## Annex I / Selected sections — Sub-Saharan Africa: Building Resilience to Climate-Related Disasters (WP/22/39)

## Source details

**Canonical URL:** [Annex I / Selected sections — Sub-Saharan Africa: Building Resilience to Climate-Related Disasters (WP/22/39)](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022039-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022039-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022039-print-pdf.pdf.json)

---

### Key impact findings (impact analysis)
- Climate-related disasters (droughts, floods, storms) have a significant negative impact on medium-term economic growth, especially for Sub-Saharan Africa (SSA).
- The impact of a drought is about three times larger in SSA than in other emerging and developing economies.
- Disaster intensity matters more than disaster frequency: intensity-driven marginal effects are consistently higher than frequency-driven marginal effects.
- Quantitative magnitudes:
  - If a drought intensifies by 10 percentage points, medium-term annual per capita growth can decline by almost 0.8 to 1 percentage points in SSA.
  - An intensification of floods by 10 percentage points takes one-fifth to one-fourth the toll on medium-term growth compared to droughts.
  - SSA’s medium-term annual per capita growth was projected at 1.8 percent in the October 2019 WEO (prior to COVID-19 and assuming no climate shocks).
  - The paper reports a specific estimate of "0.43 percentage points" as the portion of drought loss that could be avoided by closing SSA–EMDE gaps in electricity and finance for a 10 percentage point intensification.
  - When a flood intensifies by "10 percentage points", medium-term per capita annual growth declines by at least "0.15 percentage points".
  - Improving health care alone to the EMDE level can save almost "0.1 percentage points" of the damage from floods.
- Mechanism: the model incorporates potentially unrecoverable loss of human capital (deaths, malnutrition, lower school enrollment), which negatively affects long-term growth even when near-term damage is partially offset by external assistance, remittances, and reconstruction.

### Data, proxies, and model specification
- Sample and aggregation:
  - Panel of 181 countries, 1960-2018; annual figures aggregated into five-year windows yielding 12 five-year periods.
- Disaster definitions and proxies (EM-DAT/CRED):
  - Disaster defined as a climate-related hazard causing at least one of: 10 people dead, 100 people affected, a declaration of a state of emergency, or a call for international assistance.
  - Intensity proxy: dummy indicating total annual effect of disasters weighing on over 0.01 percent of the population (uses fatalities and affected people relative to population).
  - Frequency proxy (main): follows Loayza et al. (2012); robustness frequency proxy (Frequency2) = Fatalities + 0.3*Affected over Population.
- Controls and data sources:
  - Impact analysis controls follow Loayza et al. (2012); Barro (2003) controls used in robustness.
  - Data drawn from WEO, WDI, and EM-DAT.
- Econometric approach:
  - Dynamic panel growth model à la Barro (1991) with disaster intensity and frequency.
  - Five-year aggregated GMM estimation (lagged regressors used as instruments, collapsed).
  - Complementary Minimum Distance Estimation (MDE) and fixed-effects robustness checks.

### Robustness checks and alternative frequency proxy (Annex III)
- Frequency2 definition: Frequency2(i,t)^k = Fatalities(i,t)^k + 0.3*Affected(i,t)^k over Population(i,t).
- GMM estimates with Frequency2 are stated to be consistent with main findings: intensity measures (e.g., drought intensity -6.856**, flood intensity -6.795***) show negative, statistically significant effects in some specifications while frequency measures do not uniformly show significant negative effects.
- Selected coefficient estimates (as presented in Table 11 / A7, GMM):
  - Log of per capita GDP: -10.111*** (3.630); -2.468** (1.104); 5.593** (2.182); -5.790*** (2.079); -2.838* (1.521); 3.090 (2.340); 0.213 (1.327); 3.626* (1.964).
  - Drought intensity: -6.856** (3.491); -13.136* (7.042).
  - Drought frequency: 0.033 (0.142); 0.071 (0.072).
  - Flood intensity: -6.795*** (2.052); -3.216*** (1.219).
  - Flood frequency: 0.691** (0.273); -0.053 (0.115).
  - Epidemic intensity: -1.194 (1.012); -0.669 (0.975).
  - Storm intensity: -1.081 (0.897); 0.673 (0.854).
  - Education: 0.256*** (0.092); 0.061** (0.027); -0.121** (0.053); 0.154*** (0.056); 0.067 (0.047); -0.113* (0.068); -0.011 (0.035); -0.110 (0.067).
  - Investment: 0.088* (0.053); 0.124*** (0.035); 0.028 (0.046); 0.178*** (0.051); 0.043** (0.022); 0.031 (0.038); 0.037 (0.028); 0.011 (0.056).
- Diagnostics and sample reporting (as presented):
  - Observations: 2115 1331 2325 1131 631 631 5867.
  - Number of instruments: 40 68 39 40 46 46 46 44.
  - Arellano-Bond AR(1) p-values: 0.029 0.000 0.009 0.020 0.032 0.024 0.017 0.055.
  - Arellano-Bond AR(2) p-values: 0.055 0.068 0.341 0.967 0.153 0.527 0.109 0.387.
  - Hansen test p-values: 0.373 0.146 0.199 0.266 0.984 0.601 0.791 0.997.
  - Difference-in-Hansen p-values: 0.078 0.078 0.137 0.254 0.634 0.011 1.000 0.999.

