## wp18232 - conclusions are ours and not those for Eurostat, The European Commission, or any of the national authorities

## Source details

**Canonical URL:** [wp18232 - conclusions are ours and not those for Eurostat, The European Commission, or any of the national authorities](https://www.imf.org/-/media/files/publications/wp/2018/wp18232.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp18232.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp18232.pdf.json)

---

### Scope, authorship, and objectives
- Conclusions are the authors' and not those of Eurostat, The European Commission, or any of the national authorities whose data have been used.
- Acknowledged contributors: Arnout Baeyens, Craig Beaumont, Helge Berger, Dilyana Dimova, Florence Jaumotte, Davide Malacrino, Celine Piton, Anna Raggl, Antonio Spilimbergo, Petia Topalova.
- Seminar and comment venues: Swedish Ministry of Finance; IMF’s European Department; 2017 World Bank-IMF Annual Meetings’ Analytical Corner; Offices of Executive Directors for Austria, Belgium, and the United Kingdom at the IMF.
- Paper objectives:
  - Measure how successfully migrants integrate into host labor markets.
  - Compare integration speed across Europe.
  - Identify factors affecting integration, with focus on initial labor market conditions and education.

### Data, sample, and empirical framework
- Data source:
  - European Union Labor Force Survey (EU LFS) micro dataset, pooled 1998–2016; over ten million individuals in 130 NUTS 2 regions.
- Coverage and sample restrictions:
  - 13 focus countries: Austria, Belgium, Denmark, Finland, France, Ireland, Italy, the Netherlands, Norway, Portugal, Spain, Sweden, and the United Kingdom.
  - Sample restricted to individuals aged 25–64.
  - 2016 snapshot: close to 590,000 persons; share of migrants averages 15 percent (range 7 percent in Finland to 21 percent in Belgium).
  - Migrant origin shares (2016 average): 50 percent EUR-NA, 25 percent MENA, 14 percent Asia, remainder LAT-AM.
- Limitations:
  - Repeated cross-sections (cannot track individuals over time).
  - No wages/incomes, no refugee status, no language skills in LFS.
- Empirical model:
  - Probit: yi = β0 + β1 M + β2 YSM + β3 YSM^2 + β4 EDUC + β5 X + ε, where Pr(Empi = 1) = Φ(x′β).
  - M ∈ {0,1} migrant indicator; YSM = years since migration (zero for natives); EDUC = years of education; X includes age, age squared, marital status, and country × survey year and region fixed effects.
  - β1 measures initial employment gap; β2 captures integration speed; YSM^2 allows nonlinear convergence.
  - Errors clustered at sub-national regional level; separate regressions by migrant origin and gender.

### Baseline integration findings
- Main stylized results (Table 5; baseline probit):
  - Migrant status coefficients:
    - Female: -0.663 [0.062]***
    - Male: -0.608 [0.038]*** (Table excerpts report male baseline values analogous)
  - Years since migration:
    - Female: 0.027 [0.003]***
    - Male: 0.017 [0.002]***
  - Years since migration squared/100:
    - Female: -0.032 [0.004]***
    - Male: -0.014 [0.003]***
  - Education:
    - Female: 0.104 [0.003]***
    - Male: 0.074 [0.002]***
  - Age and marital status:
    - Age and age squared significant with nonlinear effects.
    - Being married: lowers employment probability for women (e.g., -0.076 [0.014]***), increases for men (e.g., 0.481 [0.007]***).
  - Observations:
    - Female: 4,993,587
    - Male: 4,698,090
  - Pseudo R-squared:
    - Female: 0.148
    - Male: 0.163
- Simulated employment profiles (Figure 1 highlights):
  - Example: 30-year old female MENA migrant with average education has on average only a 20 percent predicted employment probability versus over 70 percent for a comparable female native.
  - Gap remains substantial — more than 20 percentage points after 20 years of residence for some groups.
  - Migrants from Asia experience faster catch-up; full convergence not achieved after 20 years for most groups.
- Cross-country heterogeneity (Figures 2–3, Figure A.1):
  - Sweden and France exhibit particularly large conditional gaps for both female (~50 percent) and male (~close to 30 percent) migrants but relatively fast integration speed.
  - In Sweden, a male (female) migrant can be expected to improve employment probability by roughly one (two) percentage points with each year of residence — translating to ~20 (40) percentage points over 20 years.
  - Ireland and Austria noted as having large initial gaps but fast catch-up for MENA migrants.

