## 2.1 Agricultural sector (excerpt)

## Source details

**Canonical URL:** [2.1 Agricultural sector (excerpt)](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025233-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025233-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025233-source-pdf.pdf.json)

---

### Context and motivation
- International migrants: 281 million in 2020 (3.6% of the world population) compared to 173 million in 2000 (2.8% of the population).
- Migration drivers include economic motives, extreme weather events, and conflict; migration can be an adaptation strategy to support livelihoods, build resilience, and protect against fragility.
- Guatemala-specific context:
  - Agriculture employs around one third of the economically active population in the country (84% in rural areas).
  - Most occupations in agriculture are informal (90%).
  - Agricultural sector growth lower than total economy; sector represents less than one tenth of GDP.
  - Land markets: high concentration and segmentation, limited market information, high informality, and cumbersome and costly transaction procedures.
  - Weak governance conditions and weak institutions pervasive.
  - Guatemala among top-15 recipient countries of international remittances.
  - International remittances represented 18.9% of the GDP in 2023.
  - Net out-migration exceeded 850,000 between 2002 and 2021, equivalent to 5.2% of the 2018 census population.
  - Primary destination for Guatemalan migrants: the US (around 75% in 2019 and 85% in 2022).
  - Use of traffickers (“coyotes”) is common; migrants pay between 5.5 and 12.5 thousand US dollars depending on destination, route, and transportation method.

### Theoretical framework and model setup
- Economy:
  - Two-sector economy with J rural (agricultural) regions and one urban (non-agricultural) location; j = 1,...,J+1.
  - Agricultural good produced using land in rural regions; each region j faces distinct agricultural distortions τ_ij.
  - Non-agricultural good produced using labor in urban region j = u, without distortions.
- Individual endowments and choices:
  - Rural individuals endowed with idiosyncratic managerial ability to produce in agriculture (z_aij) and idiosyncratic ability to work in non-agriculture (z_nij).
  - Rural-born choose among: farmer in region j, migrate to urban non-agriculture, or emigrate (if welfare from emigrating exceeds staying).
  - Urban-born choose between: non-agricultural worker or emigrate.
  - Emigrants receive exogenous income independent of idiosyncratic abilities.
- Agricultural production and farmer optimization:
  - Production: y_ij = A_a z_aij l_ij^α, with α ∈ (0,1).
  - Farmer profit maximization: π(z_aij, τ_ij) = max_{l_ij} { τ_ij p A_a z_aij l_ij^α − q_j l_ij }.
  - Optimal land: l_ij = ( p A_ij / q_j )^{1/(1−α)}, where A_ij ≡ α A_a z_aij τ_ij.
  - Market clearing for land: ∫_0^1 F_ij l_ij di = L_j for j = 1,...,J.
  - Equilibrium rental price: q_j = p ( ∫_0^1 F_ij A_ij^{1/(1−α)} di / L_j )^{1−α}.
- Non-agricultural sector:
  - Y_n = A_n ∑_j ∫_0^1 W_ij z_nij di.
  - Equilibrium non-agricultural wage: w_nij = A_n z_nij.
- Household preferences and transfers:
  - Utility staying domestically: U_dij = max_{c_aij, c_nij} { ω log(c_aij − a) + (1−ω) log(c_nij) } s.t. p c_aij + c_nij ≤ I_ij + T, with subsistence a ≥ 0, ω ∈ (0,1), transfers T.
  - Transfers T include (i) government land rentals ∑_j q_j L_j / S and (ii) lump-sum transfers of tax revenues.
  - Consumption allocations: c_aij = ω (I_ij + T) / p + (1−ω) ā; c_nij = (1−ω)(I_ij + T) − (1−ω) p ā.
- Occupational and migration decisions:
  - Rural household choice: V_dij(p, I_ij, ε_ij) = max_{F_ij, W_ij ∈ {0,1}} { U_dij(...|farmer) + ε_fij, U_dij(...|worker) + ε_wij }.
  - Urban-born staying utility: V_dij(p, I_ij, ε_uij) = U_dij(p, w_nij | worker) + ε_uij.
  - Emigration utility: V_eij(U_ej, ε_eij) = U_ej + ε_eij; ε shocks i.i.d. Gumbel with mean zero and scale σ_{εj}.
  - Stay-or-emigrate decision: V_ij = max{ V_dij, V_eij }.
- Equilibrium conditions summarize prices {p, q_j}, wages w_nij = A_n z_nij, occupational choices {W_ij, F_ij}, emigrate-stay choices {E_ij}, land {l_ij}, and consumptions {c_aij, c_nij} such that markets clear.
- Aggregate relationship: p [ Y_a − S ̄a ] = ( ω / (1−ω) ) Y_n, with Y_a ≡ ∑_j ∫_0^1 F_ij y_ij di and Y_n ≡ ∑_j ∫_0^1 A_n W_ij z_nij di.

### Identified mechanisms (channels)
- Migration channel:
  - Higher distortions influence selection into emigration by increasing emigration probability of more productive individuals and reducing it for less productive ones; adverse selection worsens composition of remaining workforce and lowers aggregate productivity.
- Productivity channel:
  - Higher distortions directly reduce productivity by misallocating labor and land; resulting decline in household incomes increases incentives to emigrate from rural and urban areas.
- Interaction:
  - Migration channel dominates at lower distortion levels; productivity channel dominates at higher distortion levels.

### Theoretical results (selected)
- Theorem 1: Distortions reduce (increase) emigration of less (more) productive agents — formal inequalities comparing p_e under distortions and benchmark.
- Theorem 2: Distortions reduce incomes and raise emigration across all regions and sectors — p_eij(A_a^ℓ) > p_eij(A_a) when misallocation reduces A_a to A_a^ℓ < A_a.

---

### Empirical strategy, data, and calibration

### Data sources and estimation approach
- Micro and aggregate data for Guatemala: IV National Agricultural Census (INE, 2003), national household survey data (INE, 2019), price data (MAGA).
- Rich subnational indicators: socioeconomic, accessibility, institutional, cultural, climatic, insecurity.
- Estimation: two-stage calibration/estimation — external calibration of some parameters and estimation of remaining via Simulated Method of Moments (SMM).
- Subnational heterogeneity replicated by the model: regions with higher distortions tend to be less developed, more isolated, have limited financial penetration, and weaker government presence.

### Externally calibrated parameters (selected)
- A_n = 1
- A_a = 1
- α = 0.4 (sensitivity examined at 0.3 and 0.5)
- a = 0.25
- ω = 0.3
- L_1 = 0.5 (Region 1: Western Highlands–Verapaces–Center)
- L_2 = 0.3 (Region 2: Dry Corridor–Izabal)
- L_3 = 0.2 (Region 3: Pacifico–Bocacosta)

### Estimation targets and SMM-identified parameters
- Parameters estimated: {ψ_j, σ_τ j, U_e j, σ_ε j}_j∈J.
- Targeted moments:
  - ψ_j: match Corr(TFP_ij, TFPR_ij) within region j.
  - σ_τ j: match S.D.(TFPR_ij) for each j.
  - U_e j: match share of emigrants relative to employed population by region.
  - σ_ε j: match share of household members employed in non-agricultural sector by region.
- Targeted moments (Data vs Model):
  - Share of Emigrants: Region 1 E_1 Data 16.50 — Model 16.50; Region 2 E_2 Data 13.80 — Model 14.00; Region 3 E_3 Data 8.70 — Model 8.80.
  - Corr(TFP_j, TFPR_j): Region 1 0.48 — 0.48; Region 2 0.57 — 0.57; Region 3 0.50 — 0.50.
  - S.D.(TFPR_j): Region 1 0.82 — 0.82; Region 2 1.09 — 1.07; Region 3 0.95 — 0.90.
  - Share of Workers (W_j): Region 1 W_1 Data 52.20 — Model 54.69; Region 2 W_2 Data 54.90 — Model 58.32; Region 3 W_3 Data 62.50 — Model 62.76.

### Estimated parameter values (SMM)
- Utility of emigrating:
  - U_1 = 0.61
  - U_2 = 0.42
  - U_3 = 0.49
- Association between z_aij and τ_ij (absolute values):
  - |ψ_1| = 0.07 (Region 1)
  - |ψ_2| = 0.09 (Region 2)
  - |ψ_3| = 0.03 (Region 3)
- Dispersion of τ_ij:
  - σ_τ1 = 0.70
  - σ_τ2 = 0.90
  - σ_τ3 = 0.85
- Gumbel scale-parameter:
  - σ_ε1 = 1.21
  - σ_ε2 = 1.42
  - σ_ε3 = 0.81

### Empirical interpretations
- Distortions ranking by |ψ_j|: highest in Dry Corridor–Izabal (|ψ_2| = 0.09), then Western Highlands–Verapaces–Center (|ψ_1| = 0.07), then Pacifico–Bocacosta (|ψ_3| = 0.03).
- Dry Corridor–Izabal exhibits highest dispersion in revenue-based TFP (σ_τ2 = 0.90).
- Emigration incentives: U_1 = 0.61 (Western Highlands–Verapaces–Center) > U_3 = 0.49 (Pacifico–Bocacosta) > U_2 = 0.42 (Dry Corridor–Izabal).
- Pacifico–Bocacosta higher non-agricultural employment (~63%) corresponds to lower σ_ε3 = 0.81.

