## wpiea2023017-print-pdf

## Source details

**Canonical URL:** [wpiea2023017-print-pdf](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023017-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023017-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023017-print-pdf.pdf.json)

---

### Introduction and scope
- Geographic focus: Northern Triangle (NT): El Salvador, Guatemala, and Honduras.
- Population in U.S. in 2019: "Out of the 3 million migrants of NT origin living in the U.S in 2019, more than half were undocumented."
- Main arrival mode: undocumented entry across the U.S. Southwest border.
- Historical enforcement changes (1990–2019):
  - Border agents increased "by nearly 400 percent".
  - Border wall barrier coverage increased "by over 240 percent, to 654 miles".
- Undocumented NT migrants increased "almost four-fold, from half million in 1990 to 1.8 million by 2019."
- Remittances: "Remittances flows exceeded 20 percent in El Salvador and Honduras, and are very sizeable in Guatemala."
- Identified drivers cited in media and surveys: high crime (especially gang related), loss of livelihood due to low commodity prices, droughts, natural disasters, wages/incomes, unemployment, and job prospects.

### Research objective, theoretical framework, and data
- Aim: disentangle the role of various drivers explaining increasing undocumented migration from NT to the U.S.
- Undocumented migration proxy: U.S. border apprehensions.
- Theoretical framework: investment decision theory (Sjaastad, 1962) — migrants compare expected discounted income gains vs. moving costs and perceived probability of apprehension.
- Model extensions include non-traditional drivers: coffee production, temperature changes, homicide rates, natural disaster events; recidivism measures (deportations); perception/control variables (legal admissions, immigration wave dummies).
- Sample period: 1994-2019. Number of observations: 78.
- All variables transformed in natural logarithm.
- Data sources: DHS, U.S. CBP, Haver Analytics, ECLAC, ILO, World Bank, International Coffee Organization, FAOSTAT, EM-DAT, U.S. DHS annual statistical reports.

### Baseline empirical specification (dependent and key independent variables)
- Dependent variable: A_NT,t = annual border apprehensions of NT-born migrants at the U.S. Southwest border.
- Potential stream of income in the U.S.:
  - W_US,t = median real wage for Hispanics in the U.S.
  - U_US,t = U.S. Hispanic unemployment rate.
- Forsaken stream of income in NT countries:
  - W_NT,t = average real wage in the NT country.
  - U_NT,t = unemployment rate in the NT country.
- Probability of being apprehended indicators:
  - BAE_US,t = number of border enforcement agents at the U.S. Southwest border.
  - DEPORT_NT,t = number of deportations of undocumented migrants of NT origin from the U.S.
- Perceptions and controls:
  - WAVE1_NT,t = immigration wave 1 in FY2012-FY2014 (dummy).
  - WAVE2_NT,t = immigration wave 2 in FY2019 (dummy).
  - LA_NT,t = number of legal admissions from the NT country to the U.S.
- NT-specific factors:
  - COF_PROD_NT,t = coffee production in the NT country.
  - HOMICIDE_NT,t = homicide rate per 100,000 people.
  - TEMP_DEV_NT,t = two-year average deviation of temperature in March-August from 1951-1980 trend.
  - NAT_DISASTER_NT,t = dummy variable for specified natural disaster events.

### Estimation approach and fit
- Estimation: Panel of three countries estimated using Feasible Generalized Least Squares method with serial autocorrelation in panels.
- Robustness checks: (i) without border enforcement measures; (ii) instrumental variables regression approach; (iii) residual approach; (iv) alternative specifications including GDP per capita and immigrant-network controls.
- Model fit: fitted values comply well with actual border apprehensions; model captures long-run patterns for El Salvador, Guatemala, and Honduras.

### Baseline results (selected coefficients and significance; standard errors in parentheses)
- Constant: 5.383 (8.369)
- US: Hispanics real median wage: 5.241*** (1.693)
- US: Hispanics unemployment rate: -6.101** (2.388)
- NT: Real average wage: -1.438*** (0.236)
- NT: Unemployment rate: 16.160*** (5.230)
- US: Southwest border agents: -2.264*** (0.485)
- US: Legal admissions of NT citizens to the U.S.: 0.013 (0.230)
- US: Deportations of NT citizens from the U.S.: 0.554*** (0.092)
- NT: Immigration wave 1, FY2012-FY2014 (D): 0.831*** (0.095)
- NT: Immigration wave 2, FY2019 (D): 0.729*** (0.145)
- NT: Coffee production: -0.417*** (0.104)
- NT: Homicide rate per 100,000: 0.277** (0.117)
- NT: Temperature (deviation March-August, 2-year average): 0.197* (0.103)
- NT: Natural disaster event (D): 0.144** (0.063)
- Time trend: 1.569*** (0.418)
- Significance: * p<0.10, ** p<0.05, *** p<0.01.

### Key empirical interpretations
- U.S. labor market:
  - U.S. Hispanic real median wage has a positive and large elasticity (5.241***); migration attempts highly sensitive to U.S. wage changes.
  - U.S. Hispanic unemployment rate negative and significant (-6.101**); deterioration in U.S. employment conditions moderates apprehensions.
- Home-country labor market:
  - NT real average wage negative and significant (-1.438***); higher home wages reduce migration.
  - NT unemployment rate positive, significant, and largest coefficient (16.160***); home labor market conditions play the biggest role.
- Enforcement and recidivism:
  - U.S. Southwest border agents negative and significant (-2.264***); interpreted as higher enforcement deterring attempts.
  - U.S. deportations positive and significant (0.554***); deportations associated with recidivism or increased future apprehensions.
  - U.S. legal admissions show no effect in baseline (0.013).
- Perception-driven waves:
  - Immigration wave 1 (FY2012-FY2014): 0.831***.
  - Immigration wave 2 (FY2019): 0.729***.
- NT-specific non-income factors:
  - Coffee production negative and significant (-0.417***).
  - Homicide rate positive and significant (0.277**).
  - Temperature deviation positive and significant at 10 percent (0.197*).
  - Natural disaster dummy positive and significant (0.144**), implying on average 14 percent of border apprehensions attributed to such events in those years.