### Cohort effects
- Concern: repeated cross-section bias if cohort quality varies over time (Borjas 1985, 1995, 2015).
- Test: include arrival-decade dummies (1940s–2010s) in lieu of single migrant dummy.
- Findings (Table A.1):
  - Years since migration (YSM) coefficient increases when cohort dummies are included — baseline likely underestimates catch-up speed.
  - Increase more pronounced for women than men.
  - Cohort dummies show large negative coefficients for more recent arrival cohorts (e.g., 2010s cohort: -0.958 [0.068]***; with cohort effects: -0.765 [0.029]***), indicating heterogeneity by arrival cohort.

### Network (co-origin community) effects
- Proxies used:
  - Eurostat “Stock of Migrants” and “Share of Migrants”; UN International Migrant Stock database for broader coverage.
  - Analysis restricted to 2010–16 for some measures.
- Findings (Table 6):
  - Eurostat-based: higher stock of same-origin migrants raises employment probability; “Share of Migrants” often insignificant.
  - UN-based: both stock and share positive and significant for male migrants; insignificant for female migrants.
  - Network effects generally positive across origin regions, with exceptions: insignificant though positive coefficients for female migrants from MENA and ASIA, and for male migrants from LAT-AM.

### Initial labor market conditions at arrival
- Empirical approach:
  - Interact years since migration with labor market condition at arrival measured by estimated unemployment gap (HP-filtered; HP unemployment gap available for 1995–2017). Robustness checks use OECD unemployment gap, output gap, and labor force participation gap.
  - One-year lag of the arrival-year measure used.
- Main findings (Figures 4–5; Appendix Figure A.2):
  - Poor labor market conditions upon arrival significantly reduce subsequent employment probabilities, especially for female migrants.
  - Female migrants:
    - Marginal effect of residence years on employment probability falls from 1.7 percentage points per year for an unemployment gap of -8 percent to 0.3 percentage points per year for an unemployment gap of +8 percent.
    - This difference can translate into more than 4 percentage-points difference in employment probability after 3 years of residence.
  - Male migrants:
    - Difference between -8 and +8 percent unemployment gaps is about 0.3 percentage points per year, resulting in close to one percentage point increase in employment probability after 3 years.
  - Heterogeneity by origin:
    - Migrants from Asia are most affected by initial labor market conditions—effect almost doubles that for the average migrant.
    - Migrants from EUR-NA least affected; effect indistinguishable from zero for male migrants from EUR-NA.

### Foreign versus domestic education
- Conceptual decomposition:
  - Total education years = Ef (foreign education years) + Ed (domestic education years).
  - Years since migration decomposed into domestic education, domestic working experience (proxied by age controls), and a constant k (k = max(0, 6 − Age at arrival + Years since migration); k = 0 for majority arriving after school-entering age six).
  - Resulting estimating equation allows separate returns to foreign and domestic education and differential returns for natives and migrants.
- Data and imputation:
  - Impute foreign and domestic education years from timing of migration and year highest educational level achieved, assuming continuous education.
- Empirical results (Table 7; Figure 6):
  - Equality of returns between foreign and domestic education is rejected—foreign education matters less for employment outcomes than domestic education (columns 2 and 5).
  - Returns to domestic education are significantly lower for migrants than for natives (negative coefficient on Domestic education × Migrant status; columns 3 and 6).
  - For women, the return differential is especially pronounced at higher education levels; for men, differential broadly similar across education levels.
  - Allowing differential returns by source and migrant status:
    - Fully explains the initial native-migrant employment gap for women—the coefficient on migrant status is no longer significant in column 3 (Table 7).
    - For men, the native-migrant gap is significantly narrowed—the coefficient on migrant status becomes significantly smaller going from column 4 to column 6 (Table 7).
- Caveats:
  - LFS lacks language-skill measures; omitted language ability may drive part of the observed differential returns.
  - Subsample analysis of immigrants from countries sharing host-country language (France and the United Kingdom considered suitable) still finds negative Domestic education × Migrant status (results not reported).

