## Coping with Climate Shocks: Food Security in a Spatial Framework

## Source details

**Canonical URL:** [Coping with Climate Shocks: Food Security in a Spatial Framework](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023166-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023166-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023166-print-pdf.pdf.json)

---

### Introduction and research objectives
- 2021: "2.3 billion people suffered from moderate or severe food insecurity" (FAO, 2022).
- Research questions:
  - What is the impact of climate shocks on food security?
  - Which types of households are most vulnerable?
  - What factors build resilience, and what are the macroeconomic consequences of household coping strategies?
- Three empirical–quantitative exercises:
  - Document household and market-level responses to climate shocks in Nepal.
  - Design a quantitative spatial general equilibrium model with multiple locations and heterogeneous households.
  - Calibrate the model to 51 districts in Nepal and quantify local and macroeconomic impacts of historical district-level climate damages (2011-2022), plus counterfactuals on economic integration.

### Empirical evidence and data sources
- Definition: climate shock = incidence of flood, landslide, drought, or storm recorded as damages in the district by the Government of Nepal.
- Primary data sources:
  - Building Information Platform Against Disaster (BIPAD): municipal-level disaster events in Nepal, 2011–2022.
  - World Food Programme (WFP) Global Food Prices Database: monthly rice and wheat prices, 2001–2021, 42 markets.
  - Household Risk and Vulnerability Survey (HRVS): three-year longitudinal survey covering 6,000 households and 400 communities in non-metropolitan Nepal (2016–2018).
- Key empirical relationships:
  - Climate shocks geographically dispersed; depend on district geographic features.
  - At household level, climate shocks correlate with:
    - significantly lower yields;
    - lower farm income;
    - increased food insecurity (each additional climate shock correlated with an 11 percent increase in likelihood of being food insecure).
  - Household coping strategies:
    - shift from non-food to higher food expenditure;
    - increased migration and remittance reliance;
    - lower savings and capital.
  - Market-level: climate shocks associated with significant food price increases, approximately twice as large for remote regions.

### Modeling approach and calibration
- Model features:
  - Quantitative spatial general equilibrium model with multiple locations and heterogeneous households.
  - Captures: subsistence requirements, income diversification (farm and off-farm), temporary migration, endogenous coping channels (budget reallocation to food, asset sales, migration/remittances, inter-regional food imports).
  - Heterogeneity: household wealth and spatial frictions (costs of migrating and trading goods).
  - Remote locations face higher costs to import food and to migrate; poorer households closer to subsistence hold smaller capital buffers and dissave disproportionately.
- Calibration overview:
  - Calibrated to 51 districts (data available for 51 of 77 districts; Kathmandu, Lalitpur and Bhaktapur merged into Kathmandu Valley).
  - Uses household panel surveys, censuses, market price data, and recorded incidence of shocks.
  - Procedure to convert disaster records into productivity shocks:
    1. Estimate impulse response of local food prices to climate shocks using market price and disaster data.
    2. Normalize BIPAD economic damages by sample mean and multiply by average price impact to estimate each shock’s price impact.
    3. Convert estimated price changes into productivity shocks by estimating price–productivity elasticities via model simulations of local idiosyncratic shocks.

### Key quantitative findings (historical climate shocks, 2011–2020 unless specified)
- Aggregate and household impacts:
  - Annual welfare losses: "3.1 percent" to the average rural household.
  - Rate of undernourishment (share consuming less than 2,200 daily calories) raised by "2.8 percent" above absence of climate shocks.
  - Rural GDP in Nepal estimated to be "2.3 percent" lower due to climate shocks.
- Heterogeneity across time and districts:
  - Average annual aggregate impact relatively stable; substantial cross-district heterogeneity.
  - 2019 example: 95th percentile district experienced "almost 18 percent lower agricultural yields" corresponding to "a roughly 13 percent loss in welfare and 9 percent increase in undernourishment."
- Role of remoteness and undernourishment:
  - Top "30 percent" most remote locations: average welfare losses of "5.9 percent" and increase of "4 percent" in undernourishment rate.
  - For undernourished households, shocks cost "4.3 percent" of annual welfare.
- Empirical price effects:
  - Contemporaneous 1.9 percent increase in food prices after a climate shock; rises to 3.3 percent over subsequent 3 months; prices remain elevated for 6 months.
  - Remote districts: maximum price increase "7 percent" vs "3 percent" in connected districts; prices remain statistically above zero for 6 months in remote vs 3 months in connected districts.

### External calibration highlights and shock-to-productivity mapping
- Key externally calibrated parameters:
  - σ = 6; λ = 3.4; εF = 0.2; εN = 1.19; γ = 1.4 (Table 1); η = 0.4; kcal = 2200.
  - Farm input shares: αh = 0.50, αk = 0.15, αl = 0.35; δ = 0.05.
  - Distance decay rate ζ = 0.31.
- Population and sample:
  - Aggregated rural population across all 77 districts = 80.6 percent.
  - Model calibrated for 51 districts (roughly 78 percent of total Nepal population); if Kathmandu Valley assigned urban status, rural share = 78.4 percent.
- Productivity–price elasticity calibration:
  - Empirical peak effect from BIPAD: 3 percent increase in local food prices (peak 2 months after shock).
  - Simulated random idiosyncratic productivity shocks: median district elasticity implies 1 percent price increase ↔ 5.4 percent drop in productivity.
  - Therefore average climate shock (3 percent price increase) corresponds to a 16.2 percent reduction in productivity in the median district.
  - Average per capita damages in BIPAD panel = 37 Nepalese Rupees per capita; shocks scaled so TFP loss of event with average per capita damages equals TFP loss implied by 3 percent price effect.

### Model-simulated impulse responses (representative 10 percent temporary productivity shock)
- Simulation: 10 percent temporary negative productivity shock in a single district.
- Average district impulse responses:
  - Impact food price increase = 3.8 percent (impact; cross-district variation).
  - Average district increases share of food imported by almost 15 percent in reaction to shock.
  - Food consumption falls on average by 2.2 percent (equivalent to 0.88 percent fall in calorie consumption).
  - Welfare losses: undernourished = 3.5 percent vs average household = 2.1 percent.
  - Households increase migration: share of migrants in household rises by 5 percent on average.
- Mechanisms:
  - Districts importing larger food shares experience smaller price increases.
  - Less remote districts substitute to other food suppliers; undernourished households shift away from non-food consumption to limit food consumption fall but face larger welfare losses due to smaller capital buffers.

