## wp18136

## Source details

**Canonical URL:** [wp18136](https://www.imf.org/-/media/files/publications/wp/2018/wp18136.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp18136.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp18136.pdf.json)

---

### Introduction, objectives, and approach
- Research question: examine the impact of remittances on inequality in Mexico.
- Motivation: remittances are an important source of external financing and can facilitate investments in health, education or small businesses; their effects on inequality are less clear.
- Empirical gap: literature finds remittances increase, reduce, or have no effect on inequality.
- Data and methods:
  - Data source: National Survey of Income and Expenditure (ENIGH) by INEGI; pooled cross-sections used for various years.
  - Crisis windows:
    - Peso crisis: 1989, 1994, 2000 (no 1995 survey).
    - Global Financial Crisis window: 2002, 2008, 2014 (no 2007 or 2009 surveys).
  - Remittances measured from question: ‘What was your income from other countries last month? Two months ago? Three months ago?’, aggregated to quarterly and annual amounts.
  - Main empirical tools:
    - Probit for likelihood of receiving remittances (by decile).
    - OLS for amount received (Pesos) and remittances as percent of household income (by decile).
    - Counterfactual incomes via propensity score matching (radius matching) to impute incomes of remittance-receiving households absent remittances; Common support enforced; results robust to alternative matching estimators.
  - Covariates: household composition, urban/rural, household head age/gender/education, crisis dummies.

### Migration, remittance patterns, and descriptive facts
- Migration and destination:
  - United States accounts for 97 percent of Mexican outmigration.
- Migrant profile changes (1990 → 2013, Pew tabulations):
  - Male: 55 → 53 percent.
  - Median age: 29 → 39 years.
  - Age group shares (percent): Younger than 18 15 → 6; 18 to 29 35 → 19; 30 to 39 24 → 26; 40 to 49 13 → 24; 50 to 64 9 → 18; 65 or older 5 → 7.
  - Educational attainment (ages 25+): Less than high school diploma 76 → 58 percent; High school diploma 12 → 24 percent; Some college or more 13 → 18 percent.
  - Years in the US: 5 years or less 30 → 8; 6 to 10 years 20 → 15; 11 to 20 years 31 → 35; More than 20 years 19 → 42.
- Remittance behavior:
  - Likelihood of remitting declines with age: 40 percent for migrants below the age of 30, 30 percent after the age of 50 (Gonzales-Barrera 2015).
- Crisis characterizations:
  - 1994 Peso crisis: sudden peso devaluation in December 1994; resulted in a 6 percent contraction of the Mexican economy by 1995.
  - Global Financial Crisis: propagated to Mexico through strong trade link (around 75 percent of Mexican exports go to the United States); Mexico was better equipped to handle the crisis with smaller increases in unemployment and informal employment share.

### Who receives remittances — decile patterns and household correlates
- Shares and magnitudes:
  - Share of households receiving remittances: about 4 percent in the early 1990s, almost 6 percent by 2000, then declined slightly thereafter.
  - Remittances as share of household income: 30–40 percent in the 1990s → around 20 percent in 2014.
  - In 2014 mean remittances received per household: about 290 USD per month (mean; 140 USD median).
  - Without remittances, remittance-receiving households would be in the 4th decile on average; with remittances they are in the 6th decile; non-remittance-receiving households are on average in the 7th decile in 2014.
- Recipient household characteristics:
  - Tend to be slightly smaller households; fewer employed household members present; more children and elderly.
  - More likely to have a female household head and lower education.
  - More likely to be from rural areas.
- Regression summary across deciles (patterns from Tables 3 and 4):
  - Likelihood of receiving remittances:
    - Falls with the number of employed household members (except at the very top in later periods where coefficients suggest investment motives).
    - Increases with the number of elderly.
    - Falls with the number of children (family reunification effect), though less pronounced at the top in later period.
    - Female household head associated with higher likelihood of receiving remittances across deciles.
    - Higher education of household head associated with lower likelihood of receiving remittances across deciles (except bottom decile where not significant in later period).
    - Rural households more likely to receive remittances across the income distribution.
  - Amounts and share of income:
    - Increase with number of elderly (especially in earlier period).
    - Decrease with number of children and number of employed household members.
    - Lower for male household heads and urban households across the distribution.
    - Education reduces amounts (except at very bottom in later period).
  - Example coefficient-level highlights preserved as reported:
    - Peso crisis period example: age of household head coefficient -0.0140*** in bottom decile receiving regression; number of elderly 0.387*** in bottom decile receiving regression; 1994 dummy -0.257** for 9th decile.
    - Global Financial Crisis period example: number of employed -0.162*** in bottom decile receiving regression; 2014 dummy -0.606*** for top decile receiving regression.
  - Model fit:
    - Adjusted R-sq. values range from 0.015 to 0.100 across specifications (reported in tables).
    - Number of observations reported per decile and regression.

