Year Fixed Effects YES; R2 0.42; R2 without Disasters 0.42.
- Outflows (percent of initial population):
  - Column (4) Total Disasters 0.0159 (0.0110); Observations 664; R2 0.60; R2 without Disasters 0.60.
  - Column (5) Total Disasters 0.0180* (0.0108); Time Trend YES; R2 0.61; R2 without Disasters 0.61.
  - Column (6) Total Disasters 0.0180 (0.0119); Year Fixed Effects YES; R2 0.61; R2 without Disasters 0.61.
- Significance notation: *** p<0.01, ** p<0.05, * p<0.1.

### Main empirical findings on disasters → migration
- Climate disasters significantly increase the origin-shocks-driven portion of migration outflows.
- Effect on overall migration outflows is positive but smaller and not always statistically significant.
- Quantitative magnitudes and heterogeneity:
  - A one standard deviation increase in the number of disasters is associated with an increase in outward migration by about 1/2 percent of the initial total population in LICs and EMEs over a five-year period.
  - Translation: 3 extra climate disasters per year (one standard deviation) could translate into 1/2 percent of the overall population leaving the country of origin during the five-year window.
  - Impact large and statistically different from zero for LICs and EMEs; considerably smaller and not statistically significant for advanced economies.
- Large disasters (more than 100 deaths) generate larger and significant impacts on cross-border migration.

### Country size heterogeneity (Table 2)
- Selected estimates:
  - Column (1) Origin Factor: Disasters 0.287** (0.142); Disasters × Log Population -0.0140* (0.00767); Observations 640; Number of Countries 132; R2 0.420.
  - Column (2) Migration Outflows: Disasters 0.208 (0.127); Disasters × Log Population -0.0103 (0.00687); Observations 646; R2 0.630.
  - Column (3) Origin Factor: Disasters 0.182** (0.0923); Disasters × Log Land -0.0135* (0.00804); Observations 640; R2 0.419.
  - Column (4) Outflows: Disasters 0.124 (0.0827); Disasters × Log Land -0.00936 (0.00721); Observations 646; R2 0.629.
- Interpretation:
  - Interaction coefficients with Size (log population or log land) are negative → larger countries experience relatively smaller increases in outward migration in response to disasters.
  - Implies internal migration absorbs displacement in larger countries.

### Impact by disaster type
- Two categories:
  - Water-related: storms, landslides, floods.
  - Heat-related: wildfires, extreme temperature events, droughts.
- Findings:
  - Both water-related and heat-related disasters impact migration mainly in LICs and EMDEs; impacts for advanced economies not statistically different from zero.
  - Water-related disasters have relatively larger impact than heat-related disasters across country groups.

### Impact by sex
- Climate events increase outward migration for both males and females.
- Impact somewhat larger for males; gender difference statistically significant only for the LICs group.

### Impact of forward-looking climate risk measures (Sautner et al. (2023), 33 countries)
- Measures: overall climate risks, physical exposure, regulatory risks (basis points, size-weighted by assets).
- Table 3 (Origin Factor and Outflows; Observations 119; Number of Countries 30; R2 0.8):
  - Column (1) Origin Factor: Climate Risks 0.0713* (0.0361).
  - Column (2) Origin Factor: Climate Exposure 0.00323* (0.00163).
  - Column (3) Origin Factor: Regulatory Risks -0.00648 (0.0139).
  - Column (4) Outflows: Climate Risks 0.0406 (0.0265).
  - Column (5) Outflows: Climate Exposure 0.00234** (0.00106).
  - Column (6) Outflows: Regulatory Risks -0.00670 (0.0104).
- Interpretation:
  - Overall climate risks and physical exposure associated with higher origin-shock migration outflows; physical exposure significant for overall outflows.
  - Regulatory risks not statistically significant.

### Mediation, macroeconomic impacts, and identification (Appendix B.2 and Table 4)
- Structural equations:
  - y_o,t = α_c C_o,t + α_m ΔM_o,t + u^y_o,t
  - ΔM_o,t = β_c C_o,t + u^m_o,t
- Identification:
  - Destination factors used as instruments for migration; destination factors explain about 30 percent of cross-border migration; origin+destination+global explain about 70 percent.
- Table 4: Total impact of disasters and mediated effects (dependent variables: Agricultural GDP (% of initial GDP), Gini index, Employment as percent of 15+ population, Remittances as percent of GDP)
  - Total Impact of Disasters:
    - Agricultural GDP: -0.0559** (0.0244)
    - Gini: 0.0423 (0.0293)
    - Employment: -0.0401* (0.0219)
    - Remittances: 0.0051 (0.0173)
  - Mediated through Climate-Driven Migration:
    - Agricultural GDP: -0.0083*** (0.0026)
    - Gini: -0.023*** (0.0083)
    - Employment: -0.0010 (0.0009)
    - Remittances: 0.0202*** (0.0041)
  - Other channels (direct impact controlling for instrumented migration):
    - Agricultural GDP: -0.0476** (0.0218)
    - Gini: 0.0657** (0.0323)
    - Employment: -0.0400* (0.0211)
    - Remittances: -0.0151 (0.0214)
- Interpretations:
  - Disasters reduce agricultural output and employment (stat. sig.).
  - Climate-induced migration accounts for an important portion of climate’s total impact on Agricultural GDP, Gini, and Remittances.
  - Climate-induced migration accounts for about 15 percent of climate’s overall negative impact on agriculture (presented as agricultural GDP as percent of initial GDP).
  - Migration can mitigate increases in Gini because migrants tend to be from the bottom of the income distribution.

### Robustness: large-death disasters (Table 5)
- Disasters with more than 100 deaths:
  - Coefficients on origin/outflows: 0.0799* (0.0449), 0.0816* (0.0445), 0.0891** (0.0447), 0.0838** (0.0389), 0.0937** (0.0397), 0.0907** (0.0394) across specifications.
  - Observations: 657 or 664; Number of Countries: 135.
  - R-squared ranges: 0.449–0.644 across specs.
- Interpretation:
  - A one standard deviation increase in disasters with more than 100 deaths increases cross-border migration outflows by 0.23 percent of the initial population.