### Average Annual Climate Damages (2011–2022)
- Agricultural yields: "5.75 percent" lower on average annually across all districts due to climate shocks.
- Maximum average impact on yields: "-11 percent" in 2019.
- Average annual loss to GDP: "2.1 percent".
- Average welfare losses: "3.1 percent".
- Comparable benchmark: average annual loss of "2.65 percent" due to floods reported in the United Nations’ WESR Risk Platform.
- Rise in rate of undernourishment: "2.8 percent" due to climate shocks over the last decade.
- District/temporal heterogeneity:
  - Median household: maximum annual impact on yields of "8 percent" in 2019 and minimum of "2 percent" in 2012.
  - Eleven districts estimated to suffer average annual farm productivity losses of more than "10 percent".
  - Roughly a quarter of districts experienced increases in undernourishment rates of more than "5 percent".
- Determinants of vulnerability:
  - Remote households suffer welfare losses "2.04 times" those of non-remote districts.
  - Undernourished individuals have welfare losses of "4.3 percent", which are "1.54 times" those of food secure individuals.
  - Regression (controlling for shock size): being undernourished increases responsiveness of log utility to shocks from -0.013 to -0.020 (54 percent increase in magnitude); households in remote districts exhibit responsiveness of -0.023 (54 percent larger than non-remote).

### Policy simulations and distributional effects (infrastructure / lower spatial frictions)
- Counterfactuals:
  - "10 percent lower trade costs" or "10 percent lower migration costs".
- Aggregate impacts:
  - Welfare impact of shocks reduced on average from "3.3 percent" to "approximately 2.7 percent" (a reduction by a factor of "0.18").
  - Rates of undernourishment reduced from "2.8 percent" to "roughly 2 percent" (reduction by a factor of almost a third).
- Distributional and spillover outcomes:
  - "18 percent" of districts see an increase in undernourishment under lower trade costs.
  - "16 percent" of districts see an increase in undernourishment under lower migration costs.
  - Interpretation: improved integration can generate cross-district spillovers producing both positive and negative effects; some poorer households within districts may be harmed by infrastructure improvements due to market and wage dynamics.

### Determinants of undernourishment in simulated steady state
- Regression of simulated district-level undernourishment on characteristics:
  - A 1 percent higher agricultural yield → on average 0.26 percent fewer undernourished households.
  - Remoteness and agricultural productivity together explain 73 percent of variation in undernourishment.
  - Greater trade and migration access associated with lower undernourishment (migration access becomes significant when controlling for agricultural yields).
- Interaction effects:
  - Undernourished households benefit more from access to trade, migration, and higher agricultural productivity than food secure households because they have lower capital buffers and rely more on coping mechanisms.

### Conclusions and policy implications
- Climate shocks already impose large negative impacts on GDP, nutrition, and welfare.
- Impacts disproportionately harm remote and food-insecure households.
- Poverty and food insecurity amplify shock impacts via reduced buffers and coping capacity.
- Policies lowering trade and migration costs can reduce average impacts from climate shocks by providing alternative income and affordable food, but may have adverse distributional effects and increase vulnerability for some households due to integration-induced spillovers.
- The quantitative spatial framework enables analysis of temporary shocks, their damage and persistence, and the role of trade, migration, and remittances in shaping outcomes.

*Source: wpiea2023166-print-pdf*

### References .............................................................................................................

### Coping with Climate Shocks: Food Security in a Spatial Framework

### Introduction and research objectives
- In 2021, "2.3 billion people suffered from moderate or severe food insecurity" (FAO, 2022).
- Research questions:
  - What is the impact of climate shocks on food security?
  - Which types of households are most vulnerable?
  - What factors build resilience, and what are the macroeconomic consequences of household coping strategies?
- Three empirical–quantitative exercises:
  - Document household and market-level responses to climate shocks in a climate-vulnerable developing country (Nepal).
  - Design a quantitative spatial general equilibrium model with multiple locations and heterogeneous households.
  - Calibrate the model to 51 districts in Nepal and quantify local and macroeconomic impacts of historical district-level climate damages (2011-2022), plus counterfactuals on economic integration.

### Empirical evidence and data sources
- Definition: a climate shock is the incidence of flood, landslide, drought, or storm which is recorded damages in the district by the Government of Nepal.
- Primary data sources used:
  - Building Information Platform Against Disaster (BIPAD) database: spatially-disaggregated historical record of natural disaster events in Nepal from 2011 to 2022 at municipality-level, recording time, location, and measures of damages.
  - World Food Programme (WFP) Global Food Prices Database: monthly prices of rice and wheat in 2001-2021 for 42 Nepalese markets.
  - Household Risk and Vulnerability Survey: three-year longitudinal survey covering 6,000 households and 400 communities in non-metropolitan areas of Nepal.
- Key empirical relationships documented:
  - Climate shocks are geographically dispersed and depend on district geographic features.
  - At household level, climate shocks correlate with:
    - significantly lower yields,
    - lower farm income,
    - increased food insecurity.
  - Household coping strategies observed:
    - substitution away from non-food expenditure toward higher food expenditure,
    - increased migration,
    - lower savings and capital.
  - Market-level finding: climate shocks are associated with a significant increase in food prices, approximately twice as large for remote regions.

### Modeling approach and calibration
- Model features:
  - Quantitative spatial general equilibrium model with multiple locations and heterogeneous households.
  - Key features captured: subsistence requirements, income diversification through farm and off-farm income, temporary migration, endogenous household coping channels (budget reallocation to food, asset sales, migration/remittances, inter-regional food imports).
  - Heterogeneity sources: household wealth and spatial frictions (costs of migrating and trading goods).
  - Remote locations: higher costs to import food and to migrate; households in remote locations resort more to asset drawdown and face larger food price increases.
  - Poorer households: closer to subsistence, hold smaller capital buffers, more likely to dissave disproportionately.
- Calibration details:
  - Calibrated to 51 districts (data available for 51 of 77 districts; Kathmandu, Lalitpur and Bhaktapur merged into Kathmandu Valley).
  - Calibrated using household panel surveys, censuses, market price data, and recorded incidence of shocks at district level.
  - Conversion of disaster records to productivity shocks:
    1. Estimate impulse response function of local food prices to climate shock occurrences using market price and disaster data.
    2. Normalize reported economic damages from BIPAD by dividing by sample mean and multiply by average price impact to estimate each shock’s price impact.
    3. Convert estimated changes in food prices into productivity shocks by estimating the elasticity between prices and productivity through model simulations of local idiosyncratic shocks in each district.
  - Motivation: mitigate issues of missing data and under-reporting in damage databases by measuring losses through effects on food production and prices.

### Key quantitative findings
- Aggregate and household impacts (historical climate shocks, 2011-2020 unless otherwise specified):
  - Annual welfare losses of "3.1 percent" to the average rural household.
  - Rate of undernourishment (share consuming less than 2,200 daily calories) raised by "2.8 percent" above what it would have been absent climate shocks.
  - Rural GDP in Nepal estimated to be "2.3 percent" lower due to climate shocks.
- Heterogeneity across time and districts:
  - The average annual aggregate impact is relatively stable, with substantial cross-district heterogeneity.
  - Example for 2019: the 95th percentile district experienced "almost 18 percent lower agricultural yields" corresponding to "a roughly 13 percent loss in welfare and 9 percent increase in undernourishment."
- Role of remoteness and undernourishment:
  - Households in the top "30 percent" most remote locations suffer average welfare losses of "5.9 percent" and an increase of "4 percent" in the rate of undernourishment.
  - For undernourished households, shocks cost them "4.3 percent" of annual welfare.
- Factors that mitigate impacts:
  - Higher agricultural productivity, access to migration, and access to markets reduce climate shock impacts, particularly benefiting undernourished households closer to subsistence.

