## onlineannexch2 - Section 1

## Source details

**Canonical URL:** [onlineannexch2 - Section 1](https://www.imf.org/-/media/files/publications/reo/afr/2020/april/english/onlineannexch2.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/reo/afr/2020/april/english/onlineannexch2.pdf.md)
- [Structured JSON version](/-/media/files/publications/reo/afr/2020/april/english/onlineannexch2.pdf.json)

---

### 2.1 Nightlight regression — data and methods
- Data sources and aggregation:
  - Temperature and precipitation: Harris and others (2014), CRU TS v. 4.03, 1901-2018, 0.5°ൈ0.5° grid, monthly frequency; averaged at the state level, weighted by grid area within state boundaries.
  - Nightlights: NOAA VIIRS monthly composites of average radiance images, 15"ൈ15" grid; data used between 2013 and 2018; nightlights summed to the state level as proxy for economic activity.
  - Forest change: Hanson and others (2013) and project update (University of Maryland), 2000–2018, 30-meter resolution.
  - Water surface change: Pekel and others (2016), global surface water occurrence and intensity change from three million Landsat images at 30-meter resolution.
- Econometric specification:
  - Dependent variable: Δz_{i,t} = z_{i,t} − z_{i,t−1}, where z_{i,t} is logarithm of nightlights in state i at time t (nightlight growth).
  - Regressors: c_{i,t}, climate variable measured in deviations from its past 30-year average for the same month.
  - Fixed effects: μ_i (state) and α_t (time).
  - Estimation: ordinary least squares for 2013–2018, with standard errors clustered at the state level.
  - Units: temperature in Celsius degree; precipitation in 10 millimeters.
  - Interpretation: β_j measures impact of temperature or precipitation deviations from historical average on local economic growth.

### 2.1 Nightlight regression — key results
- Temperature deviation effects (monthly nightlight growth):
  - A 1°C deviation is associated with about 2 percentage points reduction in monthly growth of nightlights for the world on average.
  - Sub-Saharan Africa (SSA): 4.2 percentage points reduction.
  - Sahel region: 6.4 percentage points reduction.
  - The SSA result is almost entirely driven by Western Africa.
- Regression coefficient estimates (selected, exact):
  - Lagged nightlights coefficient (dependent variable: Nightlight growth):
    - World: -0.771*** (0.007)
    - SSA: -0.786*** (0.010)
    - Sahel: -0.827*** (0.025)
    - EMDE: -0.743*** (0.007)
    - Eastern Africa: -0.817*** (0.016)
    - Southern Africa: -0.755*** (0.033)
    - Western Africa: -0.774*** (0.017)
    - Central Africa: -0.805*** (0.022)
  - Temperature deviation:
    - World: -0.019*** (0.002)
    - SSA: -0.042*** (0.012)
    - Sahel: -0.064** (0.031)
    - EMDE: -0.026*** (0.002)
    - Eastern Africa: -0.006 (0.049)
    - Southern Africa: 0.028 (0.046)
    - Western Africa: -0.052** (0.026)
    - Central Africa: -0.030 (0.051)
  - Precipitation deviation:
    - World: -0.006*** (0.001)
    - SSA: -0.008*** (0.001)
    - Sahel: -0.012** (0.006)
    - EMDE: -0.008*** (0.001)
    - Eastern Africa: -0.002 (0.002)
    - Southern Africa: -0.010 (0.006)
    - Western Africa: -0.008*** (0.003)
    - Central Africa: -0.005 (0.004)
- Additional diagnostics and sample sizes (exact):
  - Observations (temperature and precipitation regressions): World 100,475; SSA 21,125; Sahel 2,953; EMDE 80,309; Eastern Africa 8,665; Southern Africa 1,599; Western Africa 8,178; Central Africa 4,138.
  - R-squared (temperature): World 0.4820; SSA 0.5900; Sahel 0.6900; EMDE 0.4780; Eastern Africa 0.7870; Southern Africa 0.7640; Western Africa 0.7590; Central Africa 0.712
  - Number of states: World 222; SSA 347; Sahel 364; EMDE 170; Eastern Africa 421; Southern Africa 636; Western Africa 161; Central Africa 199
- Duration of effects:
  - Impacts of temperature and precipitation deviations on economic activity are temporary, lasting for one or two months depending on the number of lags in the regression.
  - Some evidence that precipitation deviations have a larger and longer impact during crops’ growing seasons.