### Policy response analysis — approach and selected quantified interactions
- Model:
  - Five-year panel including interaction term b2 on (Disaster proxy) * (policy/structural variable z(i,p)); positive and significant b2 implies the policy variable improves resilience.
  - Policy variables analyzed one at a time; intensity and frequency proxies treated separately.
  - Estimation via GMM and MDE/fixed effects; focus on significant b2 estimates.
- Structural areas analyzed: Telecommunication, Access to finance (financial depth), Education, Health, Mechanization (agricultural machinery), Electricity.
- Selected significant b2 estimates (Intensity models; significance indicated as in source):
  - Telecommunication: 0.014**
  - Access to finance: 0.026*
  - Education: 0.032***
  - Health: 0.084**
  - Mechanization: 0.000*
  - Electricity: 0.057*
- Selected significant b2 estimates (Fixed effects / intensity):
  - Telecommunication: 0.017**
  - Access to finance: 0.022***
  - Education: 0.028**
  - Health: 0.113***
  - Mechanization: 0.000**
  - Electricity: 0.020*
- Additional reported interaction coefficients (GMM and selected tables):
  - Drought interactions: Disaster * Access finance: "0.026*", Disaster * Electricity: "0.057*", "0.114***", Disaster * Mechanization: "0.000", "-0.000**".
  - Flood interactions: Disaster * Telecommunication: "0.014**", "0.009**"; Disaster * Access finance: "0.022***", "0.016**"; Disaster * Health: "0.084**", "0.086**"; Disaster * Electricity (Fixed effects): "0.056***".
  - Epidemics interactions: Disaster * Telecommunication: "0.017**", "0.010"; Disaster * Electricity: "0.015**", "0.038*".
  - Storms interactions: Disaster * Telecommunication: "0.009**", "0.006"; Disaster * Education: "0.032***", "0.028**"; Disaster * Health: "0.113***", "0.083**"; Disaster * Electricity: "0.020*", "0.018".

### Policy implications, scenarios, and priority actions
- Scenario quantification method:
  - For a given disaster, resilience gain = (SSA–EMDE gap in a structural area) * b2 * 10 percentage points (increase in intensity proxy).
  - Intensity proxy used for scenarios because intensity has stronger growth impact.
  - Results show the share of per capita annual medium-term growth in SSA protected from loss when SSA improves a structural area to the EMDE average.
- Drought-focused implications:
  - Closing SSA–EMDE gaps in electricity and finance could avoid 0.43 percentage points of the medium-term per capita growth loss from a 10 percentage point drought intensification.
  - Electricity is central because it powers irrigation systems and deep tube-well pumps; hydropower generates "one fifth" of SSA’s electricity and is susceptible to droughts.
  - Conclusion: better access to electricity and finance can halve the medium-term economic loss from a drought in SSA (quantified via scenario and b2 estimates).
- Recommended policy priorities:
  - Droughts: Electrification combined with irrigation (public investment in irrigation, water, electricity systems; diversification toward geothermal, solar, wind; small dams, boreholes, solar irrigation schemes).
  - Floods and storms: Health care and education (improve health care to reduce post-flood disease incidence; building-code standards; land-use planning; widen access to quality building materials).
  - Cross-cutting agricultural resilience: Access to finance, telecommunications (mobile coverage for early warnings and weather/price information), mechanization (dikes, erosion protection, deeper seed planting).
  - Short-term/targeted measures: targeted government subsidies where household financing capacity is constrained; targeted social assistance to compensate lost income while structural improvements take effect.
- Policy sequencing and combinations:
  - The analysis identifies combinations of structural reforms most effective for specific disasters; improving multiple structural areas can yield additive resilience gains for a 10 percentage point increase in disaster intensity.