---

### Counterfactual benchmark construction and simulation results

### Benchmark construction procedure
1. Re-estimate distortions at the department level.
2. Select most “efficient” department within each region (lowest Corr(TFP_ij, TFPR_ij)).
3. Re-calibrate region-level ψ_j and σ_τ j to match correlation and S.D. of TFPR in selected benchmark departments.
4. Use re-calibrated parameters to construct benchmark and perform counterfactuals.

### Selected benchmark departments (lowest within-region Corr)
- Quetzaltenango (Region 1): Corr(TFP_ij, TFPR_ij) = 0.34; S.D.(TFPR_ij) = 0.57
- Izabal (Region 2): Corr = 0.42; S.D. = 0.72
- Escuintla (Region 3): Corr = 0.41; S.D. = 0.86

### Calibrated parameters under benchmark (Table 4)
- Association |ψ_j| (benchmark):
  - |ψ1| = 0.033
  - |ψ2| = 0.048
  - |ψ3| = 0.023
- Dispersion στ_j (benchmark):
  - στ1 = 0.52
  - στ2 = 0.62
  - στ3 = 0.76
- Targeted moments under benchmark closely match department-level statistics:
  - Corr(TFP1, TFPR1): 0.34 vs 0.34
  - Corr(TFP2, TFPR2): 0.42 vs 0.42
  - Corr(TFP3, TFPR3): 0.41 vs 0.40
  - S.D.(TFPR1): 0.57 vs 0.58
  - S.D.(TFPR2): 0.72 vs 0.73
  - S.D.(TFPR3): 0.86 vs 0.84

### Reduction in efficiency loss when moving to benchmark (Table 5)
- Reduction in efficiency loss by region:
  - Region 1: 37.3%
  - Region 2: 41.0%
  - Region 3: 14.5%
  - Total (all regions): 30.3%

### Counterfactual estimates — overall changes (Table 6)
- Changes transitioning from actual to benchmark:
  - Region 1 (Western Highlands–Verapaces–Center):
    - ∆ Share of Emigrants: -3.0 p.p.
    - ∆ Share of Workers (non-agricultural): 1.1 p.p.
    - ∆ Agricultural Productivity: 33.3%
    - ∆ Median Consumption: 11.0%
  - Region 2 (Dry Corridor–Izabal):
    - ∆ Share of Emigrants: -2.2 p.p.
    - ∆ Share of Workers: -0.1 p.p.
    - ∆ Agricultural Productivity: 45.6%
    - ∆ Median Consumption: 12.2%
  - Region 3 (Pacifico–Bocacosta):
    - ∆ Share of Emigrants: -1.8 p.p.
    - ∆ Share of Workers: 1.7 p.p.
    - ∆ Agricultural Productivity: 14.4%
    - ∆ Median Consumption: 8.8%
  - Urban:
    - ∆ Share of Emigrants: -1.6 p.p.
    - ∆ Share of Workers: 1.6 p.p.
    - ∆ Median Consumption: 14.6%
  - Total (aggregate):
    - ∆ Share of Emigrants: -2.3 p.p.
    - ∆ Share of Workers: 1.2 p.p.
    - ∆ Agricultural Productivity: 30.1%
    - ∆ Median Consumption: 12.3%
- Population context note from source:
  - The total emigration share decrease (−2.3 p.p.) equates to a 35.5% decrease in the Guatemalan population currently living in the US, given total population 16.3 million (INE, 2020) and approximately 1.1 million Guatemalan migrants in the US (US Census Bureau, 2021).

### Sensitivity to α (land elasticity)
- α = 0.3:
  - Total change in emigration share = −2.7 p.p.
  - Aggregate productivity increase = 21.7%
  - Median consumption increase = 8.4%
- α = 0.5:
  - Total change in emigration share = −1.3 p.p.
  - Aggregate productivity increase = 37.1%
  - Median consumption increase = 11.7%
- Share of non-agricultural workers change remains 1.2 p.p. across α values.

### Welfare and distributional effects
- Equivalent variation (median consumption-equivalent welfare gains):
  - Western Highlands–Verapaces–Center: 4.6%
  - Dry Corridor–Izabal: 4.5%
  - Pacifico–Bocacosta: 4.3%
  - All three regions combined with urban area: 4.5% at the median
- Distributional pattern:
  - Lower percentiles experience larger percentage increases in welfare.
  - The 10th percentile gains are more than twice those at the 90th percentile.
  - Gains particularly notable for low-income groups in Dry Corridor–Izabal and Western Highlands–Verapaces–Center.

### Contribution of migration channel to productivity gains (Table 7)
- Aggregate agricultural productivity increase between actual and benchmark when emigration is eliminated:
  - Region 1: 29.3%
  - Region 2: 40.9%
  - Region 3: 14.2%
  - Total: 30.1%
- Migration contribution (share of total gains attributable to migration channel):
  - Region 1: 11.9%
  - Region 2: 10.3%
  - Region 3: 1.3%
  - Total: 9.2%

---

### 3.1–4.3 Regional division, productivity estimation, and municipality-level correlates

### Regional grouping and agricultural structure
- Regions used in analysis:
  - Region 1: Western Highlands–Verapaces–Center (10 departments).
  - Region 2: Dry Corridor–Izabal (6 departments).
  - Region 3: Pacifico–Bocacosta (4 departments).
- Department of Guatemala considered exclusively urban and excluded from agricultural focus; Peten excluded due to limited agricultural activity.
- Agricultural structure:
  - 65% of landholdings < 1 hectare; over 3% > 10 hectares.
  - Main crops: maize, coffee, sugar cane, bananas, beans — together close to 80% of agricultural employment.
  - Most agricultural activities informal.

### Productivity distributions and estimation
- Assumed log-normal distributions:
  - log(z_aij) ∼ N(μ_a j, σ_a j) for region j.
  - log(z_nij) ∼ N(μ_n, σ_n).
- Two-stage TFP estimation (Britos et al., 2022) using IV National Agricultural Census (INE, 2003) and controls; non-agricultural TFP from 2019 National Survey of Labor and Income (INE, 2019).
- Region-specific TFP estimates (Table C.4):
  - Region 1: μ̂aj = 1.31; σ̂aj = 0.90
  - Region 2: μ̂aj = 1.39; σ̂aj = 1.05
  - Region 3: μ̂aj = 1.47; σ̂aj = 1.12
- Non-agricultural productivity: μ̂n = 0.8; σ̂n = 0.59

### Municipality-level regression of agricultural distortions (Table 8)
- Unit: municipality (N = 301). Dependent variable: agricultural distortions (ψ_j) estimated at municipality level. Variables standardized. Regional fixed effects included. R-squared = 0.062.
- Key coefficient estimates (Column (2) reported alongside Column (1) where differing):
  - Occupational precariousness index: Column (1) −0.156 (0.095); Column (2) −0.156 (0.100) — not statistically significant.
  - Per capita bank deposit accounts: −0.168** (0.074); −0.170** (0.074) — statistically significant at 5%.
  - Mobile base stations per square kilometer: 0.036 (0.058); 0.037 (0.059) — not significant.
  - Travel time to closest town of 20,000 habitants: 0.178** (0.070); 0.178** (0.071) — statistically significant at 5%.
  - Government density index: −0.141** (0.062); −0.143** (0.069) — statistically significant at 5%.
  - Share of indigenous population: 0.008 (0.073); 0.009 (0.073) — not significant.
  - Index of natural hazards: 0.003 (0.029) — not significant (in Column (2)).
  - Rate of extortions: 0.000 (0.068) — not significant (in Column (2)).
- Key municipal-level interpretations:
  - Financial penetration (per capita bank deposit accounts) negatively associated with distortions (coeff −0.168**), implying higher financial access linked to lower distortions.
  - Accessibility (travel time) positively associated with distortions (coeff 0.178**), suggesting isolation raises distortions.
  - Government presence (government density index) negatively associated with distortions (coeff −0.141**).
  - Climate vulnerability and insecurity indicators not statistically significant in these specifications.

### Land market and supplementary evidence
- Land rentals and price variation (three-year panel 2012–2014, 176 municipalities):
  - % Renting Land, Mean Price, Std. Dev., CV:
    - Region 1: 24.6% | 26.474 | 0.25 | 1.52
    - Region 2: 45.4% | 18.133 | 0.71 | 1.69
    - Region 3: 61.9% | 32.883 | 8.98 | 1.19
  - Pacifico–Bocacosta exhibits larger share of land rentals and lower coefficient of price variation → consistent with less market distortions.

---

### Key quantitative findings, interpretation, and policy implications

### Key quantitative findings (aggregate and region-specific)
- Counterfactual reducing distortions to benchmark yields:
  - 30.3% aggregate reduction in efficiency loss (definition: difference between aggregate agricultural production in a frictionless environment and actual total agricultural production under existing distortions).
  - Average decline of 2.3 percentage-points (p.p.) in the share of Guatemalans who emigrated.
  - Equivalent to a 35.5% reduction in the Guatemalan population currently living in the US (population context provided in source).
  - Total agricultural productivity increases by 30.1%.
  - Median household welfare rises by 4.5% (median equivalent variation).
- Regional highlights:
  - Dry Corridor–Izabal: largest agricultural productivity improvement (45.6%); highest initial distortion |ψ_2| = 0.09; highest σ_τ2 = 0.90.
  - Western Highlands–Verapaces–Center: largest reduction in emigration share (−3.0 p.p.).
  - Pacifico–Bocacosta: smallest productivity gain (14.4%) and lowest distortions (|ψ_3| = 0.03).