### Bounds to mediation via macroeconomic variables (Table 6)
- Identified sign, lower bound (percentage of total effect), and confidence intervals:
  - Agricultural GDP: Positive; Lower Bound 7.87; Confidence Interval (0.00 -14.32)
  - Employment: Positive; Lower Bound 6.34; Confidence Interval (0.02 -12.47)
  - Gini: Positive; Lower Bound 9.74; Confidence Interval (3.61 -15.87)
  - Remittances: Not identified; no bounds reported
- Aggregate bounds:
  - Lower Bound of Indirect Impacts: Positive 18
  - Upper Bound of Direct Impact: Positive 82
- Interpretation:
  - Identified macroeconomic outcomes could explain at least 18 percent of the impacts of climate on migration.

### Disasters in destination countries and inflows (Table 7)
- Destination disasters coefficients (select columns): -0.001140 (0.0110), 0.00439 (0.0137), -0.0136 (0.0159), -0.00490 (0.00659), -0.00511 (0.00736), -0.00536 (0.00700).
- Interaction terms (Disasters × Advanced, Disasters × EMEs) not statistically significant.
- Interpretation:
  - Point estimates are negative (disasters in destination reduce inflows) but not statistically significant.

### DIGNAD model with endogenous migration — calibration and counterfactuals
- Worker decision and migration mechanics:
  - Recursive value function V_j,t includes non-monetary location value z_j,t and moving costs m_{j,d,t}; locational shocks ε_{d,t} are Fréchet-distributed; variance governs elasticity of migration to wages.
- Budget constraints (hand-to-mouth workers):
  - C_origin_t = c_origin_t L_origin,t = W_t L_origin,t + Rem_t + T_t
  - C_abroad_t = W_abroad_t L_abroad,t − Rem_t
- Migration flows and stocks:
  - μ_migrate,t = Ω_migrate^{-1} exp(β E V_abroad,t − m_origin,abroad,t)^{1/ν}
  - Ω_migrate = exp(β E V_abroad,t+1 − m_origin,abroad,t)^{1/ν} + exp(β E V_origin,t+1)^{1/ν}
  - L_abroad,t = (1 − μ_repat,t) L_abroad,t−1 + μ_migrate,t L_origin,t−1
  - L_origin + L_abroad = L̄
- Calibration targets and baseline responses:
  - Steady state matches average cross-border migration stock to total population ratio.
  - Migration costs m calibrated to match migrant stock as percent of total population.
  - Baseline response matches three long-run impacts to a one-standard deviation increase in disasters:
    1. GDP responses: match about 1 percent decline of GDP in the long-term following a one-standard deviation shock in natural disasters (equivalent to agricultural GDP response).
    2. Climate-driven migration: match identified empirical response.
    3. Mediation through macroeconomic outcomes: match identified mediation moments.
  - Baseline assumption: about 20 percent of the migration response is driven by wage differentials.
  - Preference shocks and non-monetary benefits (z) discipline baseline responses; variance of preference shocks governs elasticity (see equation 15).
- CAPDR calibration:
  - Initial migrant stock: 4.2 percent of population for the average CAPDR country.
- Counterfactual adaptation experiment:
  - Baseline economy: no adaptation.
  - Counterfactual: 50 percent of public capital is resilient.
  - Shock: one-standard deviation increase in natural disasters, associated with a shock in real GDP of 1 percent in the baseline.
  - Key outcomes:
    - Climate-driven migration without resilient capital: about 0.5 percent of the initial population.
    - With 50 percent resilient public capital, climate-driven migration could be reduced by about 0.2 percent of the initial population.
  - Resilient infrastructure leads to lower GDP losses, more resilient private investment, and significant fiscal gains (moderation of fiscal deficit and public debt increases).

### Policy implications and recommendations
- Recommended measures:
  - Invest in climate-resilient public capital stock to reduce climate-driven cross-border migration and improve economic and fiscal outcomes.
  - Reinforce social safety nets and enhance climate-mitigating measures to contain adverse economic effects of climate shocks.
  - Tackle root causes of displacement amplified by climate change, such as poverty, inequality, and violence.
- Suggested future research:
  - Understand when climate-induced migration reinforces vs. offsets climate’s effect on macroeconomic outcomes.
  - Assess impacts of climate on internal migration and its interaction with cross-border migration and macroeconomic consequences.

### Methodological annex — identification and mediation framework (selected)
- Decomposition refinement:
  - g_{o,d,t} = O^{border}_{o,t} + O_{o,t} + D_{d,t} + G_t + ε_{o,d,t}, where O^{border}_{o,t} captures non-homotheticity during massive migration episodes (11 episodes identified).
- Structural system with feedbacks:
  - Y = β^Y_C C + β^Y_X X + ε^Y
  - X = β^X_C C + β^X_Y Y + ε^X
  - Total impact identified under exogeneity: ∆Y/∆C = m(β^Y_C + β^Y_X β^X_C), m = (1 − β^Y_X β^X_Y)^{-1}.
- Instrumental strategy:
  - Destination-level shocks as instruments for cross-border migration.
  - Relevance target: first-stage F-stat should be larger than 50; preliminary assessment gives F larger than 20.