### Robustness, control-variable patterns, and suggested research extensions
- Robustness:
  - MDE and fixed-effects alternatives confirm qualitative patterns from GMM despite potential instrument concerns (some high p-values in Hansen and Difference-in-Hansen tests).
  - Robustness to Barro (2003) controls (life expectancy, fertility, democracy) retains the finding that intensity marginal effects exceed frequency marginal effects.
- Control-variable effects (consistent with literature):
  - Education, investment, and trade openness positively impact growth; inflation negatively impacts growth.
  - Fixed-effect checks: higher fertility reduces medium-term growth prospects; better democratic systems improve growth prospects.
- Suggested further research:
  - Design epidemic-specific models (separating health-related and agriculture-related epidemics).
  - Analyze and compare disaster effects on growth across world regions.

*IMF Working Paper — Sub-Saharan Africa: Building Resilience to Climate-Related Disasters (selected sections provided).*

### Annex I. Robustness checks (impact analysis) - Different controls ......................................................

### Annex I. Robustness checks (impact analysis) - Different controls

### Key findings on the impact of climate-related disasters
- Climate-related disasters (droughts, floods, storms) have a significant negative impact on medium-term economic growth, especially for Sub-Saharan Africa (SSA).
- The impact of a drought is about three times larger in SSA than in other emerging and developing economies.
- Disaster intensity matters more than disaster frequency: higher adverse effects on medium-term growth are associated with intensity, consistent with non-linear cumulative effects of successive disasters and the finding that growth impacts are best estimated with disasters of large magnitude.
- The model accounts for potentially unrecoverable loss of human capital (from deaths, malnutrition, or lower school enrollment) that negatively affects long-term growth even if near-term damage is partially offset by foreign financial assistance, remittances, and reconstruction.

### Contextual and methodological notes
- The analysis focuses on adaptation strategies for SSA given the region’s limited contribution to greenhouse gas emissions.
- The medium-term growth model follows the approach of Barro (1991) and Loayza et al. (2012) and uses macroeconomic variables together with frequency and intensity disaster measures.
- The analysis corroborates previous literature (Cavallo et al., 2013; Fomby et al., 2013) that disaster magnitude drives growth impacts more than frequency.
- The COVID-19 pandemic and other health and agriculture-related epidemics (e.g., Ebola, locust infestations) have increased SSA’s vulnerability to climate shocks by weakening populations’ economic and health conditions and have constrained governments’ financial and human resources for recovery and resilience-building.

### Policy-relevant structural priorities for resilience (policy response analysis)
- Droughts:
  - Electrification combined with irrigation is key to building resilience to droughts.
- Floods and storms:
  - Health care and education are most important for minimizing damage from floods and storms.
- Cross-cutting agricultural resilience:
  - Access to finance, telecommunications, and use of machinery in agriculture make significant contributions to resilience-building.
- Epidemics:
  - The analysis includes epidemics given the pandemic environment, but the results were inconclusive.

### Contribution to literature and policy design
- The paper links climate-disaster impact assessment with a growth-model framework to identify which policy variables most effectively improve resilience in SSA.
- It complements global warming impact studies by focusing on country-level adaptation strategies and the medium-term growth channel through which disasters affect development prospects.

*IMF Working Paper — Sub-Saharan Africa: Building Resilience to Climate-Related Disasters*

### introduction of climate change proxies requires use of a sparse growth model. However, unlike the latter, our

### Sub-Saharan Africa: Building Resilience to Climate-Related Disasters

### Introduction and scope
- Paper simultaneously assesses the impact associated with the intensity and the frequency of climate-related disasters and contributes to strategies for improving resilience to climate-related natural disasters.
- Organization:
  - Section II: data and how climate-related disasters are quantitatively proxied.
  - Section III: impact analysis quantifying effects on medium-term growth.
  - Section IV: policy response analysis measuring how selected structural reform areas can improve resilience.
  - Section V: conclusion.