### Interpretation
- Reducing distortions and misallocation in agricultural markets can substantially reduce cross-border migration while increasing productivity and welfare.
- Migration and productivity channels operate jointly: distortions both change selection into migration and directly lower productivity, amplifying emigration incentives.
- Migration channel accounts for a non-negligible but minority share of productivity gains (Total migration contribution 9.2%; Region 1 11.9%; Region 2 10.3%; Region 3 1.3%).

### Policy implications and recommendations (from analysis)
- Policies aimed at reducing agricultural market imperfections can:
  - Enhance local productivity and welfare.
  - Mitigate emigration by lowering economic push factors associated with distorted agricultural markets.
- Targeting regions with greater distortions—characterized by isolation, limited financial access, and weak government presence—may yield larger gains in productivity and reductions in emigration.
- Specific policy areas identified by municipality regressions:
  - Promote financial inclusion and broaden access to banking services.
  - Improve road infrastructure and reduce travel times to urban centers.
  - Expand government presence and public institutions in underserved areas.

*Source: wpiea2025233-source-pdf (excerpted sections from Working Paper No. WP/2025/233).*

### 2.1    Agricultural sector  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  . 

### 2.1    Agricultural sector

### Context and motivation
- International migrants: 281 million in 2020 (3.6% of the world population) compared to 173 million in 2000 (2.8% of the population).
- Migration drivers include economic motives, extreme weather events, and conflict; migration can be an adaptation strategy to support livelihoods, build resilience, and protect against fragility.
- Guatemala-specific context:
  - Agriculture employs around one third of the economically active population in the country (84% in rural areas).
  - Most occupations in agriculture are informal (90%).
  - The agricultural sector has been growing at a lower rate than the total economy and represents less than one tenth of the Gross Domestic Product (GDP).
  - Land markets present high levels of concentration and segmentation with limited market information, high informality, and cumbersome and costly transaction procedures.
  - Weak governance conditions and weak institutions are pervasive across the country.
  - Guatemala is among the top-15 recipient countries of international remittances.
  - International remittances represented 18.9% of the GDP in 2023.
  - Net out-migration exceeded 850,000 between 2002 and 2021, equivalent to 5.2% of the 2018 census population.
  - Primary destination for Guatemalan migrants: the US (around 75% in 2019 and 85% in 2022).
  - Use of traffickers (“coyotes”) is common in Central America; migrants pay between 5.5 and 12.5 thousand US dollars depending on destination, route, and transportation method.

### Theoretical framework and model setup
- Economy:
  - Two-sector economy with a unit measure of individuals distributed across J rural (agricultural) regions and one urban (non-agricultural) location.
  - Agricultural good produced using land in rural regions; each region j faces a distinct level of agricultural distortions.
  - Non-agricultural good produced using labor in the urban region j = u, which operates without distortions.
  - Subscript j identifies both rural and urban regions such that j = 1,...,J+1.
- Individual endowments and choices:
  - Rural individuals endowed with idiosyncratic managerial ability to produce in agriculture and an idiosyncratic ability to work in non-agriculture.
  - Individuals born in rural region j choose among: (i) becoming a farmer in region j, (ii) migrating to the urban region to work in non-agriculture, or (iii) emigrating if welfare from emigrating exceeds staying.
  - Individuals born in urban region u choose between: (i) working in the non-agricultural sector, or (ii) emigrating if welfare from emigrating exceeds staying.
  - Individuals from rural or urban regions can emigrate and receive an exogenous income independent of idiosyncratic abilities.
- Agricultural distortions:
  - Farmers face a production tax that summarizes distortions in the agricultural sector across regions, denoted by τ_ij where i indicates the farmer and j the region.
  - Distortions capture inefficient regulations and interventions, information asymmetries, transaction costs, operational costs, and related factors.

### Identified mechanisms (channels)
- Migration channel:
  - Higher distortions influence selection into emigration by increasing the emigration probability of more productive individuals while reducing it for less productive ones.
  - This adverse selection worsens the composition of the remaining workforce and lowers aggregate productivity.
- Productivity channel:
  - Higher distortions directly reduce productivity by misallocating labor and land across sectors and regions.
  - The resulting decline in household incomes in agriculture and non-agriculture increases incentives to emigrate from both rural and urban areas.

### Empirical strategy and data
- Estimation strategy:
  - The model is estimated combining detailed micro and aggregate data for Guatemala: population and agricultural census microdata, national household survey data, and price data.
  - Rich subnational data are used—socioeconomic, accessibility, institutional, cultural, climatic, and insecurity indicators—to evaluate associations with agricultural distortions across areas.
  - Two-stage calibration/estimation approach for Guatemala: external calibration of some parameters and estimation of remaining parameters to reproduce heterogeneity in agricultural distortions.
- Subnational heterogeneity:
  - The estimated model reproduces heterogeneity in agricultural distortions observed in the data across and within Guatemalan regions.
  - Regions with higher distortions tend to be less developed, more isolated, have limited financial penetration, and a weaker government presence.

### Key quantitative findings and counterfactual results
- Counterfactual exercises reduce estimated distortions to benchmark scenarios within each region and yield:
  - 30.3% aggregate reduction in efficiency loss (efficiency loss defined as the difference between aggregate agricultural production in a frictionless environment and actual total agricultural production under existing distortions).
  - Average decline of 2.3 percentage-points (p.p.) in the share of Guatemalans who emigrated.
  - Equivalent to a 35.5% reduction in the Guatemalan population currently living in the US.
  - Total agricultural productivity increases by 30.1%.
  - Median household welfare rises by 4.5%.
- Interpretation:
  - Reducing distortions and misallocation in agricultural markets can lead to substantial reductions in cross-border migration while increasing productivity and general well-being.
  - The migration and productivity channels operate jointly: distortions both change selection into migration and directly lower productivity, amplifying emigration incentives.

### Policy implications (from analysis)
- Policies aimed at reducing agricultural market imperfections can:
  - Enhance local productivity and welfare.
  - Mitigate emigration by lowering the economic push factors associated with distorted agricultural markets.
- Targeting regions with greater distortions—characterized by isolation, limited financial access, and weak government presence—may yield larger gains in productivity and reductions in emigration.

*Source: wpiea2025233-source-pdf (section excerpts provided).*

### 2.1  Agricultural sector

### 2.1  Agricultural sector

### Agricultural technology and production
- Each farmer i in rural region j is endowed with ability z_aij and produces the agricultural good with the technology:
  - y_ij = A_a z_aij l_ij^α, where A_a is an aggregate productivity parameter in the agricultural sector, l_ij is the farmer’s land allocation, and α ∈ (0,1).

### Farmer profit maximization and land allocation
- Competitive market; farmer i in region j maximizes:
  - π(z_aij, τ_ij) = max_{l_ij} { τ_ij p A_a z_aij l_ij^α − q_j l_ij }, where p is the economy’s relative price of the agricultural good and q_j is the rental price of a unit of land in region j (both expressed in units of the non-agricultural good).
- Land allocation of each farmer in region j given q_j, p, and {z_aij, τ_ij}:
  - l_ij = ( p A_ij / q_j )^{1/(1−α)}, where A_ij ≡ α A_a z_aij τ_ij.
- Market clearing for land in region j:
  - ∫_0^1 F_ij l_ij di = L_j for j = 1,...,J.
- From (5) and (6), equilibrium rental price:
  - q_j = p ( ∫_0^1 F_ij A_ij^{1/(1−α)} di / L_j )^{1−α}.

### Non-agricultural sector (summary from contiguous sections)
- Aggregate non-agricultural output:
  - Y_n = A_n ∑_j ∫_0^1 W_ij z_nij di, where W_ij ∈ {0,1} indicates non-agricultural employment, z_nij is ability, and A_n is aggregate non-agricultural productivity.
- Equilibrium wage in non-agricultural sector for individual i from region j:
  - w_nij = A_n z_nij.

### Household domestic problem and consumption
- Households consume agricultural good c_aij and non-agricultural good c_nij; non-ag good is numeraire.
- Utility conditional on staying domestically:
  - U_dij = max_{c_aij, c_nij} { ω log(c_aij − a) + (1−ω) log(c_nij) } s.t. p c_aij + c_nij ≤ I_ij + T,
    - where a ≥ 0 is subsistence level of agricultural consumption, I_ij is household income, ω ∈ (0,1) is preference weight on agricultural consumption, and T denotes per capita transfers.
- Per capita transfers T include:
  - (i) transfers from land rentals accrued by the government distributed equally among individuals, and
  - (ii) lump-sum transfers of tax revenues from the idiosyncratic tax on agricultural output distributed equally.
- Transfers from land rentals equal:
  - ∑_j q_j L_j / S, where S is the number of individuals that stay in the domestic economy.
- Consumption allocations implied by maximization:
  - c_aij = ω (I_ij + T) / p + (1−ω) ā,
  - c_nij = (1−ω)(I_ij + T) − (1−ω) p ā,
    - where ā denotes the subsistence term (written ̄a in source).
- Income I_ij:
  - For urban-born households: I_ij = w_nij.
  - For rural-born households: I_ij = π(z_aij, τ_ij) if F_ij = 1 (farmer) or I_ij = w_nij if W_ij = 1 (worker).
- Rural household occupational/internal migration decision:
  - V_dij(p, I_ij, ε_ij) = max_{F_ij, W_ij ∈ {0,1}} { U_dij(p, I_ij | farmer) + ε_fij, U_dij(p, I_ij | worker) + ε_wij }.
- Urban-born staying utility:
  - V_dij(p, I_ij, ε_uij) = U_dij(p, w_nij | worker) + ε_uij.
- Idiosyncratic taste shocks ε_fij, ε_wij, ε_uij are i.i.d. Type-I extreme value (Gumbel) with mean zero and scale parameter σ_{εj}.