### Quality of employment
- Re-estimation using probability of having a full-time & permanent job (Table A.2):
  - Being immigrant:
    - Female: -0.380 [0.046]***
    - Male: -0.530 [0.046]***
  - Years since migration:
    - Female: 0.019 [0.005]***
    - Male: 0.015 [0.003]***
  - Years since migration squared/100:
    - Female: -0.023 [0.007]***
    - Male: -0.020 [0.005]***
  - Education:
    - Female: 0.084 [0.004]***
    - Male: 0.045 [0.005]***
  - Marital status effects on job quality:
    - Female being married: -0.234 [0.021]***
    - Male being married: 0.297 [0.005]***
  - Observations and fit:
    - Female: 4,993,587; Pseudo R-squared 0.0988
    - Male: 4,698,090; Pseudo R-squared 0.0826
- Interpretation:
  - Migrants less likely to hold full-time and permanent jobs on arrival, with limited convergence over time; education raises quality of employment but gaps remain.

### Other selected empirical statistics (summary)
- Summary statistics (Table 4):
  - Pr(Employment): Obs. 10,044,922; Mean 0.72
  - Migrant status: Obs. 10,044,922; Mean 0.11
  - Years since migration (migrants only; zero for natives): Obs. 1,033,342; Mean 16.0; Std. Dev. 12.90; Min 0; Max 67
  - Education years: Obs. 9,773,923; Mean 11.7; Std. Dev. 3.36; Min 2; Max 24
  - Age: Obs. 10,044,922; Mean 44.6; Std. Dev. 11.12; Min 6; Max 62
  - Married: Obs. 10,044,922; Mean 0.62

### Conclusions and policy implications
- Overall pattern:
  - Migrants in Europe tend to catch up with comparable natives in employment as they accumulate country-specific skills, but convergence is lengthy and full convergence occurs for only a few source countries.
  - Integration speed varies significantly across host countries and migrant origin groups.
  - Female migrants:
    - Tend to have larger initial conditional employment gaps but often catch up faster than men.
  - Migrants from the MENA region:
    - Integration is often significantly slower; MENA includes many who moved to Europe as refugees.
  - Recent migrants assimilate faster than earlier cohorts.
  - Informal social networks likely important for male migrants’ labor market outcomes.
  - Favorable host-country labor market conditions at arrival smooth transition—especially for female migrants.
  - Foreign-based education pays off less than host-country education; returns to domestic education are lower for migrants than for natives, pointing to non-skill barriers.
- Policy recommendations (as stated):
  - Target government support to “vulnerable” groups: women and those from the MENA region.
  - Combine integration policy with macroeconomic and/or labor market support where needed (a booming economy tends to “lift all boats”).
  - Improve transferability of foreign qualifications and put foreign-based education on a more equal footing with domestic training where possible.
  - Enhance returns to domestic education for migrants: offer adequate language training early on; remove other non-skill barriers in the labor market.
  - Design integration policy to be country-specific, considering political constraints, stakeholders’ interests, and linkages to education, housing, labor market, and financial inclusion.

*IMF Working Paper wp18232*

### conclusions are ours and not those for Eurostat, The European Commission, or any of the national authorities

### wp18232 - conclusions are ours and not those for Eurostat, The European Commission, or any of the national authorities

### Scope and authorship
- Conclusions are the authors' and not those of Eurostat, The European Commission, or any of the national authorities whose data have been used.
- Acknowledged contributors: Arnout Baeyens, Craig Beaumont, Helge Berger, Dilyana Dimova, Florence Jaumotte, Davide Malacrino, Celine Piton, Anna Raggl, Antonio Spilimbergo, Petia Topalova.
- Seminar and comment venues listed: Swedish Ministry of Finance; IMF’s European Department; 2017 World Bank-IMF Annual Meetings’ Analytical Corner; Offices of Executive Directors for Austria, Belgium, and the United Kingdom at the IMF.
- Standard IMF Working Papers disclaimer: views are those of the author(s) and do not necessarily represent the views of the IMF, its Executive Board, or IMF management.