### Data and quantitative proxies for climate-related disasters
- Sample and aggregation:
  - Panel covering 181 countries during 1960-2018.
  - Annual figures aggregated into five-year windows (averages over the windows), yielding 12 five-year periods.
  - Intensity and frequency proxies are not aggregated identically: aggregated intensity proxy captures the proportion of disruptive disasters while aggregated frequency proxy gives the ratio between disaster-related fatalities and the population.
- Definition of climate-related disasters (EM-DAT/CRED):
  - A climate-related hazard that leads minimally to one of: at least 10 people dead, at least 100 people affected, a declaration of a state of emergency, or a call for international assistance.
- Intensity proxy:
  - Dummy variable indicating whether the total annual effect of disasters weighs on over 0.01 percent of the population.
  - Follows Fomby et al. (2013); uses fatalities and affected people relative to population.
- Frequency proxy:
  - Considers total effects related to the occurrence of disasters during the year; follows Loayza et al. (2012).
  - Frequency proxy used in robustness checks also considers both fatalities and affected people (see Annex 3 and footnote: Frequency2 = Fatalities + 0.3*Affected over Population).
- Conceptual emphasis:
  - Proxies build on human capital destruction (fatalities and affected people) rather than physical capital destruction because EM-DAT provides fatalities and affected people data; physical capital deterioration may be captured through conventional growth model controls.
- Control variables and data sources:
  - Impact analysis uses controls of Loayza et al. (2012); Barro (2003) controls used in Annex 1 as robustness.
  - Data drawn from WEO, WDI, and EM-DAT.
  - Key variables listed (used in impact models): intensity/frequency of droughts, floods, epidemics and storms; Log of per capita GDP (Real per capita GDP, PPP, WEO); Education (gross secondary enrollment, WDI); Investment (gross fixed capital formation, percent of GDP, WEO); Government consumption (percent of per capita GDP, PWT); Inflation (consumer prices, percent change, WEO); Trade openness (import+export ratio, WEO); Change in terms of trade (PWT).
  - Policy-response analysis variables include: Telecommunication (mobile cellular subscriptions per 100 people, WDI); Financial depth (domestic credit to private sector, percent of GDP, WDI); Education (gross secondary enrollment, WDI); Health (life expectancy at birth, WDI); Agricultural machinery (total tractors, WDI); Electricity (access to electricity, percent of population, WDI).
  - Irrigation, sanitation, quality of fiscal policy and quality of roads excluded from policy response analysis because of data issues.

### Impact analysis: econometric strategy
- Baseline models:
  - Dynamic panel growth model à la Barro (1991) extended to include disaster intensity and frequency proxies.
  - Five-year aggregated model: intensity averaged over disasters during the 5-year period; frequency averaged over the 5 annual observations in the 5-year window.
  - Model estimated using GMM to correct for potential correlation between unobserved effects and lagged regressors; endogenous variables: per-capita GDP and disaster proxies; all available lagged regressors used as instruments then collapsed to reduce many-instrument bias.
  - Complementary estimation via minimum distance estimation (MDE) following Islam (1995) for robustness.

### Impact analysis: key quantitative findings and patterns
- General findings:
  - Significant negative impact of climate-related disasters on medium-term growth, with most significant effects originating from droughts and floods.
  - Climate-related disasters prominently weigh on SSA growth—with droughts having the strongest effect, reflecting prolonged nature and dependence on rain-fed agriculture.
  - A disaster’s intensity impacts medium-term growth more than its frequency (when both are controlled): marginal effects associated with the intensity proxy are consistently higher than those for the frequency proxy.
    - Predicted growth increase associated with a reduction of the proxies from their average values to 0 is higher for the intensity proxy.
  - Results for epidemics and storms are not conclusive due to limited data coverage; floods can help understand storms because flood data include after-effects of extreme storms such as cyclones.
- Quantitative magnitudes (MDE and interpretation):
  - If a drought intensifies by 10 percentage points, medium-term annual per capita growth can decline by almost 0.8 to 1 percentage points in SSA.
  - An intensification of floods by 10 percentage points takes one-fifth to one-fourth the toll on medium-term growth compared to droughts.
  - SSA’s medium-term annual per capita growth was projected at 1.8 percent in the October 2019 WEO—prior to the COVID-19 pandemic and assuming no climate shocks.
  - The negative impact of droughts on SSA growth can be up to three times that in emerging and developing economies.
- Robustness and instruments:
  - Null hypothesis on exogeneity of instruments is not rejected, but high p-values of some Hansen and Difference-in-Hansen tests indicate potential instrument issues; alternative MDE estimates confirm qualitative patterns.
- Control variable effects (consistent with literature):
  - Education, investment, and trade openness positively impact growth; inflation negatively impacts growth.
  - From fixed-effect robustness checks: higher fertility reduces medium-term growth prospects while better democratic systems improve them.

