## 1. Introduction (excerpt)

## Source details

**Canonical URL:** [1. Introduction (excerpt)](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1413.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1413.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1413.pdf.json)

---

### Context and motivation
- Developing countries remain extremely vulnerable to adverse exogenous shocks, including the global financial crisis and the world food price crisis of 2008.
- Natural disasters and other shocks have severe consequences for output, trade, and fiscal balances.
- Paper objective: examine how remittances are affected by structural characteristics, macroeconomic conditions, and adverse shocks in both source and recipient economies to assess whether remittances act as a counter-cyclical shock absorber or are pro-cyclical.
- Aggregate remittances to developing countries:
  - projected to reach USD 414 billion in 2013, and USD 540 billion by 2016 (The World Bank, 2013).
- Remittances have proved very resilient since the onset of the global financial crisis.

### Research question and relevance
- Core question: Should remittances be viewed as a counter-cyclical shock absorber (helping smooth consumption during downturns) or are they pro-cyclical?
- Relevant considerations:
  - Increasing migration has raised remittances relative to capital flows.
  - Ambiguous a priori effects of the global financial crisis: downturn in home country could increase remittances (altruism/insurance); recession in host country could reduce migrants’ incomes (notably temporary workers in construction).

### Data and empirical approach (overview)
- Novel panel dataset: bilateral remittances from 103 Italian provinces to 107 developing countries over the period 2005-2011.
- Estimation strategy: gravity model for remittances, exploiting cross-sectional variation across source provinces and recipient countries.
- Separately considers cyclical and trend components in GDP per capita.
- Controls for recipient-country vulnerability: natural disasters, large declines in the terms of trade, and armed conflicts.
- Focusing on remittances from Italian provinces attenuates endogeneity of recipient-country business cycle and helps address reverse-causality concerns.

### Main contributions
- Large bilateral data allow systematic analysis of the correlation between remittances and the business cycle in both source and recipient economies, covering periods before and after the 2007-08 financial crisis.
- Exploits source-province dimension to address potential endogeneity of recipient-country business cycle.
- Investigates financial development in the remittance source economy (source-province credit markets), a dimension less explored in recipient-focused literature.

### High-level empirical findings
- Remittances from Italian provinces:
  - are negatively correlated with the business cycle in recipient countries (i.e., increase when recipient-country conditions worsen).
  - increase especially strongly in response to adverse exogenous shocks, such as natural disasters or large declines in the terms of trade.
  - are positively correlated with potential GDP in recipient countries (consistent with both altruism and investment motives).
  - are positively correlated with economic conditions in the source province.
  - remain counter-cyclical with respect to the recipient country even when similar negative shocks hit both source and recipient economies.

### Italy-specific aggregate and labor market notes
- Total remittances from Italy to developing countries doubled between 2005 and 2011, reaching almost e7 billion (source text spelling preserved).
- After 2007, growth of remittances slowed, remittances declined in 2010, and 2011 saw a rapid recovery.
- Regional differences:
  - remittances as a share of regional GDP highest in Latium and Tuscany;
  - remittances rose in Campania, Liguria, Lombardy, and Veneto.
  - East Asia and Pacific is the main recipient region for both Italian and global remittances.
  - Europe and Central Asia’s share from Italy is twice its share of global remittances.
  - South Asia’s share is rising; Sub-Saharan Africa’s share is limited.
- Country-level patterns:
  - China, Romania, and the Philippines were major recipients from Italy in both 2005 and 2011.
  - Transfers to Bangladesh, Sri Lanka and Georgia increased dramatically between 2005 and 2011.
  - Colombia registered a decrease in remittances from Italy over this period.
- Labor market context (Italy):
  - Unemployment rate for foreign-born workers: 10.2 percent in 2005 and 12.1 percent in 2011.
  - Unemployment rate for native workers: 7.6 percent in 2005 and 8.0 percent in 2011.

---

### 4.1 The empirical gravity model

### Model specification and estimation
- Baseline specification (equation 1): log bilateral remittances REM_i,j,t is a function of:
  - CYCLE_i,t and CYCLE_j,t (percentage deviation of GDP per capita from trend in source province i and recipient country j).
  - TREND_i,t and TREND_j,t (log of trend GDP per capita).
  - MIGR_i,j,t (log of bilateral stocks of migrants).
  - ∆MIGR_i,j (percentage growth in the bilateral stocks of migrants).
  - D_i,j (distance between province i and country j).
  - POP_i,t and POP_j,t (log of population levels).
  - Standard error: ε_i,j,t.
- Augmented specification (equation 2) adds province-level X_i,t and country-level Z_j,t characteristics.
- Key coefficients of interest:
  - α1: correlation between remittances and the business cycle in the source province (CYCLE_i).
  - α2: correlation between remittances and the business cycle in the recipient country (CYCLE_j).
  - α4: correlation with TREND_j (long-run output trend in recipient country) as indicator of investment motive.
  - β2: correlation with ∆MIGR_i,j (growth of migrant stock) expected β2 > 0.
- Estimation method:
  - Poisson Pseudo-Maximum Likelihood (Silva and Tenreyro, 2006) used due to many zero observations and potential heteroskedasticity.
  - Country, province, and time fixed effects included.
  - Standard errors clustered by country-province pairs.