### Selected exact model parameters and summary statistics (from appendices)
- Model/calibration parameters (selected exact entries from Appendix Table A2):
  - iz i y0.0600
  - δ z a 0.03
  - λ 5 0.5
  - iz a y000 γ z i 0.5
  - r^{⋆} 0.04
  - δ x 0.05
  - share b 0.28
  - share dc 0.27
  - share bstar 0.2
  - share remit 0.15
  - imp2gdp 0.48
  - δ zi 0.075
  - ω 2.41
- Summary statistics (Appendix Table A3, exact values):
  - Total Disasters: Obs 6751, Mean 0.16741, Std. dev. 16.843580127
  - Log GDP per capita: Obs 6849, Mean 9.014431, Std. dev. 1.207397, Min 6.051807, Max 11.78874
  - Average temperature: Obs 6980, Mean 0.957702, Std. dev. 0.428512, Min -0.096, Max 2.434
  - Log Migrant Stock: Obs 7141, Mean 2.78685, Std. dev. 1.769527, Min 7.103322, Max 16.58093
  - Log Population: Obs 7141, Mean 5.73919, Std. dev. 2.106014, Min 10.12206, Max 21.04525
- Selected regression coefficients (Appendix Table A4 and A5, exact entries):
  - Appendix Table A4:
    - Column (1) Origin Factor: Total Disasters coefficient 0.0313*** (0.0115)
    - Column (2) Outflows: Total Disasters coefficient 0.0159 (0.0110)
    - Lag Event coefficients: 4.299* (2.583) in (1) and 4.539*** (0.869) in (2)
    - Log Initial Stock Migrants: -1.324*** (0.467) in (1) and -1.157*** (0.300) in (2)
    - Observations: 6576 and 6576; R-squared: 0.42 and 0.60
  - Appendix Table A5 (instrumented migration impacts):
    - Instrumented Migration on Agricultural GDP: -0.499* (0.276)
    - Instrumented Migration on Gini: -2.401*** (0.567)
    - Instrumented Migration on Remittances: 0.796* (0.449)
    - Disasters on Agricultural GDP: -0.0476** (0.0218)
    - Disasters on Gini: 0.0657** (0.0323)

*Source: Section 5 and surrounding material from the provided content unit (wpiea2023250-print-pdf).*

### Section 5.  Section 6 quantifies the impact of adaptation policies using the DIGNAD

### Section 5.  Section 6 quantifies the impact of adaptation policies using the DIGNAD general equilibrium model.

### Literature Context
- The paper contributes to three interrelated strands of literature:
  - Understanding migration patterns via a novel and simple decomposition of cross-border bilateral migration out-flows into origin-country, destination-country, and global shocks, applying the decomposition method by Amiti and Weinstein (2018) to migration.
  - Climate and cross-border migration literature through estimation of the impact of different types of climate events and risks on overall migration flows and on origin-country and destination-country-induced portions of overall migration flows (references include Piguet and Laczko, 2014; Black et al., 2011; Coniglio and Pesce, 2015; Burzyński et al., 2021; International Organization for Migration, 2022; Kaczan and Orgill-Meyer, 2020).
  - Climate impacts on macroeconomic outcomes, growth, productivity and reallocation, extending this strand by extracting the impact of climate-induced migration on macroeconomic outcomes and extending the DIGNAD general equilibrium model to allow endogenous climate-driven migration and to quantify the effect of climate adaptation policies via public investment in resilient infrastructure (references include Bustos et al., 2021; International Monetary Fund, 2021; Fernández Corugedo et al., 2023a; Fernández Corugedo et al., 2023b).

### Data
- Primary datasets:
  - United Nations Global Migration database: bilateral migration stocks in 5-year windows for the period between 1990 to 2020; data available for 238 countries and territories; breakdown by gender; paper focuses on aggregate cross-border migration at origin but uses bilateral variation for decomposition.
  - EM-DAT (CRED): number of climatological disasters including droughts, extreme temperature, floods, landslides, storms, and wildfires; disasters aggregated by 5-year windows to match migration frequency.
- Complementary datasets:
  - Sautner et al. (2023): firms’ exposures to physical and regulatory risks of climate change; country aggregates computed using asset-weighted averages; data available for 33 countries and between 2000 to 2022.
  - FAOSTAT: average temperature data and land-use to control for initial exposures to climate.
  - World Bank WDI: other economic outcomes and population.
- Sample and outcomes:
  - Sample limited to a maximum of 135 countries for which additional control variables are available.
  - Sample composition: 44 LICs, 63 EMEs, and 28 Advanced Economies.
  - Estimated impacts on: agricultural GDP and remittances as percent of GDP, employment as percent of total labor force, and the Gini index.
  - Controls include agricultural GDP growth, GDP per capita, and initial population.

### Decomposition Method
- Model setup:
  - Bilateral migration stock notation: L_{o,d,t} is stock of migrants from origin o in destination d at time t.
  - Bilateral migration outflow equation:
    - L_{o,d,t} − L_{o,d,t−1} = [O_{o,t} + D_{d,t} + G_{t} + ε_{o,d,t}] L_{o,d,t−1}
  - Growth rate formulation:
    - g_{o,d,t} = O_{o,t} + D_{d,t} + G_{t} + ε_{o,d,t}
  - Interpretation:
    - O_{o,t} (origin factor): push factors or outward migration market access from origin (e.g., productivity shocks, wages, amenities); climate-related shocks likely affect migration via O_{o,t}.
    - D_{d,t} (destination factor): pull factors or inward migration access into destination.
    - G_{t} (global factor): global trends affecting migration including openness, global immigration policies, and common shocks.
- Decomposition approach:
  - Applies Amiti and Weinstein (2018) decomposition to bilateral migration growth rates, producing an exact decomposition of aggregate cross-border migration flows at origin and destination.
  - Aggregate origin growth decomposition:
    - gOrigin_{o,t} ≡ (Σ_d L_{o,d,t} − L_{o,d,t−1}) / (Σ_d L_{o,d,t−1}) = O_{o,t} + Σ_d φ_{o,d,t−1} D_{d,t} + G_{t}
    - Defines ˜D_{o,t} ≡ Σ_d φ_{o,d,t−1} D_{d,t}, where φ_{o,d,t−1} is initial share of migrants in destination d.
    - Alternate expression: gOrigin_{o,t} = O_{o,t} + ˜D_{o,t} + G_{t}
  - Aggregate destination growth decomposition:
    - gDestination_{d,t} ≡ (Σ_o L_{o,d,t} − L_{o,d,t−1}) / (Σ_o L_{o,d,t−1}) = Σ_o θ_{o,d,t−1} O_{o,t} + D_{d,t} + G_{t}
    - θ_{o,d,t−1} is initial share of migrants from origin o.
  - Identification notes:
    - The definition of origin-destination interaction terms implies that, when aggregated, bilateral idiosyncratic shocks do not explain on average the growth in cross-border migration stocks at origin and destination levels.
    - While origin-destination specific shocks are useful for acceleration in bilateral migration, origin and destination factors can be estimated consistently without modeling interaction terms.
    - Appendix B.1 discusses identification assumptions and refinements when interaction terms are worrisome; Appendix B.1 in Amiti and Weinstein (2018) provides further detail.