### 2.2 Country-level panel regression — data and design
- Dataset and aggregation:
  - Panel of 181 countries from 1960 to 2018.
  - Yearly data aggregated into 5-year windows (final dataset has 12 five-year periods).
- Disaster proxies:
  - Intensity proxy: proportion of disruptive disasters in the 5-year period (intensity dummy = 1 if disaster in year t affects over 0.01 percent of population, where ratio (death + 0.3*affected)/population is used).
  - Frequency proxy: average effects of disasters via average ratio of deaths to population.
- Control variables (impact analysis, “sparse” growth model) and policy-response variables:
  - Impact-analysis controls (selected): Log of per capita GDP (Real per capita GDP, PPP (WEO)), Education (Gross rate of enrollment in the secondary (WDI)), Investment (Gross fixed capital formation, percent of GDP (WEO)), Government consumption (Percent of per capita GDP government consumption (PWT)), Inflation (Consumer Prices, period average, percent change (WEO)), Trade openness (Ratio (Import+Export)-GDP (WEO)), Change in terms of trade (Change, ratio price export-price import (PWT))
  - Policy-response variables: Telecommunication (Mobile cellular subscriptions per 100 people (WDI)), Financial depth (Domestic credit to private sector, percent of GDP (WDI)), Education (Gross rate of enrollment in the secondary (WDI)), Health (Life expectancy at birth (WDI)), Agri. Machinery (Agricultural machinery, total tractors (WDI)), Electricity (Access to electricity, percent of population (WDI))
  - Other variables with no significant effect detected: Irrigation, Sanitation, Quality of fiscal policy, Roads
- Impact model estimated:
  - Dependent variable: Gr_{i,p} = average per capita growth over 5-year period p.
  - Regressors: lagged log GDP, intensity and frequency disaster proxies by type j, additional controls X_{i,p}, time fixed effects c_p, country fixed effects d_i.
- Policy-response model:
  - Interaction specification: includes Disasters * z_{i,p} interaction to assess whether policy variable z_{i,p} mitigates growth impact of disasters; b1 negative and b2 positive interpreted as policy improving resilience if b2 significant and positive.
  - Annex Table 2.5 reports changes in SSA per capita growth if average conditions improve to EMDE levels.

### 2.2 Country-level panel regression — selected econometric results
- Selected coefficients from Annex Table 2.4 (exact values reported):
  - Log of per capita GDP:
    - Column (1) EMDEs: -2.09** (0.98)
    - Column (2) EMDEs: -2.03*** (0.65)
    - Column (3) EMDEs: -2.52** (0.99)
    - Column (4) EMDEs: -1.06 (1.16)
    - SSA columns (examples): -2.37 (1.43); -1.57 (1.37); -1.51 (1.07); 6.72 (3.94)
  - Intensity drought:
    - EMDEs: -2.66*** (0.74)
    - SSA: -7.81*** (2.16)
  - Frequency drought:
    - EMDEs: -0.34*** (0.05)
    - SSA: -0.41*** (0.04)
  - Intensity flood:
    - EMDEs: -1.31*** (0.44)
    - SSA: -1.49** (0.54)
  - Frequency flood:
    - EMDEs: -0.06** (0.03)
    - SSA: -0.18** (0.08)
  - Intensity epidemic and frequency epidemic: coefficients reported but not statistically significant in shown columns (Intensity epidemic -0.20 (0.33); Frequency epidemic -0.00 (0.13) for EMDEs; Intensity epidemic -0.53 (0.41); Frequency epidemic -0.01 (0.16) for SSA)
  - Frequency storm: EMDEs -0.19** (0.08); SSA -12.26 (11.27)
  - Education (examples):
    - EMDEs: -0.07** (0.03), -0.07*** (0.02), -0.05* (0.03), -0.07*** (0.02)
    - SSA: -0.05 (0.06), -0.07* (0.04), -0.06 (0.05), -0.16 (0.16)
  - Investment (examples):
    - EMDEs: 0.03*** (0.01), 0.09*** (0.02), 0.07*** (0.02), 0.02 (0.04)
    - SSA: 0.03*** (0.01), 0.05* (0.02), 0.05** (0.02), 0.09 (0.06)
  - Inflation and intercepts: intercept examples 21.37*** (7.53), 20.31*** (5.23), 20.61** (7.98), 14.34 (10.14); SSA intercepts 26.85*** (9.38), 12.29 (9.11), 7.99 (8.79), -46.39 (27.15)
- Notes:
  - Dependent variable is per capita GDP growth in the following year.
  - Robust standard errors in parenthesis.
  - Country and year fixed effects included in reported specifications.
  - Number of observations vary across columns, e.g., EMDEs columns: 2115, 1331, 2323, 2511; SSA columns: 316, 315, 867.