### Contributions to the FY2019 increase in border apprehensions (change relative to FY2011)
- Aggregate increase in border apprehensions in FY2019 relative to FY2011: 230 percent.
- Potential stream of income (U.S. labor market conditions):
  - Together explained 100 percentage point increase in FY2019 relative to FY2011.
  - By FY2019, U.S. Hispanic real median wage increased by 12 percentage points relative to FY2011, and U.S. Hispanic unemployment rate declined by 6.6 percentage points.
- Perceptions of migration policy (immigration waves):
  - Immigration wave 1 (FY2012-FY2014) accounted for an additional 83 percent relative to FY2011.
  - Immigration wave 2 (FY2019) accounted for an additional 73 percent relative to FY2011.
- Probability of being apprehended (indirect measures):
  - Changes in indirect measures contributed 46 percentage points to the FY2019 increase relative to FY2011.
  - Context: number of border agents declined by 10 percentage points by FY2019, and deportations increased by 40 percentage points in the same period.
- NT-specific and forsaken stream factors:
  - Variable contributions across countries and time; overall smaller role than potential stream and perception effects but important in specific country-year episodes.

### Climate, natural disasters, and NT-specific dynamics
- Temperature trends:
  - Average May-August temperature deviation accelerated since 2015, averaging around 30 percent annually.
  - Model suggests warmer temperatures resulted in an additional 6 percent of border apprehensions since FY2015.
  - A one standard deviation increase in temperature equivalent of 1.7 degrees would increase undocumented migration by 10 percentage points.
- Natural disaster events:
  - Events in Honduras (2016, 2018) and Guatemala (2019) contributed an additional 14 percent of border apprehensions per the model.
- Country heterogeneity:
  - El Salvador: rising homicide rate and decline in coffee production most important in explaining apprehensions in certain years.
  - Guatemala and Honduras: coffee production increases and homicide declines in some periods contributed negatively to apprehensions.
  - Forsaken stream indicators contributed positively to apprehensions in 2016-2019, with heterogeneity across countries (El Salvador’s real wage increase reduced apprehensions; Guatemala and Honduras saw opposite effects).

### Robustness and alternative specifications (selected results)
- Robustness strategy tested:
  1. Baseline with and without border agents.
  2. Border apprehensions per capita specifications.
  3. Real GDP per capita instead of real wages.
  4. IV regression instrumenting for lagged border agents and total Southwest border apprehensions.
  5. Residual enforcement regression defining border agents as excess/residual.
  6. Controls for pre-existing immigrant network.
- Core findings preserved across specifications: income effects, unemployment, deportations, immigration waves, coffee production, homicide rate, temperature, and natural disaster effects remain significant in many specifications.
- Examples from alternative tables:
  - IV and residual approaches: US Hispanics real median wage remains positive and significant (e.g., 6.438***), NT unemployment rate remains large and significant (e.g., 23.098***), coffee production remains negative and significant (e.g., -0.524***), temperature positive and significant (e.g., 0.313***).
  - Pre-existing immigrant network specifications: immigrant population (t-1) sometimes positive and significant (e.g., 1.949*** in one column) but network controls do not overturn baseline results.

### Key unit-root and cointegration test statistics (selected ADF results)
- Border apprehensions: Level -2.78, Difference -4.43***
- Border agents: Level 0.05, Difference -3.59**
- U.S. Hispanics real median wage: Level -2.20, Difference -4.11***
- NT real wage: Level -2.33, Difference -7.07***
- NT coffee production: Level -3.89**, Difference -8.70***
- NT temperature deviation: Level -4.32***, Difference -6.22***
- Table notes: ADF test statistics reported; *** indicates rejection at 1 percent; ** at 5 percent; * at 10 percent.

### Policy implications and quantitative magnitudes
- Summary policy implications:
  - Lift economic conditions in NT countries: ensure higher real wages, low unemployment, and create more and better-paid jobs to disincentivize undocumented migration.
  - Help traditional producers adapt to climate change and diversify income opportunities for coffee producers, especially in El Salvador; invest in climate-change resilient agricultural production.
  - Enhance security measures to tackle crime in NT countries to reduce migration push factors.
  - Broaden legal migration pathways for NT migrants, including seasonal and temporary jobs, to help regulate flows, particularly if home-country conditions improve.
  - Continue dialogue and cooperation with the U.S., USAID, and other development partners to address root causes of migration.
- Quantitative policy-relevant magnitudes highlighted:
  - A 10 percent increase in home country real wages would decrease undocumented migration by 14 percentage points.
  - A one standard deviation increase in temperature (1.7 degrees) would increase undocumented migration by 10 percentage points.

*International Monetary Fund — IMF WORKING PAPERS Northern Triangle Undocumented Migration to the U.S. (excerpts and annexes contained in wpiea2023017-print-pdf).*

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

### wpiea2023017-print-pdf - References .............................................................................................................

### Annexes and Supporting Material
- Annex I. Investment Decision Theory of Undocumented Migration — page 17.  
- Annex II. Border Apprehensions and Undocumented Migration — page 18.  
- Annex III. Evolution of the Factors Driving Migration — page 22.  
- Annex IV. Country Results — page 27.  
- Annex V. Unit Root and Cointegration Tests — page 29.  
- Annex VI. Robustness Tests — page 30.  
- Annex VII. Discussion on the Role of Border Enforcement Measures — page 35.

### Figures (main text)
- Figure 1. Evolution of Apprehensions at U.S. Southwest Border — page 10.  
- Figure 2. Fitness of Estimated Model — page 13.