### Counterfactuals, behavioral responses, and inequality
- Methods: propensity score matching to impute counterfactual incomes for remittance-receiving households in absence of remittances; enforces common support and does not impose functional form.
- Gini comparisons (preserving reported magnitudes):
  - Gini based on actual income (including remittances) is lower than Gini excluding remittances — suggesting remittances reduce inequality.
  - Counterfactual incomes (matching) produce a Gini coefficient lower than income excluding remittances but still higher than actual income — indicating inequality would be higher in the absence of remittances even allowing for behavioral responses.
- Quantitative magnitudes and elasticity implications:
  - Between 2002 and 2012 the Gini coefficient fell by about 0.01.
  - Differences between Gini coefficients in the later period:
    - Actual income vs income without remittances: about 0.006.
    - Actual income vs counterfactual income: about 0.004.
  - Behavioral response (difference between income excluding remittances and counterfactual income) reduces the impact of remittances on inequality by more than two thirds, implying very elastic household labor supply to remittance income.
  - Contextual figure: in 2014 remittances accounted for about 20 percent of household income in the median decile.

### Crisis effects on remittance targeting and amounts
- Overall pattern: crises showed increasingly pro-poor remittance targeting over time.
- During the 1994 Peso crisis:
  - Likelihood of receiving remittances and remittance amounts as a share of income fell at the top deciles; little change in lower deciles.
- During the Global Financial Crisis (2008–09 window):
  - Likelihood of receiving remittances and remittance amounts as a share of income fell at the top deciles and increased at the bottom deciles.
- Interpretation:
  - Decline of remittances to richer households during crises consistent with reduced investment motives.
  - Increase in remittances to poorer households during the Global Financial Crisis may reflect more widespread migration opportunities and longer-established migrant communities better able to cushion shocks in the United States.
  - Insurance effect occurred even when both sending and receiving countries were hit by a common shock.
- Robustness:
  - Including household income (excluding remittances) as a regressor yields qualitatively similar results.
  - Heckman two-step estimator selection correction yields similar results (see Annex Table 1).

### Annex Table 1 — selection-corrected amount equations (Global Financial Crisis; pooled 2002, 2008, 2014)
- Dataset: pooled cross-section of households in 2002, 2008 and 2014; 2002 is treated as the base year.
- Selection variable: nonlinear transformations (the logarithm and the square respectively) of the predicted probability of sending as a control in the amount equation.
- Number of observations by decile: 6563, 6556, 6563, 6579, 6569, 6564, 6582, 6585, 6524, 6596.
- Selection-corrected amount equation (selection variable = Pred. prob. (ln)) — selected coefficients and standard errors as reported:
  - No. of hh members:
    - Bottom decile: 95.36*** (26.62)
    - 5th decile: 403.3*** (99.80)
    - 6th decile: 229.0*** (59.14)
    - 7th decile: 395.4*** (137.7)
  - No. of children:
    - 5th decile: -674.9*** (160.4)
    - 7th decile: -558.7** (220.8)
    - Top decile: -835.3* (504.9)
  - No. of elderly:
    - Bottom decile: 587.5*** (114.7)
    - 5th decile: 619.7*** (223.6)
    - 6th decile: 1346.8*** (234.9)
    - 8th decile: 1183.9* (605.2)
  - No. of employed:
    - Bottom decile: -762.3*** (93.65)
    - 2nd decile: -1245.1*** (261.4)
    - 3rd decile: -1209.2*** (259.9)
    - Top decile: -2850.3** (1123.5)
  - Age of hh head:
    - Bottom decile: -21.06*** (4.081)
    - 2nd decile: -46.12*** (13.41)
    - Top decile: -125.9*** (13.72)
  - Male hh head:
    - Bottom decile: -1044.0*** (107.6)
    - 9th decile: -4525.0*** (1543.3)
    - Top decile: -3713.1** (1701.6)
  - Education of hh head:
    - Bottom decile: -120.0*** (27.32)
    - 5th decile: -1052.7*** (98.06)
    - Top decile: -1541.1*** (446.5)
  - Urban hh:
    - Bottom decile: -861.7*** (109.2)
    - 5th decile: -5153.0*** (454.5)
    - Top decile: -5025.2*** (1611.2)
  - 2008 dummy:
    - 2nd decile: 569.6* (320.2)
    - 3rd decile: 1744.4*** (405.6)
    - Top decile: -844.3*** (125.4)
  - 2014 dummy:
    - 3rd decile: 856.2* (460.4)
    - 9th decile: 4372.2** (1729.3)
    - Top decile: -1620.7*** (543.0)
  - Pred. prob. (ln):
    - Bottom decile: 0.158*** (0.0197)
    - 2nd decile: 0.451*** (0.0343)
    - Top decile: 0.115*** (0.0250)
- Selection-corrected amount equation (selection variable = Pred. prob. (square)) — selected coefficients and standard errors as reported:
  - No. of hh members:
    - 5th decile: 371.2* (203.1)
    - 7th decile: 665.2** (282.9)
    - 9th decile: 596.4** (248.9)
    - Top decile: 3221.5** (1279.9)
  - No. of children: top decile: -3066.1 (1943.2)
  - No. of elderly: 8th decile: 830.2 (985.9); 9th decile: 961.2 (789.7)
  - No. of employed:
    - Bottom decile: -526.6*** (126.6)
    - 3rd decile: -1283.3*** (232.2)
    - Top decile: -6511.8*** (1601.7)
  - Age of hh head:
    - Bottom decile: -22.22** (10.98)
    - 4th decile: -53.74*** (20.44)
    - 9th decile: -158.1*** (26.22)
  - Male hh head:
    - Bottom decile: -559.7*** (192.7)
    - 4th decile: -2724.5*** (514.2)
    - 9th decile: -4873.4*** (739.2)
  - Education of hh head:
    - 4th decile: -215.8* (116.3)
    - 8th decile: -260.8* (148.8)
    - 9th decile: -319.8*** (88.73)
  - Urban hh:
    - 2nd decile: -881.7** (377.5)
    - 4th decile: -1578.2*** (594.4)
    - 7th decile: -2660.8*** (720.6)
  - 2008 dummy:
    - 2nd decile: 711.0** (307.7)
    - 3rd decile: 1738.8*** (405.4)
    - 7th decile: 2958.2*** (869.7)
    - 9th decile: -1674.4* (889.6)
  - 2014 dummy:
    - 3rd decile: 857.6* (460.6)
    - Top decile: 9633.7 (.)
  - Pred. prob. (square):
    - Bottom decile: 33.15*** (4.200)
    - 2nd decile: 75.89*** (5.709)
    - Top decile: 88.65*** (13.02)