### Complementary robustness checks
- Alternative specifications:
  - Minimum distance estimation (MDE) used as robustness; results consistent with GMM for qualitative patterns.
  - Robustness to using Barro (2003) growth controls: life expectancy, fertility and democracy added—Tables A1 and A2 in Annex 1 confirm robustness; intensity marginal effect remains higher than frequency.
  - Robustness to frequency proxy that considers both fatalities and affected people reported in Annex 3.
- Notes on interpretation:
  - Difference between intensity and frequency proxies may not be accurately characterized for very extreme disasters (dead toll higher than 1 percent of the population); such extreme disasters represent less than 2.6 percent of floods and less than 1.6 percent of droughts in the sample.

### Policy response analysis: econometric approach
- Model:
  - Five-year panel model including interaction between a policy/structural variable z(i,p) and disaster proxy Dis(i,p)^k (intensity or frequency).
  - Focus on parameter b2 (slope on the interaction term): a positive and significant b2 implies the policy variable improves resilience to that type of disaster.
  - Policy variables analyzed one at a time; intensity and frequency proxies included separately to mitigate multicollinearity.
  - Estimation via GMM and MDE/fixed effects; focus on cases where b2 is significant.
- Structural areas analyzed:
  - Telecommunication, access to finance (financial depth), education, health, mechanization (agricultural machinery), electricity.

### Policy response analysis: quantitative estimates (b2) and significant results
- Selected significant b2 estimates (Intensity models; significance indicated exactly as in source):
  - Telecommunication: 0.014**
  - Access to finance: 0.026*
  - Education: 0.032***
  - Health: 0.084**
  - Mechanization: 0.000*
  - Electricity: 0.057*
- Selected significant b2 estimates (Fixed effects / intensity):
  - Telecommunication: 0.017**
  - Access to finance: 0.022***
  - Education: 0.028**
  - Health: 0.113***
  - Mechanization: 0.000**
  - Electricity: 0.020*
- Additional patterns:
  - For epidemics (intensity proxy) most policy variables (except mechanization) tend to be positively associated with resilience, though b2 not always significant.
  - GMM and fixed effects are consistent for the parameter of interest b2.

### Policy implications and scenario quantification
- Scenario approach:
  - For a given climate-related disaster, compute resilience gain by multiplying SSA–EMDE gap in each structural area by the estimated b2 and by an increase in the intensity proxy by 10 percentage points.
  - Intensity proxy used (not frequency) since intensity has stronger impact on growth.
  - Results show per capita annual medium-term growth in SSA that is protected from loss when SSA improves a given structural area to the EMDE average.
  - Impacts illustrated are separate from each structural area’s direct impact on growth through other channels (represented by b3 in the model).
- Example: droughts
  - When a drought intensifies by 10 percentage points, medium-term per capita annual growth declines by at least 0.8 percentage points (Section III).
  - Closing gaps with EMDEs in electricity and finance could avoid 0.43 percentage points of this loss—electricity plays a central role because it powers irrigation systems and deep tube-well pumps critical during prolonged dry spells.
  - Conclusion: better access to electricity and finance can halve the medium-term economic loss from a drought in SSA (quantified via the scenario and the b2 estimates).
- Combinations:
  - The analysis identifies combinations of structural reform areas most effective for specific types of disasters; Figure 2 (in source) shows reductions in the impact on per capita annual medium-term growth when structural factors are improved to EMDE averages for a 10 percentage point increase in intensity.