### Migration and emigration decision
- Utility of emigrating:
  - V_eij(U_ej, ε_eij) = U_ej + ε_eij, where U_ej is exogenous deterministic component (net foreign wage net of migration costs), allowed to be region-specific; ε_eij i.i.d. Gumbel mean-zero.
- Individual’s stay-or-emigrate decision:
  - V_ij(p, I_ij, ε_ij, U_ej, ε_eij) = max{ V_dij(p, I_ij, ε_ij), V_eij(U_ej, ε_eij) }.
- Emigration indicator E_ij ∈ {0,1}; number of individuals that stay domestically:
  - S = 1 − ∑_j E_j, where E_j ≡ ∫_0^1 E_ij di is the share of emigrants from region j.

### Equilibrium conditions (summary)
- Competitive equilibrium consists of prices {p, q_j} for j = 1,...,J, wages w_nij = A_n z_nij for j = 1,...,J+1; occupational choices {W_ij, F_ij}; emigrate-stay choices {E_ij}; land allocations {l_ij}; consumption allocations {c_aij, c_nij}, given distortions {τ_ij}, such that markets clear.
- Non-agricultural labor market clearing:
  - ∑_j ∫_0^1 W_ij di = W, for j = 1,...,J+1.
- Agricultural goods market clearing:
  - ∑_j ∫_0^1 c_aij di = ∑_j ∫_0^1 F_ij y_ij di.
- Non-agricultural goods market clearing:
  - ∑_j ∫_0^1 c_nij di = ∑_j ∫_0^1 A_n W_ij z_nij di.
- Combined agricultural/non-agricultural equilibrium relationship:
  - p [ Y_a − S ̄a ] = ( ω / (1−ω) ) Y_n,
    - where Y_a ≡ ∑_j ∫_0^1 F_ij y_ij di and Y_n ≡ ∑_j ∫_0^1 A_n W_ij z_nij di.

### Model mechanisms: migration and productivity channels
- Distortions τ_ij are assumed negatively correlated with agricultural ability z_aij (more productive farmers face lower τ_ij).
- Two primary channels by which agricultural distortions affect emigration and production:
  1. Migration (micro) channel — reduced productivity via selection:
     - Distortions increase the likelihood that more productive individuals emigrate and reduce that likelihood for less productive individuals.
     - Holding total emigration constant, distortions produce a more positively selected emigrant pool, lowering aggregate domestic productivity beyond misallocation losses.
     - This channel alters who migrates: in high-distortion regions, selection into emigration is more strongly positive (most productive individuals more likely to leave).
  2. Productivity (macro) channel — increased emigration via income effects:
     - Distortions cause factor misallocation within agriculture, reducing output and lowering household incomes in rural and urban areas.
     - Lower domestic incomes increase attractiveness of opportunities abroad, raising the overall emigration rate.
     - To isolate this channel: simulate a closed-economy scenario with no emigration to compute productivity decline from misallocation; then in a distortion-free baseline (τ_ij = 1) replicate same productivity loss by reducing A_a and compare emigration shares.

- Choice probabilities under Gumbel idiosyncratic shocks follow multinomial logit:
  - p_ki = exp( U_ki / σ_ε ) / ∑_{h∈{f,w,e}} exp( U_hi / σ_ε ), where k ∈ {f,w,e} and σ_ε > 0 is the scale parameter.

- Interaction of channels:
  - Migration channel: distortions lower emigration probability of less productive agents (who are relatively more likely to emigrate) and raise it for more productive agents (who are relatively less likely to emigrate); tends to reduce overall emigration by encouraging those more likely to leave to stay.
  - Productivity channel: distortions reduce aggregate productivity/incomes across distribution, raising emigration probabilities across all groups.
  - Dominance: migration channel dominates at lower distortion levels; productivity channel dominates at higher distortion levels — as distortions increase, income losses raise emigration probability across the entire distribution.

### Theoretical results (Theorems)
- Theorem 1 (Distortions reduce (increase) emigration of less (more) productive agents).
  - Let p_eij be emigration probability of an agent with agricultural productivity z_aij in region j with distortions; p_e^∗ij is corresponding probability in benchmark without distortions.
  - For two identical regions j = H (high distortions) and j = L (low distortions), and two productivity types z_a^l (low) and z_a^h (high):
    - For low-productivity agents:
      - p_e^{lH} − p_e^{∗ lH} < p_e^{lL} − p_e^{∗ lL}.
      - Equivalently, since p_e^{∗ lH} = p_e^{∗ lL}, we have p_e^{lH} − p_e^{lL} < 0.
    - For high-productivity agents:
      - p_e^{hH} − p_e^{∗ hH} > p_e^{hL} − p_e^{∗ hL}.
      - Equivalently, since p_e^{∗ hH} = p_e^{∗ hL}, we have p_e^{hH} − p_e^{hL} > 0.
- Theorem 2 (Distortions reduce incomes and raise emigration across all regions and sectors).
  - Let p_eij(A_a) denote emigration probability when agricultural TFP is A_a.
  - If distortions reduce A_a to A_a^ℓ < A_a (pure misallocation), then for all i and j:
    - p_eij(A_a^ℓ) > p_eij(A_a).

### Simulated illustration (Figure summary)
- Figure shows change in probability of being a farmer p_fij (upper panels) and an emigrant p_eij (lower panels) between distorted and benchmark cases across distribution of log(z_aij) for low- versus high-distortion regions.
  - Low-distortion region:
    - Small changes in p_fij for most abilities; slight reductions at higher ability.
    - Small increases in p_eij at higher abilities; overall modest changes.
  - High-distortion region:
    - Pronounced reductions in p_fij as abilities increase — high distortions discourage farming especially among more productive agents.
    - Decrease in p_eij among low-to-medium ability agents and important increase in p_eij among higher-ability individuals — high distortions strongly encourage emigration among more productive agents.

### Model extension
- Appendix B extends the model to allow individuals to work in agriculture as wage earners in any rural region (choices: be farmer in own region, agricultural worker in any of J regions, migrate to urban non-ag worker, or emigrate).
- Extension permits modeling internal (rural–rural) migration; mechanisms (migration and productivity channels) are preserved in the extended model.
- Authors do not estimate the extended setting due to data limitations (rural-to-rural migration data partial; rural-to-rural migration mostly temporary in Guatemala).

*Source: wpiea2025233-source-pdf - 2.1  Agricultural sector*

### 3.1  Regional division of Guatemala

### 3.1  Regional division of Guatemala

### Regional grouping and agricultural structure
- Guatemala exhibits heterogeneity in geography, weather, ethnic composition, and rural development.
- Agricultural activities are mainly small-scale: 65% of landholdings have less than one hectare and only over 3% have more than ten hectares (INE, 2003; Durr, 2016).
- Main crops: maize, coffee, sugar cane, bananas, and beans — together generate close to 80% of the total agricultural employment (MAGA, 2011; MAGA, 2013).
- Most agricultural activities are informal.

### Regions used in the analysis
- Regions (see map in Figure 3):
  - Western Highlands-Verapaces-Center: comprises ten departments.
  - Dry Corridor-Izabal: comprises six departments.
  - Pacifico-Bocacosta: comprises four departments.
- The department of Guatemala is considered exclusively urban and excluded from agricultural focus.
- The department of Peten is excluded from the study due to predominance of natural grasslands and forest and very limited agricultural activity.

### Regional characteristics and vulnerabilities
- ANOVA results (Appendix Table D.1) indicate departments within each region are more similar than across regions on indicators including travel time to the closest town of 20,000 habitants, risk of frosts, droughts, floodings, geodynamic and geophysical disasters, density of public institutions, prevalence of indigenous population and Spanish as main language reported by household head, poverty rate, and chronic malnutrition rate.
- Region-specific highlights:
  - Western Highlands-Verapaces-Center: important presence of indigenous population, high risk of frosts, poorest, and highest rate of malnutrition.
  - Dry Corridor-Izabal: important risk of droughts and floodings; relatively less accessible.
  - Pacifico-Bocacosta: more developed and better connected; important risk of geodynamic and geophysical disasters and floodings.
- Footnote: Pacifico-Bocacosta is characterized by a larger prevalence of commercial farming relative to the other two regions.

---

### 3.2  Externally calibrated parameters

### Calibration approach and key externally set values
- Elasticity of output with respect to land (α): set to 0.4 (midrange value in the literature).
- Subsistence level of consumption for the agricultural good (a): calibrated to 0.25.
- Weight of agricultural consumption in utility (ω): set to 0.3 so model average employment share in agriculture matches Guatemala’s 30%.
- Aggregate land size in each region (L_j) obtained from IV National Agricultural Census (INE, 2003) and normalized as percentage of total land:
  - L_1 = 0.5 (Region 1: Western Highlands-Verapaces-Center)
  - L_2 = 0.3 (Region 2: Dry Corridor-Izabal)
  - L_3 = 0.2 (Region 3: Pacifico-Bocacosta)
- Aggregate agricultural and non-agricultural productivities normalized to one.
- Sensitivity: α examined for alternative values 0.3 and 0.5.