### Conclusions and policy-relevant implications
- Main empirical conclusions:
  - Remittances lower income inequality in Mexico: counterfactual-income-based Gini is higher than actual-income-based Gini, implying inequality would be higher in the absence of remittances even after accounting for behavioral responses.
  - Remittances became increasingly pro-poor over time: from concentration in middle deciles in earlier years to reaching poorer deciles in later periods.
  - During crises (1994 Peso crisis and Global Financial Crisis), remittances to top deciles fell while remittances to bottom deciles fell less or increased (notably during the Global Financial Crisis), indicating a cushioning/insurance role for poorer households.
- Mechanisms and behavioral notes:
  - Migrants’ better integration in host labor markets and wider migration opportunities likely lowered migration costs and expanded remittance reach.
  - Home-country labor supply responds elastically to remittance income; this behavioral response moderates remittances’ inequality-reducing effect (behavioral response reduces the impact by more than two thirds in reported estimates).
- Tentative policy implications:
  - Reducing remittance transaction costs and expanding access can strengthen remittances’ pro-poor impact.
  - Social insurance and labor-market policies should account for remittances’ role in household income smoothing, especially during external shocks.
  - Monitoring migrant integration and migration opportunities is important for anticipating distributional effects of remittance flows over time.

*IMF Working Paper (Annex Table 1, wp18136)*

### References ________________________________________________________________18

### References

### Figures
- Figure 1. Migration and Remittances ____________________________________________7
- Figure 2. Remittances before, during and after the Peso Crisis _______________________10
- Figure 3. Remittances before, during and after the Global Financial Crisis______________10
- Figure 4. Gini coefficients ___________________________________________________14
- Figure 5. Crisis Effects ______________________________________________________15

### Tables
- Table 1. Demographic Characteristics of Mexican Immigrants in the United States ________8
- Table 2. Characteristics of Remittance-Receiving and Non-Remittance-Receiving  
  Households ________________________________________________________11
- Table 3. Determinants of Remittances by Decile; Before, During and After the  
  Peso Crisis ________________________________________________________12
- Table 4. Determinants of Remittances by Decile; Before, During and After the Global  
  Financial Crisis ____________________________________________________13

### Annex
- Annex Table 1. Determinants of Remittances by Decile, Heckman Two-Step Estimator, 
  Global Financial Crisis ______________________________________________________17