### Document structure (table of contents with page references)
- I. INTRODUCTION — page 4
- II. ANALYSIS OF INTEGRATION SPEED: EMPIRICAL FRAMEWORK — page 5
- III. DATA AND SAMPLE — page 6
- IV. RESULTS ON INTEGRATION SPEED — page 8
- V. THE ROLE OF FOREIGN AND DOMESTIC EDUCATION — page 13
- VI. CONCLUDING REMARKS — page 15
- REFERENCES — page 17

### Empirical outputs and visual materials (figures)
- Figure 1. Simulated Employment Probabilities of a 30-Year Old Migrant with Average Education, by Gender and Origin — page 18
- Figure 2. Estimates of Conditional Employment Gaps and Integration Speed for All Migrants, By Country — page 19
- Figure 3. Estimates of Conditional Employment Gaps and Integration Speed for MENA Migrants, By Country — page 20
- Figure 4. Effects of Initial Labor Market Conditions — page 20
- Figure 5. Effects of Initial Labor Market Conditions, by Migrant Origin — page 22
- Figure 6. Effects of Domestic Education for Natives vs. Migrants — page 23

### Empirical outputs and tabular materials (tables)
- Table 1. Distribution by Country of Birth, EU LFS 2016 Sample — page 24
- Table 2. Distribution by Demographics and Education, EU LFS 2016 Sample — page 24
- Table 3. Distribution by Labor Market Outcome, EU LFS 2016 Sample — page 25
- Table 4. Summary Statistics of Variables Entering Regression Analysis — page 25
- Table 5. Baseline Results from Probit Regression — page 26
- Table 6. The Effects of Networks of Migrants — page 27
- Table 7. The Impact of Foreign and Domestic Education — page 28

### Key themes emphasized by the document organization
- Integration speed: empirical framework and results (Sections II and IV).
- Role of education: distinction between foreign and domestic education (Section V and Figure 6; Table 7).
- Importance of initial labor market conditions and migrant origin in shaping outcomes (Figures 4 and 5).
- Use of EU LFS 2016 sample for distributions and regression analysis (Tables 1–4).

*IMF Working Paper wp18232*

### APPENDIX TABLES AND FIGURES ________________________________________29

### APPENDIX TABLES AND FIGURES ________________________________________29

### I. INTRODUCTION
- 2015: a record 1.3 million people sought asylum in European Union (EU) states—the largest annual flow into the continent in 30 years.
- Integration of refugees and other migrants into host labor markets is a public policy challenge; slow progress may fuel political divides in the EU.
- Demographic pressures (rapidly ageing population, low fertility rates) motivate better tapping migrants to stem projected labor force decline.
- Economic benefits of faster integration: higher growth and productivity and lower budgetary costs (Aiyar and others 2016; Jaumotte, Koloskova, and Saxena 2016).
- Paper objectives:
  - Measure how successfully migrants integrate into host labor markets.
  - Compare integration speed across Europe.
  - Identify factors affecting integration, with focus on initial labor market conditions and education.
- Dataset and scope:
  - European Union Labor Force Survey (EU LFS) micro dataset.
  - Covers over ten million individuals surveyed between 1998 and 2016 in 13 European countries.
- Limitations noted:
  - Repeated cross-sections (cannot track individuals over time) — integration estimate is an approximation and may be biased by changing cohort quality.
  - No wages/incomes in dataset — labor market integration proxied by employment outcome only.
  - Refugee status not identified; special focus on MENA region but results do not necessarily extend to refugees.

*Key organization: empirical framework (Section II), data and estimation sample (Section III), integration speed results (Section IV), role of education (Section V), conclusion (Section VI).*

### II. ANALYSIS OF INTEGRATION SPEED: EMPIRICAL FRAMEWORK
- Model:
  - Probit model where employment indicator Empi = 1 if latent yi > 0.
  - yi = x′β + ε, ε ~ N(0,1), Pr(Empi = 1) = Φ(x′β).
- Core specification (equation (1)):
  - yi = β0 + β1 M + β2 YSM + β3 YSM^2 + β4 EDUC + β5 X + ε
    - M ∈ {0,1} indicator for being a migrant.
    - YSM = years since migration (equal to zero for natives).
    - EDUC = years of education.
    - X includes age, age squared, marital status, and full set of fixed effects (country × survey year and region).
- Interpretation:
  - β1 (migrant status) measures initial employment gap upon arrival relative to comparable natives.
  - β2 (years since migration) captures rate at which the migrant-native employment gap narrows over time (integration speed).
  - YSM^2 allows for nonlinear convergence.
- Identification features:
  - Country-year fixed effects absorb time-varying host-country factors (macroeconomic/labor market conditions, reforms).
  - Region fixed effects capture sub-national unobserved heterogeneity.
  - Errors clustered at sub-national regional level.
- Heterogeneity:
  - Separate regressions run by migrant origin and by gender to explore differential integration profiles.