### Table of externally calibrated parameters (selected)
- A_n = 1 (Aggregate non-agricultural productivity; Normalized)
- A_a = 1 (Aggregate agricultural productivity; Normalized)
- α = 0.4
- a = 0.25
- ω = 0.3
- L_1 = 0.5
- L_2 = 0.3
- L_3 = 0.2

---

### 3.3  Productivity and distortions

### Productivity distributions and estimation
- Assumption: agricultural and non-agricultural productivity follow log-normal distributions.
  - Agricultural: log(z_aij) ∼ N(μ_a j, σ_a j) for region j.
  - Non-agricultural: log(z_nij) ∼ N(μ_n, σ_n).
- Agricultural TFP estimation (two-stage approach, Britos et al., 2022):
  1. Derive TFP from agricultural census (INE, 2003) using production and land and labor inputs.
  2. Regress TFP on controls not in the theoretical framework: farmer’s education, share of family labor force, use of machinery, equipment, enhanced seeds, fertilizer, pesticides, irrigation system, livestock ownership, number of different crops produced, farmer’s age and gender; include village fixed effects. Agricultural productivity is the regression residual adjusted for productivity changes based on age and sex.
  3. Compute mean and standard deviation for each region (μ_a j, σ_a j).
- Non-agricultural TFP estimated similarly using 2019 National Survey of Labor and Income (INE, 2019) with workers’ non-agricultural income, education, age, and gender to obtain residuals and parameters (μ_n, σ_n).

### Modeling agricultural distortions
- Distortions follow Adamopoulos & Restuccia (2014) framework:
  - log(τ_ij) = ψ_j z_aij + ε_ij
  - τ_ij: idiosyncratic distortion faced by producer i in region j.
  - z_aij: agricultural productivity of producer i.
  - ε_ij ∼ N(0, σ_τ j).
  - ψ_j captures correlation between distortions and agricultural productivity in region j.
- Interpretation:
  - ψ_j < 0 indicates market interventions/imperfections drive factor misallocation.
  - Larger |ψ_j| implies greater distortions and higher inefficiency in land use.

---

### 3.4  Estimation of remaining parameters and empirical results

### Estimation method and targeted moments
- Parameters estimated via Simulated Method of Moments (SMM): {ψ_j, σ_τ j, U_e j, σ_ε j}_j∈J.
- Moment choices and parameter identification:
  - ψ_j identified by matching correlation within region j between total factor productivity (TFP_ij) and revenue-based total factor productivity (TFPR_ij).
  - σ_τ j estimated by targeting standard deviation of TFPR_ij for each j.
  - U_e j (utility of emigrating) targeted by share of emigrants relative to total employed population in each region j.
  - σ_ε j identified by matching share of household members employed in the non-agricultural sector from region j.

### Targeted moments (data vs model) and parameter estimates (Table 2)
- Share of Emigrants:
  - Region 1, E_1: Data 16.50 — Model 16.50
  - Region 2, E_2: Data 13.80 — Model 14.00
  - Region 3, E_3: Data 8.70 — Model 8.80
- Corr(TFP_j, TFPR_j):
  - Region 1: Data 0.48 — Model 0.48
  - Region 2: Data 0.57 — Model 0.57
  - Region 3: Data 0.50 — Model 0.50
- S.D.(TFPR_j):
  - Region 1: Data 0.82 — Model 0.82
  - Region 2: Data 1.09 — Model 1.07
  - Region 3: Data 0.95 — Model 0.90
- Share of Workers (W_j):
  - Region 1, W_1: Data 52.20 — Model 54.69
  - Region 2, W_2: Data 54.90 — Model 58.32
  - Region 3, W_3: Data 62.50 — Model 62.76

- Estimated parameter values:
  - Utility of emigrating:
    - U_1 = 0.61
    - U_2 = 0.42
    - U_3 = 0.49
  - Association between z_aij and τ_ij (absolute values):
    - |ψ_1| = 0.07 (Region 1)
    - |ψ_2| = 0.09 (Region 2)
    - |ψ_3| = 0.03 (Region 3)
  - Dispersion of τ_ij:
    - σ_τ1 = 0.70
    - σ_τ2 = 0.90
    - σ_τ3 = 0.85
  - Gumbel scale-parameter:
    - σ_ε1 = 1.21
    - σ_ε2 = 1.42
    - σ_ε3 = 0.81

### Key empirical interpretations and patterns
- Distortions ranking by |ψ_j|:
  - Highest in Dry Corridor-Izabal (|ψ_2| = 0.09)
  - Next Western Highlands-Verapaces-Center (|ψ_1| = 0.07)
  - Lowest Pacifico-Bocacosta (|ψ_3| = 0.03)
- Dry Corridor-Izabal exhibits highest dispersion in revenue-based TFP (σ_τ2 = 0.90), indicating significant factor misallocation.
- Emigration incentives:
  - Western Highlands-Verapaces-Center shows strongest incentives to migrate (U_1 = 0.61), followed by Pacifico-Bocacosta (U_3 = 0.49) and Dry Corridor-Izabal (U_2 = 0.42).
  - Noted disparity: Pacifico-Bocacosta U_3 = 0.49 comparable to Dry Corridor-Izabal U_2 = 0.42 despite higher emigration share in Dry Corridor-Izabal (13.8 versus 8.7) — possibly due to greater agricultural distortions in Dry Corridor-Izabal driving emigration.
- Non-agricultural employment and idiosyncratic decision variance:
  - Pacifico-Bocacosta higher share of non-agricultural workers (~63%) corresponds to lower Gumbel scale-parameter (σ_ε3 = 0.81 versus σ_ε1 = 1.21 and σ_ε2 = 1.42), suggesting reduced variance in idiosyncratic (non-economic) decision factors.

### Model-implied distributions and land market evidence
- Kernel density of log(TFPR) by region (Figure 4):
  - Pacifico-Bocacosta: most concentrated distribution → lowest agricultural market distortions.
  - Dry Corridor-Izabal: flattest distribution → largest factor misallocation.
  - Western Highlands-Verapaces-Center: intermediate dispersion.
- Appendix Figure D.1 (actual vs no-distortion output):
  - Pacifico-Bocacosta observations closest to 45-degree line → minimal distortions.
  - Western Highlands-Verapaces-Center moderate scatter.
  - Dry Corridor-Izabal greatest scatter → substantial distortions.
- Land market complementary evidence (Appendix Table D.2) from three-year panel (2012–2014) covering 176 municipalities:
  - Pacifico-Bocacosta: significantly larger share of land rentals and considerably lower coefficient of price variation of best land compared to other regions → consistent with less market distortions.
  - Western Highlands-Verapaces-Center and Dry Corridor-Izabal: fewer transactions and higher price dispersion consistent with greater distortions.

---

### 4.1  Counterfactual scenario (benchmark construction)

### Procedure to construct benchmark scenario
1. Re-estimate distortions at the department level for more granular estimates.
2. Identify the most “efficient” department within each region by selecting the department with the lowest correlation between TFPR_ij and TFP_ij.
3. Re-calibrate region-level distortion distribution parameters ψ_j and σ_τ j to match correlation between TFPR and TFP and standard deviation of TFPR observed in the selected benchmark departments.
4. Use model outcomes from re-calibrated parameters to construct benchmark scenario and perform counterfactual exercises.

### Selected benchmark departments (Table D.3)
- Quetzaltenango (Region 1): Corr(TFP_ij, TFPR_ij) = 0.34; S.D.(TFPR_ij) = 0.57
- Izabal (Region 2): Corr(TFP_ij, TFPR_ij) = 0.42; S.D.(TFPR_ij) = 0.72
- Escuintla (Region 3): Corr(TFP_ij, TFPR_ij) = 0.41; S.D.(TFPR_ij) = 0.86
- Note: These departments have the lowest within-region correlations between TFP_ij and TFPR_ij and are among the most developed in their regions.

### Calibration outcomes under benchmark scenario (reported outcomes up to truncation)
- For Western Highlands-Verapaces-Center (Region 1) under benchmark:
  - |ψ_1| decreases from 0.07 to 0.03.
  - σ_τ1 decreases from 0.70 to 0.52.
- For Dry Corridor-Izabal (Region 2) under benchmark:
  - |ψ_2| declines (text truncated before providing full post-calibration value for Region 2 and Region 3 in the source excerpt).

*Source: wpiea2025233-source-pdf - 3.1  Regional division of Guatemala*

### 0.09 to 0.05 andσ

### 0.09 to 0.05 andσ τ 2 drops from 0.90 to 0.62.

### Model-implied distortions and TFPR distributions
- Figure 5 compares kernel density plots of log(TFPR) for three regions under the actual scenario (solid red lines) and the benchmark scenario (dashed green lines).
- Western Highlands–Verapaces–Center:
  - Benchmark scenario shows a narrower distribution (lower variance) and a higher peak versus actual, indicating more concentrated TFPR around the mean and improved allocation.
- Dry Corridor–Izabal:
  - Benchmark scenario also yields a narrower distribution and a higher peak, reflecting significantly reduced distortions relative to the actual scenario.
- Pacifico–Bocacosta:
  - Actual and benchmark distributions are closely aligned; benchmark exhibits only a slightly narrower spread and marginally higher peak, indicating limited gains in resource reallocation efficiency.
- Overall summary: transitioning to the benchmark scenario leads to a reduced dispersion of log(TFPR), particularly in Western Highlands–Verapaces–Center and Dry Corridor–Izabal; Pacifico–Bocacosta shows limited gains.