### 2.3 Household surveys on food security — data and descriptive findings
- Surveys and countries:
  - Nationally representative household surveys in Ethiopia, Malawi, Mali, Niger, and Tanzania, collected with the World Bank LSMS.
  - Households report shocks experienced in the last 1 to 5 years depending on country (Ethiopia, Malawi, Niger: last 12 months; Mali: last 3 years; Tanzania: last 5 years).
  - Analysis focuses on shocks likely to increase with climate variability, rising temperatures and sea levels.
- Prevalence of shocks (summary):
  - Households in Malawi reported the most shocks overall, followed by Mali, Tanzania, Ethiopia and Niger.
  - Major price changes affected over 30 percent of households in all countries.
  - Droughts, water shortages and irregular rainfall affected a large number of households.
  - Severe illness, crop damage, and loss of livestock frequently experienced; flood and conflict or violence less frequent; land loss and fire affected a small percentage of households (land loss not asked in Niger and Mali; fire asked only in Ethiopia).
- Coping strategies (Annex Figure 2.2 summary):
  - Most common coping strategy: draw on savings.
  - Other common strategies: relying on friends and family, selling assets, adjusting consumption.
  - Government support reported by 17 percent of households in Ethiopia and 15 percent in Mali.
  - Additional strategies: migration, reducing health and education expenditure, taking out a loan, seeking additional work, NGO support.

### 2.3 Household surveys — econometric results on food security
- Main food-security regression (linear probability model via OLS):
  - Dependent variable: Food security = 1 if household reports not recently worried about having enough to eat.
  - Key regressor: shock indicator = 1 if household reports experiencing any listed shock.
  - Controls: household wealth index (constructed via principal component analysis, standardized mean 0 SD 1), region dummies, rural dummy.
- Estimated impact of shocks on food security (exact coefficients from Annex Table 2.6):
  - Shock coefficient (effect on probability of reporting food security):
    - Ethiopia: -0.229*** ( -0.011)
    - Malawi: -0.185*** ( -0.023)
    - Mali: -0.110*** ( -0.016)
    - Niger: -0.168*** ( -0.016)
    - Tanzania: -0.059*** ( -0.017)
  - Household wealth index:
    - Ethiopia: 0.102*** ( -0.006)
    - Malawi: 0.152*** ( -0.004)
    - Mali: 0.049*** ( -0.007)
    - Niger: 0.137*** ( -0.008)
    - Tanzania: 0.090*** ( -0.023)
  - Constant terms:
    - Ethiopia: 0.870*** ( -0.013)
    - Malawi: 0.531*** ( -0.028)
    - Mali: 0.762*** ( -0.019)
    - Niger: 0.614*** ( -0.02)
    - Tanzania: 0.700*** ( -0.018)
  - Adjusted R-squared:
    - Ethiopia 0.1520; Malawi 0.2040; Mali 0.1630; Niger 0.2390; Tanzania 0.075
  - Number of observations:
    - Ethiopia 4954; Malawi 12447; Mali 38133; Niger 61336; Tanzania 173352
- Interpretation of main results:
  - Shocks are associated with a 5 to 20 percent higher incidence of food insecurity across the five countries, with the estimated impact stronger for more recently experienced shocks (Ethiopia, Malawi and Niger).
  - A one standard deviation increase in the household wealth index increases the probability of reporting food security by 5 to 15 percent.
- Coping strategies regression (sample limited to households reporting a shock):
  - Dependent variable: food security among shocked households.
  - Regressors include access to banking, mobile communication devices, robust housing, improved sanitary facilities, literacy of household head.
  - Results (Annex Table 2.7 summary): results broadly support importance of access to finance, mobile communication, robust housing, improved sanitation and literacy in helping build buffers against food insecurity for households facing climate-related shocks.
  - Wealth index remains positively associated with food security among shocked households.

*Source: IMF staff estimates, Annex Chapter 2—Adapting to Climate Change in Sub-Saharan Africa (onlineannexch2 - Section 1).*

### onlineannexch2 - Section 2

### Household determinants of food security (Dependent variable: Food security conditional on experiencing a shock)
- Estimation details:
  - Includes region and rural dummies. Robust standard errors in parenthesis. ***, **, and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively.
- adj_R-squared by country:
  - Ethiopia: 0.136
  - Malawi: 0.206
  - Mali: 0.17
  - Niger: 0.26
  - Tanzania: 0.09
- Number of observations (N):
  - Ethiopia: 2,836
  - Malawi: 1,204
  - Mali: 3,284
  - Niger: 7,182
  - Tanzania: 2,157
- Constant terms:
  - Ethiopia: 0.370*** ( -0.031)
  - Malawi: 0.150*** ( -0.021)
  - Mali: 0.520*** ( -0.038)
  - Niger: 0.326*** ( -0.04)
  - Tanzania: 0.421*** ( -0.045)
- Key estimated coefficients (coefficient (robust standard error)):
  - Household wealth index:
    - Ethiopia: 0.070*** ( -0.011)
    - Malawi: 0.115*** ( -0.006)
    - Mali: 0.031*** ( -0.01)
    - Niger: 0.164*** ( -0.021)
    - Tanzania: 0.056* ( -0.03)
  - Bank account:
    - Ethiopia: 0.082*** ( -0.022)
    - Malawi: 0.006 ( -0.01)
    - Mali: 0.072*** ( -0.025)
    - Niger: 0.074** ( -0.037)
    - Tanzania: (no coefficient reported)
  - Mobile phone:
    - Ethiopia: 0.138*** ( -0.021)
    - Malawi: 0.036*** ( -0.009)
    - Mali: 0.015 ( -0.025)
    - Niger: 0.060*** ( -0.008)
    - Tanzania: 0.064** ( -0.03)
  - Robust housing:
    - Ethiopia: 0.130*** ( -0.014)
    - Malawi: 0.191*** ( -0.006)
    - Mali: 0.078** ( -0.011)
    - Niger: 0.061 ( -0.018)
    - Tanzania: 0.105*** ( -0.013)
  - Improved sanitation:
    - Ethiopia: 0.054*** ( -0.019)
    - Malawi: 0.061*** ( -0.022)
    - Mali: 0.065*** ( -0.02)
    - Niger: 0.027 ( -0.042)
    - Tanzania: -0.001 ( -0.026)
  - Literacy:
    - Ethiopia: 0.048** ( -0.021)
    - Malawi: 0.049*** ( -0.009)
    - Mali: 0.02 ( -0.019)
    - Niger: -0.005 ( -0.024)
    - Tanzania: 0.068*** ( -0.026)