### Annex Figures
- Figure III.1. Economic Factors Driving Undocumented Migration — page 24.  
- Figure III.2. Homicide Rate — page 25.  
- Figure III.3. Socioeconomic Factors Driving Migration — page 26.  
- Figure IV.1. Country Specific Results: Goodness of Fit and Role of Various Factors Explaining the Increase in Migration in FY2019 compared to FY2011 — page 27.  
- Figure IV.2. Country Specific Nontraditional Factors — page 28.

### Annex Tables
- Table V.1. Unit Root Test — page 29.  
- Table V.2. Cointegration Test between Border Apprehensions and Various Indicators — page 29.  
- Table VI.1. Determinants of Evolution of Border Apprehensions of NT Citizens (FY1994 - FY2019) — page 31.  
- Table VI.2. Determinants of Evolution of Border Apprehensions of NT Citizens (FY1994 - FY2019) — page 32.  
- Table VI.3. Instrumental Variable and Residual Enforcement Regressions (FY1994 - FY2019) — page 33.  
- Table VI.4. Pre-existing Network and Border Apprehensions (FY1994 - FY2019) — page 34.

### Glossary (selected acronyms)
- CBP – U.S. Customs and Border Protection  
- CMS – Center for Migration Studies  
- DACA – Deferred Action for Childhood Arrivals  
- DHS – U.S. Department of Homeland Security  
- ECLAC – Economic Commission for Latin America and the Caribbean  
- FAOSTAT – Statistics Department of the Food and Agriculture Organization of the United Nations  
- IDA – Institute for Defense Analyses  
- ILO – International Labor Organization of the United Nations  
- IRCA – Immigration Reform and Control Act  
- NT – Northern Triangle  
- UNODC – United Nations Office on Drugs and Crime  
- USDA – U.S. Department of Agriculture

### Introduction — scope and context
- Geographic focus: Northern Triangle (NT): El Salvador, Guatemala, and Honduras.  
- Population in U.S. in 2019: "Out of the 3 million migrants of NT origin living in the U.S in 2019, more than half were undocumented."  
- Main arrival mode: undocumented entry across the U.S. Southwest border.  
- Historical enforcement changes (1990–2019):  
  - Border agents increased "by nearly 400 percent".  
  - Border wall barrier coverage increased "by over 240 percent, to 654 miles".  
- undocumented NT migrants increased "almost four-fold, from half million in 1990 to 1.8 million by 2019."  
- Remittances: "Remittances flows exceeded 20 percent in El Salvador and Honduras, and are very sizeable in Guatemala."  
- Identified drivers cited in media and surveys: high crime (especially gang related), loss of livelihood due to low commodity prices, droughts, natural disasters, wages/incomes, unemployment, and job prospects.

### Research objective and approach
- Aim: disentangle the role of various drivers explaining increasing undocumented migration from NT to the U.S.  
- Undocumented migration proxy: U.S. border apprehensions (following Hanson and Spilimbergo, 1999).  
- Theoretical framework: investment decision theory (Sjaastad, 1962) — migrants compare expected discounted income gains vs. moving costs and perceived probability of apprehension.  
- Model extensions include: non-traditional drivers (coffee production, temperature changes, homicide rates, natural disaster events), recidivism measures (deportations), and perception/control variables (legal admissions, immigration wave dummies).

### Baseline empirical specification (variables and definitions)
- Dependent variable: A_NT,t = annual border apprehensions of NT-born migrants at the U.S. Southwest border.  
- Potential stream of income in the U.S.:  
  - W_US,t = the median real wage for Hispanics in the U.S.  
  - U_US,t = the U.S. Hispanic unemployment rate.  
- Forsaken stream of income in NT countries:  
  - W_NT,t = the average real wage in the NT country.  
  - U_NT,t = the unemployment rate in the NT country.  
- Probability of being apprehended indicators:  
  - BAE_US,t = the number of border enforcement agents at the U.S. Southwest border.  
  - DEPORT_NT,t = the number of deportations of undocumented migrants of NT country origin from the U.S.  
- Perceptions of changes in U.S. migration policy and waves:  
  - WAVE1_NT,t = immigration wave 1 in FY2012-FY2014 (dummy).  
  - WAVE2_NT,t = immigration wave 2 in FY2019 (dummy).  
  - LA_NT,t = the number of legal admissions from the NT country to the U.S.  
- NT-specific factors:  
  - COF_PROD_NT,t = coffee production in the NT country.  
  - HOMICIDE_NT,t = homicide rate per 100,000 people in the NT country.  
  - TEMP_DEV_NT,t = two-year average deviation of temperature in March-August from historical trend (1951-1980) in the NT country.  
  - NAT_DISASTER_NT,t = dummy variable for natural disaster event in the NT country (specified earthquakes and volcano eruptions).  
- Additional model components: time trend TRENDT and country fixed effect W_NT.  
- Transformations: "All variables have been transformed in natural logarithm."

### Data and stylized facts
- Sample period: 1994-2019 (sample used is 1994-2019, given availability and structural breaks).  
- Notable historical events affecting region:  
  - Volcano eruption in Guatemala; Mitch Hurricane in 1998 with severe impact in Honduras; 2001 earthquakes in El Salvador; devastating floods; prolonged drought affecting coffee production.  
- Apprehensions dynamics:  
  - FY2005: "a total of 112,000 citizens of NT countries were apprehended at the U.S. Southwest border."  
  - Post-FY2005 average: "plateaued at 40,000 people per year on average until September 2017."  
  - FY2019 peak: "reaching a record high of 608,000 people in FY2019."  
  - Increase after October 2017: "U.S. Southwest border apprehensions of NT citizens increased 170 percent until September 2019."  
  - 2019 impact: "apprehensions reached 2.6 percent of home population for Honduras in 2019."  
  - NT share of Southwest border apprehensions in FY2019: "NT citizens made up 62 percent of all US Southwest border apprehensions, up from 1 percent in FY1994."  
- Post-COVID-19 dynamics:  
  - FY2020: "sharp decrease in border apprehensions in FY2020—due to travel restrictions and lockdowns."  
  - 2021: "apprehensions of NT citizens rebounded in 2021, they remained well below FY2019 levels (left chart)."  
  - Since FY2021: NT citizens "represent the third largest group, below the number of apprehensions of citizens of Mexico ... and other countries (including Haiti, Cuba, and Nicaragua)."  
- Data sources used: DHS, U.S. CBP, Haver Analytics, ECLAC, ILO, World Bank (homicide rates), International Coffee Organization, FAOSTAT (temperature deviations), EM-DAT International Disaster Database, U.S. CBP annual border apprehensions by nationality, U.S. DHS annual statistical reports (legal admissions and deportations).