### Additional controls and vulnerability indicators (Z_j,t)
- DIS_j,t: indicator = 1 if country j experienced natural disasters in year t.
- WAR_j,t: indicator = 1 if armed conflicts occurred in country j at time t.
- TT_j,t: indicator = 1 if country j experienced a major negative shock to the terms of trade (lowest 5 percent of annual variation).

### Financial development measures
- Recipient-country financial development:
  - FINDEV_j,t = log(share of credit to the private sector over GDP).
  - A priori ambiguous effect: could increase remittances via lower transaction costs or be a substitute if remittances relax liquidity constraints.
- Source-province financial development:
  - BANK_i,t: number of local bank branches per inhabitant (proxy for banking-sector penetration).
  - FD_i,t: functional distance between banks and local economies (based on banks headquartered in the relevant province).
  - Expectation: greater provincial banking penetration reduces transaction costs and encourages remittance outflows; lower functional distance improves local information and service provision to migrants.

### Data sources and construction (selected)
- Remittances dataset: bilateral outward remittances from 103 Italian provinces to 107 developing countries, annual, constant prices, 2005-11, compiled by the Bank of Italy; covers formal channels (money-transfer operators and postal system); banking system included only since 2010 and accounts for 5 to 10 percent of total remittances; informal transfers not included.
- Migrant stocks: bilateral stocks 2005-11 from ISTAT; represent foreign resident population in each province by citizenship at the beginning of each year.
- Recipient-country GDP (1950-2012): IMF World Economic Outlook; cyclical and trend components via Hodrick-Prescott filter.
- Natural disasters: EM-DAT (CRED). Armed conflicts: UCDP/PRIO Armed Conflict Dataset.
- Provincial real value added (1995-2010): ISTAT and Istituto Guglielmo Tagliacarne; HP filter for cycles/trends.
- Number of bank branches: Bank of Italy.

---

### 5.4 Robustness checks

### Samples and exclusions
- Including China (original baseline excluded China as outlier) confirms earlier results (Table 6, column 1).
- Splitting by recipient income level:
  - Low-income countries: coefficient on CYCLE_j remains negative and significant (Table 6, column 2).
  - Middle-income countries: coefficient on CYCLE_j remains negative and significant (Table 6, column 3).
- Excluding small migrant communities:
  - Exclusion rule: MIGR_ijt <100 reduces original sample by almost three-quarters; baseline results remain valid (Table 6, column 4).
  - Results robust to alternative thresholds up to MIGR_ijt <500.
- Dropping largest recipient countries (population above the 95th percentile): countries dropped: Bangladesh, Brazil, China, India, Indonesia, Nigeria and Pakistan. General findings confirmed.

### Additional covariates and macro controls
- Controlling for foreign aid (aid per capita) indicates aid and remittances are substitutes.
  - Inclusion of aid reduces magnitude of the coefficient on CYCLE_j compared to baseline.
- Adding macroeconomic stability and institutional measures:
  - Fiscal balance (as ratio of GDP): coefficient statistically significant and positive.
  - Ratio of external debt over GDP: coefficient statistically significant and negative.
  - Constraints on the executive: positive coefficient indicates remittances are larger for countries with stronger institutions.
  - Elasticity of remittances to CYCLE_j is almost unaffected by these additions; reduction in coefficient mainly reflects sample size reduction rather than institutional quality control.
- Jointly controlling for these covariates still yields a negative effect of the cyclical component of output in recipient countries on remittances.

### Alternative estimator
- Fixed Effects Poisson estimator used to account for panel dimension.
  - Variables that vary only across province-country pairs and not over time (e.g., D_i,j and ΔMIGR_i,j) cannot be identified.
  - Main results largely confirmed.
  - Negative coefficient on CYCLE_j remains evident even with province-country pair fixed effects.

### Robustness implications (summary)
- The counter-cyclical behavior of remittances with respect to recipient-country output (negative and significant coefficient on CYCLE_j) is robust to:
  - Inclusion of China in the sample.
  - Splitting the sample by recipient income level.
  - Excluding small-migrant communities (MIGR_ijt <100, robust up to MIGR_ijt <500).
  - Dropping largest recipient countries (population above the 95th percentile).
  - Adding controls for AID_j,t, FISC BAL_j,t, EXT DEBT_j,t, and EXEC CONST_j,t.
  - Alternative estimation via Fixed Effects Poisson with province-country pair fixed effects.