### Introduction — key objectives and approach
- Research question: examine the impact of remittances on inequality in Mexico.
- Motivation: remittances are an important source of external financing at the macro level and can facilitate investments in health, education or small businesses at the micro level; their effects on inequality are less clear.
- Empirical gap: literature is inconclusive on the impact of remittances on inequality, with studies finding positive, negative, or no significant effects.
- Data and methodology:
  - Uses several cross-sections of a large-scale representative household survey.
  - Examines determinants of remittances across income deciles to assess whether remittances play an insurance role.
  - Analyzes changes in likelihood of receiving remittances and amounts received around domestic and external crises: the 1994 Mexican Peso crisis and the Global Financial Crisis of 2008–09.
  - Constructs counterfactual income distributions in the absence of remittances to assess impact on inequality measured using Gini coefficients.
- Paper structure: Section 2 reviews empirical literature with emphasis on counterfactual income distribution studies; Section 3 introduces the household surveys used in the analysis.

*Source: wp18136 - References*

### Section 4 presents a brief overview of Mexico’s migration history. Section 5 outlines the

### Section 4–7 (Mexico: Migration, Empirical Strategy, Results, Conclusions)

### Literature review: remittances, poverty, and inequality
- Large literature finds remittances reduce poverty; cited studies include Acosta and others 2006, Loritz 2008, Taylor and others 2009.
- Empirical findings on inequality are mixed:
  - Studies reporting remittances increase inequality: Adams 2006; Adams and others 2008; Barham and Boucher 1998; Bouoiyour and Miftah 2014; Möllers and Meyer 2014.
  - Studies reporting remittances reduce inequality: Acosta and others 2006; Brown and Jimenez 2007; Gubert and others 2010; Jones 1998; Loritz 2008; Margolis and others 2013; Mughal and Anwar 2012; Taylor and others 2009.
  - Studies reporting no significant effect: Beyene 2014; Yang and Martinez 2005.
- Migration timing matters: ‘Pioneer’ migrants may come from relatively richer households; as migrant networks expand and migration costs fall, later migrants may come from poorer households (Lipton 1980; Stark, Taylor and Yitzhaki 1986; Portes and Rumbaut 1990; Massey, Goldring, and Durand 1994). Cross-sectional implication: positive link between outmigration and inequality in sending countries with more recent migration history (Stark, Taylor and Yitzhaki 1988).
- Methodological note: simple comparisons of income distributions with and without remittances or income-source decompositions may be biased because remittances substitute for earnings the migrant would have earned at home.
- Counterfactual approaches: impute what emigrant households would have earned absent migration (e.g., Barham and Boucher 1998; Adams 2006; Brown and Jimenez 2007; Acosta and others 2008). Acosta and others (2008) for Latin America (including Mexico) find inequality based on imputed income is lower than inequality based on actual income, so remittances lower inequality relative to a no-remittances, no-migration scenario.
- Contribution of this paper: adopt a non-parametric propensity score matching approach to estimate counterfactual incomes and examine how remittances’ effects on inequality change during crises (1994 Mexican Peso crisis; Global Financial Crisis of 2008–09).

### Data
- Data source: National Survey of Income and Expenditure (ENIGH) by Instituto Nacional de Geografia, Estadistica e Informatica (INEGI).
- Survey timing: broadly every second year in the 3rd and 4th quarter.
- Remittances measured from question: ‘What was your income from other countries last month? Two months ago? Three months ago?’, then aggregated to quarterly and annual amounts.
- Years used to capture the Peso crisis: 1989, 1994, 2000 (no survey available for 1995).
- Years used to proxy pre-crisis, during, and post-Global Financial Crisis: 2002, 2008, 2014 (no surveys available for 2007 or 2009).
- Note: reported remittance amount could in theory include investment income from abroad; such income is likely negligible for the typical household in Mexico.

### Migration from Mexico (overview and crisis context)
- Historical trends:
  - Migration to the United States increased in the 1970s due to push and pull factors, including Immigration and Nationality Act (1965) family reunification incentives and Mexican economic crises in 1976 and 1983 (Rosenblum and Brick 2011).
  - Migration flows slowed recently with increased enforcement since 2005 (Gonzales-Barrera 2015) and reversed during the Global Financial Crisis and the subsequent slow recovery in the United States.
- Destination concentration: United States accounts for 97 percent of Mexican outmigration.
- Migrant profile changes (1990 to 2013, Pew Research Center tabulations):
  - Male: 55 (1990) to 53 (2013) percent.
  - Median age: 29 (1990) to 39 (2013) years.
  - Age group shares changed (percent): Younger than 18 15 → 6; 18 to 29 35 → 19; 30 to 39 24 → 26; 40 to 49 13 → 24; 50 to 64 9 → 18; 65 or older 5 → 7.
  - Educational attainment (ages 25 and older): Less than high school diploma 76 → 58 percent; High school diploma 12 → 24 percent; Some college or more 13 → 18 percent.
  - Years in the US: 5 years or less 30 → 8; 6 to 10 years 20 → 15; 11 to 20 years 31 → 35; More than 20 years 19 → 42.
- Remittance behavior:
  - Likelihood of remitting declines with age: 40 percent for migrants below the age of 30, 30 percent after the age of 50 (Gonzales-Barrera 2015).
- Crisis characterization:
  - 1994 Peso crisis: sudden peso devaluation in December 1994; resulted in a 6 percent contraction of the Mexican economy by 1995; increased unemployment, informal employment share and formal-informal wage gap (IMF 2010).
  - Global Financial Crisis: originated in the United States; propagated to Mexico through strong trade link (around 75 percent of Mexican exports go to the United States); Mexico was better equipped to handle the crisis (smaller increases in unemployment, informal employment share, and wage gap).