### III. DATA AND SAMPLE
- EU LFS:
  - Covers 28 EU member states plus Iceland, Norway and Switzerland from 1983 onwards; harmonized by Eurostat.
  - 2016 quarterly EU LFS sample size across the EU was about 1.5 million individuals.
  - Rich individual and household information: demographics, labor market status, employment characteristics, education.
  - For migrants (foreign-born): country of birth/nationality (aggregated to about 15 country groups) and length of residence available.
- Limitations reiterated:
  - Public microdata are repeated cross-sections (household IDs randomized) — cannot track respondents over time.
  - Missing variables: wages/incomes, migrant type (asylum vs economic), language skill.
- Estimation sample:
  - Pool data from 19 survey waves in 1998–2016.
  - Focus on 13 countries: Austria, Belgium, Denmark, Finland, France, Ireland, Italy, the Netherlands, Norway, Portugal, Spain, Sweden, and the United Kingdom.
    - Note: Data for Germany were not made available for this research. Data for Ireland and Italy are available as of 2010.
  - Sample restricted to individuals aged 25–64.
  - Final sample: over ten million individuals in 130 NUTS 2 regions.
- 2016 snapshot (Table 1 summary):
  - Close to 590,000 persons surveyed in 2016.
  - Share of migrants (foreign-born) averages 15 percent, ranging from 7 percent in Finland to 21 percent in Belgium.
  - On average: 50 percent of migrants from EUR-NA (Europe, North America, Australia, and Oceania), 25 percent from MENA, 14 percent from Asia, remainder from LAT-AM.
  - Countries with large shares of MENA migrants include Belgium, France, Portugal, and Sweden.
- Summary statistics (Table 4, all years 1998–2016):
  - About 10.04 million respondents.
  - 72 percent employed (rest unemployed or inactive).
  - Average education: slightly less than 12 years.
  - Average age: almost 45 years.
  - Probability of being married: 62 percent.
  - Migrants account for 11 percent of individuals.
  - Average length of stay among migrants: about 16 years.

### IV. RESULTS ON INTEGRATION SPEED
#### Baseline results
- Main findings from baseline probit regressions (Table 5):
  - Coefficient on migrant status (M) is negative and statistically significant.
    - Magnitude larger for females than males — larger initial employment gap for migrant women.
  - Coefficient on years since migration (YSM) is positive and significant; YSM^2 coefficient is negative — employment probability improves with each year of residence but with diminishing returns.
  - Education increases probability of employment for both sexes.
  - Age increases employment probability with nonlinear effect (age squared).
  - Marital status has differential effects: being married lowers employment probability for women, increases it for men.
- Simulated employment profiles (Figure 1):
  - Simulation: 30-year old migrant with average education upon arrival relative to comparable native.
  - Heterogeneity by origin and gender:
    - MENA migrants experience the largest employment gaps upon arrival, especially females.
      - Example: a 30-year old female MENA migrant with average education has on average only a 20 percent predicted employment probability versus over 70 percent for a comparable female native.
      - Gap remains substantial — more than 20 percentage points after 20 years of residence.
    - Migrants from Asia experience particularly fast catch-up; full convergence not achieved even after 20 years for most groups.
- Robustness to sample period:
  - Repeating analysis for 2010–16 (seven most recent waves) yields qualitatively unchanged results: migrants less likely to have a job on arrival, larger gap for women, gaps narrow over time.
- Cross-country heterogeneity (Figure 2):
  - Country-specific probit regressions (1998–2016) with year fixed effects show variation in:
    - Conditional native-migrant employment gaps upon arrival (panel A).
    - Average migrant integration speed (panel B).
  - Examples:
    - Sweden and France have particularly large conditional gaps for both female (~50 percent) and male (~close to 30 percent) migrants, but relatively fast integration speed.
    - In Sweden, a male (female) migrant can be expected to improve employment probability by roughly one (two) percentage points with each year of residence — translating to ~20 (40) percentage points over 20 years.
  - Composition effects addressed by focusing on MENA migrants (Figure 3):
    - Ireland and Austria (in addition to Sweden) stand out as countries with relatively large employment gaps but fast catch-up for MENA migrants.
  - Country-level employment path simulations presented in Appendix Figure A.1.