### Household surveys on crop yields (Data and descriptive findings)
- Dataset:
  - Nationally representative household surveys from Ethiopia, Rwanda, and Uganda collected with the World Bank LSMS.
  - Objective: assess extent to which weather affects crop yields and family consumption; assess farmer protective actions (erosion mitigation, access to investment finance, agricultural inputs).
- Selected descriptive findings from Annex Table 2.8:
  - Percent of households reporting crop damage:
    - Rwanda: 38
    - Ethiopia: 38
  - Of households reporting crop damage, causes (percent of households reporting crop damage):
    - Change of climate (too much/little rain):
      - Rwanda: 72.2
      - Ethiopia: 66
    - Landslide:
      - Rwanda: 6
    - Erosion/insects/crop disease:
      - Rwanda: 10.2
      - Ethiopia: 21.8
    - Destructive rains/hail:
      - Rwanda: 5.5
      - Ethiopia: 5.2
    - Loss of soil fertility:
      - Rwanda: 6.1
      - Ethiopia: 2.7
    - Other:
      - Rwanda: 4.3
  - Narrative: Surveys for Ethiopia and Rwanda show almost 40 percent of farmers have been affected by crop damage during the agricultural season associated with the survey. 70 percent of the farmers attribute the crop damage to climate change with an additional 10-20 percent associating crop damage with erosion and crop disease.

### Econometric results on determinants of crop yields (Annex Table 2.9)
- Dependent variable: Crop yield in log kilograms per hectare.
- Control variables (not reported): age, gender, plot ownership, soil quality, elevation, slope, the use of plot (partial versus whole), log surface area of plot and farm, type of settlement (urban versus rural), literacy, educational level, household size, primary occupation, and ownership of non-agricultural enterprises.
- Robust standard errors in parenthesis. ***, ** and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively.
- R-squared and number of observations:
  - Ethiopia: R-squared 0.088; Number of observations 836
  - Rwanda: R-squared 0.344; Number of observations 87338
- Estimated coefficients (coefficient (robust standard error)):
  - Ethiopia (1):
    - Fertilizer: 0.728*** ( -0.083)
    - Insecticide: 0.863*** ( -0.231)
    - Improved seeds: 1.225*** ( -0.132)
    - Irrigation: 1.246*** ( -0.241)
    - Protection against erosion: 0.525*** ( -0.109)
    - Access to finance: -0.092 ( -0.094)
  - Rwanda (2):
    - Fertilizer: -0.148 ( -0.093)
    - Insecticide: 0.387*** ( -0.044)
    - Improved seeds: 0.06 ( -0.041)
    - Irrigation: -0.068 ( -0.085)
    - Protection against erosion: -0.541 ( -0.04)
    - Access to finance: 0.091*** ( -0.047)
- Key interpretation points:
  - Use of inputs is important for raising crop yields.
  - Insecticide and improved seeds are significant determinants in both Ethiopia and Rwanda.
  - Fertilizer is important in Ethiopia.
  - Irrigation plays a major role in improving crop yields in Ethiopia (positive and significant); coefficients differ by country.
  - Access to finance shows different signs across countries (Ethiopia: negative coefficient not significant; Rwanda: 0.091***).