### Policy simulations and distributional effects
- Infrastructure and integration counterfactuals:
  - Under either "10 percent lower trade costs" or "10 percent lower migration costs":
    - The welfare impact of shocks is reduced on average from "3.3 percent" to "approximately 2.7 percent" (a reduction by a factor of "0.18").
    - Rates of undernourishment are reduced by a factor of almost a third, falling from "2.8 percent" to "roughly 2 percent" rise due to climate shocks.
  - Distributional outcomes and spillovers:
    - "18 percent" of districts see an increase in undernourishment under lower trade costs.
    - "16 percent" of districts see an increase in undernourishment under lower migration costs.
    - Interpretation: improved integration can generate cross-district spillovers producing both positive and negative effects; some poorer households within districts may be harmed by infrastructure improvements due to market and wage dynamics.

### Contributions to literature and framing
- Links to three strands:
  - Empirical literature on household coping mechanisms to adverse income shocks (migration/remittances, asset sales, off-farm labor), with novelty in embedding household decision-making within a general equilibrium spatial framework.
  - Quantitative spatial literature with realistic geography, trade and migration frictions, focusing here on temporary productivity shocks and household responses particularly relevant for low-income countries.
  - Food security literature linking nutrition to productivity, human capital, and forecasting of food crises; focus here on quantifying spatial spillovers of localized climate shocks.
- Methodological contribution:
  - Novel methodology to infer productivity shocks from observed movements in local food prices and disaster records, addressing under-reporting and missingness in reported damages.

### Paper organization (as presented)
- Section 2: empirical relationships to inform the model.
- Section 3: the model.
- Section 4: calibration procedure.
- Section 5: quantitative results.
- Section 6: conclusion.

*Source: wpiea2023166-print-pdf*

### 2.1  Data

### 2.1  Data

### Data sources and construction
- Household dataset:
  - Nepal Household Risk and Vulnerability Survey: three-year longitudinal household survey covering 6,000 households and 400 communities in non-metropolitan areas of Nepal.
  - Sample frame: all households in non-metropolitan areas per the 2010 Census definition, excluding households in the Kathmandu valley (Kathmandu, Lalitpur and Bhaktapur districts).
  - Sampling: 50 of the 75 districts in Nepal were selected with probability proportional to the number of households.
  - Survey content: detailed food and non-food expenses, sources of farm and off-farm income, migration, assets, exposure to shocks, community-level market prices of key consumption items.
  - Food insecurity measure: survey-estimated measure compiled from a food insecurity index score (Household Food Insecurity Access Scale).
  - Timing: data collected over three waves during 2016-2018; unit of observation is household-year.
- District / market dataset:
  - Monthly district-level market prices from the World Food Programme’s (WFP) market price database covering 41 Nepalese districts in 2001-2021.
  - Climate shocks from the Building Information Platform Against Disaster (BIPAD) database: historical record of natural disaster events in Nepal from 2011 to 2022 (earthquakes, floods, landslides, droughts, storms); records time and location at municipal level and measures of damages (fatalities, people affected, estimated cost).
  - Climate-shock sample restriction: include all climate shocks (landslides, storms, cold, heavy rainfall, and floods) with non-missing and positive estimated economic damages, yielding an annual average of 125 shocks per year and a district-level probability of experiencing at least one shock of 34 percent in any given year.
- Remoteness variable:
  - District-level remoteness generated using the population-weighted average distance to all districts.
  - Dummy remote defined as district in the top 30 percent of remoteness score.

### Context: exposure and economic relevance
- Climate and trends:
  - Average temperature in Nepal in the last decade is over 0.6 degrees Celsius higher than the baseline of 1950-1980.
  - Monsoon has become increasingly unpredictable; number of climate shocks related to floods, storms, and landslides has steadily risen.
- Economic structure and vulnerabilities:
  - Agricultural sector: 65 percent of total employment and 24 percent of GDP (ILO, 2020; Nepal National Statistics Office, 2023).
  - World Bank estimate in a severe climate change scenario: GDP would be 7 percent lower (World Bank, 2022).
  - Food security trends: prevalence of moderate or severe food insecurity in Nepal has grown every year between 2015 and 2020; food security currently affects more than one third of the population (Baptista et al., 2023).
  - Recent shock indicator: 18 percent of the population reported to not have consumed an adequate diet in October 2022 (World Food Programme, 2022).
- Migration and remittances:
  - Nepal among top 10 countries in the world in terms of remittance inflows; remittances accounted for 22.7 percent of GDP in 2020.
  - High domestic and temporary migration despite being among the 10 least urbanised countries.

### Key empirical relationships and quantified facts
- Empirical approach:
  - Use household-level correlations (OLS) between household outcomes and number of district-level climate shocks; unit caution: results presented as correlations, not causal estimates.
  - Local projection regressions (Jordà, 2005) used to estimate dynamic effects of shocks on district food prices with controls including three month lags of shock variables and fixed effects (district, product, time).
  - Remote districts defined as bottom 30 percent of districts in terms of population-weighted driving time to all other Nepalese districts.
- Fact 1 — Impacts on agricultural outcomes and food insecurity:
  - Each additional climate shock is correlated with statistically significant lower yields, and lower crop, livestock and farm income.
  - Each additional climate shock is correlated with an 11 percent increase in the likelihood of being food insecure and a higher food insecurity index score.
- Fact 2 — Household coping responses:
  - Households subject to an additional climate shock have:
    - higher expenditure on food;
    - lower expenditure on non-food consumption;
    - higher likelihood to migrate and a higher number of migrants;
    - greater reliance on remittances in income;
    - lower savings and lower capital equipment (consistent with drawing down savings to cover income shortfalls).
- Fact 3 — Heterogeneity by food insecurity status:
  - Among food insecure households, the average savings rate is 5 percent lower for those subject to a climate shock.
  - Food secure households subject to a climate shock do not have statistically different savings rates.
- Fact 4 — Price effects and remoteness:
  - Panel estimates for all districts show a contemporaneous 1.9 percent increase in food prices following a climate shock, which continues to rise over the subsequent 3 months to reach 3.3 percent, with prices remaining elevated for a total of 6 months.
  - Remote vs connected districts:
    - Remote districts see a maximum increase in prices of 7 percent compared to 3 percent in connected districts.
    - In remote districts prices remain statistically above zero for 6 months after the shock, compared to 3 months for connected districts.