### Methodology (estimation specifics)
- Empirical model:
  - Starts from a general spatial model of migration that exhibits network effects, can accommodate utility-maximizing migration, migration costs, and migration frictions typical of quantitative equilibrium models.
  - Does not impose functional forms or additional assumptions on the structure of relevant markets; assumes bilateral migration incentives summarized by origin, destination, and global factors plus bilateral idiosyncratic shocks.
- Novelty:
  - First application of the Amiti and Weinstein (2018) decomposition to cross-border migration, allowing assignment of variation in bilateral growth rates to origin, destination, and global factors.

### Regional and Global Patterns (decomposition findings)
- Advanced economies:
  - Migration outflows have been mainly driven by global factors.
  - Idiosyncratic domestic factors (origin bars) have contributed to lower outward migration, likely reflecting relatively better domestic economic conditions.
  - Combined, these forces have been roughly balanced over past decades, resulting in very limited outward migration from advanced economies.
- Emerging and developing economies (EMDEs):
  - Global, origin-country, and destination-country shocks have all worked in the same direction, contributing to migration outflows from EMDEs.
  - Origin-country shocks gained importance over 2005-2015, likely reflecting challenging conditions in some EMDEs that pushed more migrants out of their countries of origin.

*Italic: Source: Section 5 and surrounding material from the provided content unit.*

### 5.1    Impact of Climate on Cross-Border Migration

### 5.1    Impact of Climate on Cross-Border Migration

### 5.1.1 Empirical Setting
- Main specification (unit: country of origin o and period t, measured in 5-year windows):
  - ∆M_o,t = β C_o,t + x_o,t Θ′ + δ_o + u_o,t
- Dependent variables ∆M_o,t are the change in migration outflow and the origin-level factor as percent of initial population.
- Origin-level factor identified from growth in bilateral claims; dependent variable constructed as O_o,t × M_o,t−1 / P_o,t−1 where O_o,t is the origin-factor, M_o,t−1 is the initial migrant stock, and P_o,t−1 is the initial population.
- Controls included in regressions:
  - fixed effects for countries δ_o
  - initial population, previous population change, initial migrant stock as percent of population, previous migration outflows, initial origin factor
  - first and second lags of natural disasters
  - lagged average temperature difference with respect to 1950 levels
  - initial altering land
  - robustness checks: initial GDP per capita, GDP growth, agricultural GDP growth; time-trends and year-fixed effects
- Main explanatory variable C_o,t: frequency of natural disasters measured as the number of climatological disasters in EM-DAT in the 5-year window.
  - Coefficient of interest β measures the change in migration outflows due to natural disasters (OLS intended to capture total impact, including mediated economic outcomes).
- Heterogeneity by country size assessed with:
  - ∆M_o,t = β_0 C_o,t + β_1 Size_o,t × C_o,t + x_o,t Θ′ + δ_o + u_o,t
  - Size measured by land size and initial population; expectation: β_1 negative (country size mitigates climate impact on cross-border migration).