*IMF Regional Economic Outlook: Sub‑Saharan Africa — April 2020, Annex Section 2*

### Section 1

### onlineannexch2 - Section 1

### 2.1 Nightlight regression — data and methods
- Data sources and aggregation:
  - Temperature and precipitation: Harris and others (2014), CRU TS v. 4.03, 1901-2018, 0.5°ൈ0.5° grid, monthly frequency. For analysis, temperature and precipitation are averaged at the state level, weighted by the area of each grid within state boundaries.
  - Nightlights: NOAA Visible Infrared Imaging Radiometer Suite (VIIRS) monthly composites of average radiance images, 15"ൈ15" grid; nightlights data used between 2013 and 2018; nightlights summed to the state level as a proxy for economic activity.
  - Forest change: Hanson and others (2013) and project update (University of Maryland), 2000–2018, 30-meter resolution.
  - Water surface change: Pekel and others (2016), global surface water occurrence and intensity change from three million Landsat images at 30-meter resolution.
- Econometric specification:
  - Dependent variable: Δz_{i,t} = z_{i,t} − z_{i,t−1}, where z_{i,t} is logarithm of nightlights in state i at time t; dependent variable interpreted as nightlight growth.
  - Regressors: c_{i,t}, climate variable measured in deviations from its past 30-year average for the same month.
  - Fixed effects: μ_i (state) and α_t (time).
  - Estimation: ordinary least squares for the period between 2013 and 2018, with standard errors clustered at the state level.
  - Units: temperature measured in Celsius degree; precipitation measured in 10 millimeters.
  - Interpretation: β_j measures impact of temperature or precipitation deviations from historical average on local economic growth.

### 2.1 Nightlight regression — key results
- Temperature deviation effects (monthly nightlight growth):
  - A 1°C deviation is associated with about 2 percentage points reduction in monthly growth of nightlights for the world on average.
  - Sub-Saharan Africa (SSA): 4.2 percentage points reduction.
  - Sahel region: 6.4 percentage points reduction.
  - The SSA result is almost entirely driven by Western Africa.
- Regression coefficient estimates (selected, exact):
  - Lagged nightlights coefficient (dependent variable: Nightlight growth):
    - World: -0.771*** (standard error 0.007)
    - SSA: -0.786*** (0.010)
    - Sahel: -0.827*** (0.025)
    - EMDE: -0.743*** (0.007)
    - Eastern Africa: -0.817*** (0.016)
    - Southern Africa: -0.755*** (0.033)
    - Western Africa: -0.774*** (0.017)
    - Central Africa: -0.805*** (0.022)
  - Temperature deviation:
    - World: -0.019*** (0.002)
    - SSA: -0.042*** (0.012)
    - Sahel: -0.064** (0.031)
    - EMDE: -0.026*** (0.002)
    - Eastern Africa: -0.006 (0.049)
    - Southern Africa: 0.028 (0.046)
    - Western Africa: -0.052** (0.026)
    - Central Africa: -0.030 (0.051)
  - Precipitation deviation:
    - World: -0.006*** (0.001)
    - SSA: -0.008*** (0.001)
    - Sahel: -0.012** (0.006)
    - EMDE: -0.008*** (0.001)
    - Eastern Africa: -0.002 (0.002)
    - Southern Africa: -0.010 (0.006)
    - Western Africa: -0.008*** (0.003)
    - Central Africa: -0.005 (0.004)
- Additional diagnostics and sample sizes (exact):
  - Observations (temperature and precipitation regressions): World 100,475; SSA 21,125; Sahel 2,953; EMDE 80,309; Eastern Africa 8,665; Southern Africa 1,599; Western Africa 8,178; Central Africa 4,138.
  - R-squared (temperature): World 0.4820; SSA 0.5900; Sahel 0.6900; EMDE 0.4780; Eastern Africa 0.7870; Southern Africa 0.7640; Western Africa 0.7590; Central Africa 0.712
  - Number of states: World 222; SSA 347; Sahel 364; EMDE 170; Eastern Africa 421; Southern Africa 636; Western Africa 161; Central Africa 199
- Duration of effects:
  - The impact of temperature and precipitation deviations on economic activity is temporary, lasting for one or two months depending on the number of lags introduced in the regression.
  - Some evidence that precipitation deviations have a larger and longer impact during crops’ growing seasons.