### Modeling implications and environment
- Empirical relationships motivate a quantitative spatial model to reproduce observed facts and estimate general equilibrium impacts and household response mechanisms.
- Model environment (sketch from Figure 7):
  - Economy: n = 1,...,N locations; continuum of rural households ω ∈ Ω of size L_R and a mass of size L_U of identical urban households.
  - Goods: two goods (food and non-food); both traded subject to bilateral trade costs.
  - Household production and labor:
    - Rural households own and operate farms (food production) and allocate labor between farm work and supplying labor to non-food firms (off-farm).
    - Urban households supply labor exclusively to non-food sector and do not own farms.
    - Off-farm labor may be supplied in a different location (migration) subject to bilateral movement costs.
  - Markets and frictions:
    - Food production uses labor, land and capital; land productivity differs across locations.
    - Trade and migration costs include monetary and non-monetary frictions (tariffs, visas, search frictions, cultural/language barriers).
    - Households cannot borrow; capital for agricultural production financed through savings.
    - Prices are fully flexible; no money or exchange-rate considerations in the model.
  - Policy levers (modelled): infrastructure, trade and migration costs.
- Preference and savings specification:
  - Household utility: u_it(ω) = (c_it(ω) − C)^β k_it+1(ω)^(1−β), with 0 < β < 1, where C is a consumption subsistence requirement.
  - Per-period budget constraint: P_it c_it + P^k_it k_it+1 = y_it + (1−δ) P^k_it k_it, with 0 < δ < 1.
  - First-order solutions:
    - Optimal consumption: c_it = (1−β) C + β ( y_it + (1−δ) P^k_it k_it / P_it ).
    - Optimal bequest: k_it+1 = −(1−β) P_it / P^k_it C + (1−β) ( y_it + (1−δ) P_it k_it / P^k_it ).
  - Non-homotheticity and vulnerability:
    - Capital share in wealth: P^k_it k_it+1 / W_it = −(1−β) P_it C / W_it + (1−β), implying poorer households (lower W_it) have a smaller capital-to-wealth share.
    - Relative changes show households closer to subsistence (P C relative to W) exhibit larger reductions in capital-to-wealth ratio after income shocks, implying slower capital rebuilding and larger welfare impacts.
  - Modeling choice rationale:
    - Savings enter directly into utility (parsimonious) to avoid modeling forward-looking expectations and occasionally binding borrowing constraints, while generating an asset-holdings-to-wealth ratio increasing in wealth in steady state.

*Source: wpiea2023166-print-pdf - 2.1  Data*

### 3.3  Final Goods Consumption and Food Security

### 3.3  Final Goods Consumption and Food Security

### Final goods demand specification and spending shares
- Final good C_it is composed of consumption of a food and a non-food good and is modeled with a non-homothetic CES demand specification following Comin et al. (2021), in which food is a necessity good and non-food is a luxury good.
- Preferences for the final good are defined implicitly as:
  - Ω_F^{1/ρ} C^{ε_F}_{it} c^{ρ−1}_{Fit} + Ω_N^{1/ρ} C^{ε_N}_{it} c^{ρ−1}_{Nit} = 1 (equation (7)).
- Parameters:
  - Ω_F, Ω_N: sectoral taste parameters for food and non-food goods.
  - ρ: price elasticity of substitution between food and non-food goods.
  - ε_F, ε_N: utility elasticities governing responsiveness of sectoral consumption to changes in utility for given relative prices.
- Household budget constraint: p_Fit c_Fit + p_Nit c_Nit = P_it C_it.
- Optimal share of spending on food ω_F is given by:
  - ω_f ≡ p_F · c_F / (P · C) = Ω_F (p_F / P)^{1−ρ} C^{ε_F − (1−ρ)} (equation (8)).
- Average cost index P satisfies a composite expression (equation (9)) derived from sectoral price components and ω_F.
- Implication: Since 0< ε_F <1 (food is a necessity) the share of spending on food falls as consumption C rises; since ε_N >1 (non-food is a luxury) the non-food share rises.

### Calorie consumption mapping and food insecurity measure
- Calorie consumption is modeled as a constant-elasticity function of real food consumption:
  - kcal_it(ω) = η_0 c_{Fit}(ω)^{η_1}, with η_0 >0 and 0< η_1 <1.
- Interpretation and empirical motivation:
  - As food spending rises households shift toward higher cost-per-calorie foods (e.g., meat, oils, beverages), so calories rise slower than food consumption.
  - The constraint 0< η_1 <1 ensures calories rise slower than food consumption.
- Food insecurity definition:
  - A threshold kcal level of 2200 calories is defined; households with calorie consumption below this threshold are classified as undernourished.
  - Prevalence of food insecurity in a district = share of households who are undernourished.
- Note on terminology:
  - The framework primarily uses "undernourishment" rather than "food insecure" because food security is multidimensional; the model links closest to the access dimension.

### Household heterogeneity and the undernourishment threshold
- Each rural household has human capital z(ω) with cdf F(·), iid across households and districts.
- Farm and off-farm per-unit incomes are y_Fit and y_Nit respectively; household income of type ω is:
  - y_it(ω) = z(ω) (y_Fit + y_Nit).
- Land endowment is proportional to human capital: h(ω) = χ z(ω), χ >0.
- There exists a district-specific threshold z_i such that households with z(ω) below z_i cannot generate enough income to choose food consumption that yields at least 2200 calories and are undernourished.
- Shocks:
  - Negative shocks (e.g., flood) push the threshold z_i up, increasing the share of undernourished households.
- Heterogeneity generates a non-degenerate distribution of outcomes at the district level.

### Food markets: supply chain, trade costs, and import shares
- Supply chain agents: farms (household producers), wholesalers (local aggregators), retailers (final good aggregators).
- Local households produce differentiated local varieties and sell all output to a perfectly competitive local wholesaler at price p_Fi (farm-gate).
- Retailers aggregate intermediate goods from wholesalers using an Armington aggregator:
  - C^F_i = (Σ_{n=1}^N c_{Fin}^{σ−1/σ})^{σ/σ−1}, with σ >1 (equation (10)).
- Retailers face CES price index:
  - P^F_i = (Σ_{n=1}^N (p_{Fin})^{1−σ})^{1/(1−σ)}.
- Trade costs and prices:
  - Shipping intermediate goods from n into i incurs iceberg trade cost τ_in ≥1, τ_ii = 1, and p_{Fin} = τ_in p_{Fn}.
  - Assumption: τ_in ≤ τ_mn τ_im for m≠n and m≠i (no profitable indirect shipping).
  - Rest of world price p^F_* is exogenous.
- Wholesaler import share π_in:
  - π_in ≡ p_{Fin} c_{Fin} / X^F_i = (τ_in p_{Fn} / P^F_i)^{1−σ} (equation (11)).
  - Import share falls with farm-gate price p_{Fn} and shipping cost τ_in, and rises with the CES price index P^F_i.
- Final food bundle C^F_i is non-tradable—households buy final food goods from local retailer at local price index P^F_i.