### Estimation strategy
- Descriptive analysis: document socioeconomic characteristics of remittance-receiving vs non-remittance-receiving households.
- Regression analysis by income decile to examine determinants of:
  - Receiving remittances (binary variable) — probit model.
  - Amount received (in Pesos) — OLS with robust standard errors.
  - Remittances received as percent of household income — OLS with robust standard errors.
- Covariates: household composition, location (urban/rural), characteristics of household head (age, gender, education), and crisis dummies (e.g., 1994, 2000; 2008, 2014 depending on period).
- Counterfactual income construction:
  - Use propensity score matching (radius matching) to impute what remittance-receiving households’ incomes would be in the absence of remittances, enforcing common support and not imposing functional form.
  - Matches remittance-receiving households only to non-remittance-receiving households with similar characteristics.
  - Results robust to alternative matching estimators.

### Results — who receives remittances and determinants by decile
- Trends and magnitudes:
  - Share of households receiving remittances increased in the 1990s from about 4 percent to almost 6 percent by 2000, then declined slightly thereafter.
  - Remittances declined as a share of household income from 30–40 percent in the 1990s to around 20 percent in 2014.
  - In 2014 mean remittances received per household: about 290 USD per month (mean; 140 USD median).
  - Poorer households are generally more likely to receive remittances, though in the 1990s the poorest decile was less likely to receive remittances.
  - Without remittances, remittance-receiving households would be in the 4th decile on average; with remittances they are in the 6th decile; non-remittance-receiving households are on average in the 7th decile in 2014.
- Household characteristics of remittance recipients:
  - Tend to be slightly smaller households, fewer employed household members present, more children and elderly.
  - More likely to have a female household head and lower education.
  - More likely to be from rural areas.
- Determinants across deciles (summary patterns from Tables 3 and 4):
  - Likelihood of receiving remittances:
    - Falls with the number of employed household members (except at the very top in later periods, where positive coefficients suggest investment motives).
    - Increases with the number of elderly.
    - Falls with the number of children (family reunification effect), though less pronounced at the top in later period.
    - Female household head associated with higher likelihood of receiving remittances across deciles.
    - Higher education of household head associated with lower likelihood of receiving remittances across deciles (except bottom decile where not significant in later period).
    - Rural households more likely to receive remittances across the income distribution.
  - Amount of remittances and remittances as share of income:
    - Increase with number of elderly (especially in earlier period).
    - Decrease with number of children and number of employed household members.
    - Lower for male household heads and urban households across the distribution.
    - Education reduces amounts (except at very bottom in later period).
  - Regression specifics (examples of coefficients preserved as reported):
    - In the Peso crisis period regressions (Table 3), age of household head coefficients like -0.0140*** in bottom decile receiving regression; number of elderly 0.387*** in bottom decile receiving regression; 1994 dummy coefficients vary (e.g., -0.257** for 9th decile).
    - In the Global Financial Crisis period regressions (Table 4), number of employed shows negative and significant coefficients across deciles (e.g., -0.162*** bottom decile receiving regression); 2008 and 2014 dummy coefficients often negative at the top deciles (e.g., 2014 dummy -0.606*** for top decile receiving regression).
  - Adjusted R-squared and number of observations provided in tables indicate model fit and sample sizes (e.g., Adjusted R-sq. values range from 0.015 to 0.100 across specifications; number of obs. reported per decile and regression).

### Results — counterfactuals and inequality
- Simple comparison of Gini coefficients:
  - Gini based on actual income (including remittances) is lower than Gini excluding remittances — suggesting remittances reduce inequality.
  - Counterfactual incomes constructed via propensity score matching (taking behavioral responses into account) produce a Gini coefficient lower than income excluding remittances but still higher than actual income — indicating inequality would be higher in the absence of remittances, even allowing for behavioral responses.
- Quantitative magnitudes (relative scale examples provided):
  - Between 2002 and 2012 the Gini coefficient fell by about 0.01.
  - Differences between Gini coefficients in the later period:
    - Actual income vs income without remittances: about 0.006.
    - Actual income vs counterfactual income: about 0.004.
  - Behavioral response (difference between income excluding remittances and counterfactual income) reduces the impact of remittances on inequality by more than two thirds, implying very elastic household labor supply to remittance income.
  - Contextual figure: in 2014 remittances accounted for about 20 percent of household income in the median decile.