### 5.1.2 Results — Impact of Climate Disasters on Cross-Border Migration
- Sample and estimation:
  - Panel of 135 countries; observation level country-time with 5-year windows from 1990 to 2020; sample selected based on data availability since 1995.
- Table 1 key estimates (Outflows and Origin Factor as percent of Initial population):
  - Column (1) Origin Factor: Total Disasters coefficient 0.0313*** (standard error 0.0115); Observations 657; Number of Countries 135; Country Fixed Effects YES; Controls YES; Time Trend NO; Year Fixed Effects NO; R2 0.42; R2 without Disasters 0.41.
  - Column (2) Origin Factor: Total Disasters 0.0318*** (0.0115); Time Trend YES; Year FE NO; R2 0.42; R2 without Disasters 0.41.
  - Column (3) Origin Factor: Total Disasters 0.0264** (0.0123); Year Fixed Effects YES; Time Trend - ; R2 0.42; R2 without Disasters 0.42.
  - Column (4) Outflows: Total Disasters 0.0159 (0.0110); Observations 664; R2 0.60; R2 without Disasters 0.60.
  - Column (5) Outflows: Total Disasters 0.0180* (0.0108); Time Trend YES; R2 0.61; R2 without Disasters 0.61.
  - Column (6) Outflows: Total Disasters 0.0180 (0.0119); Year Fixed Effects YES; R2 0.61; R2 without Disasters 0.61.
  - Significance notation: *** p<0.01, ** p<0.05, * p<0.1.
- Main empirical findings:
  - Climate disasters have a strong and statistically significant impact on the origin-shocks-driven portion of migration outflows (higher disasters → higher migration out of affected countries).
  - Effect on overall migration outflows remains positive but is smaller and not always statistically significant.
  - Interpretation: climate disasters affecting migrants’ origin countries are primarily associated with the portion of migration assigned to origin-country shocks.
- Quantitative magnitude and heterogeneity by country group (Figure 2):
  - A one standard deviation increase in the number of disasters is associated with an increase in outward migration by about 1/2 percent of the initial total population in LICs and EMEs over a five-year period.
  - Translation: 3 extra climate disasters per year (one standard deviation) could translate into 1/2 percent of the overall population leaving the country of origin during the five-year window.
  - Impact is particularly large and statistically different from zero for Low Income Countries (LICs) and Emerging Market Economies (EMEs); impact is considerably smaller and not statistically significant for advanced economies.
  - Robustness: large disasters (more than 100 deaths) generate significant and larger impacts on cross-border migration compared with the baseline estimates.
- Country size heterogeneity (Table 2 and Figure 3):
  - Table 2 key estimates:
    - Column (1) Origin Factor: Disasters 0.287** (0.142); Disasters × Log Population -0.0140* (0.00767); Observations 640; Number of Countries 132; Country FE YES; R-squared 0.420.
    - Column (2) Migration Outflows: Disasters 0.208 (0.127); Disasters × Log Population -0.0103 (0.00687); Observations 646; R-squared 0.630.
    - Column (3) Origin Factor: Disasters 0.182** (0.0923); Disasters × Log Land -0.0135* (0.00804); Observations 640; R-squared 0.419.
    - Column (4) Outflows: Disasters 0.124 (0.0827); Disasters × Log Land -0.00936 (0.00721); Observations 646; R-squared 0.629.
    - Significance notation: *** p<0.01, ** p<0.05, * p<0.1.
  - Findings:
    - More frequent climate disasters associated with larger migration outflows, especially for the portion attributed to origin-country shocks.
    - Interaction coefficients with Size (log population or log land) are negative, implying larger countries experience relatively smaller increases in outward migration in response to climate disasters.
    - Interpretation: larger economies (by population or land area) can absorb displacement via internal migration, mitigating cross-border migration responses.
  - Conclusion: country size mitigates the climate-to-cross-border-migration channel.

### Impact by Type of Disaster
- Two disaster types defined:
  - Water-related: storms, landslides, floods.
  - Heat-related: wildfires, extreme temperature events, droughts.
- Findings (Figure 4):
  - Both water-related and heat-related disasters have impacts that are especially important for LICs and EMDEs; impacts for advanced economies are not statistically different from zero.
  - The impact of water-related disasters appears relatively larger than heat-related disasters for each country group.

### Impact by Sex
- Findings (Figure 5):
  - Climate events increase outward migration of both males and females.
  - Impact is somewhat larger for males.
  - The gender difference is statistically significant only for the LICs group.

### Impact of Climate Risks (Forward-Looking Measures) on Cross-Border Migration
- Climate risk measures aggregated from firm-level data in Sautner et al. (2023) (dataset covers 33 countries; mainly advanced economies and few emerging markets).
- Measures considered: (i) overall climate; (ii) physical exposure; (iii) regulatory risks.
- Table 3 key estimates (Outflows and Origin Factor as percent of Initial population):
  - Column (1) Origin Factor: Climate Risks 0.0713* (0.0361); Observations 119; Number of Countries 30; Country Fixed Effects YES; Controls YES; R2 0.8.
  - Column (2) Origin Factor: Climate Exposure 0.00323* (0.00163); Observations 119; R2 0.8.
  - Column (3) Origin Factor: Regulatory Risks -0.00648 (0.0139); Observations 119; R2 0.8.
  - Column (4) Outflows: Climate Risks 0.0406 (0.0265); Observations 119; R2 0.8.
  - Column (5) Outflows: Climate Exposure 0.00234** (0.00106); Observations 119; R2 0.8.
  - Column (6) Outflows: Regulatory Risks -0.00670 (0.0104); Observations 119; R2 0.8.
  - Risks and exposures measured in basis points; aggregated using size-weighted average using total assets.
  - Clustered standard errors in parenthesis. Significance notation: *** p<0.01, ** p<0.05, * p<0.1.
- Findings:
  - Overall climate risks and physical exposure risks are associated with higher origin-shock migration outflows; both statistically significant.
  - For overall migration outflows, physical exposure risks remain statistically significant; overall climate risks decrease by about half and lose (marginal) statistical significance.
  - Regulatory risks are not statistically significant.
- Interpretation:
  - Anticipation of climate impacts (as captured by firms’ perceptions and assessments) is associated with higher migration outflows, suggesting forward-looking risk perceptions matter in addition to observed disasters.
  - Lack of significance of regulatory risks may reflect ambiguous effects: higher perceived regulatory risk could correlate with jurisdictions that either emit migrants (due to weak regulation) or receive migrants (due to swift regulatory adjustment).

*Source: IMF working paper section 5.1 "Impact of Climate on Cross-Border Migration" from wpiea2023250-print-pdf*

### Appendix B.2 shows how results are robust to the inclusion of additional control variables.

### wpiea2023250-print-pdf - Appendix B.2 shows how results are robust to the inclusion of additional control variables.

### Identification and theoretical setup
- Structural equations:
  - y_o,t = α_c C_o,t + α_m ΔM_o,t + u^y_o,t
  - ΔM_o,t = β_c C_o,t + u^m_o,t
- Identification assumptions and instrument:
  - Assume C ⟂ u^m, u^y but E(u^y u^m) ≠ 0.
  - There exists Z such that E(Z ΔM) ≠ 0 and E(Z u^y) = 0.
- Key identified parameters:
  - Total impact of natural disasters on economic outcomes:
    - ∂y/∂C = α_c + α_m β_c (equation (9))
    - Identified via ˜β = E(C y) / E(C′ C) (equation (10))
  - Mediation effect of climate-driven migration:
    - ∂y/∂ΔM ∂ΔM/∂C = α_m β_c (equation (11))
    - Identified using the moment conditions:
      - E((y − α_c C − α_m ΔM) C) = 0
      - E((y − α_c C − α_m ΔM) Z) = 0
    - Interpretation: 2SLS using Z as instrument for migration when regressing y on C and ΔM identifies α_c, hence the mediation effect is identified.
- Extensions and interpretation:
  - Proposition 1 shows how to identify mediation effects when an instrument for migration is available; can be extended to allow feedback from economic outcomes to migration. Appendix B.2 provides proofs and further details.