### Key findings and policy implications (as stated)
- Drivers: undocumented migration from NT origin is well explained by both traditional factors (real wages or GDP per capita, unemployment rates, especially U.S. indicators) and non-traditional factors (coffee production, temperature changes, homicide rates, natural disasters).  
- Economic considerations: "economic considerations (real wages and unemployment rates, especially in the U.S.) play an important role in driving undocumented migration."  
- Border enforcement: "border enforcement measures act as a deterrent for undocumented migration."  
- Apprehension spikes: "large and sudden increases in border apprehensions are at times linked to favorable changes in the perception of probability of being apprehended."  
- Robustness: results are robust to alternative specifications, estimation methods, and different measures of income, probability of apprehension, decline in livelihoods due to coffee production and changes in temperature, and to additional explanatory variables.  
- Policy recommendations for NT countries: "Changes at home should be centered around providing conditions for sustained inclusive growth, to create enough jobs, and advancing economic transformation to ensure higher wages as well as livelihood resilience to climate change and natural disasters, along with reducing crime."

*International Monetary Fund — IMF WORKING PAPERS Northern Triangle Undocumented Migration to the U.S.*

### 2019. As was documented in several official and media reports, the post-pandemic increase of apprehensions

### Northern Triangle Undocumented Migration to the U.S.

### IV. Estimation Strategy and Results
- Estimation approach:
  - Panel of three countries—El Salvador, Guatemala, and Honduras—estimated using the Feasible Generalized Least Squares method with serial autocorrelation in panels.
  - Robustness checks: (i) without border enforcement measures; (ii) instrumental variables regression approach; (iii) residual approach.
- Sample and fit:
  - Period: FY1994-FY2019 (Table 1).
  - Number of observations: 78.
  - All variables are in natural log form.
  - Model estimated with serial autocorrelation in panels; fitted values comply well with actual border apprehensions.

### Determinants of Border Apprehensions (Baseline results, Table 1)
- Constant: 5.383 (8.369)
- U S: Hispanics real median wage: 5.241*** (1.693)
- US: Hispanics unemployment rate: -6.101** (2.388)
- NT: Real average wage: -1.438*** (0.236)
- NT: Unemployment rate: 16.160*** (5.230)
- US: Southwest border agents: -2.264*** (0.485)
- US: Legal admissions of NT citizens to the U.S.: 0.013 (0.230)
- US: Deportations of NT citizens from the U.S.: 0.554*** (0.092)
- NT: Immigration wave 1, FY2012-FY2014 (D): 0.831*** (0.095)
- NT: Immigration wave 2, FY2019 (D): 0.729*** (0.145)
- NT: Coffee production: -0.417*** (0.104)
- NT: Homicide rate per 100,000: 0.277** (0.117)
- NT: Temperature (deviation March-August, 2-year average): 0.197* (0.103)
- NT: Natural disaster event (D): 0.144** (0.063)
- Time trend: 1.569*** (0.418)
- Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01

Key empirical interpretations:
- Potential stream of income (U.S. labor market):
  - U.S. Hispanic real median wage: positive and large elasticity; migration attempts highly sensitive to changes.
  - U.S. Hispanic unemployment rate: negative and significant; deterioration in U.S. employment conditions moderates apprehensions.
  - These two coefficients are the second and third highest in the regression.
- Forsaken stream of income (home country conditions):
  - NT real average wage: negative and significant; magnitude smaller than U.S. Hispanic real median wage.
  - NT unemployment rate: positive, significant, and largest coefficient (16.160***), indicating home labor market conditions play the biggest role.
- Probability of being apprehended and enforcement:
  - US: Southwest border agents: negative and significant (-2.264***); interpreted as higher enforcement increasing probability of apprehension and deterring attempts.
  - US: Legal admissions: no effect (0.013).
  - US: Deportations: positive and significant (0.554***); deportations increase perceived probability of future apprehension and can lead to recidivism.
- Perception-driven immigration waves:
  - NT immigration wave 1 (FY2012-FY2014): 0.831***; associated with DACA announcement and favorable perceptions.
  - NT immigration wave 2 (FY2019): 0.729***; linked to tightening enforcement and anti-immigration rhetoric that prompted forward-moving migration decisions.
- NT-specific and non-income factors:
  - Coffee production: negative and significant (-0.417***); higher production disincentivizes migration.
  - Homicide rate: positive and significant (0.277**); higher crime pushes migration.
  - Temperature deviation: positive and significant at 10 percent (0.197*); warmer temperatures increased undocumented migration.
  - Natural disaster event dummy: positive and significant (0.144**); in years of such events, 14 percent of border apprehensions could be attributed to these events on average.