### Targeted moments and calibrated parameters (Table 4)
- Targeted moments (Data vs Model):
  - Corr(TFP1, TFPR1): 0.34 vs 0.34
  - Corr(TFP2, TFPR2): 0.42 vs 0.42
  - Corr(TFP3, TFPR3): 0.41 vs 0.40
  - S.D.(TFPR1): 0.57 vs 0.58
  - S.D.(TFPR2): 0.72 vs 0.73
  - S.D.(TFPR3): 0.86 vs 0.84
- Calibrated parameters (Value):
  - Association between zaij and τij in Region 1, |ψ1| 0.033
  - Association between zaij and τij in Region 2, |ψ2| 0.048
  - Association between zaij and τij in Region 3, |ψ3| 0.023
  - Dispersion of τij in Region 1, στ1 0.52
  - Dispersion of τij in Region 2, στ2 0.62
  - Dispersion of τij in Region 3, στ3 0.76
- Note: Region 1 = Western Highlands–Verapaces–Center; Region 2 = Dry Corridor–Izabal; Region 3 = Pacifico–Bocacosta.

### Reduction in the agricultural gap (Table 5)
- Reduction in efficiency loss (difference between distortion-free and actual production) when moving to benchmark:
  - Region 1: 37.3%
  - Region 2: 41.0%
  - Region 3: 14.5%
  - Total (all regions): 30.3%
- Interpretation: Nationwide reduction in the gap is substantial—close to one-third of the total agricultural gap.

### Counterfactual estimates — Overall changes (Table 6)
- Changes when transitioning from actual to benchmark by region:
  - Region 1 (Western Highlands–Verapaces–Center):
    - ∆ Share of Emigrants: -3.0 p.p.
    - ∆ Share of Workers (non-agricultural): 1.1 p.p.
    - ∆ Agricultural Productivity: 33.3%
    - ∆ Median Consumption: 11.0%
  - Region 2 (Dry Corridor–Izabal):
    - ∆ Share of Emigrants: -2.2 p.p.
    - ∆ Share of Workers: -0.1 p.p.
    - ∆ Agricultural Productivity: 45.6%
    - ∆ Median Consumption: 12.2%
  - Region 3 (Pacifico–Bocacosta):
    - ∆ Share of Emigrants: -1.8 p.p.
    - ∆ Share of Workers: 1.7 p.p.
    - ∆ Agricultural Productivity: 14.4%
    - ∆ Median Consumption: 8.8%
  - Urban:
    - ∆ Share of Emigrants: -1.6 p.p.
    - ∆ Share of Workers: 1.6 p.p.
    - ∆ Median Consumption: 14.6%
  - Total (aggregate including three agricultural regions and urban):
    - ∆ Share of Emigrants: -2.3 p.p.
    - ∆ Share of Workers: 1.2 p.p.
    - ∆ Agricultural Productivity: 30.1%
    - ∆ Median Consumption: 12.3%
- Additional points:
  - Dry Corridor–Izabal shows largest improvement in agricultural productivity (45.6%).
  - Western Highlands–Verapaces–Center shows largest reduction in emigration share (-3.0 p.p.).
  - Pacifico–Bocacosta shows smallest productivity gain (14.4%) but largest labor reallocation to non-agriculture (1.7 p.p.).
  - Using population context: total emigration share decrease (-2.3 p.p.) equates to a 35.5% decrease in the Guatemalan population currently living in the US, given total population 16.3 million (INE, 2020) and approximately 1.1 million Guatemalan migrants in the US (US Census Bureau, 2021).
- Sensitivity to α (elasticity of output w.r.t. land):
  - For α = 0.3: total change in emigration share = −2.7 p.p.; aggregate productivity increase = 21.7%; median consumption increase = 8.4%.
  - For α = 0.5: total change in emigration share = −1.3 p.p.; aggregate productivity increase = 37.1%; median consumption increase = 11.7%.
  - Share of non-agricultural workers change remains 1.2 p.p. across α values.

### Welfare (equivalent variation and distributional effects)
- Equivalent variation approach yields consumption-equivalent welfare gains at the median:
  - Western Highlands–Verapaces–Center: 4.6%
  - Dry Corridor–Izabal: 4.5%
  - Pacifico–Bocacosta: 4.3%
  - All three regions combined with urban area: 4.5% at the median
- Distributional findings:
  - Lower percentiles experience larger percentage increases in welfare across all regions.
  - The 10th percentile exhibits the greatest gains; the 90th percentile the smallest.
  - The welfare gain at the 10th percentile is more than twice that at the 90th percentile across the three regions.
  - Gains are particularly notable for low-income groups in Dry Corridor–Izabal and Western Highlands–Verapaces–Center.

### Probability of emigration (Figure 7 and related findings)
- Change in individual probability of emigrating (p e* ij − p e ij) plotted across percentiles of log(zaij) shows:
  - An inverted-u-shaped relationship between agricultural productivity and changes in migration probabilities, with regional variation.
  - Among higher-ability individuals, decreases in emigration probability under the benchmark are most pronounced in Western Highlands–Verapaces–Center, then Dry Corridor–Izabal, then Pacifico–Bocacosta.
- Emigrants’ productivity under the benchmark scenario (average) is lower than in the actual case by:
  - Western Highlands–Verapaces–Center: 11.9%
  - Dry Corridor–Izabal: 12.3%
  - Pacifico–Bocacosta: 13.9%
- Median declines (alternative metric) for emigrants:
  - Western Highlands–Verapaces–Center: 4.8%
  - Dry Corridor–Izabal: 2.4%
  - Pacifico–Bocacosta: 8.4%
- Comparison using ENCOVI 2014:
  - Households with higher expenditures show a notably lower likelihood of emigration in benchmark departments versus other departments.
  - Western Highlands–Verapaces–Center shows lower probability only at the 90th percentile; other regions show steadily growing differences as income increases.

### Contribution of the migration channel to productivity gains (Table 7)
- Aggregate agricultural productivity increase between actual and benchmark when emigration is eliminated:
  - Region 1: 29.3%
  - Region 2: 40.9%
  - Region 3: 14.2%
  - Total: 30.1%
- Migration contribution (share of total gains attributable to migration channel):
  - Region 1: 11.9%
  - Region 2: 10.3%
  - Region 3: 1.3%
  - Total: 9.2%
- Interpretation: Migration accounts for a non-negligible but minority share of productivity gains; the remainder is driven by “pure” misallocation improvements.

*Source: wpiea2025233-source-pdf*

### 4.3  Agricultural distortions and observable local characteristics

### 4.3  Agricultural distortions and observable local characteristics

### Regression framework and data
- Unit of analysis: municipality level (each observation is a municipality).
- Dependent variable: Agricultural distortions (ψ_j) estimated at the municipality level.
- Two model specifications:
  - Column (1): regressors include socioeconomic, accessibility, institutional, and cultural indicators:
    - Occupational precariousness index
    - Per capita bank deposit accounts
    - Mobile base stations (towers) per square kilometer
    - Travel time to the closest town of 20,000 habitants
    - Government (public institutions) density index
    - Share of indigenous population
  - Column (2): adds climate vulnerability and insecurity indicators:
    - Index of natural hazards
    - Rate of extortions
- Data sources (as described in the text):
  - Occupational precariousness index and government density index: Food Insecurity and Malnutrition Vulnerability Index (IVISAN), 2012 (SESAN).
  - Per capita bank deposit accounts: Superintendente of Banks of Guatemala (SIB), 2018.
  - Mobile base stations per square kilometer: Superintendence of Telecommunications of Guatemala (SIT), 2018.
  - Travel time to closest town of 20,000 habitants: IFPRI typology of micro-regions exercise, 2021.
  - Share of indigenous population: 2018 Population and Housing Census (INE).
  - Index of natural hazards: INSIVUMEH, period 1530-2015.
  - Rate of extortions: INFOSEGURA-Guatemala database, 2018.
- All variables were standardized prior to the regression.
- Regressions include regional fixed effects; reported standard errors are robust.

### Estimated relationships (Table 8 coefficients)
- Table 8: Relationship between agricultural distortions and observable characteristics at the municipality level (N = 301). R-squared = 0.062 in both specifications. Robust standard errors in parentheses. *, **, *** denote significance at 10%, 5%, and 1% levels.

- Coefficients (Column (1), Column (2)):
  - Occupational precariousness index: -0.156  (-0.095) ; -0.156  (0.100)
  - Per capita bank deposit accounts: -0.168**  (0.074) ; -0.170**  (0.074)
  - Mobile base stations per square kilometer: 0.036  (0.058) ; 0.037  (0.059)
  - Travel time to closest town of 20,000 habitants: 0.178**  (0.070) ; 0.178**  (0.071)
  - Government density index: -0.141**  (0.062) ; -0.143**  (0.069)
  - Share of indigenous population: 0.008  (0.073) ; 0.009  (0.073)
  - Index of natural hazards: (not in column (1)) 0.003  (0.029)
  - Rate of extortions: (not in column (1)) 0.000  (0.068)
  - Constant: -0.155**  (0.073) ; -0.169  (0.165)
  - Regional fixed effects: Yes, Yes
  - Observations: 301, 301
  - R-squared: 0.062, 0.062

### Key empirical findings
- Financial penetration:
  - Per capita bank deposit accounts is negatively associated with agricultural distortions (coefficient -0.168** in column (1), -0.170** in column (2)), indicating higher financial penetration is associated with lower distortions.
  - Interpretation: limited financial access in some locations could be contributing to distortions in agricultural markets.
- Accessibility:
  - Travel time to the closest town of 20,000 habitants is positively correlated with agricultural distortions (coefficient 0.178** in both specifications), suggesting reduced accessibility and higher transaction costs are associated with larger distortions.
- Government presence:
  - Government density index is negatively associated with agricultural distortions (coefficient -0.141** in column (1), -0.143** in column (2)), pointing to the importance of public sector and institutional presence to reduce market inefficiencies.
- Climate vulnerability and insecurity:
  - Index of natural hazards (0.003) and rate of extortions (0.000) are not statistically significant in column (2).
  - Implication: the estimated contribution of modeled agricultural distortions on emigration is not necessarily driven by climatic or insecurity events.
- Other indicators:
  - Occupational precariousness index and share of indigenous population show no statistically significant correlation with agricultural distortions in these specifications.