---

### Key empirical results and selected statistics (from tables)

### Baseline Poisson Pseudo Maximum Likelihood results (Table 3 highlights)
- CYCLE_i,t coefficients (columns 1–3): 1.122, 1.258, 1.329. [Clustered SEs: [0.822], [0.830], [0.830]]
- CYCLE_j,t coefficients (columns 1–3): -3.510***, -3.491***, -4.699***. [SEs: [0.639], [0.649], [0.615]]
- MIGR_i,j,t coefficients (columns 1–5): 0.805***, 0.813***, 0.814***, 0.801***, 0.803***. [SEs: [0.035], [0.037], [0.037], [0.036], [0.036]]
- POP_j,t coefficients (columns 1–4): 8.648***, 9.373***, 9.974***, 10.719***. [SEs: [1.193], [1.304], [1.361], [1.328]]
- Observations: 57,248; 48,608; 46,760; 38,994; 38,994.
- R2: 0.741; 0.763; 0.771; 0.795; 0.796.

### Vulnerability factors (Table 4 highlights)
- CYCLE_i,t (cols 1–4): 1.316, 1.329, 1.391, 1.359. [SEs: [0.831], [0.830], [0.940], [0.946]]
- CYCLE_j,t (cols 1–4): -4.680***, -4.854***, -2.266***, -2.373***. [SEs: [0.616], [0.600], [0.667], [0.671]]
- TREND_j,t (cols 1–4): 3.194***, 3.247***, 1.721***, 1.824***. [SEs: [0.534], [0.536], [0.522], [0.531]]
- DIS_j,t (cols 3–4): 0.045**, 0.046**. [SEs: [0.019], [0.018]]
- TT_j,t (cols 3–4): 0.094***, 0.105***. [SEs: [0.033], [0.035]]
- Observations: 38,994; 36,764; 37,414; 35,555.
- R2: 0.795; 0.796; 0.895; 0.896.

### Financial development (Table 5 highlights)
- FINDEV_j,t coefficients (cols 1–2): -0.920***, -0.936***. [SEs: [0.138], [0.132]]
- BANK_i,t coefficients (cols 3–4): 1.047**, 1.180***. [SEs: [0.429], [0.425]]
- FD_i,t coefficients (cols 3–4): -0.389**, -0.454***. [SEs: [0.195], [0.161]]
- CYCLE_i,t (cols 1–4): 1.725**, 1.491*, 0.680, 1.153*. [SEs: [0.725], [0.878], [0.643], [0.600]]
- Observations: 37,810; 38,994; 38,994; 37,810.
- R2: 0.807; 0.795; 0.798; 0.810.

### Robustness tables (selection)
- Table 6 (sample definition) observations and R2 vary:
  - Including China: Observations 39,610. R2 0.711.
  - Low Income Countries: Observations 8,919. R2 0.959.
  - Middle Income Countries: Observations 30,943. R2 0.717.
  - Large migrant communities: Observations 11,069. R2 0.808.
  - No large recipients: Observations 36,439. R2 0.797.
  - CYCLE_j,t coefficients across columns include -3.055***, -2.182**, -3.205***, -5.370***, -4.975***.
- Table 7 (additional covariates) key coefficients:
  - AID_j,t (cols 1–2): -0.115***, -0.243***. [SEs: [0.012], [0.041]]
  - FISC BAL_j,t (cols 1–2): 4.612***, 1.633**. [SEs: [0.622], [0.784]]
  - EXEC CONST_j,t (cols 4–5): 0.115***, 0.252***. [SEs: [0.028], [0.039]]
  - Observations vary: 35,938; 26,270; 38,079; 17,530; 10,951. R2 up to 0.911.
- Tables 8–10 (Fixed Effects Poisson estimator) confirm main patterns:
  - Table 8 baseline: CYCLE_i,t coefficients 1.418*, 1.522**, 1.815** in shown columns. Observations: 55,140; 46,286; 45,381; 38,549; 38,549. Number of pairs: 8,457; 8,199; 8,044; 6,512; 6,512.
  - Table 9 vulnerability: DIS_j,t and TT_j,t remain relevant; TT_j,t 0.090*** and 0.102*** in columns. Observations: 38,549; 36,346; 37,019; 35,173. Number of pairs: 6,512; 6,139; 6,253; 5,940.
  - Table 10 financial development: FINDEV_j,t -0.745***, -0.752***; FD_i,t -0.402**, -0.427***. Observations: 37,305; 38,549; 38,549; 37,305. Number of pairs: 6,364; 6,512; 6,512; 6,364.