### 2.2 Country-level panel regression — data and design
- Dataset and aggregation:
  - Panel of 181 countries from 1960 to 2018.
  - Yearly data aggregated into 5-year windows (final dataset has 12 five-year periods).
- Disaster proxies:
  - Intensity proxy: proportion of disruptive disasters in the 5-year period (intensity dummy = 1 if disaster in year t affects over 0.01 percent of population, where ratio (death + 0.3*affected)/population is used).
  - Frequency proxy: average effects of disasters via average ratio of deaths to population.
- Control variables (impact analysis, “sparse” growth model) and policy-response variables:
  - Impact-analysis controls (selected): Log of per capita GDP (Real per capita GDP, PPP (WEO)), Education (Gross rate of enrollment in the secondary (WDI)), Investment (Gross fixed capital formation, percent of GDP (WEO)), Government consumption (Percent of per capita GDP government consumption (PWT)), Inflation (Consumer Prices, period average, percent change (WEO)), Trade openness (Ratio (Import+Export)-GDP (WEO)), Change in terms of trade (Change, ratio price export-price import (PWT))
  - Policy-response variables: Telecommunication (Mobile cellular subscriptions per 100 people (WDI)), Financial depth (Domestic credit to private sector, percent of GDP (WDI)), Education (Gross rate of enrollment in the secondary (WDI)), Health (Life expectancy at birth (WDI)), Agri. Machinery (Agricultural machinery, total tractors (WDI)), Electricity (Access to electricity, percent of population (WDI))
  - Other variables listed with no significant effect detected: Irrigation, Sanitation, Quality of fiscal policy, Roads
- Impact model estimated:
  - Dependent variable: Gr_{i,p} = average per capita growth over 5-year period p.
  - Regressors include lagged log GDP, intensity and frequency disaster proxies by type j, additional controls X_{i,p}, time fixed effects c_p, country fixed effects d_i.
- Policy-response model:
  - Interaction specification: assesses whether policy variable z_{i,p} mitigates growth impact of disasters by including Distasters * z_{i,p} interaction; b1 negative and b2 positive interpreted as policy improving resilience if b2 significant and positive.
  - Annex Table 2.5 reports changes in SSA per capita growth if average conditions improve to EMDE levels.

### 2.2 Country-level panel regression — selected econometric results
- Selected coefficients from Annex Table 2.4 (exact values reported):
  - Log of per capita GDP:
    - Column (1) EMDEs: -2.09** (0.98)
    - Column (2) EMDEs: -2.03*** (0.65)
    - Column (3) EMDEs: -2.52** (0.99)
    - Column (4) EMDEs: -1.06 (1.16)
    - Additional columns for SSA: -2.37 (1.43); -1.57 (1.37); -1.51 (1.07); 6.72 (3.94)
  - Intensity drought:
    - EMDEs: -2.66*** (0.74)
    - SSA: -7.81*** (2.16)
  - Frequency drought:
    - EMDEs: -0.34*** (0.05)
    - SSA: -0.41*** (0.04)
  - Intensity flood:
    - EMDEs: -1.31*** (0.44)
    - SSA: -1.49** (0.54)
  - Frequency flood:
    - EMDEs: -0.06** (0.03)
    - SSA: -0.18** (0.08)
  - Intensity epidemic and frequency epidemic: coefficients reported but not statistically significant in shown columns (Intensity epidemic -0.20 (0.33); Frequency epidemic -0.00 (0.13) for EMDEs; Intensity epidemic -0.53 (0.41); Frequency epidemic -0.01 (0.16) for SSA)
  - Frequency storm: EMDEs -0.19** (0.08); SSA  -12.26 (11.27)
  - Education:
    - EMDEs: -0.07** (0.03), -0.07*** (0.02), -0.05* (0.03), -0.07*** (0.02)
    - SSA columns include: -0.05 (0.06), -0.07* (0.04), -0.06 (0.05), -0.16 (0.16)
  - Investment:
    - EMDEs: 0.03*** (0.01), 0.09*** (0.02), 0.07*** (0.02), 0.02 (0.04)
    - SSA: 0.03*** (0.01), 0.05* (0.02), 0.05** (0.02), 0.09 (0.06)
  - Inflation: numbers vary by column; EMDEs show small coefficients with significance reported (e.g., 0.00*** (0.00) and negative variants)
  - Intercept examples: 21.37*** (7.53), 20.31*** (5.23), 20.61** (7.98), 14.34 (10.14); SSA intercepts reported as 26.85*** (9.38), 12.29 (9.11), 7.99 (8.79), -46.39 (27.15)
- Notes:
  - Dependent variable is per capita GDP growth in the following year.
  - Robust standard errors in parenthesis.
  - Country and year fixed effects included in reported specifications.
  - Number of observations vary across columns, e.g., EMDEs columns: 2115, 1331, 2323, 2511; SSA columns: 316, 315, 867 (as reported in the table).

### 2.3 Household surveys on food security — data and descriptive findings
- Surveys and countries:
  - Nationally representative household surveys in Ethiopia, Malawi, Mali, Niger, and Tanzania, collected with the World Bank Living Standards Measurement Study (LSMS).
  - Households report shocks experienced in the last 1 to 5 years depending on country (Ethiopia, Malawi, Niger: last 12 months; Mali: last 3 years; Tanzania: last 5 years).
  - Analysis focuses on shocks likely to increase with climate variability, rising temperatures and sea levels.
- Prevalence of shocks (summary):
  - Households in Malawi reported the most shocks overall, followed by Mali, Tanzania, Ethiopia and Niger.
  - In all countries, major price changes (increases in food or input prices or decreases in crop output prices) affected over 30 percent of households.
  - Droughts, water shortages and irregular rainfall also affected a large number of households.
  - Severe illness, crop damage, and loss of livestock are other frequently experienced shocks; flood and conflict or violence less frequent; land loss and fire affected a small percentage of households (land loss not asked in Niger and Mali; fire asked only in Ethiopia).
- Coping strategies (Annex Figure 2.2 summary):
  - Most common coping strategy: draw on savings.
  - Other common strategies: relying on friends and family, selling assets, adjusting consumption.
  - Government support reported by 17 percent of households in Ethiopia and 15 percent in Mali.
  - Additional strategies reported: migration, reducing health and education expenditure, taking out a loan, seeking additional work, NGO support.
- Selected percentages from figure (exact values shown in figure headers and bars):
  - (Figure bars list percentages per strategy by country; the table in text highlights government support: Ethiopia 17 percent; Mali 15 percent.)