*Source: IMF Working Paper — Sub-Saharan Africa: Building Resilience to Climate-Related Disasters (selected sections provided).*

### 0.43 percentage points estimated in this paper—where the benefits from greater access to electricity are

### wpiea2022039-print-pdf - 0.43 percentage points estimated in this paper—where the benefits from greater access to electricity are

### Major findings on climate-related disasters and growth
- Climate-related disasters, especially droughts, have a substantial impact on medium-term growth in SSA—much more than in other regions of the world.
- A disaster’s intensity matters much more than its frequency because of non-linear cumulative effects of successive disasters.
- The paper estimates "0.43 percentage points" in a context where benefits from greater access to electricity are assessed based on existing irrigation and pumping systems.
- Hydropower generates "one fifth" of SSA’s electricity and is susceptible to droughts.
- When a flood intensifies by "10 percentage points", medium-term per capita annual growth declines by at least "0.15 percentage points".
- Based on the policy response analysis, improving health care alone to the EMDE level can save almost "0.1 percentage points" of the damage from floods.

### Policy priorities and recommended interventions
- Electrification combined with irrigation is key to building resilience to droughts.
  - Governments can prioritize public investment in appropriate irrigation, water, and electricity systems.
  - Diversification of electricity sources toward geothermal, solar, and wind power is a major component of increasing access to electricity.
  - Reduced reliance on hydroelectricity facilitates water management; building and rehabilitating small dams and boreholes and setting up solar irrigation schemes are highlighted.
- Health care and education are most important for minimizing damage from floods and storms.
  - Policies include widening accessibility to quality building materials for the poor; appropriate building-code standards; and effective land-use planning and zoning.
  - Improved health care reduces disease incidence after floods and preserves household income flows and savings.
- Access to finance, telecommunications, and mechanization in agriculture are significant contributors to resilience-building.
  - Access to finance allows households and SMEs to invest in weather-resilient infrastructure and provide post-disaster buffers.
  - Modern telecommunications (solid mobile phone coverage) broaden early warning reach and provide weather and price information to support climate-smart agriculture.
  - Mechanization enables construction of dikes, erosion protection, and deeper seed planting.
- Targeted government subsidies may be necessary where household financing capacity is constrained by low-income levels and asset values.
- In the short term, targeted social assistance can compensate for lost income and purchasing power post-disaster while structural improvements take effect.

### Policy response analysis — key quantified interaction results (intensity proxies)
- Droughts (Table 7 / A3, GMM columns):
  - Lagged growth: "0.157**", "0.133*", "0.138*", "0.080", "0.208**", "0.029", "0.061", "0.042", "-0.015", "0.066", "0.078", "0.142".
  - Intensity Disaster: "-1.790***", "-2.349***", "-2.056*", "-3.339", "-1.671**", "-6.390**", "-1.885***", "-2.920**", "-2.341*", "-10.069", "-1.383", "-14.198***".
  - Disaster * Access finance: "0.026*", "0.026*".
  - Disaster * Electricity: "0.057*", "0.114***".
  - Disaster * Mechanization: "0.000", "-0.000**".
- Floods (Table 8 / A4, GMM columns):
  - Lagged growth: "0.132**", "0.097", "0.100", "0.121**", "0.030", "0.206***", "0.101***", "0.062*", "0.110**", "0.092**", "0.127***", "0.073*".
  - Intensity Disaster: "-1.063***", "-1.315***", "-0.951", "-5.933***", "-0.684**", "-0.552", "-1.296***", "-1.524***", "-1.103*", "-6.594***", "-0.963***", "-0.048".
  - Disaster * Telecommunication: "0.014**", "0.009**".
  - Disaster * Access finance: "0.022***", "0.016**".
  - Disaster * Health: "0.084**", "0.086**".
  - Disaster * Electricity (Fixed effects): "0.056***".
- Epidemics (Table 9 / A5, GMM columns):
  - Lagged growth: "0.362***", "0.319***", "0.366***", "0.372***", "0.254***", "0.336***", "0.160***", "0.119**", "0.137**", "0.158***", "0.135**", "0.137*".
  - Intensity Disaster: "-0.775", "-0.671", "-0.688", "-4.261", "-0.859*", "0.301", "-0.133", "-0.090", "-0.787", "-2.402", "-0.088", "-0.172".
  - Disaster * Telecommunication: "0.017**", "0.010".
  - Disaster * Electricity: "0.015**", "0.038*".
- Storms (Table 10 / A6, GMM columns):
  - Lagged growth: "0.085", "0.054", "0.204***", "0.093", "0.039", "0.279***", "0.086", "0.035", "0.114*", "0.083", "0.064", "0.084".
  - Intensity Disaster: "-0.599", "-0.804*", "-2.449***", "-7.709***", "-0.309", "-1.521*", "-0.441", "-0.263", "-2.232**", "-5.855**", "-0.406", "-1.770*".
  - Disaster * Telecommunication: "0.009**", "0.006".
  - Disaster * Education: "0.032***", "0.028**".
  - Disaster * Health: "0.113***", "0.083**".
  - Disaster * Electricity: "0.020*", "0.018".