---

### Key data constructs and summary statistics (from Table 1)
- REM_ijt: Mean 652969.4. St. Dev.: 1.14E+07.
- CYCLE_i,t: Mean 0.000. St. Dev.: 0.029.
- CYCLE_j,t: Mean 0.001. St. Dev.: 0.031.
- TREND_i,t: Mean 9.994. St. Dev.: 0.251.
- TREND_j,t: Mean 10.837. St. Dev.: 2.493.
- MIGR_ijt: Mean 3.220. St. Dev.: 2.096.
- ΔMIGR_ij: Mean 0.037. St. Dev.: 0.107.
- WAR_j,t: Mean 0.170. St. Dev.: 0.376.
- DIS_j,t: Mean 0.742. St. Dev.: 0.437.
- TT_j,t: Mean 0.048. St. Dev.: 0.214.
- DIST_ij: Mean 8.415. St. Dev.: 0.764.
- FINDEV_j,t: Mean 3.257. St. Dev.: 0.798.
- BANK_i,t: Mean 1.744. St. Dev.: 0.370.
- FD_i,t: Mean 1.176. St. Dev.: 0.341.
- AID_j,t: Mean 3.296. St. Dev.: 1.386.
- FISC BAL_j,t: Mean -0.015. St. Dev.: 0.043.
- EXT DEBT_j,t: Mean 0.465. St. Dev.: 0.591.
- EXEC CONST_j,t: Mean 4.733. St. Dev.: 1.904.

*Italic source attribution: Excerpt from IMF working paper _wp1413 (sections 1, 4.1, 5.4 and References as provided).*

### 1. Introduction....................................................................................... 3

### 1. Introduction

### Context and motivation
- Developing countries remain extremely vulnerable to adverse exogenous shocks, including the global financial crisis and the world food price crisis of 2008 (Figure 1, left panel).
- Over recent decades developing countries have been increasingly subject to natural disasters with severe consequences for output, trade, and fiscal balances (Raddatz, 2007; Noy, 2009).
- The paper examines how remittances are affected by structural characteristics, macroeconomic conditions, and adverse shocks in both source and recipient economies to assess whether remittances act as a counter-cyclical shock absorber (smoothing consumption during downturns) in contrast to typically pro-cyclical private capital flows.
- Remittances to developing countries:
  - projected to reach USD 414 billion in 2013, and USD 540 billion by 2016 (The World Bank, 2013).
  - have proved very resilient since the onset of the global financial crisis.

### Research question and relevance
- Core question: Should remittances be viewed as a counter-cyclical shock absorber (helping smooth consumption during downturns) or are they pro-cyclical?
- Salient reasons this matters:
  - Increasing migration has raised remittances relative to capital flows.
  - The global financial crisis affected both migrants’ home and host countries, creating ambiguous a priori effects on remittances:
    - Downturn in home country could increase remittances via altruism/insurance.
    - Recession in host country could reduce migrants’ incomes (notably temporary workers in construction).

### Data and empirical approach (overview)
- Uses a novel panel dataset covering bilateral remittances from 103 Italian provinces to 107 developing countries over the period 2005-2011.
- Estimates a gravity model for remittances, exploiting large cross-sectional variation across source provinces and recipient countries.
- Separately considers cyclical and trend components in GDP per capita.
- Controls for specific recipient-country vulnerability factors: natural disasters, large declines in the terms of trade, and armed conflicts.
- By focusing on remittances from Italian provinces rather than aggregate remittance inflows, the dataset attenuates endogeneity of recipient-country business cycle and helps address reverse-causality concerns.

### Main contributions
- First: Large bilateral data allow systematic analysis of the correlation between remittances and the business cycle in both source and recipient economies, covering periods before and after the 2007-08 financial crisis.
- Second: Addresses potential endogeneity of recipient-country business cycle more satisfactorily than many previous studies by exploiting the source-province dimension.
- Third: Investigates the role of financial development in the remittance source economy (source-province credit markets) — a dimension less explored in the literature focused on recipient-country financial development.

### Key empirical findings (high-level)
- Remittances from Italian provinces:
  - are negatively correlated with the business cycle in recipient countries (i.e., increase when recipient-country conditions worsen).
  - increase especially strongly in response to adverse exogenous shocks, such as natural disasters or large declines in the terms of trade.
  - are positively correlated with potential GDP in recipient countries (consistent with both altruism and investment motives).
  - are positively correlated with economic conditions in the source province.
  - remain counter-cyclical with respect to the recipient country even when similar negative shocks hit both source and recipient economies.
- Financial development relationships:
  - remittances are positively associated with financial development in the source province (suggesting reduced transaction costs and eased access to financial services for migrants).
  - remittances are negatively associated with financial development in the recipient country (suggesting remittances help alleviate credit constraints in recipient countries).