### Results — crisis effects
- Both the 1994 Peso crisis and the Global Financial Crisis showed an increasingly pro-poor pattern of remittances:
  - During the Peso crisis:
    - Likelihood of receiving remittances and remittance amounts as a share of income fell at the top deciles; little change in lower deciles.
  - During the Global Financial Crisis:
    - Likelihood of receiving remittances and remittance amounts as a share of income fell at the top deciles and increased at the bottom deciles.
- Interpretation:
  - Decline of remittances to richer households during crises is consistent with reduced investment motives.
  - Increase in remittances to poorer households during the Global Financial Crisis could reflect:
    - More widespread migration opportunities in Mexico (remittances reaching lower deciles).
    - Longer-established migrant communities better able to cushion shocks in the United States.
  - Noteworthy that the insurance effect occurred even when both sending and receiving countries were hit by a common shock.
- Robustness:
  - Results qualitatively similar when including household income (excluding remittances) as a regressor.
  - Heckman two-step estimator accounting for non-random selection into receiving remittances yields similar results (Annex Table 1).

### Conclusions and policy-relevant implications
- Main findings:
  - Remittances lower income inequality in Mexico at the macro level: counterfactual-income-based Gini is higher than actual-income-based Gini, implying inequality would be higher in the absence of remittances even after accounting for behavioral responses.
  - Remittances became increasingly pro-poor over time: earlier years saw remittances concentrated more in middle deciles; later periods show remittances shifting to poorer deciles.
  - During crises (1994 Peso crisis and Global Financial Crisis), remittances to top deciles fell while remittances to bottom deciles fell less or even increased (particularly during the Global Financial Crisis), indicating a cushioning/insurance role for poorer households.
- Mechanisms and interpretation:
  - Changes consistent with migrants becoming better integrated in host country labor markets and migration opportunities becoming more widespread, lowering migration costs and expanding remittance reach.
  - Behavioral responses in home-country labor supply are significant; imputed counterfactual incomes show labor supply is elastic to remittance income, substantially moderating remittances’ inequality-reducing effect.
- Tentative policy implications (derived from empirical findings):
  - Remittance channels and policies that lower transaction costs and expand access may strengthen remittances’ pro-poor impact.
  - Social insurance and labor-market policies should account for remittances’ role in household income smoothing, especially during external shocks.
  - Monitoring migrant integration and migration opportunities is important for anticipating distributional effects of remittance flows over time.

*Source: wp18136 (Sections 4–7).*

### Annex Table 1. Determinants of Remittances by Decile, Heckman Two-Step Estimator, Global Financial

### Annex Table 1. Determinants of Remittances by Decile, Heckman Two-Step Estimator, Global Financial Crisis

### Selection-corrected amount equation (selection variable = Pred. prob. (ln))
- No. of hh members:
  - Bottom decile: 95.36***
  - 2: 132.6
  - 3: 59.12
  - 4: 107.0
  - 5: 403.3***
  - 6: 229.0***
  - 7: 395.4***
  - 8: 6.287
  - 9: 485.3
  - Top decile: 798.7
  - Standard errors (in order): (26.62), (150.9), (152.9), (160.1), (99.80), (59.14), (137.7), (155.6), (429.6), (495.7)

- No. of children:
  - Bottom decile: -61.90
  - 2: -124.5
  - 3: 38.23
  - 4: -36.13
  - 5: -674.9***
  - 6: -9.001
  - 7: -558.7**
  - 8: 157.4
  - 9: -145.6
  - Top decile: -835.3*
  - Standard errors: (39.46), (228.8), (229.0), (259.0), (160.4), (30.68), (220.8), (301.3), (754.4), (504.9)

- No. of elderly:
  - Bottom decile: 587.5***
  - 2: 288.1
  - 3: -323.8
  - 4: 303.8
  - 5: 619.7***
  - 6: 1346.8***
  - 7: 262.0
  - 8: 1183.9*
  - 9: -2098.6
  - Top decile: 356.6
  - Standard errors: (114.7), (388.5), (439.6), (407.9), (223.6), (234.9), (247.6), (605.2), (1382.3), (755.8)