### Empirical strategy and instruments
- Destination factors:
  - Destination factors are used as instruments for cross-border migration because they are by construction orthogonal to origin-level factors and plausibly exogenous to unobserved origin-level variation.
  - Destination factors explain about 30 percent of cross-border migration.
  - When including origin-country and global factors, the authors can explain 70 percent of the variation of cross-border migration.
- Controls used in regressions:
  - Initial population and previous population change, initial migrant stock, 1st and 2nd lag of natural disasters, lagged average temperature difference with respect to 1950 levels, initial altering land, lagged outflows, and lag of the dependent variable.
  - Observation level: country-time measured by 5-year windows from 1990 to 2020. Sample selected based on data availability since 1995.
- Natural disasters measurement:
  - Natural disasters measured as the number of climatological disasters in EM-DAT in the 5-year window.
  - Robustness check: limiting disasters to those with reported deaths above 100.

### Main empirical results (Table 4: Impacts of Climate Disasters on Macroeconomic Outcomes)
- Dependent variables: Agricultural GDP (measured as percentage of initial GDP (0-100)), Gini index (0-100), Employment as percent of 15+ population (0-100), Remittances as percent of GDP (0-1000).
- Total Impact of Disasters:
  - Agricultural GDP: -0.0559** (standard error 0.0244)
  - Gini: 0.0423 (standard error 0.0293)
  - Employment: -0.0401* (standard error 0.0219)
  - Remittances: 0.0051 (standard error 0.0173)
- Of which, Mediated through Climate-Driven Migration:
  - Agricultural GDP: -0.0083*** (standard error 0.0026)
  - Gini: -0.023*** (standard error 0.0083)
  - Employment: -0.0010 (standard error 0.0009)
  - Remittances: 0.0202*** (standard error 0.0041)
- Other Channels (direct impact controlling for migration outflows instrumented using destination-level shocks):
  - Agricultural GDP: -0.0476** (standard error 0.0218)
  - Gini: 0.0657** (standard error 0.0323)
  - Employment: -0.0400* (standard error 0.0211)
  - Remittances: -0.0151 (standard error 0.0214)
- Sample and estimation details:
  - Observations: 576, 453, 643, 629 (for columns 1–4 respectively)
  - Number of Countries: 130, 120, 130, 131
  - Country Fixed Effects: YES for all specifications
  - Significance notation: *** p<0.01, ** p<0.05, * p<0.1
- Key empirical interpretations:
  - Climate disasters reduce agricultural output and employment (statistically significant).
  - Climate disasters increase income inequality and remittances in point estimates (not always statistically significant).
  - Climate-induced migration accounts for an important portion of climate’s overall impact for Agricultural GDP, Gini, and Remittances.
  - Specific magnitudes:
    - Climate-induced migration accounts for about 15 percent of climate’s overall negative impact on agriculture (presented as agricultural GDP as percent of initial GDP).
  - Mechanism insights:
    - Migration can reinforce or mitigate overall climate impacts depending on the outcome (e.g., migrants tend to be from the bottom of the income distribution, which can mitigate increases in Gini).

### Robustness: disasters with more than 100 deaths (Table 5)
- Regression of migration outflows and origin-level shocks on natural disasters excluding disasters with less than 100 deaths.
- Disasters (above 100 deaths) coefficients on origin/outflows:
  - Column values: 0.0799* (0.0449), 0.0816* (0.0445), 0.0891** (0.0447), 0.0838** (0.0389), 0.0937** (0.0397), 0.0907** (0.0394)
- Sample and estimation details:
  - Observations: 657, 657, 657, 664, 664, 664
  - Number of Countries: 135 for all columns
  - Country FE: YES for all columns
  - Time FE and Time Trend vary across columns (see table for specification)
  - R-squared: 0.449, 0.450, 0.453, 0.637, 0.643, 0.644
- Interpretation:
  - The estimated impact of disasters (more than 100 deaths) is positive and significant on both migration outflows and the origin-level factor.
  - A one standard deviation increase in disasters with more than 100 deaths increases cross-border migration outflows by 0.23 percent of the initial population.

### Bounds to mediation via macroeconomic variables (Table 6)
- Purpose: Provide upper and lower bounds for the mediated effect of macroeconomic outcomes on migration using exogenous variation from disasters and destination factors in a simultaneous equation approach (extensions of Dippel et al. (2021)). Appendix B.2 provides methodological details.
- Identified sign, lower bound (percentage of total effect), and confidence intervals:
  - Agricultural GDP: Positive; Lower Bound 7.87
    - Confidence Interval (0.00 -14.32)
  - Employment: Positive; Lower Bound 6.34
    - Confidence Interval (0.02-12.47)
  - Gini: Positive; Lower Bound 9.74
    - Confidence Interval (3.61-15.87)
  - Remittances: Not identified; no bounds reported
- Aggregate bounds:
  - Lower Bound of Indirect Impacts: Positive 18
  - Upper Bound of Direct Impact: Positive 82
- Interpretation:
  - Identified macroeconomic outcomes could explain at least 18 percent of the impacts of climate on migration.
  - Mediated impacts through productivity (agricultural GDP, employment) are likely large.
  - Results suggest unskilled-labor intensive sectors are likely more affected (evidence from inequality/Gini findings).