### Food production, factor allocation, and returns
- Household farm production function (constant returns to scale):
  - q^F_it(ω) = z(ω) A_it h(ω)^{α_h} [k_it(ω)]^{α_k} [l^F_it(ω)]^{α_L},
  - where A_it is district farm productivity, k_it is farm capital, l_it is farm labor, and h(ω) = χ z(ω).
- Capital is broadly interpreted to include fertilizer, seeds, tools, machinery, irrigation.
- Labor allocation:
  - Households choose farm labor l^F_it(ω) by equating marginal product of farm labor to off-farm wage w_it.
  - Resulting labor supply:
    - l^F_it = (α_l p_Fit A_it K_it^{α_k} / w_it)^{1/(1−α_l)} (equation (12)).
- Return on capital r_it:
  - r_it = z^{α_k} p_Fit A_it q^F_it / k_it.
- Farms use only owner-supplied land and labor; no separate farm wage or land rental rate introduced.

### Non-food sector production, wages, and migration
- Non-food production in location i:
  - Q^N_it = A^N_i (ψ L^U_it + L^R_it (1−l^F_it) ∫_ω z(ω) dF(ω) ), with ψ >1 (urban productivity premium).
- Non-food goods trade follows same structure as food (same σ).
- Firms pay wage w_it and zero profits imply marginal revenue product of labor equals wage:
  - w_it = p^N_it z^N_i (equation (13)).
- Migration and remittances:
  - Households may send migrants to other locations or foreign economy offering exogenous wage w^*.
  - Idiosyncratic migration costs κ_in(ω) follow independent Frechet distributions with scale parameter D_{i|n} and common share parameter λ (dispersion).
  - Normalization: D_{n|n} = 1 for all n.
  - Net real wage for a household choosing destination i:
    - V_nt(ω) = max_i κ_in(ω) w_it / P_it|n, where P_it|n = P_it^{φ} P_nt^{1−φ}, 0< φ <1.
  - Remittances from destination i to origin n:
    - rmt_{ni} = z(ω) φ_n (ψ P_n w_i − (1−ψ) P_i w_n) / (ψ P_n + (1−ψ) P_i).
  - Migration probabilities ξ_int:
    - ξ_int = D_{i|n} (w_it / P_it|n)^{λ} / Σ_{m=1}^{N+1} D_{m|n} (w_mt / P_mt|n)^{λ} (equation (14)).
  - Law of large numbers implies bilateral migration flows match probabilities: ξ_it|n = L^R_it|n / L^R_nt.
  - Dispersion λ governs elasticity of migration flows to real wages: lower λ → smaller inflow response; λ→∞ → infinite elasticity and equalization of net real wage.
  - Expected net real wage:
    - E[V_nt] = (Σ_{i=1}^{N+1} D_{i|n} (w_it / P_it|n)^{λ})^{1/λ} (equation (15)).

### Market clearing and calibration overview
- Exports to the rest of the world X^j_{*,n} follow:
  - X^j_{*,n} = b_j · (τ_{*,n} p^j_n)^{1−σ}, for j = F, N (equation (16)); σ is same variety elasticity.
- Intermediate goods market clearing:
  - Y^j_i = Σ_{n=1}^N π_{ni} X^j_n + X^j_{*,i}, for j = F, N.
- Final goods market clearing with price-sales identity:
  - p^j_i Q^j_i(ω) = Σ_{n=1}^N π_{ni} X^j_i + X^j_{*,i}, for j = F, N (equation (17)); this pins down equilibrium prices {p^j_i}_{i=1}^N given π_{ni} and exports X^F_{*,i}.
- Non-food labor market clearing:
  - w_it (ψ L^U_it + L^R_it) = Σ_{n=1}^N π_{ni} X^N_i + X^N_{*,i}, which pins down sector wages w_it.
- Rural labor supply consistency:
  - L^R_it = Σ_{n=1}^N φ_n ξ_it|n L^R_nt.
- Calibration and simulation approach:
  - Calibration targets a steady-state equilibrium with constant household-level capital stocks k_it(ω) = k_{it−1}(ω) = k_i(ω).
  - Simulation procedure:
    - Draw independent human capital draws for 1000 households per location.
    - Initialize capital stocks arbitrarily, solve for equilibrium, update capital based on household optimal choices, iterate until household-level capital converges to steady-state.
  - Model parameters are calibrated externally or internally to match targeted steady-state data moments.
  - The calibration includes approximating the magnitude of historical climate shocks, validating the model via determinants of prevalence of undernourishment in simulated data, and demonstrating mechanisms by simulating responses to a geographically isolated normalized climate shock.

*Source: IMF working paper chapter text (3.3–4 excerpt).*

### 4.1  External Calibration

### 4.1 External Calibration

### Preferences
- Variety elasticity of substitution σ = 6 (consistent with Eaton and Kortum (2002) range translating to 4.6 to 13.86).
- Migration dispersion parameter λ = 3.4 (Monte et al. (2018)).
- Non-homothetic CES utility parameters:
  - Income elasticity of demand for food εF = 0.2 (Comin et al. (2021), non-OECD sample).
  - Income elasticity of demand for non-food εN = 1.19 (Comin et al. (2021), non-OECD sample).
  - Price elasticity of substitution γ = 0.48 (estimate used in text from Comin et al. (2021)).
  - Note: Table 1 in the source lists γ = 1.4 as "Price Elasticity of Substitution."
- Elasticity of calorie intake with respect to food consumption η = 0.4 (Subramanian and Deaton (1996); range 0.3-0.5).
- Calorie intake threshold for undernourishment kcal = 2200 (Government of Nepal).

### Population
- Urban/rural assignment rule: sub-administrative region with population > 120,000 = urban; all others = rural.
- Aggregated rural population across all 77 districts = 80.6 percent (close to World Bank reported 79.4 percent in 2020).
- Model calibrated for 51 districts (accounts for roughly 78 percent of total Nepal population).
- If calibration restricted to 51 districts and Kathmandu Valley assigned urban status, rural share = 78.4 percent.
- Across 51 districts:
  - 38 entirely rural.
  - 12 mixed (both rural and urban population).
  - 1 (Kathmandu Valley) entirely urban, containing 35.3 percent of all urban population in the sample.

### Farm production and household human capital
- Data source: Nepal Household Risk and Vulnerability Survey (HRVS) 2016-18 used to estimate local average agricultural yields and household farm profits.
- Farm production shares adopted from Bergquist et al. (2019) average across nine crops:
  - αh = 0.50 (Farm Land Share)
  - αk = 0.15 (Farm Capital Share)
  - αl = 0.35 (Farm Labor Share)
- Capital depreciation rate δ = 0.05.
- Household human capital z(ω) assumed log-normal with log z(ω) ∼ N(0, σ^2_z_i), iid across households; district-specific dispersion σ^2_z calibrated to match observed variance of real consumption among rural households:
  - σ^2_z ranges between 0.24 and 1.07 across districts; median district σ^2_z = 0.55.
- Method for farm profits: compute nominal crop revenues using volume-weighted average selling prices times production volume, add net livestock revenues, subtract hired labor and input costs, divide by land × labor hours to obtain household-level farm profits; aggregate to district average productivity.