### Policy implications and recommendations
- Targeted policy areas to reduce agricultural distortions and improve local outcomes:
  - Promote financial inclusion and broaden access to banking services (given the negative association between per capita bank deposit accounts and distortions).
  - Improve road infrastructure and reduce travel times to urban centers (addressing the positive correlation between travel time and distortions).
  - Expand government presence and public institutions in underserved areas (given the negative association between government density index and distortions).
- Expected local outcomes from reducing agricultural distortions:
  - Increase local productivity and attenuate emigration by reducing market inefficiencies that encourage relocation.
  - Specifically, development programs addressing agricultural market distortions can prevent productive workers from leaving communities while improving agricultural productivity and local welfare.

### Broader conclusions from the paper (contextual numeric results)
- Simulations reported in the paper (aggregate results summarized in the concluding remarks):
  - Reducing distortions in the agricultural sector to the most efficient department in each region can:
    - Decrease the share of Guatemalan emigrants by 2.3 p.p.
    - Increase aggregate agricultural productivity by 30.1%
    - Increase median household welfare by 4.5%
  - The simulated decrease in emigrants by 2.3 p.p. is described as primarily among more productive workers—an amount equivalent to over 35% of the Guatemalan population currently residing in the US.

*Source: IMF working paper chapter 4.3 (municipality-level analysis and Table 8).*

### References

### References and Appendix A — Proofs of the migration and productivity channels

### Migration channel — main propositions and logical steps
- Proposition for low-productivity agricultural individuals (denoted by z_a_l):
  - The change in the probability of emigration satisfies: p_e_lj − p_e*_lj < 0, for all j.
  - Emigration probability expression used (Gumbel-distributed utility random terms):
    - p_e_ij = exp(U_e/σ_ε) / [ exp(U_e/σ_ε) + exp(U_w_ij/σ_ε) + exp(U_f_ij/σ_ε) ].
  - Key comparisons and algebraic rearrangements produce:
    - exp(U_w*_lj/σ_ε) + exp(U_f*_lj/σ_ε) < exp(U_w_lj/σ_ε) + exp(U_f_lj/σ_ε).
  - Wage in the non-agricultural sector unaffected by distortions: w_n*_lj = w_n_lj, hence U_w*_lj ≈ U_w_lj for sufficiently mild distortions.
  - Conclusion: exp(U_f*_lj/σ_ε) < exp(U_f_lj/σ_ε) which holds when U_f*_lj < U_f_lj.
  - Mechanism: distortions are weakest for low-ability agents because τ_ij decreases with productivity; these individuals receive higher income from profits and lump-sum transfers—and attain higher utility—under distortions.

- Proposition for high-productivity agricultural individuals (denoted by z_a_h):
  - The change in the probability of emigration satisfies: p_e_hj − p_e*_hj > 0, for all j.
  - Comparison leads to:
    - exp(U_w*_hj/σ_ε) + exp(U_f*_hj/σ_ε) > exp(U_w_hj/σ_ε) + exp(U_f_hj/σ_ε).
  - Because w_n*_hj = w_n_hj but p* < p, it follows that U_w*_hj ≥ U_w_hj, yielding:
    - exp(U_f*_hj/σ_ε) > exp(U_f_hj/σ_ε) whenever U_f*_hj > U_f_hj.
  - Mechanism: distortions are strongest for high-ability agents (τ_ij decreases with productivity), so high-productivity individuals receive lower income from profits and lump-sum transfers under distortions; agricultural utility falls and p_e_hj > p_e*_hj.

- Comparative-static for two identical regions differing in distortion degree (j = H high distortions, j = L low distortions):
  - Emigration probability inequalities:
    - p_e_lH < p_e_lL
    - p_e_hH > p_e_hL
  - Proof outline:
    - For low-productivity individuals (z_a_l): w_n_lL = w_n_lH ⇒ U_w_lL = U_w_lH; rearrangement yields U_f_lL < U_f_lH because τ_lH > τ_lL and τ_hH < τ_hL (distortions weaker for low-ability agents in the high-distortion region).
    - Therefore, p_e_lH < p_e_lL.
    - Symmetric logic yields U_f_hL > U_f_hH and p_e_hH > p_e_hL.

### Productivity channel — setup and main insight
- Core statement:
  - Agricultural distortions reduce household incomes in both agricultural and non-agricultural sectors by inducing factor misallocation.
  - This decline in incomes enhances the relative attractiveness of foreign opportunities in equilibrium, thereby increasing emigration from both rural and urban areas.
  - Even absent selection effects, distortions that depress aggregate productivity increase the overall emigration rate.

- Two-step comparison to isolate the productivity channel (as stated in the text):
  1. Compute a closed-economy equilibrium with agricultural distortions where international emigration is not allowed.
     - Let A_a denote the agricultural productivity parameter in this economy.
     - Let ̄Y_a be the resulting aggregate agricultural output.
     - In this scenario, aggregate productivity is reduced due to misallocation.

- Analytical role of the productivity channel:
  - The productivity channel captures the effect of income losses (from misallocation-induced productivity declines) on the share of emigrants, distinct from the migration channel which affects selection into migration.
  - By first computing the closed-economy equilibrium with distortions (step 1), the framework isolates how reduced A_a and ̄Y_a alter the returns to staying versus emigrating when international mobility is reintroduced.

*Excerpted content unit: wpiea2025233-source-pdf - References and Appendix A (Proofs of the migration and productivity channels).*

### 2.  Next, consider a counterfactual open-economy equilibrium with no agricultural dis-

### 2. Next, consider a counterfactual open-economy equilibrium with no agricultural distortions (i.e., τij = 1 for all i,j)

### Proof that reducing Aa (aggregate agricultural productivity) increases emigration probabilities
- Setup:
  - Compare two equilibria: one with agricultural distortions and aggregate productivity Aa, and a counterfactual with no agricultural distortions (τij = 1 for all i,j) but lower aggregate agricultural productivity Aal < Aa, chosen so aggregate agricultural output Ȳa is the same.
  - Emigration probability under the multinomial logit structure:
    - peij(Aa) = exp(Uej/σε) / [exp(Uej/σε) + exp(Uwij(Aa)/σε) + exp(Ufij(Aa)/σε)]
    - Similarly defined for Aal < Aa.
  - Uej is exogenous and unchanged across the two equilibria.
- Key monotonicity arguments:
  - Non-agricultural utility: A decrease in Aa raises the relative price p of the agricultural good, making subsistence agricultural consumption ā more costly, tightening the household budget constraint and lowering non-agricultural utility:
    - Uwij(Aa) > Uwij(Aal).
  - Agricultural utility: Farming profits are increasing in Aa, so reducing Aa reduces farmer incomes:
    - Ufij(Aa) > Ufij(Aal).
- Conclusion:
  - exp[Uwij(Aa)/σε] + exp[Ufij(Aa)/σε] > exp[Uwij(Aal)/σε] + exp[Ufij(Aal)/σε], implying
    - peij(Aal) > peij(Aa) for all i and j.
  - Therefore, replicating productivity loss purely via a decline in Aa (holding emigration open and abstracting from selection effects) leads to a strictly higher probability of emigration for every individual. Misallocation, operating through the productivity channel, increases the emigration rate.

### Model extension with agricultural labor and rural-rural migration — setup and choices
- Individual choices (born in region j) conditional on productivity draw:
  - (i) farm in region j (use land l and labor n to produce agricultural good);
  - (ii) work as an agricultural wage earner in any region;
  - (iii) work in the non-agricultural sector;
  - (iv) emigrate, if welfare from emigrating exceeds staying.
- Farmer production technology (farmer i in region j):
  - yij = Aa zaij lαij nβij, with Aa aggregate agricultural productivity, α, β ∈ (0,1), α + β < 1, and zaij the farmer’s productivity draw.
- Farmer profit maximization (competitive market):
  - π(zaij, τij) = max_{lij, nij} τij p Aa zaij lαij nβij − qj lij − waj nij,
    - p = relative price of agricultural good,
    - qj = rental price of land in region j,
    - waj = agricultural-sector wage in region j,
    - τij = region- and individual-specific distortion.
- Optimal factor allocations (given prices qj, waj, p and Aij ≡ Aa zaij τij):
  - lij = [(α/β)(waj/qj)]^{(1−β)/(1−α−β)} [β(p/waj) Aij]^{1/(1−α−β)}
  - nij = [(α/β)(waj/qj)]^{α/(1−α−β)} [β(p/waj) Aij]^{1/(1−α−β)}