### Additional empirical and dataset notes
- The dataset exhibits significant variability in remittances both over time and across source provinces and recipient countries.
- Aggregate patterns for Italy:
  - Total remittances from Italy to developing countries doubled between 2005 and 2011, reaching almoste7 billion, in line with growth in the stock of foreign residents in Italy (Figure 2).
  - After 2007, growth of remittances slowed, remittances declined in 2010, and 2011 saw a rapid recovery.
  - Regional differences: remittances as a share of regional GDP are highest in Latium and Tuscany; remittances rose in Campania, Liguria, Lombardy, and Veneto.
  - Regional/destination patterns largely mimic global distribution: East Asia and Pacific is the main recipient region for both Italian and global remittances; Europe and Central Asia’s share from Italy is twice its share of global remittances; South Asia’s share is rising; Sub-Saharan Africa’s share is limited.
  - Country-level: China, Romania, and the Philippines were the major recipients of remittances from Italy in both 2005 and 2011. Transfers to Bangladesh, Sri Lanka and Georgia increased dramatically between 2005 and 2011. Colombia registered a decrease in remittances from Italy over this period.
  - The stock of resident migrants by country of origin is positively correlated with remittances to that country in 2011, with some outliers (notably China).
- Labor market context (Italy):
  - Unemployment rate for foreign-born workers: 10.2 percent in 2005 and 12.1 percent in 2011.
  - Unemployment rate for native workers: 7.6 percent in 2005 and 8.0 percent in 2011.

### Structure of the paper (as presented)
- Section 2: literature review on macroeconomic determinants of remittances.
- Section 3: selected statistics about remittances outflows from Italian provinces to developing countries.
- Section 4: data and the estimated model.
- Section 5: empirical results.

*Source: IMF working paper, "1. Introduction" (pages 3–16 of the provided content).*

### 4.1  The empirical gravity model

### 4.1  The empirical gravity model

### Model specification and estimation
- Baseline specification (equation 1): the log of bilateral remittances REM_i,j,t is a function of:
  - CY CLE_i,t and CY CLE_j,t (percentage deviation of GDP per capita from trend in source province i and recipient country j)
  - TREND_i,t and TREND_j,t (log of trend GDP per capita)
  - MIGR_i,j,t (log of bilateral stocks of migrants)
  - ∆MIGR_i,j (percentage growth in the bilateral stocks of migrants)
  - D_i,j (distance between province i and country j)
  - POP_i,t and POP_j,t (log of population levels)
  - Standard error term: ε_i,j,t
- Augmented specification (equation 2) adds province-level X_i,t and country-level Z_j,t characteristics.
- Key coefficients of interest:
  - α1: correlation between remittances and the business cycle in the source province (CY CLE_i)
  - α2: correlation between remittances and the business cycle in the recipient country (CY CLE_j)
  - α4: correlation with TREND_j (long-run output trend in recipient country) as indicator of investment motive
  - β2: correlation with ∆MIGR_i,j (growth of migrant stock) expected β2 > 0
- Estimation method:
  - Poisson Pseudo-Maximum Likelihood (Silva and Tenreyro, 2006) used due to many zero observations in the dependent variable and potential heteroskedasticity.
  - Country, province, and time fixed effects included.
  - Standard errors clustered by country-province pairs.

### Additional controls and vulnerabilities
- Country-level vulnerability indicators included in Z_j,t:
  - DIS_j,t: indicator equal to 1 if country j experienced natural disasters in year t
  - WAR_j,t: indicator equal to 1 if armed conflicts occurred in country j at time t
  - TT_j,t: indicator equal to 1 if country j experienced a major negative shock to the terms of trade, defined as an observation falling in the lowest 5 percent of the distribution of the annual variation in the terms-of-trade index
- Financial development measures:
  - Recipient-country financial development: FINDEV_j,t = log(share of credit to the private sector over GDP)
    - A priori ambiguous effect: could increase remittances via lower transaction costs or channel profit-driven remittances to projects; or be a substitute to remittances if remittances relax liquidity constraints.
  - Source-province financial development:
    - BANK_j,t: number of local bank branches per inhabitant (proxy for banking-sector penetration)
    - FD_j,t: functional distance between banks and local economies (based on banks headquartered in the relevant province)
    - Expectation: greater provincial banking penetration reduces transaction costs and encourages remittance outflows; lower functional distance (banks headquartered locally) improves local information and service provision to migrants.

### Data sources and construction
- Main remittances dataset: detailed panel on bilateral outward remittances from 103 Italian provinces to 107 developing countries, annual, constant prices, period 2005-11, compiled by the Bank of Italy; covers formal channels (money-transfer operators and postal system); banking system included only since 2010 and accounts for 5 to 10 percent of total remittances; informal transfers not included.
- Migrant stocks: bilateral stocks 2005-11 from ISTAT, representing foreign resident population in each province by citizenship at the beginning of each year; age structure unavailable; use total growth rate of migrants over 2005-11 as proxy for recency of migrant community.
- Distances: bilateral distances (kilometers) computed from geographical coordinates of administrative capitals.
- Recipient-country GDP (1950-2012): IMF World Economic Outlook; cyclical and trend components extracted via Hodrick-Prescott filter.
- Population and FINDEV: World Development Indicators.
- Natural disasters: EM-DAT (Centre for Research on the Epidemiology of Disasters).
- Armed conflicts: UCDP/PRIO Armed Conflict Dataset.
- Terms of trade: IMF World Economic Outlook.
- Provincial real value added (1995-2010): ISTAT and Istituto Guglielmo Tagliacarne; cyclical and trend components via Hodrick-Prescott filter.
- Provincial population (2005-2011): ISTAT.
- Number of bank branches: Bank of Italy.