### Contributions to the FY2019 Increase in Border Apprehensions (Change relative to FY2011)
- Aggregate finding:
  - Total increase in border apprehensions in FY2019 relative to FY2011: 230 percent.
- Contribution by factor groups (model-based decomposition):
  - Potential stream of income (U.S. labor market conditions):
    - Together (U.S. Hispanic real median wage increase and U.S. Hispanic unemployment rate decline) explained 100 percentage point increase in FY2019 relative to FY2011.
    - By FY2019, U.S. Hispanic real median wage increased by 12 percentage points relative to FY2011, and U.S. Hispanic unemployment rate declined by 6.6 percentage points.
  - Perceptions of migration policy (immigration waves):
    - Immigration wave 1 (FY2012-FY2014) accounted for an additional 83 percent of apprehensions relative to FY2011.
    - Immigration wave 2 (FY2019) accounted for an additional 73 percent of apprehensions relative to FY2011.
  - Probability of being apprehended (indirect measures):
    - Changes in indirect measures contributed 46 percentage points to the FY2019 increase relative to FY2011.
    - Context: number of border agents declined by 10 percentage points by FY2019, and deportations increased by 40 percentage points in the same period.
  - NT-specific and forsaken stream factors:
    - Variable contributions across countries and time; overall smaller role than potential stream and perception effects but important in specific country-year episodes.

### Climate, Natural Disasters, and NT-specific Dynamics
- Temperature trends:
  - Average May-August temperature deviation accelerated since 2015, averaging around 30 percent annually.
  - Model suggests warmer temperatures resulted in an additional 6 percent of border apprehensions since FY2015.
  - A one standard deviation increase in temperature equivalent of 1.7 degrees would increase undocumented migration by 10 percentage points.
- Natural disaster events:
  - Events in Honduras (2016, 2018) and Guatemala (2019) contributed an additional 14 percent of border apprehensions per the model.
- Country heterogeneity:
  - El Salvador: rising homicide rate and decline in coffee production were most important in explaining apprehensions in certain years.
  - Guatemala and Honduras: coffee production increases and homicide declines in some periods contributed negatively to apprehensions.
  - Forsaken stream indicators contributed positively to apprehensions in 2016-2019, with heterogeneity across countries (El Salvador’s real wage increase reduced apprehensions; Guatemala and Honduras saw opposite effects).

### V. Concluding Remarks, Outlook, and Policy Implications
- Summary of findings:
  - Undocumented migration from NT countries to the U.S. is well explained by traditional economic factors (U.S. and NT incomes/unemployment), perceptions of migration policy, probability of apprehension, and nontraditional factors (temperature, coffee production, homicides, natural disasters).
  - Border enforcement measures act as a deterrent, but sudden favorable perceptions of migration probability can trigger migration surges.
  - Results are robust to various specifications, estimation methods, variable measures, sample sizes, and include checks with GDP per capita, immigrant stock as network effect, instrumental variables, and residual enforcement estimations.
- Outlook:
  - Migration pressures expected to persist given climate change trends and relative economic developments between the U.S. and NT countries.
  - Medium-term U.S. outlook remains favorable for labor market conditions; NT outlook subject to downside risks from tighter global financial conditions affecting private investment.
- Quantitative policy-relevant magnitudes:
  - A 10 percent increase in home country real wages would decrease undocumented migration by 14 percentage points.
  - A one standard deviation increase in temperature (1.7 degrees) would increase undocumented migration by 10 percentage points.
- Policy implications and recommendations:
  - Lift economic conditions in NT countries: ensure higher real wages, low unemployment, and create more and better-paid jobs to disincentivize undocumented migration.
  - Help traditional producers adapt to climate change and diversify income opportunities for coffee producers, especially in El Salvador; invest in climate-change resilient agricultural production.
  - Enhance security measures to tackle crime in NT countries to reduce migration push factors.
  - Broaden legal migration pathways for NT migrants, including seasonal and temporary jobs, to help regulate flows, particularly if home-country conditions improve.
  - Continue dialogue and cooperation with the U.S., USAID, and other development partners to address root causes of migration.

*IMF Working Paper excerpt: Northern Triangle Undocumented Migration to the U.S.*

### Annex I. Investment Decision Theory of

### Annex I. Investment Decision Theory of Undocumented Migration

### Investment decision framework
- Individuals decide to migrate if expected discounted net benefits of the U.S. location exceed current home wage and perceived migration costs.
- Decision depends on:
  - Expected future utility/value in new location: V_U,i,t+1
  - Expected future utility/value in old location: V_N,i,t+1
  - Discount factor: 1/(1+ρ)
  - Current home wage: w_N,i,t
  - Perceived migration costs: C_i,t
  - Expected probability of being apprehended today and in the future: P_A,t
- Formal individual indicator (I_i,t) equals 1 if w_N,i,t + C_i,t < (1 − P_A,t)/(1+ρ) * (V_U,i,t+1 − V_N,i,t+1), and 0 otherwise.

### Aggregation and determinants of undocumented migration
- Aggregate attempts to migrate undocumented, M_NN,t, are a function:
  - M_NN,t = M(W_US,t, W_NN,t, P_A,t, Ω_NN,t, Θ_NN,t)
  - Where W_US,t = expected future U.S. earnings; W_NN,t = home earnings; Ω_NN,t = information predicting future paths; Θ_NN,t = personal characteristics.
- Apprehensions at the border, A_NN,t, depend on the number of attempts and the individual apprehension probability:
  - A_NN,t = P(BPA_US,t, M_NN,t) * M(·)  (apprehension probability function of border enforcement effort BPA_US,t and total attempts M_NN,t)

### Apprehensions function and reduced form
- Because P_A,t and M_NN,t are not directly observed, the paper uses a reduced-form specification for border apprehensions:
  - A_NN,t = γ1 BPA_US,t + γ2 W_US,t + γ3 W_NN,t + γ4 L_A_NN,t + γ5 U_US,t + ε_NN,t
  - Variables:
    - BPA_US,t = border enforcement efforts measured by border agents
    - W_US,t = US real minimum wage
    - W_NN,t = home country real minimum wage
    - L_A_NN,t = number of legal admissions to the US
    - U_US,t = US unemployment rate of Hispanics
- Extended specification adds controls relevant to Northern Triangle (NT) migration: homicide rates, international coffee prices, temperatures, and disaster events.