### Externally calibrated parameter highlights (as reported)
- σ = 6 (Elasticity of Substitution)
- λ = 3.4 (Migration Dispersion Param.)
- εF = 0.2 (Income Elasticity of Demand)
- εN = 1.19 (Income Elasticity of Demand)
- γ = 1.4 (Price Elasticity of Substitution) — listed in Table 1
- Z_i (Local Farm Productivity) from HRVS
- σ^2_z_i (Dispersion of Human Capital) from HRVS
- δ = 0.05 (Depreciation Rate)
- αh = 0.50, αk = 0.15, αl = 0.35 (Farm input shares)
- ζ = 0.31 (Distance Decay Rate, Disdier and Head (2008))
- η = 0.4 (Elasticity of Kcal wrt Food)
- kcal = 2200 (Undernourishment Threshold)

### Internal calibration: consumption, trade costs, non-food productivities, migration
- Consumption preferences:
  - β calibrated to match capital stock value to wealth ratio for average household = 0.5.
  - Subsistence requirement C calibrated to 80 percent of the 5th percentile of consumption level C_i across households in baseline economy (80% of 5th percentile used to approximate survival threshold).
  - Consumption weights ΩF, ΩN calibrated to match average household food budget share = 0.58 (CBS 2015-16).
- Trade costs:
  - τ_ni = a_n dist_ni^ζ with a_n > 0, ζ > 0.
  - Distances dist_ni measured as driving distance using Google Maps Distance Matrix API for 51 districts.
  - Own-district distance dist_nn set equal to 0.25 times the smallest bilateral distance that n has with other locations i ≠ n.
  - a_n set so normalization τ_nn = 1 for all n.
  - Distance decay rate ζ chosen to be in the range implied by meta-analysis: Disdier and Head (2008) central estimate -0.9 for elasticity of trade flows w.r.t. distance; with σ = 6 this implies ζ range between 0.056 and 0.31. The model sets ζ = 0.31 to reflect Nepal’s mountainous geography.
- Import/export price and shares:
  - Import prices p* and ROW demand X* calibrated to match 2019 share of imports in total food expenditures consistent with World Bank data and WITS export shares.
- Non-food sector productivities:
  - A_N_i calibrated to match observed district-level farm labor shares l_F_it from HRVS (hours spent on agriculture vs non-agriculture).
  - Calibration implies A_N_i = α_l p_F_it A_F_it K_it^α_k p_N_it (l_F_it)^{α_l − 1} (as derived from equations in text context).
  - Urban household relative productivity ψ calibrated to match urban-rural wage gap.
- Migration:
  - Bilateral migration scale parameters D_i|n estimated from HRVS district-to-district and international migration flows, matching model-implied bilateral flows to empirical matrix.
  - International destinations pooled into ROW including India, Gulf Region, Malaysia → total of 52 potential migration destinations.
- Internally calibrated parameter targets:
  - β → Capital-to-Wealth Ratio target.
  - C → HRVS 80% of 5th Percentile Consumption.
  - ΩF, ΩN → HRVS Average Food Share.
  - X* (ROW Demand), p* (Import Price) → WITS Export/Import Shares.
  - A_N_i → HRVS Farm Labor Shares.
  - D_i|n → HRVS Bilateral Migration Shares.

### Climate shocks: construction and calibration
- Large climate shocks considered: floods, landslides, cold weather, storms (large shock defined as any shock with recorded damages > 0).
- Empirical average treatment effect from BIPAD data (Section 2.3): peak effect (2 months after shock) = 3 percent increase in local food prices (average across districts).
- Productivity-price relationship estimation:
  - Simulate random T = 100 series of local idiosyncratic productivity shocks per district (decreases x% in agricultural productivity drawn from log-normal with log x ∼ N(0, 0.1^2)), feed into calibrated model, estimate implied change in prices, compute average elasticity of productivity w.r.t. prices per district.
  - Median district: 1 percent increase in local prices associated with 5.4 percent drop in productivity.
  - Therefore, the average climate shock (3 percent price increase) corresponds to a 16.2 percent reduction in productivity in the median district.
  - Substantial heterogeneity: some districts show elasticities higher than 12 percent.
- Panel of climate damages:
  - Monthly large climate events from BIPAD period 2011-2022; damages divided by district population to obtain per capita damages.
  - Average per capita damages = 37 Nepalese Rupees per capita.
  - Calibration: set TFP loss of an event with average per capita damages equal to TFP loss equivalent to the 3 percent price average treatment effect; scale other shocks proportionally and aggregate monthly TFP damages within a year by compounding TFP losses.

### Determinants of food security in calibrated steady state (model-simulated)
- Analysis: regress district-level proportion of undernourished on district characteristics using model-simulated data for 51,000 households across 51 districts.
- Key empirical findings from regressions:
  - A 1 percent higher agricultural yield → on average 0.26 percent fewer undernourished households.
  - More remote regions → significantly more undernourished.
  - Remoteness and agricultural productivity together explain 73 percent of variation in undernourishment (column 5).
  - Greater trade and migration access associated with lower undernourishment (migration access not statistically significant alone; becomes significant when controlling for agricultural yields).
- Interpretation:
  - Controlling for low yields, household coping strategies of importing and migration are strongly effective in lowering undernourishment.
  - Even without climate shocks, remoteness and agricultural productivity are important determinants of undernourishment.

### Model simulation in response to shocks (representative impulse responses)
- Simulation design: for each district, simulate a 10 percent temporary negative productivity shock lasting one period with other districts at baseline (capturing isolated climate shock reducing district harvest by 10 percent).
- Average district impulse responses (average household vs undernourished household):
  - Impact food price increase = 3.8 percent on impact (considerable cross-district variation).
  - Average district increases share of food imported from other regions by almost 15 percent in reaction to shock.
  - Food consumption falls on average by 2.2 percent (equivalent to 0.88 percent fall in calorie consumption).
  - Undernourished households suffer larger consumption and welfare losses than average households:
    - Welfare loss undernourished = 3.5 percent vs average household welfare loss = 2.1 percent.
  - Drivers of larger losses for undernourished:
    - Higher budget share spent on food → more sensitive to food price increases.
    - Non-homothetic subsistence term implies undernourished households hold smaller capital stock buffer relative to income → current resources fall relatively more.
  - Real non-food sector wages decline due to higher food prices and depressed local non-food demand.
  - Households send additional members to migrate: share of migrants in household rises by 5 percent on average.
  - Cross-district variation in responses driven by remoteness and migration/trade costs.

- Notes on heterogeneity and mechanisms:
  - Districts importing larger food shares experience smaller price increases.
  - Less remote districts more easily substitute to other food suppliers, mitigating price increases.
  - Undernourished households mitigate consumption losses partly by shifting away from non-food consumption, limiting food consumption fall.