#### Cohort effects
- Concern: repeated cross-section estimates may be biased if migrant cohort quality varies over time (Borjas 1985, 1995, 2015).
  - If earlier cohorts have higher employment potential than recent ones, YSM coefficient biased upward (overstates catch-up); opposite if cohort quality improves.
- Test implemented:
  - Replace single migrant dummy M with dummies for arrival decades (1940s, 1950s, ..., 2000s, 2010s) — eight dummies; natives are excluded group.
  - Results (Table A.1):
    - Coefficient on years since migration (YSM) increases when cohort dummies are included — suggests baseline underestimates catch-up speed.
    - Increase more pronounced for women than men.
    - Interpretation: migrant “quality” may have improved over time or more recent immigrants assimilate faster than earlier ones.

#### Network effects
- Hypothesis: larger co-origin migrant communities facilitate job finding via informal networks.
- Data and proxies:
  - Eurostat data on region of birth of migrants in recipient countries (available for more recent period); complemented by UN International Migrant Stock database (data available for 2010, 2015, 2017; interpolation used).
  - Network proxies: “Stock of Migrants” (stock of migrants born in same region as respondent) and “Share of Migrants” (their share in total migrants).
  - Analysis restricted to 2010–16 for Eurostat coverage; UN dataset used for broader coverage.
- Findings (Table 6):
  - Eurostat-based results: higher stock of same-origin migrants raises employment probability, but “Share of Migrants” often insignificant.
  - UN-based results: both stock and share coefficients positive and significant for male migrants; insignificant for female migrants.
    - Suggests larger existing communities of same origin raise employment probability for male migrants but not for female migrants.
  - Further heterogeneity:
    - Running regressions by origin region (EUR-NA, MENA, ASIA, LAT-AM) with “Stock of Migrants” as proxy shows positive and significant coefficients across most specifications, indicating network effects are present.
    - Exceptions: insignificant though positive coefficients for female migrants from MENA and ASIA, and for male migrants from LAT-AM.

#### Quality of employment (introductory note)
- Concern: migrants more likely to take part-time or temporary jobs with lower pay/benefits.
- Authors re-estimate baseline specification using probability of having a full-time and permanent job (analysis continues beyond provided excerpt).

*Final source attribution: the content above is drawn from the paper’s appendix section and main text provided in the source PDF.*

### 2016. Results—reported in Table A.2 in the appendix—show that labor market integration is

### wp18232 - 2016. Results—reported in Table A.2 in the appendix—show that labor market integration is

### Initial labor market conditions
- Question explored: Do cyclical conditions in the host country around migrants’ arrival shape integration outcomes?
- Empirical approach:
  - Add interaction between years since migration and labor market conditions upon arrival to equation (1).
  - Labor market condition measure: estimated economy-wide unemployment gap in the year immediately after arrival, where the natural unemployment rate is proxied using the Hodrick-Prescott (HP) filter measure (HP unemployment gap available for 1995–2017).
  - Robustness checks: alternative unemployment gap from OECD Economic Outlook, output gap, and labor force participation gap (OECD measures available for 1985–2017).
  - One-year lag of the arrival-year measure is used to mitigate demand-pull selection bias.
- Main empirical findings (full sample 1998–2016):
  - Poor labor market conditions upon arrival significantly reduce subsequent probabilities of employment, especially for female migrants (Figure 4).
  - For female migrants:
    - Marginal effect of residence years on probability of employment falls from 1.7 percentage points per year for an unemployment gap of -8 percent to 0.3 percentage points per year for an unemployment gap of +8 percent.
    - The -8 percent to +8 percent range is roughly the range of gap values observed in the data.
    - This difference is highly statistically significant (95-percent confidence intervals reported).
    - Economically, this can translate into more than 4 percentage-points difference in employment probability after 3 years of residence.
  - For male migrants:
    - The effect is less pronounced but still sizable: the corresponding difference between -8 and +8 percent unemployment gaps is about 0.3 percentage points per year, resulting in close to one percentage point increase in employment probability after 3 years of residence.
  - Alternative measures of initial conditions give similar qualitative results (Appendix Figure A.2).
- Heterogeneity by origin (Figure 5):
  - Migrants from Asia are most affected by initial labor market conditions—effect almost doubles that for the average migrant.
  - Migrants from Europe, North America, Australia, and Oceania are least affected; effect is indistinguishable from zero for male migrants from these regions.
  - Possible explanation: these migrants are more often highly qualified or work in jobs less influenced by economic cycles.