### Empirical patterns and key statistics (Annex II and III material referenced)
- Undocumented population and country shares:
  - Total undocumented migrants living in the US by FY2017: 10.6 million
  - NT-born undocumented migrants represented a sixth of that total stock by FY2017.
  - Mexico represented about half of total undocumented population in the U.S.
- NT country undocumented stock changes to FY2017:
  - Honduras-born undocumented population increased nine times to 380,000.
  - Guatemala-born increased nearly five times to 550,000.
  - El Salvador-born increased slightly more than twice to 670,000.
  - Migrants from Honduras and Guatemala came primarily as undocumented; their share in the total foreign-born population increased to 60 percent. Share of undocumented migrants from El Salvador declined from about two thirds to less than half.
- Apprehension rates and estimation:
  - Studies report an estimated rate of apprehension remained stable at 40-50 percent of attempts up until FY2011 and increased since FY2011 to 70 percent.
  - Derived rates of apprehension from census and departure data differ marginally from other prominent estimates.
- Border enforcement and capacity (1994–2019):
  - US Border Patrol budget increased from US$400 million in FY1994 to US$4,696 million in FY2019.
  - Number of agents increased from 3.7 thousand in FY1994 to 16.6 thousand in FY2019.
  - Wall construction expanded from 203 miles in FY1994 to 654 miles in FY2019.
  - Since FY2011, deployment of accessible surveillance technology expanded (drones, mobile cameras, sensor alarm systems).
- Enforcement effects and migrant responses:
  - Increased enforcement narrowed areas of potential crossings along the ~2,000 mile US-Mexico border, decreasing probability of successful entry.
  - NT migrants increasingly relied on coyote smugglers; coyote fees rose, sharply following the 2012-2014 NT immigration episode.
- Uncertainty around apprehension rate estimates arises from:
  - (i) sampling skewed to Mexican citizens who had fewer attempts since FY2012;
  - (ii) ignoring significant demographic composition changes of apprehended migrants since FY2012 (families and unaccompanied children);
  - (iii) choice of mathematical optimization model.
- Despite uncertainty, border apprehensions are considered the best proxy for gross arrivals of undocumented NT-born migrants because undetected border entries are the main channel of arrival for NT nationals and estimates of attempts and entries are consistently proportional to apprehensions.

### Drivers of migration (selected economic and environmental factors)
- Labor market and wage disparities:
  - Real average wages in private sector of Honduras declined by 30 percentage points, from an index of 100 percent in 1991 to 70 percent in 2018.
  - In Guatemala, real average wages halved by 2019 after peaking in 2004.
  - In El Salvador, real average wages in private sector remained relatively unchanged during 1991-2018.
  - Median real wage of US Hispanics increased by 18 percentage points relative to 1991 by 2019.
  - US Hispanics unemployment rate: declined to 5 percent in 2007, peaked at 12.5 percent in 2010, then rapidly declined.
- GDP per capita and cost of living:
  - Ratio of NT countries GDP per capita to US GDP per capita declined from 1991 to 2005; temporary acceleration in 2005-2008; stagnation afterward for Guatemala and Honduras; slow convergence in El Salvador.
  - Cost of living relative to US increased substantially in Guatemala and Honduras, and to a lesser extent in El Salvador.
- Crime and homicide rates:
  - Organized crime contributed to high homicide rates; El Salvador experienced a marked rise after large deportations of gang members in the 1990s.
  - From 1999 to 2011, El Salvador homicide rate remained high at 62 homicides per 100,000 people.
  - Temporary truce in 2012 reduced homicide rates but rates surged again after 2014.
- Agricultural shocks and coffee:
  - International coffee prices fell from 185 cents per pound in 1997 to 60 cents per pound in 2002, reducing coffee production and investment.
  - Recovery in prices from 2002 to 2010 did not restore production in El Salvador and left Guatemala stable; Honduras increased production partly via scale expansion.
  - 2012 international coffee price collapse and 2011-2012 coffee leaf rust outbreak further reduced production and prompted farmer displacement.
- Climate and disasters:
  - March-May seasonal temperatures: 1991-2000 = 0.7°C above 1951-1980 baseline; 2001-2010 = 0.71°C above; 2011-2018 = 0.83°C above.
  - Droughts and volatile rainfall increased; UN WFP reported repeated droughts since 2014 causing unprecedented food insecurity in the Central American Dry Corridor (58% of El Salvador, south Honduras, southeast Guatemala).
  - EM-DAT/author summary (last 30 years): 186 natural disaster events in NT countries, causing 21.5 thousand deaths and affecting livelihoods of 21 million people.
  - Hurricane Mitch (October 1998): nearly 15 thousand deaths, impacted >2 million people in Honduras, estimated damage US$ 2 billion (16 percent of GDP); Honduran apprehensions in FY1999 doubled to 16 thousand from 7.8 thousand in FY1998.
  - Hurricane Stan (2005): 1.5 thousand deaths in Guatemala, damages US$ 1 billion (1 percent of GDP).
  - 2001 earthquake in El Salvador: 844 deaths, affected 1.3 million people, damages US$ 1.5 billion (12 percent of GDP).
- Policy and migration-status changes:
  - Changes in U.S. migration policy are reflected in shifts in apprehensions status beginning with 2020 (text chart referenced).