*Source: wpiea2023166-print-pdf - 4.1 External Calibration*

### 5.1  Annual Climate Damages

### 5.1  Annual Climate Damages

### Average Annual Impacts (2011–2022)
- Agricultural yields are 5.75 percent lower on average annually across all districts due to climate shocks.
- Maximum average impact on yields of -11 percent in 2019.
- Average annual loss to GDP: 2.1 percent.
- Average welfare losses: 3.1 percent.
- Comparable benchmark: average annual loss of 2.65 percent due to floods reported in the United Nations’ WESR Risk Platform.
- Rise in the rate of undernourishment of 2.8 percent due to climate shocks over the last decade.

### District and Temporal Heterogeneity (2011–2022)
- Median household: maximum annual impact on agricultural yields of 8 percent in 2019 and minimum of 2 percent in 2012.
- Extreme district impacts: in 2019 the 95th percentile district experienced almost 18 percent lower agricultural yields, corresponding to a roughly 13 percent loss in welfare and 9 percent increase in undernourishment.
- Eleven districts estimated to suffer average annual farm productivity losses of more than 10 percent.
- Roughly a quarter of districts experienced increases in undernourishment rates of more than 5 percent.
- Not all districts with similar exposure have equal welfare losses—some districts cope better, indicating district-level resilience heterogeneity.

### Determinants of Vulnerability
- Remoteness definition: population-weighted average distance to other districts (remoteness_n = sum_{i≠n} dist_{ni} L_i); a district is remote if remoteness is in the top 30 percent of districts.
- Remote households suffer welfare losses 2.04 times those of non-remote districts.
- Undernourished individuals have welfare losses of 4.3 percent, which are 1.54 times those of food secure individuals.
- Regression findings (controlling for shock size):
  - Being undernourished increases the responsiveness of log utility to shocks from -0.013 to -0.020 (a 54 percent increase in magnitude).
  - Households in remote districts exhibit responsiveness of -0.023, which is 54 percent larger than non-remote counterparts.
  - Higher local agricultural productivity reduces the magnitude of the shock (households have larger buffers).
- Market access (following Redding and Sturm, 2008) and migration access are constructed from model equilibrium equations:
  - marketaccess_n = sum_{i=1}^{N+1} τ_{ni}^{1−σ} (P^F_i)^{σ−1} X^F_i (weights distances by food expenditures; includes rest of the world).
  - migrationaccess_n = (sum_{i≠n}^{N+2} B_{i|n} (w_i/P_i)^λ)^{1/λ} (expected net wage conditional on migration; includes foreign economy).
- Interaction effects:
  - Undernourished households benefit more from access to trade, migration, and higher agricultural productivity than food secure households, because they have lower capital buffers and rely more on these coping mechanisms.

### Policy Counterfactuals (Infrastructure / Lower Spatial Frictions)
- Counterfactuals modeled:
  - 10 percent reduction in migration costs D_{i|n} for all migration destinations (domestic and abroad).
  - 10 percent reduction in iceberg trade costs τ_{in} for all pairs of Nepali districts.
- Mechanisms:
  - Lower trade/migration costs raise baseline real incomes (move households away from subsistence) and reduce costs of accessing alternative income sources and goods.
  - Spatial integration may increase exposure to external shocks via cross-district spillovers.
- Aggregate effects (average outcomes):
  - Under either 10 percent lower trade costs or 10 percent lower migration costs, the welfare impact of shocks is reduced on average from 3.3 percent to approximately 2.7 percent (a reduction by a factor of 0.18).
  - Rates of undernourishment are reduced from 2.8 percent to roughly 2 percent (reduction by a factor of almost a third).
- Distributional and district-level effects:
  - All districts experience smaller welfare losses on average, though relative reductions vary by initial openness to trade and migration.
  - Undernourishment does not fall in all districts:
    - 18 percent of districts see an increase in undernourishment with lower trade costs.
    - 16 percent of districts see an increase in undernourishment with lower migration costs.
  - These distributional effects reflect complex price and income effects and cross-district spillovers; poorer, more food-insecure households can be harmed by infrastructure improvements if exposure to external shocks increases.

### Conclusion (key takeaways)
- Climate shocks already impose large negative impacts on GDP, nutrition, and welfare.
- Impacts disproportionately harm remote and food-insecure households.
- Poverty and food insecurity exacerbate shock impacts through reduced buffers and coping capacity.
- Policies lowering trade and migration costs can reduce average impacts from climate shocks by providing alternative income and affordable food, but may have adverse distributional effects and increase vulnerability for some households due to integration-induced spillovers.
- The quantitative spatial framework allows analysis of temporary shocks, their damage and persistence, and the role of trade, migration, and remittances in shaping outcomes.

*Source: wpiea2023166-print-pdf - 5.1  Annual Climate Damages.*

### References

### wpiea2023166-print-pdf - References

### Bibliographic references
- Contains a comprehensive list of cited works spanning topics including rural–urban linkages, trade costs, volatility and trade, agricultural productivity, climate impacts, migration, food security measurement, and spatial economic modeling.
- Notable citations include empirical and theoretical contributions by Adam, Allen and Arkolakis, Anderson and Van Wincoop, Dell et al., Donaldson and Hornbeck, Eaton and Kortum, Gollin et al., Redding and Rossi-Hansberg, Samuelson, and many discipline-spanning reports from FAO, ILO, World Bank, World Food Programme, and IMF.

### Empirical Appendix — Key regression results and sample details (Tables 3–5)
- Table 3: Determinants of Prevalence of Undernourishment
  - Agricultural productivity: -0.162 ∗∗∗ (0.0147); -0.148 ∗∗∗ (0.0147); -0.130 ∗∗∗ (0.00977)
  - Remoteness: 0.0881 ∗∗∗ (0.0234); 0.0389 ∗∗∗ (0.0142)
  - Trade access: -0.109 ∗∗∗ (0.0156); -0.0655 ∗∗∗ (0.00785)
  - Migration access: -0.0471 (0.0449); -0.0402 ∗∗∗ (0.0148)
  - N 50 (for all columns)
  - r2: 0.717; 0.227; 0.504; 0.0224; 0. 756; 0.900
  - Robust standard errors in parentheses. Significance: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
  - Note: Regression uses model-simulated data for each of the districts in the sample. All variables are in steady-state at the district-level. Outcome: proportion of undernourished. Definitions: agricultural productivity = log of farm productivity; remoteness = population weighted-distance to all other Nepali districts; trade access = weighted-distance to other Nepali districts including a decay rate of distance based on the elasticity of trade; migration access = expected net wage earned by a household living in district n if they were to migrate.