### Impact of disasters in destination countries on inflows (Table 7)
- Regression of migration inflows and destination-level shocks on natural disasters in destination countries.
- Disasters (in destination) coefficients across columns:
  - Column values: -0.001140 (0.0110), 0.00439 (0.0137), -0.0136 (0.0159), -0.00490 (0.00659), -0.00511 (0.00736), -0.00536 (0.00700)
- Interaction terms:
  - Disasters × Advanced: -0.0192 (0.0482), -0.00674 (0.0261)
  - Disasters × EMEs: -0.0127 (0.0144), -0.00512 (0.00639)
- Sample and estimation details:
  - Observations: 672 for all columns
  - Number of Countries: 135 for all columns
  - Country FE: YES for all columns
  - Time Trend and Year FE vary across columns
  - R-squared: reported across specifications (e.g., 0.217, 0.221, 0.226, 0.588, 0.591)
- Interpretation:
  - Point estimates are negative, indicating disasters in destination countries reduce migration inflows, but estimates are not statistically significant.

### Model implications and calibration (DIGNAD model with endogenous migration)
- Macro model: DIGNAD (Debt-Investment-Growth and Natural Disasters) augmented with endogenous locational choices and migration.
- Worker decision problem and migration mechanics:
  - Recursive value function for worker in country j:
    - V_j,t = ln c_j,t + z_j,t + max_{d ∈ {origin, abroad}} β E [ V_d,t+1 − m_{j,d,t} + ν ε_{d,t} ]
  - Non-monetary location value z_j,t and moving costs m_{j,d,t} (zero if staying).
  - Idiosyncratic locational shocks ε_{d,t} drawn from a Fréchet distribution; variance determines aggregate elasticity of migration to wages.
- Budget constraints for hand-to-mouth workers:
  - Home-country aggregated: C_origin_t = c_origin_t L_origin,t = W_t L_origin,t + Rem_t + T_t
  - Abroad aggregated: C_abroad_t = W_abroad_t L_abroad,t − Rem_t
  - W_abroad assumed exogenous; remittances assumed exogenous in baseline (endogenous remittances allowed in robustness).
- Migration flows and stocks:
  - μ_migrate,t = Ω_migrate^{-1} exp(β E V_abroad,t − m_origin,abroad,t)^{1/ν}
  - Ω_migrate = exp(β E V_abroad,t+1 − m_origin,abroad,t)^{1/ν} + exp(β E V_origin,t+1)^{1/ν}
  - L_abroad,t = (1 − μ_repat,t) L_abroad,t−1 + μ_migrate,t L_origin,t−1
  - L_origin + L_abroad = L̄
- Calibration targets and baseline responses:
  - Steady state matches average cross-border migration stock to total population ratio.
  - Migration costs m calibrated to match stock of migrants as percent of total population.
  - Baseline response matches three long-run impacts to a one-standard deviation increase in disasters:
    1. GDP responses: match about 1 percent decline of GDP in the long-term following a one-standard deviation shock in natural disasters (equivalent to the response of the agricultural GDP).
    2. Climate-driven migration: match identified empirical response of cross-border migration.
    3. Mediation through macroeconomic outcomes: match identified mediation moments.
- Purpose of model:
  - Quantify impacts of adaptation investments on alleviating effects of disasters on cross-border migration.
  - Introduce endogenous locational choices to study adaptation, migration, and macroeconomic feedbacks.

*Italic: Source — wpiea2023250-print-pdf - Appendix B.2 shows how results are robust to the inclusion of additional control variables.*

### 0.53 percent of the initial population in the long-run in the baseline response.  Third,

### wpiea2023250-print-pdf - 0.53 percent of the initial population in the long-run in the baseline response.  Third,

### Model calibration and assumptions
- Baseline assumption: about 20 percent of the migration response is driven by wage differentials.
- Preference shocks and non-monetary benefits of migration (z) are instrumental to discipline the baseline response.
  - The variance of preference shocks governs the elasticity of cross-border migration to migration trade-offs (see equation 15).
  - Non-monetary benefits of migration (z) balance the share of migration attributable to macroeconomic outcomes.
- Model calibrated for CAPDR:
  - Initial migrant stock: 4.2 percent of population for the average CAPDR country.
  - Remaining parameters are standard and match the average CAPDR economy.
  - Appendix A presents the parameters used in the model.

### Counterfactual analysis: effects of adaptation on cross-border migration
- Experiment setup:
  - Compare two economies differing in public capital adaptation levels.
  - Baseline economy: does not invest in adaptation.
  - Counterfactual economy: 50 percent of public capital is resilient.
  - Shock considered: one-standard deviation increase in natural disasters, associated with a shock in real GDP of 1 percent in the baseline.
- Key quantitative outcomes from Figure 6 (counterfactual exercise):
  - Climate-driven migration without resilient capital: about 0.5 percent of the initial population.
  - With 50 percent resilient public capital, climate-driven migration could be reduced by about 0.2 percent of the initial population.
- Transmission channels and macro outcomes:
  - Resilient infrastructure leads to lower GDP losses (panel b).
  - Important impacts on private investment resiliency (panel d).
  - Significant fiscal gains: fiscal deficit and public debt increases moderate significantly when government invests in resilient public capital (panels e and f).