*Source: Annex I–III material from the IMF working paper unit provided.*

### Annex IV. Country Results

### Annex IV. Country Results

### Country-specific goodness of fit and decomposition of increase in migration (FY2019 vs FY2011)
- Figures compare Actual vs Fitted border apprehensions (natural logarithm) for El Salvador, Guatemala, and Honduras (series covering 1994–2019; detrended series with contribution charts relative to FY2011).
- Decomposition components shown in figures (contribution to detrended series of border apprehensions, percentage points relative to FY2011):
  - Potential stream of income
  - Forsaken stream of income
  - Probability of apprehension
  - NT-specific factors (nontraditional factors)
  - Migration waves (NT: Immigration wave 1, FY2012–FY2014 (D); Immigration wave 2, FY2019 (D))
  - Unexplained
- Charts indicate fitted model captures long-run patterns; total change shown as aggregation of components for each country.

### Country-specific nontraditional factors (contributions to change in border apprehensions relative to FY2011)
- Components reported separately for El Salvador, Guatemala, Honduras (percentage points to change in border apprehensions relative to FY2011), including:
  - Coffee production
  - Homicides
  - Natural disaster
  - Temperature (deviation)
  - Total (sum of NT-specific factors)
- Visual series presented for 2012–2019 showing year-by-year contributions from each NT-specific factor to border apprehensions for each country.

### Key unit-root and cointegration test statistics (Annex V)
- Table V.1. Augmented Dickey Fuller test statistics (levels and first differences). Examples:
  - Border apprehensions: Level -2.78, Difference -4.43***
  - Border agents: Level 0.05, Difference -3.59**
  - U.S. Hispanics real median wage: Level -2.20, Difference -4.11***
  - NT real wage: Level -2.33, Difference -7.07***
  - NT coffee production: Level -3.89**, Difference -8.70***
  - NT temperature deviation: Level -4.32***, Difference -6.22***
  - Border apprehensions per capita: Level -2.79, Difference -4.43***
  - Border agents per capita: Level 0.02, Difference -3.61**
  - Legal admissions per capita: Level -1.40, Difference -2.39
- Table notes: ADF test statistics reported; *** indicates rejection at 1 percent; ** at 5 percent; * at 10 percent.
- Additional ADF-on-residuals statistics (note form with trend). Examples:
  - Border agents: El Salvador -3.04**, Guatemala -3.09**, Honduras -2.66*
  - U.S. Hispanics real median wage: El Salvador -2.91**, Guatemala -1.82, Honduras -2.68*
  - NT coffee production: El Salvador -3.54***, Guatemala -1.58, Honduras -2.26
  - NT temperature deviation: El Salvador -3.22**, Guatemala -1.68, Honduras -2.80*
- Table notes: ADF test on residuals of OLS regressions; significance codes as above.

### Robustness tests and specifications (Annex VI)
- Robustness strategy enumerated (specifications tested):
  1. Baseline specification with and without variable for border agents (Table VI.1, columns 1 and 2).
  2. Baseline for border apprehensions per capita and border agents per capita of U.S. population (Table VI.1, columns 3 and 4).
  3. Same specifications with real GDP per capita (U.S. and NT) instead of real wage variables (Table VI.2).
  4. IV regression instrumenting for lagged values of border agents and total apprehensions in Southwest border (Table VI.3, columns 1 and 2).
  5. Residual enforcement regression with border agents defined as excess/residual beyond structural equation (Table VI.3, columns 3 and 4).
  6. Specification that controls for pre-existing immigrant network (Table VI.4).
- Summary conclusion: Across specifications, baseline coefficient estimates are preserved. The coefficient for legal admissions was estimated as negative and statistically significant in specification VI.1.4 and VI.3.4.

### Determinants of border apprehensions—selected estimates (Table VI.1, FY1994–FY2019; all variables in natural log form; FGLS)
- Dependent variables: Border apprehensions; Border apprehensions per capita.
- Constant: (1) -40.942; (2) -63.163**; (3) -54.727**; (4) -62.459** (standard errors reported in table).
- US: Hispanics real median wage: (1) 5.241***; (2) 7.789***; (3) 4.554***; (4) 7.580***.
- US: Hispanics unemployment rate: (1) -6.101**; (2) -14.250***; (3) -7.099***; (4) -14.645***.
- NT: Real average wage: (1) -1.438***; (2) -1.384***; (3) -1.330***; (4) -1.319***.
- NT: Unemployment rate: (1) 16.160***; (2) 24.009***; (3) 18.124***; (4) 24.127***.
- US: Southwest border agents: (1) -2.264***; (2) -2.073*** (only present in columns with agents).
- US: Legal admissions of NT citizens to the U.S.: (1) 0.013; (2) -0.224; (3) -0.159; (4) -0.449*.
- US: Deportations of NT citizens from the U.S.: (1) 0.554***; (2) 0.401***; (3) 0.478***; (4) 0.322***.
- NT: Immigration wave 1 (FY2012–FY2014, D): (1) 0.831***; (2) 0.826***; (3) 0.808***; (4) 0.821***.
- NT: Immigration wave 2 (FY2019, D): (1) 0.729***; (2) 0.799***; (3) 0.723***; (4) 0.789***.
- NT: Coffee production: (1) -0.417***; (2) -0.493***; (3) -0.469***; (4) -0.552***.
- NT: Homicide rate per 100,000: (1) 0.277**; (2) 0.247*; (3) 0.235**; (4) 0.223*.
- NT: Temperature (deviation March–August, 2-year average): (1) 0.197*; (2) 0.323***; (3) 0.190*; (4) 0.305***.
- NT: Natural disaster event (D): (1) 0.144**; (2) 0.123*; (3) 0.133**; (4) 0.115*.
- Number of observations: 78 in all columns.