### Robustness checks and model specifications
- Alternative controls from Barro (2003) added: life expectancy, fertility (total fertility rate), and democracy (Polity4).
- Baseline GMM results (Table 5 / A1) show:
  - Log of per capita GDP coefficients such as "-9.976***", "-4.179***", "1.779", "-2.048", "-0.868", "4.735**", "5.887***", "5.343***".
  - Intensity drought coefficients include "-6.597***", "-8.713**".
  - Frequency drought coefficients include "-0.451***", "-0.426***".
  - Intensity flood coefficients include "-7.040***", "-1.582".
  - Selected control variable coefficients: Education "0.112*", " -0.153***", "-0.169**", "-0.132**"; Investment "0.072*", "0.129***", "0.065**", "0.077**".
  - Observations vary across models: "204", "495", "303", "305", "106", "155", "150", "67".
- Fixed effects alternative (Table 6 / A2) reports:
  - Log of per capita GDP values like "-2.562**", "-2.099***", "-2.684***", "-0.767", "-3.436***", "-2.283*", "-1.816", "6.390*".
  - Intensity drought "-2.392***", "-5.730***".
  - Frequency drought "-0.403***", "-0.441***".
  - Observations: "204", "495", "303", "305", "106", "155", "150", "67".
  - R2 values: "0.549", "0.402", "0.420", "0.408", "0.698", "0.523", "0.603", "0.592".

### Conclusions and suggested further research
- Urgent policy action is needed to build SSA’s resilience to rapidly growing climate-related disasters, with priority setting required in the post-COVID-19 resource-constrained environment.
- The paper’s policy response analysis finds:
  - Electrification + irrigation best for drought resilience.
  - Health care + education best for floods and storms.
  - Access to finance, telecommunications, and mechanization also significantly strengthen resilience.
- Recommended extensions: design epidemic-specific models (separately for health-related and agriculture-related epidemics) and analyze and compare disaster effects on growth across world regions.

*IMF WORKING PAPERS Sub-Saharan Africa: Building Resilience to Climate-Related Disasters*

### Annex III. Impact analysis with a different

### Annex III. Impact analysis with a different frequency proxy

### Methodology and frequency proxy
- The impact analysis is replicated with a frequency proxy that accounts for both the fatalities and the affected people:
  - 퐹푟푒푞푢푒푛푐푦2
    푖,푡
    푘
    =
    퐹푎푡푎푙푖푡푖푒푠
    푖,푡
    푘
    + 0.3∗퐴푓푓푒푐푡푒푑
    푖,푡
    푘
    푃표푝푢푙푎푡푖표푛
    푖,푡
- The results reported use a GMM specification and are stated to be consistent with the results obtained in the main paper.

### Table 11 (A7): Growth models with disaster indicators (different frequency proxy, GMM) — model mapping
- (1), (2), (3) and (4) represent models for droughts, floods, epidemics and storms, respectively.
- *, ** and *** indicate statistical significance at 10, 5 and 1 percent, respectively.
- Year-dummy parameters are not presented.