- No. of employed:
  - Bottom decile: -762.3***
  - 2: -1245.1***
  - 3: -1209.2***
  - 4: -1517.0***
  - 5: -2347.0***
  - 6: -1892.7***
  - 7: -1571.8***
  - 8: -1094.6***
  - 9: -2268.2***
  - Top decile: -2850.3**
  - Standard errors: (93.65), (261.4), (259.9), (274.5), (225.0), (185.2), (312.6), (283.2), (666.7), (1123.5)

- Age of hh head:
  - Bottom decile: -21.06***
  - 2: -46.12***
  - 3: -12.91
  - 4: -43.29**
  - 5: -71.99***
  - 6: -63.76***
  - 7: -72.98***
  - 8: -65.27**
  - 9: -41.18
  - Top decile: -125.9***
  - Standard errors: (4.081), (13.41), (18.76), (20.42), (11.43), (10.76), (18.69), (27.77), (61.67), (13.72)

- Male hh head:
  - Bottom decile: -1044.0***
  - 2: -1389.4***
  - 3: -1548.4***
  - 4: -2313.3***
  - 5: -2795.3***
  - 6: -1600.9***
  - 7: -1777.8***
  - 8: -2059.4***
  - 9: -4525.0***
  - Top decile: -3713.1**
  - Standard errors: (107.6), (426.5), (477.9), (512.2), (322.1), (191.5), (441.1), (741.5), (1543.3), (1701.6)

- Education of hh head:
  - Bottom decile: -120.0***
  - 2: -189.9**
  - 3: -2.185
  - 4: -73.30
  - 5: -1052.7***
  - 6: -1037.9***
  - 7: -928.1***
  - 8: -1100.9***
  - 9: -88.69
  - Top decile: -1541.1***
  - Standard errors: (27.32), (90.12), (132.2), (159.6), (98.06), (105.4), (150.8), (179.6), (284.2), (446.5)

- Urban hh:
  - Bottom decile: -861.7***
  - 2: -2042.4
  - 3: -916.9
  - 4: -888.8
  - 5: -5153.0***
  - 6: -3919.2***
  - 7: -2118.1***
  - 8: -2421.7***
  - 9: 1117.4
  - Top decile: -5025.2***
  - Standard errors: (109.2), (.), (680.0), (727.1), (454.5), (381.7), (373.2), (642.3), (1520.2), (1611.2)

- 2008 dummy:
  - Bottom decile: -9.782
  - 2: 569.6*
  - 3: 1744.4***
  - 4: 78.62
  - 5: 613.1**
  - 6: -419.8***
  - 7: -213.1
  - 8: -256.5
  - 9: 401.8
  - Top decile: -844.3***
  - Standard errors: (126.1), (320.2), (405.6), (538.0), (266.4), (81.97), (449.7), (641.9), (1525.9), (125.4)

- 2014 dummy:
  - Bottom decile: 50.03
  - 2: 289.4
  - 3: 856.2*
  - 4: -513.7
  - 5: 428.8
  - 6: -590.2***
  - 7: -329.6
  - 8: -29.87
  - 9: 4372.2**
  - Top decile: -1620.7***
  - Standard errors: (118.3), (368.9), (460.4), (566.9), (307.4), (91.42), (446.0), (713.6), (1729.3), (543.0)

- Pred. prob. (ln):
  - Bottom decile: 0.158***
  - 2: 0.451***
  - 3: 0.313***
  - 4: 0.325***
  - 5: 0.259***
  - 6: 0.200***
  - 7: 0.0770***
  - 8: 0.0753***
  - 9: 0.121***
  - Top decile: 0.115***
  - Standard errors: (0.0197), (0.0343), (0.0228), (0.0238), (0.0196), (0.0195), (0.0124), (0.0127), (0.0165), (0.0250)

- Number of obs. (for each decile): 6563, 6556, 6563, 6579, 6569, 6564, 6582, 6585, 6524, 6596

### Selection-corrected amount equation (selection variable = Pred. prob. (square))
- No. of hh members:
  - Bottom decile: -0.864
  - 2: 46.43
  - 3: 53.96
  - 4: 120.9
  - 5: 371.2*
  - 6: 115.0
  - 7: 665.2**
  - 8: 297.5
  - 9: 596.4**
  - Top decile: 3221.5**
  - Standard errors: (94.88), (129.3), (152.6), (161.8), (203.1), (255.4), (282.9), (345.2), (248.9), (1279.9)

- No. of children:
  - Bottom decile: -144.7
  - 2: 98.73
  - 3: 27.51
  - 4: -151.9
  - 5: -88.22
  - 6: 350.2
  - 7: -450.2
  - 8: -455.9
  - 9: -356.6
  - Top decile: -3066.1
  - Standard errors: (149.8), (194.2), (229.3), (268.5), (315.3), (367.0), (429.6), (566.8), (440.0), (1943.2)