### The role of foreign and domestic education
- Conceptual decomposition:
  - Total education years = foreign education years (Ef) + domestic education years (Ed).
  - Years since migration decomposed into domestic education, domestic working experience (proxied by age in controls), and a constant term k (k = max(0, 6 − Age at arrival + Years since migration); k = 0 for majority arriving after school-entering age six).
  - Substitute into baseline equation and collect terms to obtain equation (4):
    - yi = α0 + α1 M + α2 E_f + α3 E_d + α4 E_d × M + α5 X + ε
  - Allows returns to education to differ by source (α2 ≠ α3) and allows returns to domestic education to differ for natives and migrants (α4 ≠ 0).
- Data and imputation:
  - EU LFS provides timing of migration and year highest educational level achieved → used to impute foreign and domestic education years assuming continuous education.
- Empirical results (Table 7):
  - Equality of returns between foreign and domestic education is rejected—foreign education matters less for employment outcomes than domestic education (columns 2 and 5).
  - Formal tests (not reported) show for both men and women the coefficient on foreign education is statistically significantly different from that on domestic education.
  - Unrestricted estimation rejects equality of returns to domestic education for natives and migrants—the returns are significantly lower for migrants, indicated by negative coefficient on Domestic education × Migrant status (columns 3 and 6).
  - Figure 6:
    - For women, the return differential is especially pronounced at higher levels of education.
    - For men, return differential is broadly similar across education levels.
  - Allowing for differential impact of human capital across sources and migrant status:
    - Fully explains the initial native-migrant employment gap for women—the coefficient on migrant status is no longer significant in column 3 (Table 7).
    - For men, the native-migrant gap is significantly narrowed—the magnitude of the coefficient on migrant status becomes significantly smaller going from column 4 to column 6 (Table 7).
- Caveats and robustness:
  - LFS does not provide information on migrants’ language skills; omission could influence results because language ability may drive differential returns.
  - Subsample analysis of immigrants from countries sharing host-country language (France and the United Kingdom considered suitable) likewise finds a negative coefficient on Domestic education × Migrant status (results not reported).

### Other empirical findings (selected)
- Baseline probit (Table 5):
  - Migrant status coefficients:
    - Female: -0.663 [0.062]***
    - Male: -0.608 [0.038]***
  - Years since migration:
    - Female: 0.027 [0.003]***
    - Male: 0.017 [0.002]***
  - Years since migration squared/100:
    - Female: -0.032 [0.004]***
    - Male: -0.014 [0.003]***
  - Education:
    - Female: 0.104 [0.003]***
    - Male: 0.074 [0.002]***
  - Observations:
    - Female: 4,993,587
    - Male: 4,698,090
  - Pseudo R-squared:
    - Female: 0.148
    - Male: 0.163
- Networks of migrants (Table 6 highlights):
  - Stock of Migrants coefficients positive and significant for some specifications (e.g., female: 0.002 [0.000]***; male: 0.002 [0.000]***).
  - Share of Migrants positive and significant in some male specifications (e.g., 0.159 [0.034]***).
  - Years since migration coefficients remain positive and significant across specifications.
  - Notes: Data not available for France, Ireland, Portugal, and the UK in some network measures.
- Summary statistics (Table 4):
  - Pr(Employment): Obs. 10,044,922; Mean 0.72
  - Migrant status: Obs. 10,044,922; Mean 0.11
  - Years since migration (calculated for migrants only, zero for natives): Obs. 1,033,342; Mean 16.0; Std. Dev. 12.90; Min 0; Max 67
  - Education years: Obs. 9,773,923; Mean 11.7; Std. Dev. 3.36; Min 2; Max 24
  - Age: Obs. 10,044,922; Mean 44.6; Std. Dev. 11.12; Min 6; Max 62
  - Married: Obs. 10,044,922; Mean 0.62