### 2.3 Household surveys — econometric results on food security
- Main food-security regression (linear probability model via OLS):
  - Dependent variable: Food security = 1 if household reports not recently worried about having enough to eat.
  - Key regressor: shock indicator = 1 if household reports experiencing any listed shock.
  - Controls: household wealth index (constructed via principal component analysis of durable goods, standardized mean 0 SD 1), region dummies, rural dummy.
- Estimated impact of shocks on food security (exact coefficients from Annex Table 2.6):
  - Shock coefficient (effect on probability of reporting food security):
    - Ethiopia: -0.229*** (standard error -0.011)
    - Malawi: -0.185*** (-0.023)
    - Mali: -0.110*** (-0.016)
    - Niger: -0.168*** (-0.016)
    - Tanzania: -0.059*** (-0.017)
  - Household wealth index:
    - Ethiopia: 0.102*** (-0.006)
    - Malawi: 0.152*** (-0.004)
    - Mali: 0.049*** (-0.007)
    - Niger: 0.137*** (-0.008)
    - Tanzania: 0.090*** (-0.023)
  - Constant terms reported:
    - Ethiopia: 0.870*** (-0.013)
    - Malawi: 0.531*** (-0.028)
    - Mali: 0.762*** (-0.019)
    - Niger: 0.614*** (-0.02)
    - Tanzania: 0.700*** (-0.018)
  - Adjusted R-squared (exact):
    - Ethiopia 0.1520; Malawi 0.2040; Mali 0.1630; Niger 0.2390; Tanzania 0.075
  - Number of observations (exact):
    - Ethiopia 4954; Malawi 12447; Mali 38133; Niger 61336; Tanzania 173352 (as reported in table row: Number of observations 4954 12447 38133 61336 173352)
- Interpretation of main results:
  - Shocks are associated with a 5 to 20 percent higher incidence of food insecurity across the five countries, with the estimated impact stronger for more recently experienced shocks (Ethiopia, Malawi and Niger).
  - A one standard deviation increase in the household wealth index increases the probability of reporting food security by 5 to 15 percent (Annex Table 2.6).
- Coping strategies regression (sample limited to households reporting a shock):
  - Dependent variable: food security among shocked households.
  - Regressors X include access to banking, mobile communication devices, robust housing, improved sanitary facilities, literacy of household head (as available).
  - Results (Annex Table 2.7 summary): although not all variables significant across surveys, results broadly support importance of access to finance, mobile communication, robust housing, improved sanitation and literacy in helping build buffers against food insecurity for households facing climate-related shocks.
  - Wealth index remains positively associated with food security among shocked households.

_Italic line: Source: IMF staff estimates, Annex Chapter 2—Adapting to Climate Change in Sub-Saharan Africa (onlineannexch2 - Section 1)._

### Section 2

### onlineannexch2 - Section 2

### Household determinants of food security (Dependent variable: Food security conditional on experiencing a shock)
- Estimation includes region and rural dummies. Robust standard errors in parenthesis. ***, **, and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively.
- adj_R-squared by country:
  - Ethiopia: 0.136
  - Malawi: 0.206
  - Mali: 0.17
  - Niger: 0.26
  - Tanzania: 0.09
- Number of observations (N):
  - Ethiopia: 2,836
  - Malawi: 1,204
  - Mali: 3,284
  - Niger: 7,182
  - Tanzania: 2,157
- Constant terms:
  - Ethiopia: 0.370*** (-0.031)
  - Malawi: 0.150*** (-0.021)
  - Mali: 0.520*** (-0.038)
  - Niger: 0.326*** (-0.04)
  - Tanzania: 0.421*** (-0.045)

Key estimated coefficients (coefficient (robust standard error)):

- Household wealth index:
  - Ethiopia: 0.070*** (-0.011)
  - Malawi: 0.115*** (-0.006)
  - Mali: 0.031*** (-0.01)
  - Niger: 0.164*** (-0.021)
  - Tanzania: 0.056* (-0.03)