*Italic source attribution: Excerpt from _wp1413 - 4.1  The empirical gravity model*

### 5.4  Robustness checks

### 5.4  Robustness checks

### Different samples
- Baseline sample originally excludes remittances to China (outlier; see footnote 8). Re-estimating the baseline model including China confirms earlier results (Table 6, column 1).
- Splitting the sample by recipient income level:
  - Low-income countries: coefficient on CY_CLE_j remains negative and significant (Table 6, column 2).
  - Middle-income countries: coefficient on CY_CLE_j remains negative and significant (Table 6, column 3).
- Excluding province-country pairs with small migrant communities:
  - Exclusion rule: MIGR_ijt <100.
  - This threshold reduces the original sample by almost three-quarters, yet baseline results remain valid (Table 6, column 4).
  - Results are robust to alternative specifications of the threshold up to MIGR_ijt <500.
- Dropping largest recipient countries (those with total population above the 95th percentile of the sample distribution):
  - Countries dropped: Bangladesh, Brazil, China, India, Indonesia, Nigeria and Pakistan.
  - General findings are confirmed.

### Additional covariates
- Augmenting the baseline model with additional regressors (Table 7):
  - Controlling for foreign aid (aid per capita) yields that aid and remittances are substitutes (similar to Amuedo-Dorantes et al. (2007)).
  - Inclusion of aid reduces the magnitude of the coefficient on CY_CLE_j compared to the baseline (Table 3, column 5), suggesting aid’s role as a shock absorber weakens the counter-cyclical pattern of remittances (Presbitero, 2013).
- Adding measures of macroeconomic stability (columns 2 to 4):
  - Remittances are larger in countries with better macroeconomic and institutional conditions.
  - Fiscal balance (as a ratio of GDP): coefficient is statistically significant and positive.
  - Ratio of external debt over GDP: coefficient is statistically significant and negative.
  - The elasticity of remittances to CY_CLE_j is almost unaffected by these additions.
  - Constraints on the executive: positive coefficient indicates remittances are larger for countries with stronger institutions (Singh et al., 2011).
  - Reduction in the coefficient on CY_CLE_j mainly reflects significant reduction in sample size, rather than the impact of controlling for institutional quality.
- Jointly controlling for these covariates (column 5) still yields a negative effect of the cyclical component of output in recipient countries on remittances.

### Different estimator
- Use of Fixed Effects Poisson estimator to account for panel dimension (Tables 8-10):
  - Variables that vary only across province-country pairs and not over time (e.g., D_i,j and ΔMIGR_i,j) cannot be identified.
  - Main results from baseline and augmented specifications are largely confirmed.
  - The negative coefficient on CY_CLE_j remains evident even when including province-country pair fixed effects.

### Implications within robustness exercises
- The counter-cyclical behavior of remittances with respect to recipient-country output (negative and significant coefficient on CY_CLE_j) is robust to:
  - Inclusion of China in the sample.
  - Splitting the sample by recipient income level.
  - Excluding small-migrant communities (MIGR_ijt <100, robust up to MIGR_ijt <500).
  - Dropping largest recipient countries (population above the 95th percentile).
  - Adding controls for aid, fiscal balance (ratio of GDP), external debt (ratio of GDP), and institutional constraints.
  - Alternative estimation via Fixed Effects Poisson with province-country pair fixed effects.

*Source: _wp1413 - 5.4  Robustness checks*

### References

### _wp1413 - References

### Major themes in the cited literature
- Determinants of international remittances, including altruism, insurance, migrant characteristics, and household composition (Adams Jr., 2009; Agarwal and Horowitz, 2002; de la Briere et al., 2002; Lucas and Stark, 1985).
- Macroeconomic impacts of remittances: effects on output volatility, growth, poverty, and business cycle transmission (Amuedo-Dorantes et al., 2007; Bugamelli and Paternò, 2011; Chami et al., 2012; Giuliano and Ruiz-Arranz, 2009; Gupta et al., 2009).
- Remittances as resilience or insurance against shocks: natural disasters, food price shocks, and external shocks (Combes et al., 2014; Mohapatra et al., 2012; Naudé and Bezuidenhout, 2012; Noy, 2009; Yang, 2008).
- Financial development, transaction costs, formality, and the banking channel for remittances (Albareto and Mistrulli, 2011; Freund and Spatafora, 2008; Silva and Tenreyro, 2006).
- Measurement and estimation issues: gravity models, endogeneity, sample selection, instruments, and estimation methods (Anderson, 2011; Bettin et al., 2012; Lueth and Ruiz-Arranz, 2008; Roodman, 2009; Silva and Tenreyro, 2006).
- Conflict and political economy factors influencing remittance flows (Themnér and Wallensteen, 2013; Rodrik, 2006; Docquier et al., 2012).