- No. of elderly:
  - Bottom decile: -1.238
  - 2: -206.1
  - 3: -187.4
  - 4: 462.5
  - 5: -138.1
  - 6: -96.72
  - 7: -634.4
  - 8: 830.2
  - 9: 961.2
  - Top decile: -327.6
  - Standard errors: (198.5), (272.8), (385.5), (406.8), (504.4), (593.9), (698.8), (985.9), (789.7), (4684.6)

- No. of employed:
  - Bottom decile: -526.6***
  - 2: -830.0***
  - 3: -1283.3***
  - 4: -1745.0***
  - 5: -1767.2***
  - 6: -1631.7***
  - 7: -2401.8***
  - 8: -2777.9***
  - 9: -2025.1***
  - Top decile: -6511.8***
  - Standard errors: (126.6), (189.9), (232.2), (290.6), (308.1), (342.7), (375.8), (479.0), (362.1), (1601.7)

- Age of hh head:
  - Bottom decile: -22.22**
  - 2: -28.00**
  - 3: -18.49
  - 4: -53.74***
  - 5: -14.12
  - 6: -18.15
  - 7: -65.15**
  - 8: -102.1**
  - 9: -158.1***
  - Top decile: -339.7
  - Standard errors: (10.98), (12.12), (15.47), (20.44), (24.51), (26.37), (32.13), (45.33), (26.22), (257.3)

- Male hh head:
  - Bottom decile: -559.7***
  - 2: -738.5**
  - 3: -1717.1***
  - 4: -2724.5***
  - 5: -3287.1***
  - 6: -3117.7***
  - 7: -4086.2***
  - 8: -3888.7***
  - 9: -4873.4***
  - Top decile: -2643.2
  - Standard errors: (192.7), (338.8), (398.8), (514.2), (551.3), (614.9), (692.9), (1004.7), (739.2), (4550.3)

- Education of hh head:
  - Bottom decile: 54.81
  - 2: -87.66
  - 3: -58.56
  - 4: -215.8*
  - 5: -95.63
  - 6: 17.15
  - 7: -186.6
  - 8: -260.8*
  - 9: -319.8***
  - Top decile: -288.1
  - Standard errors: (62.40), (66.19), (61.69), (116.3), (123.4), (121.9), (127.1), (148.8), (88.73), (508.4)

- Urban hh:
  - Bottom decile: -278.0
  - 2: -881.7**
  - 3: -1198.8***
  - 4: -1578.2***
  - 5: -1477.7***
  - 6: -1341.2**
  - 7: -2660.8***
  - 8: -2322.5**
  - 9: -1827.0**
  - Top decile: -4746.6
  - Standard errors: (237.9), (377.5), (443.7), (594.4), (568.8), (666.6), (720.6), (1009.9), (769.0), (4368.8)

- 2008 dummy:
  - Bottom decile: 380.5
  - 2: 711.0**
  - 3: 1738.8***
  - 4: 60.49
  - 5: 1630.7***
  - 6: 1883.2**
  - 7: 2958.2***
  - 8: 1145.4
  - 9: -1674.4*
  - Top decile: 1744.2
  - Standard errors: (265.0), (307.7), (405.4), (540.5), (631.4), (761.4), (869.7), (1230.3), (889.6), (3519.8)

- 2014 dummy:
  - Bottom decile: 356.5
  - 2: 455.4
  - 3: 857.6*
  - 4: -527.7
  - 5: 948.8
  - 6: 1357.5
  - 7: 1470.8
  - 8: 99.27
  - 9: -518.8
  - Top decile: 9633.7
  - Standard errors: (267.8), (329.0), (460.6), (568.7), (733.7), (844.0), (934.7), (1429.4), (1044.0), (.)

- Pred. prob. (square):
  - Bottom decile: 33.15***
  - 2: 75.89***
  - 3: 43.96***
  - 4: 74.48***
  - 5: 75.98***
  - 6: 60.14***
  - 7: 53.60***
  - 8: 34.48***
  - 9: 26.93***
  - Top decile: 88.65***
  - Standard errors: (4.200), (5.709), (3.600), (5.251), (7.584), (5.534), (6.667), (4.979), (4.367), (13.02)

- Number of obs. (for each decile): 6563, 6556, 6563, 6579, 6569, 6564, 6582, 6585, 6524, 6596

### Notes and dataset
- Note: The selection variable is a nonlinear transformations (the logarithm and the square resp ectively ) of the p redicted p robability  of sending as a control in the amount equation, acting as a selection correction.
- Dataset: pooled cross-section of households in 2002, 2008 and 2014; 2002 is treated as the base year.

*Italic source:* IMF Working Paper (Annex Table 1, wp18136)

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp18136.pdf_