### Selected coefficient estimates and standard errors (as presented)
- Log of per capita GDP:
  - -10.111*** (3.630)
  - -2.468** (1.104)
  - 5.593** (2.182)
  - -5.790*** (2.079)
  - -2.838* (1.521)
  - 3.090 (2.340)
  - 0.213 (1.327)
  - 3.626* (1.964)
- Drought intensity:
  - -6.856** (3.491)
  - -13.136* (7.042)
- Drought frequency:
  - 0.033 (0.142)
  - 0.071 (0.072)
- Flood intensity:
  - -6.795*** (2.052)
  - -3.216*** (1.219)
- Flood frequency:
  - 0.691** (0.273)
  - -0.053 (0.115)
- Epidemic intensity:
  - -1.194 (1.012)
  - -0.669 (0.975)
- Epidemic frequency:
  - 0.006 (0.623)
  - 0.864 (0.846)
- Storm intensity:
  - -1.081 (0.897)
  - 0.673 (0.854)
- Storm frequency:
  - -0.061 (0.146)
  - -0.095 (0.145)
- Education:
  - 0.256*** (0.092)
  - 0.061** (0.027)
  - -0.121** (0.053)
  - 0.154*** (0.056)
  - 0.067 (0.047)
  - -0.113* (0.068)
  - -0.011 (0.035)
  - -0.110 (0.067)
- Investment:
  - 0.088* (0.053)
  - 0.124*** (0.035)
  - 0.028 (0.046)
  - 0.178*** (0.051)
  - 0.043** (0.022)
  - 0.031 (0.038)
  - 0.037 (0.028)
  - 0.011 (0.056)
- Government consumption:
  - -0.043 (0.044)
  - -0.048** (0.023)
  - 0.009 (0.041)
  - -0.117*** (0.037)
  - 0.015 (0.028)
  - -0.002 (0.039)
  - -0.029 (0.021)
  - -0.044 (0.039)
- Inflation:
  - 0.002 (0.001)
  - -0.003*** (0.001)
  - -0.002** (0.001)
  - -0.010*** (0.002)
  - 0.001 (0.001)
  - -0.002*** (0.000)
  - -0.002*** (0.001)
  - 0.073 (0.054)
- Trade openness:
  - 3.214** (1.529)
  - 0.551 (0.548)
  - -0.700 (1.461)
  - 0.404 (0.621)
  - 2.867* (1.484)
  - -0.081 (1.483)
  - 3.363** (1.588)
  - 1.954* (1.139)
- Change in terms of trade:
  - 0.048 (0.113)
  - 0.075 (0.065)
  - -0.030 (0.084)
  - 0.271** (0.119)
  - -0.154 (0.099)
  - 0.084 (0.093)
  - 0.029 (0.058)
  - -0.090 (0.238)
- Intercept:
  - 74.438*** (25.789)
  - 21.400** (8.326)
  - -37.729** (14.758)
  - 41.322*** (14.392)
  - 32.758** (16.126)
  - -15.730 (14.860)
  - -2.125 (8.794)
  - -23.335* (14.173)

### Model diagnostics and sample details (as presented)
- Observations: 2115 1331 2325 1131 631 631 5867 (presented as contiguous digits in table)
- Number of instruments: 40 68 39 40 46 46 46 44 (presented as contiguous digits in table)
- Arellano-Bond test for AR(1): 0.029 0.000 0.009 0.020 0.032 0.024 0.017 0.055
- Arellano-Bond test for AR(2): 0.055 0.068 0.341 0.967 0.153 0.527 0.109 0.387
- Hansen test: 0.373 0.146 0.199 0.266 0.984 0.601 0.791 0.997
- Difference-in-Hansen test: 0.078 0.078 0.137 0.254 0.634 0.011 1.000 0.999
- Labeling in table indicates panel breakdown EMDEs SSA

### Core findings (as expressed by the table and text)
- The GMM estimation using the alternative frequency proxy yields results that the authors state are consistent with the main paper’s findings.
- Several disaster intensity measures (e.g., drought intensity, flood intensity) show negative and statistically significant coefficients in some specifications (for example, drought intensity -6.856** and flood intensity -6.795***).
- Frequency measures (constructed to combine fatalities and affected people) do not uniformly show significant negative effects across hazards in these GMM specifications (examples: drought frequency 0.033, flood frequency 0.691** in one specification, epidemic frequency 0.006).
- Standard control variables (education, investment, trade openness, inflation, government consumption, change in terms of trade) show a mix of statistically significant and insignificant estimates across specifications; examples include Education 0.256***, Investment 0.124***, Trade openness 3.214**, Inflation -0.003*** in some specifications.

*IMF Working Paper — Annex III. Impact analysis with a different frequency proxy, Sub-Saharan Africa: Building Resilience to Climate-Related Disasters (Working Paper No. WP/22/39)*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022039-print-pdf.pdf_