- Table 4: Which Factors Impact Climate Vulnerability?
  - Climate shock: -0.0129 ∗∗∗ (0.0000725); -0.0150 ∗∗∗ (0.0000573); -0.0415 ∗∗∗ (0.000157); -0.0394 ∗∗∗ (0.000188)
  - Undernourished=1 × Climate shock: -0.00691 ∗∗∗ (0.000118); -0.00364 ∗∗∗ (0.000114)
  - Climate shock × Remoteness: -0.00808 ∗∗∗ (0.000109); 0.000167 (0.000116)
  - Climate shock × Agricultural productivity: 0.0312 ∗∗∗ (0.000177); 0.0304 ∗∗∗ (0.000199)
  - HouseholdFE: Yes (in all columns)
  - N 351000 (in all columns)
  - r2: 0.724; 0.725; 0.747; 0.747
  - Standard errors in parentheses. Significance: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
  - Note: Regression uses model-simulated data for each of the 1000 households in each of the 51 districts over each of year 2010-2022 in the sample. Outcome: household utility at time t. Definitions: climate shock = household change in productivity due to climate shock in period t; undernourished = dummy equal to 1 if household is below the calorie threshold; agricultural productivity = log of farm productivity; remoteness = population weighted-distance to all other Nepali districts; trade access = weighted-distance with decay based on elasticity of trade; migration access = expected net wage if migrate.

- Table 5: How Does Being Undernourished alter Effectiveness of Coping Strategies?
  - Climate shock: -0.0370 ∗∗∗ (0.000212); -0.0130 ∗∗∗ (0.0000701); -0.0137 ∗∗∗ (0.0000744); -0.0342 ∗∗∗ (0.000268)
  - Undernourished=1 × Climate shock: -0.00904 ∗∗∗ (0.000316); -0.00389 ∗∗∗ (0.000117); -0.00625 ∗∗∗ (0.000120); -0.00717 ∗∗∗ (0.000403)
  - Climate shock × Agricultural productivity: 0.0276 ∗∗∗ (0.000230); 0.0248 ∗∗∗ (0.000310)
  - Undernourished=1 × Climate shock × Agricultural productivity: 0.00670 ∗∗∗ (0.000366); 0.00580 ∗∗∗ (0.000475)
  - Climate shock × Trade access: 0.0106 ∗∗∗ (0.000105); 0.00562 ∗∗∗ (0.000118)
  - Undernourished=1 × Climate shock × Trade access: 0.00373 ∗∗∗ (0.000174); 0.00142 ∗∗∗ (0.000199)
  - Climate shock × Migration access: -0.0108 ∗∗∗ (0.000250); 0.00594 ∗∗∗ (0.000288)
  - Undernourished=1 × Climate shock × Migration access: 0.00846 ∗∗∗ (0.000389); 0.00300 ∗∗∗ (0.000427)
  - HouseholdFE: Yes (in all columns)
  - N 351000 (in all columns)
  - r2: 0.748; 0.741; 0.725; 0.753
  - Standard errors in parentheses. Significance: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
  - Note: Regression uses model-simulated data for each of the 1000 households in each of the 51 districts over each of year 2010-2022. Outcome: household utility at time t. Variable definitions as in Table 4.

### Variable definitions and descriptive statistics (Tables 6–7)
- Table 6: Variable Description (selected)
  - Crop yield: Natural log of household crop yield
  - Crop income: Natural log of household income from crops
  - Livestock income: Natural log of household income from livestock
  - Food insecure dummy: =1 if household is categorised by World Bank as food insecure
  - Food insecurity score: Index of food security which assigns a value between 0 and 3 on four categories. High number indicates less secure.
  - Expenditure: Natural log of total household expenditure
  - Food expenditure: Natural log of household food expenditure
  - Non-food expenditure: Natural log of household food expenditure
  - Food expenditure share: Food expenditure divided by total expenditure
  - Migrant dummy: =1 if any household member is a migrant in the sample period
  - Number of migrants: Count of migrants in household
  - Remittance share of income: Remittances divided by total income
  - Savings: Natural log of savings
  - Capital equipment: Natural log of farm equipment holdings
  - Capital Livestock: Natural log of livestock holdings
  - Savings rate: Savings divided by income
  - Price: Natural log of price in NPR
  - Climate shock: =1 if district has recorded flood, heavy rainfall, landslide, or storm in month

- Table 7: Descriptive Statistics (selected household- and district-level means, SD, min, max, N)
  - Household-level (Mean SD Min Max N)
    - Crop yield 2.02 0.73 -4.53 9.28 13,977
    - Crop income 8.70 1.35 -2.75 15.51 9,890
    - Livestock income 9.53 1.28 2.08 14.81 3,407
    - Food insecure dummy 0.14 0.35 0.00 1.00 16,951
    - Food insecurity score 0.69 2.10 0.00 24.00 16,951
    - Expenditure 11.17 0.83 7.76 16.18 16,951
    - Food expenditure 9.58 0.83 5.56 13.60 16,774
    - Non-food expenditure 10.83 0.94 7.28 16.18 16,951
    - Food expenditure share 0.26 0.17 0.00 0.94 16,951
    - Migrant dummy 0.46 0.50 0.00 1.00 16,951
    - Number of migrants 0.35 0.52 0.00 2.89 8,380
    - Remittance share of income 0.10 0.18 0.00 1.00 16,276
    - Savings 10.59 1.45 0.75 15.50 5,452
    - Capital equipment 7.91 1.28 1.61 14.65 9,442
    - Capital Livestock 8.73 1.45 3.40 14.00 2,305
    - Savings rate 0.48 0.25 0.00 1.00 5,452
  - District-level
    - Price 4.50 0.84 1.95 6.91 14,394
    - Climate shock 0.14 0.35 0.00 1.00 14,394

### Interaction of Remoteness and Undernourishment (Appendix B)
- Reported analysis and figure (Figure 15) examine how household wealth and district remoteness interact to determine resilience to shocks.
- Key descriptive findings reported in text:
  - The figure shows model simulated average changes in welfare due to historical climate shocks across wealth quintiles and remoteness.
  - "For instance, the bar on the far left of the figure shows that the bottom wealth quintile of non-remote households has on average 8(?) percent larger decline in welfare than the average household."
  - All other non-remote households perform better than the average household.
  - Connected (non-remote) households can use alternative coping strategies including accessing trade and migration and avoiding more damaging responses.
  - Quintiles 1-4 among remote households all have larger welfare losses than the average.
  - "Those households which are both the most remote and poorest see, on average, 40 percent larger losses in welfare than the average in the whole sample, compared to the poorest households in non-remote districts."
- Figure notes:
  - The figure shows model simulated average changes in welfare due to climate shocks relative to the mean change in welfare among different wealth quintiles disaggregated in the top XX percent of remoteness districts.
  - Example note: the bar on the far left shows that the bottom wealth quintile in non-remote districts were more negatively impacted than the average household in the country, whereas the next four quintiles in non-remote districts were less negatively impacted than the average household in the country.
  - Note: these households still see a decline in welfare in absolute values.

*Coping with Climate Shocks: Food Security in a Spatial Framework — Working Paper No. WP/2023/166*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023166-print-pdf.pdf_