- Bank account:
  - Ethiopia: 0.082*** (-0.022)
  - Malawi: 0.006 (-0.01)
  - Mali: 0.072*** (-0.025)
  - Niger: 0.074** (-0.037)
  - Tanzania: (no coefficient reported)

- Mobile phone:
  - Ethiopia: 0.138*** (-0.021)
  - Malawi: 0.036*** (-0.009)
  - Mali: 0.015 (-0.025)
  - Niger: 0.060*** (-0.008)
  - Tanzania: 0.064** (-0.03)

- Robust housing:
  - Ethiopia: 0.130*** (-0.014)
  - Malawi: 0.191*** (-0.006)
  - Mali: 0.078** (-0.011)
  - Niger: 0.061 (-0.018)
  - Tanzania: 0.105*** (-0.013)

- Improved sanitation:
  - Ethiopia: 0.054*** (-0.019)
  - Malawi: 0.061*** (-0.022)
  - Mali: 0.065*** (-0.02)
  - Niger: 0.027 (-0.042)
  - Tanzania: -0.001 (-0.026)

- Literacy:
  - Ethiopia: 0.048** (-0.021)
  - Malawi: 0.049*** (-0.009)
  - Mali: 0.02 (-0.019)
  - Niger: -0.005 (-0.024)
  - Tanzania: 0.068*** (-0.026)

### Household surveys on crop yields (Data and descriptive findings)
- Dataset: nationally representative household surveys from Ethiopia, Rwanda, and Uganda collected with the World Bank LSMS.
- Objective: assess extent to which weather adversely affects crop yields and family consumption; assess farmer protective actions (erosion mitigation, access to investment finance, agricultural inputs).
- Selected descriptive findings from Annex Table 2.8:
  - Percent of households reporting crop damage:
    - Rwanda: 38
    - Ethiopia: 38
  - Of households reporting crop damage, causes (percent of households reporting crop damage):
    - Change of climate (too much/little rain):
      - Rwanda: 72.2
      - Ethiopia: 66
    - Landslide:
      - Rwanda: 6
    - Erosion/insects/crop disease:
      - Rwanda: 10.2
      - Ethiopia: 21.8
    - Destructive rains/hail:
      - Rwanda: 5.5
      - Ethiopia: 5.2
    - Loss of soil fertility:
      - Rwanda: 6.1
      - Ethiopia: 2.7
    - Other:
      - Rwanda: 4.3
  - Narrative: Surveys for Ethiopia and Rwanda show almost 40 percent of farmers have been affected by crop damage during the agricultural season associated with the survey. 70 percent of the farmers attribute the crop damage to climate change with an additional 10-20 percent associating crop damage with erosion and crop disease.

### Econometric results on determinants of crop yields (Annex Table 2.9)
- Dependent variable: Crop yield in log kilograms per hectare.
- Control variables (not reported): age, gender, plot ownership, soil quality, elevation, slope, the use of plot (partial versus whole), log surface area of plot and farm, type of settlement (urban versus rural), literacy, educational level, household size, primary occupation, and the ownership of non-agricultural enterprises.
- Robust standard errors in parenthesis. ***, ** and * indicate statistical significance at the 1, 5, and 10 percent levels, respectively.
- R-squared and number of observations:
  - Ethiopia: R-squared 0.088; Number of observations 836
  - Rwanda: R-squared 0.344; Number of observations 87338 (as printed)

Estimated coefficients (coefficient (robust standard error)):

- Ethiopia (1):
  - Fertilizer: 0.728*** (-0.083)
  - Insecticide: 0.863*** (-0.231)
  - Improved seeds: 1.225*** (-0.132)
  - Irrigation: 1.246*** (-0.241)
  - Protection against erosion: 0.525*** (-0.109)
  - Access to finance: -0.092 (-0.094)

- Rwanda (2):
  - Fertilizer: -0.148 (-0.093)
  - Insecticide: 0.387*** (-0.044)
  - Improved seeds: 0.06 (-0.041)
  - Irrigation: -0.068 (-0.085)
  - Protection against erosion: -0.541 (-0.04)
  - Access to finance: 0.091*** (-0.047)

- Key interpretation points from the results:
  - Use of inputs is important for raising crop yields.
  - Both insecticide and improved seeds are significant determinants in both Ethiopia and Rwanda (noting coefficients and significance levels).
  - Fertilizer is important in Ethiopia.
  - Irrigation plays a major role in improving crop yields in Rwanda (as asserted in the text; coefficients indicate irrigation positive and significant in Ethiopia, negative and not significant in Rwanda).
  - Access to finance shows different signs across countries (Ethiopia: negative coefficient not significant; Rwanda: 0.091***).

*IMF Regional Economic Outlook: Sub‑Saharan Africa — April 2020, Annex Section 2*

---


_Source: https://www.imf.org/-/media/files/publications/reo/afr/2020/april/english/onlineannexch2.pdf_