### Key data constructs and summary statistics (from Table 1)
- REM_ijt: Total official remittances at constant prices from province i to country j in year t (Source: Bank of Italy). Mean: 652969.4. St. Dev.: 1.14E+07.
- CYCLE_i,t: Logarithm of actual GDP over potential GDP in province i in year t (potential GDP via H-P filter). Source: ISTAT and Istituto Tagliacarne. Mean: 0.000. St. Dev.: 0.029.
- CYCLE_j,t: Logarithm of actual GDP over potential GDP in country j in year t (World Economic Outlook (IMF)). Mean: 0.001. St. Dev.: 0.031.
- TREND_i,t: Logarithm of potential GDP in province i in year t. Mean: 9.994. St. Dev.: 0.251.
- TREND_j,t: Logarithm of potential GDP in country j in year t. Mean: 10.837. St. Dev.: 2.493.
- MIGR_ijt: Logarithm of the stock of migrants living in province i and coming from country j in year t (ISTAT). Mean: 3.220. St. Dev.: 2.096.
- ΔMIGR_ij: Growth rate of the migrant stock MIG_ij over 2003-2011 (ISTAT). Mean: 0.037. St. Dev.: 0.107.
- WAR_j,t: Indicator = 1 if country j experienced armed conflicts in year t (UCDP/PRIO Armed Conflict Dataset). Mean: 0.170. St. Dev.: 0.376.
- DIS_j,t: Indicator = 1 if country j experienced natural disasters in year t (EM-DAT, CRED). Mean: 0.742. St. Dev.: 0.437.
- TT_j,t: Indicator = 1 if country j experienced a large negative terms-of-trade shock (lowest 5 percent of annual variation in the terms-of-trade index). Mean: 0.048. St. Dev.: 0.214.
- DIST_ij: Logarithm of kilometric distance between province i and country j. Mean: 8.415. St. Dev.: 0.764.
- FINDEV_j,t: Logarithm of the ratio of domestic credit to the private sector over GDP in country j in year t (World Development Indicators). Mean: 3.257. St. Dev.: 0.798.
- BANK_i,t: Logarithm of the number of bank branches per 10,000 inhabitants in province i in year t. Mean: 1.744. St. Dev.: 0.370.
- FD_i,t: Logarithm of the ratio of the number of branches in province i weighted by log(1 + distance) to total branches in province i in year t (Bank of Italy). Mean: 1.176. St. Dev.: 0.341.
- AID_j,t: Logarithm of official aid per capita received in country j in year t. Mean: 3.296. St. Dev.: 1.386.
- FISC BAL_j,t: Fiscal balance as a share of GDP in country j in year t. Mean: -0.015. St. Dev.: 0.043.
- EXT DEBT_j,t: External debt stocks as a share of GDP in country j in year t. Mean: 0.465. St. Dev.: 0.591.
- EXEC CONST_j,t: Constraint on the executives’ index (Polity IV; 1 = unlimited authority; 7 = Executive parity or subordination). Mean: 4.733. St. Dev.: 1.904.

### Countries covered (Table 2)
- Country sample includes: Afghanistan, Albania, Algeria, Angola, Argentina, Armenia, Azerbaijan, Bangladesh, Belarus, Benin, Bolivia, Bosnia and Herzegovina, Brazil, Bulgaria, Burkina Faso, Burundi, Cambodia, Cameroon, Cape Verde, Central African Rep., Chad, Chile, China, Colombia, Congo, Congo, Dem. Rep., Costa Rica, Cote d’Ivoire, Cuba, Dominica, Dominican Rep., Ecuador, Egypt, El Salvador, Eritrea, Ethiopia, Gabon, Gambia, Georgia, Ghana, Guatemala, Guinea, Guinea-Bissau, Haiti, Honduras, India, Indonesia, Iran, Iraq, Jamaica, Jordan, Kazakhstan, Kenya, Korea, North, Kyrgyzstan, Laos, Lebanon, Liberia, Libya, Lithuania, Macedonia, Madagascar, Malaysia, Mali, Mauritania, Mauritius, Mexico, Moldova, Mongolia, Morocco, Mozambique, Myanmar, Nepal, Nicaragua, Niger, Nigeria, Pakistan, Panama, Paraguay, Peru, Philippines, Romania, Russia, Rwanda, Senegal, Seychelles, Sierra Leone, Somalia, South Africa, Sri Lanka, Sudan, Syria, Tanzania, Thailand, Togo, Tunisia, Turkey, Turkmenistan, Uganda, Ukraine, Uruguay, Uzbekistan, Venezuela, Vietnam, Yemen, Zambia, Zimbabwe.