### Alternative specification using GDP per capita (Table VI.2)
- Constant: (1) -145.237**; (2) -213.233**; (3) -141.786**; (4) -208.903**.
- US: Real GDP per capita: (1) 7.974**; (2) 11.789***; (3) 7.700**; (4) 11.906***.
- US: Hispanics unemployment rate: (1) 0.775; (2) -3.577; (3) -0.754; (4) -4.188.
- NT: Real GDP per capita: (1) 0.760; (2) 0.854; (3) 0.046; (4) 0.478.
- NT: Unemployment rate: (1) 15.990**; (2) 23.817***; (3) 16.318***; (4) 23.801***.
- US: Southwest border agents: (1) -2.080***; (2) -1.974***.
- US: Legal admissions of NT citizens: (1) 0.050; (2) -0.083; (3) -0.098; (4) -0.276.
- US: Deportations of NT citizens: (1) 0.342**; (2) 0.091; (3) 0.299**; (4) 0.045.
- NT: Coffee production: consistently negative and significant (e.g., -0.440***, -0.518***, -0.466***, -0.574***).
- NT: Temperature: positive and significant in several columns (e.g., 0.270**, 0.410***, 0.269**, 0.399***).
- Number of observations: 78 in all columns.

### Instrumental variable and residual enforcement regressions (Table VI.3)
- IV regression (columns 1–2) and Residual approach regression (columns 3–4).
- Constant: (1) -65.459**; (2) -60.633**; (3) -43.771; (4) -45.919*.
- US: Hispanics real median wage: (1) 6.438***; (2) 5.713***; (3) 6.545***; (4) 6.343***.
- US: Hispanics unemployment rate: (1) -10.039***; (2) -11.037***; (3) -13.015***; (4) -13.439***.
- NT: Real average wage: (1) -1.453***; (2) -1.365***; (3) -1.655***; (4) -1.591***.
- NT: Unemployment rate: (1) 23.098***; (2) 22.121***; (3) 19.636***; (4) 20.438***.
- US: Southwest border agents (instrumented or residual): (1) -1.192*; (2) -1.018*; (3) -2.891***; (4) -2.726***.
- US: Legal admissions: (1) -0.101; (2) -0.279; (3) -0.233; (4) -0.434**.
- US: Deportations: (1) 0.487***; (2) 0.384***; (3) 0.652***; (4) 0.545***.
- NT: Coffee production: negative and significant across specifications (e.g., -0.524***, -0.546***, -0.409***, -0.460***).
- NT: Homicide rate per 100,000: positive and significant (e.g., 0.295**, 0.268**, 0.259**, 0.245**).
- NT: Temperature: positive and significant (e.g., 0.313***, 0.283***, 0.260**, 0.248**).
- NT: Natural disaster event (D): (1) 0.181***; (2) 0.168**; (3) 0.097; (4) 0.095.
- Number of observations: 78; reported R-square for panel regressions: 0.959 (columns shown); IV diagnostics: Underidentification test 27.934 and 25.544; Weak identification test 33.491 and 37.712; Hansen J-statistic 1.725 and 0.263.

### Pre-existing immigrant network (Table VI.4)
- Controls for NT: Immigrant population in U.S. (t-1) included.
- Constant: (1) -61.995**; (2) -55.901*; (3) -68.599***; (4) -59.726**.
- US: Hispanics real median wage: (1) 4.077**; (2) 5.112**; (3) 7.246***; (4) 7.061***.
- US: Hispanics unemployment rate: (1) -9.775***; (2) -14.662***; (3) -12.144***; (4) -14.644***.
- NT: Real average wage: (1) -1.243***; (2) -1.426***; (3) -1.321***; (4) -1.341***.
- NT: Unemployment rate: (1) 19.363***; (2) 23.941***; (3) 21.893***; (4) 24.094***.
- US: Southwest border agents: (1) -1.288***; (2) -0.879***.
- US: Legal admissions: (1) -0.028; (2) -0.283; (3) -0.316; (4) -0.462*.
- US: Deportations: (1) 0.461***; (2) 0.244**; (3) 0.459***; (4) 0.288***.
- NT: Immigration wave 1 (D): (1) 0.775***; (2) 0.819***; (3) 0.819***; (4) 0.818***.
- NT: Immigration wave 2 (D): (1) 0.735***; (2) 0.886***; (3) 0.744***; (4) 0.810***.
- NT: Coffee production: large negative coefficients (e.g., -0.685***, -0.559***, -0.554***, -0.552***).
- NT: Immigrant population in U.S. (t-1): (1) 1.949***; (2) 0.717; (3) 1.082*; (4) -0.006.
- Number of observations: 78 in all columns.

### Discussion on the role of border enforcement measures (Annex VII)
- Elasticity decomposition (notation preserved from text):
  - ∂P/∂B = ∂P/∂B + (1 + ∂P/∂A) ∂A/∂B  (equation structure shown in Annex VII; variables: probability of apprehension P, border agents B, attempts A)
- Interpretation of terms (signs given in source):
  - ∂P/∂B ≥ 0: elasticity of probability of apprehension with respect to border agents (direct effect).
  - ∂A/∂B ≤ 0: elasticity of migration attempts with respect to border agents (indirect deterrent effect).
  - ∂P/∂A ≤ 0: elasticity of probability of apprehension with respect to attempts (direct effect of attempts on apprehensions).
- Comparison to Hanson and Spilimbergo (1999):
  - Hanson and Spilimbergo find ∂P/∂B > 0 (interpreted as a lower bound of marginal product of enforcement).
  - The present model finds elasticity of apprehension with respect to border agents is negative—evidence of a strong deterrent effect.
- Reconciliation via nonlinearity and equilibrium enforcement level (notation preserved):
  - Define equilibrium border enforcement level B_A^*.
  - If B_A < B_A^*: ∂P/∂B > 0 and ∂A/∂B = 0, consistent with Hanson and Spilimbergo.
  - If B_A ≥ B_A^*: ∂P/∂B = 0 and ∂A/∂B < 0, leading to overall ∂P/∂B < 0 (net deterrent).
- Visual conceptual diagram elements presented: Migration attempts ↔ Border agents ↔ Probability of apprehension.

*Northern Triangle Undocumented Migration to the United States Working Paper No. WP/23/17 — Annex IV–VII (figures, tables, and text as provided).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023017-print-pdf.pdf_