### Main empirical findings and implications
- Decomposition of cross-border migration outflows:
  - Origin-country, destination-country, and global shocks all contribute to understanding migration patterns over the past three decades.
  - Relative importance differs by country group:
    - EMDEs: origin-country, destination-country, and global shocks all contribute to outflows.
    - Advanced economies: mainly global shocks are important.
  - Origin-country migration shocks have gained importance as drivers of cross-border migration in EMDEs over 2005-2015.
- Role of climate disasters:
  - Climate disasters play an important role in explaining origin-country migration shocks, though less for overall migration outflows.
  - Effect is particularly large and statistically significant for LICs and EMEs; considerably smaller and not statistically significant for advanced economies.
  - Quantitative effect: on average, a one standard-deviation increase in the number of climate disasters (3 extra climate disasters per year) is associated with an increase in outward migration by about 1/2 percent of the initial total population in LICs and EMEs over a five-year period.
- Country size effects:
  - Smaller economies (by population size or land area) experience relatively larger increases in outward migration as a result of climate disasters.
- Gender and disaster type heterogeneity:
  - Climate disasters raise migration for both genders; impact seems particularly important for males from LICs.
  - Water-related disasters have a relatively larger impact on migration outflows than heat-related disasters.
- Anticipation effects:
  - Firms’ perceptions and assessments of climate-related risks (anticipation of climate events) are associated with higher migration outflows.
- Interaction with economic channels:
  - Climate-driven migration is associated with lower (agricultural) output and higher remittances.
  - Migration plays an important role in mitigating climate’s adverse impact on income inequality.
- Policy-relevant model implication:
  - Investment in resilient infrastructure reduces the impact of climate on cross-border migration and yields economic gains via lower climate-driven migration.

### Policy recommendations and future research directions
- Policy measures suggested by the analysis:
  - Invest in climate-resilient public capital stock to reduce climate-driven cross-border migration and improve economic and fiscal outcomes.
  - Reinforce social safety nets and enhance climate-mitigating measures to contain adverse economic effects of climate shocks.
  - Tackle root causes of displacement amplified by climate change, such as poverty, inequality, and violence.
- Areas for future research:
  - Understand why climate-induced migration reinforces climate’s effect on macroeconomic outcomes in some cases but offsets it in others.
  - Assess impacts of climate on internal migration, its interaction with cross-border migration, and macroeconomic consequences.

### Methodological annex: identification and mediation framework
- Decomposition refinement (equation 18):
  - g_{o,d,t} = O^{border}_{o,t} + O_{o,t} + D_{d,t} + G_t + ε_{o,d,t}, where O^{border}_{o,t} captures non-homotheticity during massive migration episodes (11 episodes identified).
- Identification framework:
  - Treatment variable C: natural disasters.
  - Outcome Y: cross-border migration outflows.
  - Mediator X: macro variable impacted by C and affecting Y.
  - Structural equations:
    - Y = β^Y_C C + β^Y_X X + ε^Y
    - X = β^X_C C + β^X_Y Y + ε^X
  - Total impact of natural disasters on Y identified by regressing Y on C under exogeneity assumptions (Proposition 2):
    - ∆Y/∆C = m(β^Y_C + β^Y_X β^X_C), where m = (1 − β^Y_X β^X_Y)^{-1}.
- OLS and mediation bias results:
  - OLS coefficient on X after regressing Y on C and X converges to (Proposition 3):
    - ˜β → β^Y_X σ^2_{ε_x} + β^X_Y σ^2_{ε_y} over (σ^2_{ε_x} + (β^X_Y)^2 σ^2_{ε_y}).
  - Proposition 4 provides bounds and sign identification conditions for β^Y_X given knowledge of β^X_Y and σ^2_{ε_x}, and enumerates regions (I–VI) with corresponding bounds (equations 33–36).
- Instrumental strategy:
  - Use destination-level shocks as instruments for cross-border migration flows.
  - Required assumptions:
    - Exogeneity: destination-level shocks orthogonal to unobserved confounders of domestic variables and impact macro-outcomes only through migration.
    - Relevance: first-stage F-stat should be larger than 50; a preliminary assessment gives F larger than 20.

### Key model parameters and summary statistics (selected exact values from appendices)
- Model and calibration parameters (selected exact entries from Appendix Table A2):
  - iz i y0.0600
  - δ z a 0.03
  - λ 5 0.5
  - iz a y000 γ z i 0.5
  - r^{⋆} 0.04
  - δ x 0.05
  - share b 0.28
  - share dc 0.27
  - share bstar 0.2
  - share remit 0.15
  - imp2gdp 0.48
  - δ zi 0.075
  - ω 2.41
- Summary statistics (Appendix Table A3, exact values):
  - Total Disasters: Obs 6751, Mean 0.16741, Std. dev. 16.843580127
  - Log GDP per capita: Obs 6849, Mean 9.014431, Std. dev. 1.207397, Min 6.051807, Max 11.78874
  - Average temperature: Obs 6980, Mean 0.957702, Std. dev. 0.428512, Min -0.096, Max 2.434
  - Log Migrant Stock: Obs 7141, Mean 2.78685, Std. dev. 1.769527, Min 7.103322, Max 16.58093
  - Log Population: Obs 7141, Mean 5.73919, Std. dev. 2.106014, Min 10.12206, Max 21.04525
- Selected regression results (Appendix Tables A4 and A5, exact coefficients and significance):
  - Appendix Table A4 (Impact of Total Disasters):
    - Column (1) Origin Factor: Total Disasters coefficient 0.0313*** (0.0115)
    - Column (2) Outflows: Total Disasters coefficient 0.0159 (0.0110)
    - Lag Event coefficients: 4.299* (2.583) in (1) and 4.539*** (0.869) in (2)
    - Log Initial Stock Migrants: -1.324*** (0.467) in (1) and -1.157*** (0.300) in (2)
    - Observations: 6576 and 6576; R-squared: 0.42 and 0.60
  - Appendix Table A5 (Instrumented migration impacts on other channels, selected coefficients):
    - Instrumented Migration on Agricultural GDP: -0.499* (0.276)
    - Instrumented Migration on Gini: -2.401*** (0.567)
    - Instrumented Migration on Remittances: 0.796* (0.449)
    - Disasters on Agricultural GDP: -0.0476** (0.0218)
    - Disasters on Gini: 0.0657** (0.0323)
    - Observations and country counts: varying by column (e.g., Observations 576, 453, 643, 629; Number of Countries 130, 120, 130, 131)
    - Standard errors in parentheses; significance: *** p<0.01, ** p<0.05, * p<0.1

*Italic: Source — wpiea2023250-print-pdf (content supplied).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023250-print-pdf.pdf_