### Selected empirical results and robustness (regression tables)
- Table 3: Baseline results (Poisson Pseudo Maximum Likelihood estimator)
  - CYCLE_i,t coefficients (columns 1–5): 1.122, 1.258, 1.329, (column 4 blank in source), (column 5 blank in source). [Brackets show clustered robust standard errors: [0.822], [0.830], [0.830]]
  - CYCLE_j,t coefficients (columns 1–3 shown): -3.510***, -3.491***, -4.699***. [Standard errors: [0.639], [0.649], [0.615]]
  - MIGR_i,j,t coefficients (columns 1–5): 0.805***, 0.813***, 0.814***, 0.801***, 0.803***. [SEs: [0.035], [0.037], [0.037], [0.036], [0.036]]
  - POP_j,t coefficients (columns 1–4): 8.648***, 9.373***, 9.974***, 10.719***. [SEs: [1.193], [1.304], [1.361], [1.328]]
  - Observations: 57,248; 48,608; 46,760; 38,994; 38,994.
  - R2: 0.741; 0.763; 0.771; 0.795; 0.796.
  - Notes: Dependent variable REM_i,j,t; constant and province (i), country (j) and year (t) dummies included.

- Table 4: Extended specification — factors of vulnerability
  - CYCLE_i,t (cols 1–4): 1.316, 1.329, 1.391, 1.359. [SEs: [0.831], [0.830], [0.940], [0.946]]
  - CYCLE_j,t (cols 1–4): -4.680***, -4.854***, -2.266***, -2.373***. [SEs: [0.616], [0.600], [0.667], [0.671]]
  - TREND_j,t (cols 1–4): 3.194***, 3.247***, 1.721***, 1.824***. [SEs: [0.534], [0.536], [0.522], [0.531]]
  - DIS_j,t (cols 3–4): 0.045**, 0.046**. [SEs: [0.019], [0.018]]
  - TT_j,t (cols 3–4): 0.094***, 0.105***. [SEs: [0.033], [0.035]]
  - Observations: 38,994; 36,764; 37,414; 35,555.
  - R2: 0.795; 0.796; 0.895; 0.896.

- Table 5: Extended specification — the role of financial development
  - FINDEV_j,t coefficients (cols 1–2): -0.920***, -0.936***. [SEs: [0.138], [0.132]]
  - BANK_i,t coefficients (cols 3–4): 1.047**, 1.180***. [SEs: [0.429], [0.425]]
  - FD_i,t coefficients (cols 3–4): -0.389**, -0.454***. [SEs: [0.195], [0.161]]
  - CYCLE_i,t (cols 1–4): 1.725**, 1.491*, 0.680, 1.153*. [SEs: [0.725], [0.878], [0.643], [0.600]]
  - Observations: 37,810; 38,994; 38,994; 37,810.
  - R2: 0.807; 0.795; 0.798; 0.810.

- Table 6–10: Robustness checks
  - Table 6: Robustness — sample definition. Observations and R2 vary by sample:
    - Including China: Observations 39,610. R2 0.711.
    - Low Income Countries: Observations 8,919. R2 0.959.
    - Middle Income Countries: Observations 30,943. R2 0.717.
    - Large migrant communities: Observations 11,069. R2 0.808.
    - No large recipients: Observations 36,439. R2 0.797.
    - CYCLE_j,t coefficients across columns include -3.055***, -2.182**, -3.205***, -5.370***, -4.975***. [Associated SEs reported in table.]
  - Table 7: Robustness — additional covariates (AID_j,t, FISC BAL_j,t, EXT DEBT_j,t, EXEC CONST_j,t included)
    - AID_j,t coefficients (cols 1–2): -0.115***, -0.243***. [SEs: [0.012], [0.041]]
    - FISC BAL_j,t (cols 1–2): 4.612***, 1.633**. [SEs: [0.622], [0.784]]
    - EXEC CONST_j,t (cols 4–5): 0.115***, 0.252***. [SEs: [0.028], [0.039]]
    - Observations vary: 35,938; 26,270; 38,079; 17,530; 10,951. R2 up to 0.911.
  - Tables 8–10: Estimation method robustness using Fixed effects Poisson estimator
    - Table 8 (baseline): CYCLE_i,t (cols shown) 1.418*, 1.522**, 1.815**. Observations: 55,140; 46,286; 45,381; 38,549; 38,549. Number of pairs: 8,457; 8,199; 8,044; 6,512; 6,512.
    - Table 9 (factors of vulnerability): CYCLE_i,t 1.789**, 1.800**, 1.707*, 1.659*. DIS_j,t and TT_j,t remain relevant: DIS_j,t reported 0.022 and 0.038** in columns; TT_j,t 0.090*** and 0.102***. Observations: 38,549; 36,346; 37,019; 35,173. Number of pairs: 6,512; 6,139; 6,253; 5,940.
    - Table 10 (financial development): FINDEV_j,t -0.745***, -0.752***. FD_i,t -0.402**, -0.427***. Observations: 37,305; 38,549; 38,549; 37,305. Number of pairs: 6,364; 6,512; 6,512; 6,364.

*Source: _wp1413 - References (tables and references as provided in the source PDF).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1413.pdf_