### Non-agricultural sector and migration decision
- Non-agricultural production:
  - Yn = An ∑_j ∫_0^1 Wij znij di, where Wij ∈ {0,1} indicator of non-agricultural worker, znij individual non-agricultural productivity, An aggregate non-agricultural productivity.
  - Individual non-agricultural wage in equilibrium: wnij = An znij.
- Migration decision (individual i born in j):
  - Choose between staying (Sij ∈ {0,1}) and emigrating (Eij ∈ {0,1}) by maximizing {Vdij, Veij} where:
    - Vdij = Udij + εdij, and Udij is the deterministic utility of staying,
    - Veij = Uej + εeij, with Uej foreign wage net of moving costs,
    - εdij and εeij follow Gumbel distributions.
- Conditional staying utility household problem:
  - Udij = max_{caij, cnij} { ω log(caij − a) + (1 − ω) log(cnij) } subject to p caij + cnij ≤ Iij + T,
    - a ≥ 0 subsistence agricultural consumption,
    - ω ∈ (0,1) weight on agricultural consumption,
    - Iij = max over occupational incomes {π(zaij, τij), wnij − bij, waj, wak̸=j − κik},
      - κik commuting cost in non-agricultural goods,
      - T per capita transfers (land rental from government and lump-sum redistributed tax revenues).
  - Consumption solutions:
    - caij = ω(Iij + T)/p + (1 − ω) ā
    - cnij = (1 − ω)(Iij + T) − (1 − ω) p ā

### Equilibrium definitions and market clearing conditions
- Competitive equilibrium consists of prices {p, qj, waj} for all j and wnij = An znij for all (i,j); occupational choices {Wij, Fij, Njij, Nkij}; emigrate-stay choices {Sij, Eij}; farmers’ factor allocations {lij, nij}; and consumption allocations {caij, cnij} such that markets clear.
- Market clearing:
  - Land: ∫_0^1 Fij lij di = Lj for all j.
  - Agricultural labor (labor demand from farmers in j equals total labor supply to region j):
    - ∫_0^1 nij di |_{Labor demand from farmers in j} = ∑_k ∫_0^1 Njik di |_{Total labor supply to region j} ≡ Nj for all j.
  - Non-agricultural labor: ∑_j ∫_0^1 Wij di = W.
  - Agricultural good: ∑_j ∫_0^1 caij di = ∑_j ∫_0^1 τij Fij yij di.
  - Non-agricultural good: ∑_j ∫_0^1 [cnij + κik Nkij] di = An ∑_j ∫_0^1 Wij znij di, where ∑_j ∫_0^1 κik Nkij di (for k̸=j) denotes total commuting costs.

### Mechanisms: migration and productivity channels in the extended model
- Distortions operate through two channels:
  - Migration (selection) channel: Higher agricultural distortions increase emigration probability of more productive agents and reduce that of less productive agents via selection.
  - Productivity channel: Distortions generate factor misallocation and reduce aggregate incomes, increasing emigration incentives broadly.
- Extended model differences relative to base model:
  - Expanded choice set allows agents to become agricultural wage workers and engage in internal rural mobility.
  - These additional margins primarily affect medium- and high-ability agents, who can substitute away from farming without emigrating.
  - Qualitative implications for aggregate emigration and productivity remain consistent with the base model: aggregate emigration response to distortions is non-monotonic due to interaction of migration and productivity channels.
  - At low distortions: selection effects can dominate, making emigration more positively selected and potentially lowering aggregate emigration as low-productivity individuals become less likely to emigrate.
  - At high distortions: income and price effects dominate, causing broad increases in emigration due to declining productivity and rising food prices.

### Data and calibration — key constructions and externally calibrated parameters
- Employment and migration shares:
  - Farmers, NA-workers, and international migrants per region j are constructed as:
    - Number of farmers: 2018 Population and Housing Census (INE, 2020) occupational categories (market-oriented farming, skilled agricultural labor, subsistence agricultural workers, fishers, hunters, gatherers, and agricultural laborers).
    - Number of non-agricultural workers (NA): entrepreneurs and workers in occupations outside agriculture, forestry, and fishing (INE, 2020).
    - Number of international migrants: stock of Guatemalan migrants working in the US from 2021 American Community Survey (US Census Bureau, 2021), allocated across regions using 2002–2018 regional migration shares from Guatemalan Census (INE, 2020).
    - Number of employed people = farmers + NA-workers + international migrants.
- Agricultural production and income data:
  - Microdata from IV National Agricultural Census (INE, 2003) for 2002–03 crop year: crop output, land allocation, inputs, labor use, machinery, socioeconomic characteristics.
  - Crop-level revenues computed using 2002 average monthly prices from MAGA (Food Price Monitoring and Analysis Tool (MAGA, 2025)).
  - Land input = land allocated to permanent and seasonal crops.
  - Labor input = number of people involved in farm work (family + hired), measured at farm level.
  - Non-agricultural income from 2019 National Survey of Labor and Income (INE, 2019).
- Externally calibrated parameters (Table C.3):
  - An = 1 (Aggregate non-agricultural productivity; Normalized)
  - Aa = 1 (Aggregate agricultural productivity; Normalized)
  - α = 0.4 (Land share in agricultural production; Midrange value in the literature)
  - a = 0.25 (Subsistence consumption of agricultural good; Adamopoulos et al. (2024))
  - ω = 0.3 (Weight of agricultural consumption in utility; Employment share in agriculture)
  - L1 = 0.5 (Normalized aggregate land size in Region 1; IV National Agricultural Census)
  - L2 = 0.3 (Region 2)
  - L3 = 0.2 (Region 3)
  - Note: Region 1 = Western Highlands-Verapaces-Center, Region 2 = Dry Corridor-Izabal, Region 3 = Pacifico-Bocacosta.
- Agricultural productivity estimation (TFP/T FP R):
  - TFPij = yij / lαij, where yij = farm total revenue per unit of labor and lij = farm total land used per unit of labor. In this setting TFPij ≡ zaij.
  - Two-stage estimation (Britos et al. (2022) approach):
    - First derive zaij from census data and MAGA prices.
    - Second, regress log(yipj / lαipj) = β1j X1ipj + β2j X2ipj + ηp + εipj where:
      - X1ipj controls: age and sex of farmer i in village p region j.
      - X2ipj controls: years of education, ratio of household farm labor, machinery/equipment indicator, high-performance seeds indicator, fertilizer/pesticide use, irrigation system indicator, livestock, number of crops produced.
      - ηp = village fixed effect, εipj residual.
    - Predicted log productivity:
      - log(ẑaij) = β̂1j X1ij + ε̂ij.
    - Remove top and bottom percentiles as outliers.
  - Region-specific estimates (Table C.4):
    - Region 1: Mean (μ̂aj) = 1.31; Std. Dev. (σ̂aj) = 0.90
    - Region 2: Mean (μ̂aj) = 1.39; Std. Dev. (σ̂aj) = 1.05
    - Region 3: Mean (μ̂aj) = 1.47; Std. Dev. (σ̂aj) = 1.12
  - TFPRij (revenue-based TFPR) defined as TFPRij = yij / lij and computed by plugging ẑaij into yij.
- Non-agricultural productivity:
  - Estimate log(NA Incomeip) = λ1 W1ip + λ2 W2ip + φp + νip where:
    - W1ip controls: age and sex,
    - W2ip: education,
    - φp: sampling unit fixed effect.
  - Predicted non-agricultural productivity:
    - log(ẑnai) = λ̂1 W1i + ν̂i.
  - Keep urban individuals, remove top and bottom percentiles.
  - Resulting estimates:
    - μ̂n = 0.8
    - σ̂n = 0.59

### Selected supplemental empirical results and calibration moments
- Region characteristics (Table D.1) — selected means (standard deviations in parentheses):
  - Travel time to closest town of 20,000 habitants (hours): Region 1 = 1.317 (0.514); Region 2 = 1.503 (0.603); Region 3 = 0.761 (0.122); ANOVA p-value = 0.091.
  - Share of indigenous population (0-100): Region 1 = 70.088 (24.290); Region 2 = 18.046 (14.772); Region 3 = 20.490 (14.888); ANOVA p-value = 0.000.
  - Poverty rate (0-100): Region 1 = 69.172 (12.696); Region 2 = 60.443 (7.735); Region 3 = 59.990 (4.384); ANOVA p-value = 0.185.
- Land rentals and perceived prices of best land (Table D.2):
  - % Renting Land, Mean Price, Std. Dev., CV:
    - Region 1: 24.6% | 26.474 | 0.25 | 1.52
    - Region 2: 45.4% | 18.133 | 0.71 | 1.69
    - Region 3: 61.9% | 32.883 | 8.98 | 1.19
- Targeted moments from selected benchmark departments (Table D.3):
  - Quetzaltenango, Region 1: Corr(TFPij, TFPRij) = 0.34; S.D.(TFPRij) = 0.57
  - Izabal, Region 2: Corr = 0.42; S.D. = 0.72
  - Escuintla, Region 3: Corr = 0.41; S.D. = 0.86
- Robustness: sensitivity to land elasticity α (Table D.4):
  - Panel A: α = 0.3 — Changes from actual to benchmark (Region, ∆ Share of Emigrants (p.p.), ∆ Share of Workers (p.p.), ∆ Agricultural Productivity (%), ∆ Median Consumption (%)):
    - Region 1: -3.0 | 0.6 | 19.2 | 4.5
    - Region 2: -2.4 | -0.6 | 32.8 | 8.5
    - Region 3: -2.2 | 2.4 | 19.6 | 7.8
    - Urban: -2.4 | 2.4 | – | 13.3
    - Total: -2.7 | 1.2 | 21.7 | 8.4
  - Panel B: α = 0.5:
    - Region 1: -1.5 | 1.0 | 37.6 | 13.1
    - Region 2: -1.2 | 0.8 | 49.4 | 12.9
    - Region 3: -0.9 | 2.7 | 24.8 | 10.0
    - Urban: -1.0 | 1.0 | – | 11.2
    - Total: -1.3 | 1.2 | 37.1 | 11.7

*Source: Agricultural Distortions and International Migration — Working Paper No. WP/2025/233.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025233-source-pdf.pdf_