### Concluding remarks
- Overall pattern:
  - Migrants in Europe tend to catch up with comparable natives in labor market status as they spend time in the host country and accumulate country-specific skills, but convergence is lengthy and full convergence occurs for only a few source countries.
  - Integration speed varies significantly across host countries and migrant origin groups.
  - Female migrants:
    - Tend to be more disadvantaged initially (larger initial conditional employment gaps) but catch up faster than men.
  - Migrants from the MENA region:
    - Integration is often significantly slower; MENA is the origin region for most people who moved to Europe as refugees.
  - Recent migrants assimilate faster than previous cohorts.
  - Informal social networks likely important for male migrants’ labor market outcomes.
  - Favorable host-country labor market or macroeconomic conditions around arrival smooth migrants’ transition—especially for female migrants.
  - Foreign-based education pays off less than host-country education; returns to domestic education are lower for migrants than for natives, suggesting non-skill barriers.

### Policy implications (as stated)
- Target government support to “vulnerable” groups:
  - Women and those from the MENA region—groups with most disadvantaged initial conditions and largest scope for improvement.
- Combine integration policy with macroeconomic and/or labor market support where needed:
  - A booming economy tends to “lift all boats,” improving job prospects for natives and migrants alike.
- Improve transferability of foreign qualifications:
  - Put foreign-based education on a more equal footing with domestic training where possible.
- Enhance returns to domestic education for migrants:
  - Offer adequate language training early on.
  - Remove other non-skill barriers in the labor market.
- Design of integration policy should be country-specific:
  - Consider political constraints, stakeholders’ interests, and linkages to education, housing, labor market, and financial inclusion.

*International Monetary Fund working paper wp18232 (extract).*

### Appendix Tables and Figures

### Appendix Tables and Figures (wp18232)

### Cohort effects (Table A.1)
- Dependent variable: probability of being employed. Robust standard errors clustered at the regional level.
- Regression specifications: Baseline and With cohort effects; separate columns for Female and Male.
- Key coefficient estimates (with clustered standard errors in brackets):
  - Migrant status:
    - Female baseline: -0.663 [0.062]***
    - Female with cohort effects: -0.608 [0.038]***
    - Male baseline: -0.663 (implied same layout) — table presents male columns (3)(4) analogous; see full table.
  - Years since migration:
    - Female baseline: 0.027 [0.003]***
    - Female with cohort effects: 0.026 [0.003]***
    - Male baseline: 0.017 [0.002]***
    - Male with cohort effects: 0.017 [0.004]***
  - Years since migration squared/100:
    - Female baseline: -0.032 [0.004]***
    - Female with cohort effects: -0.042 [0.006]***
    - Male baseline: -0.014 [0.003]***
    - Male with cohort effects: -0.024 [0.006]***
  - Education:
    - Female baseline: 0.104 [0.003]***
    - Female with cohort effects: 0.105 [0.003]***
    - Male baseline: 0.074 [0.002]***
    - Male with cohort effects: 0.074 [0.002]***
  - Age:
    - Female baseline: 0.193 [0.005]***
    - Female with cohort effects: 0.193 [0.005]***
    - Male baseline: 0.226 [0.005]***
    - Male with cohort effects: 0.225 [0.005]***
  - Age squared:
    - Female baseline: -0.002 [0.000]***
    - Female with cohort effects: -0.002 [0.000]***
    - Male baseline: -0.003 [0.000]***
    - Male with cohort effects: -0.003 [0.000]***
  - Being married:
    - Female baseline: -0.076 [0.014]***
    - Female with cohort effects: -0.076 [0.014]***
    - Male baseline: 0.481 [0.007]***
    - Male with cohort effects: 0.481 [0.007]***
  - Cohort indicators (With cohort effects columns report cohort dummies):
    - 1940s cohort: -0.311 [0.155]** ; with cohort effects: -0.436 [0.057]***
    - 1950s cohort: -0.156 [0.089]* ; with cohort effects: -0.246 [0.069]***
    - 1960s cohort: -0.333 [0.069]*** ; with cohort effects: -0.373 [0.064]***
    - 1970s cohort: -0.547 [0.054]*** ; with cohort effects: -0.515 [0.067]***
    - 1980s cohort: -0.590 [0.038]*** ; with cohort effects: -0.556 [0.056]***
    - 1990s cohort: -0.604 [0.042]*** ; with cohort effects: -0.591 [0.052]***
    - 2000s cohort: -0.642 [0.065]*** ; with cohort effects: -0.590 [0.050]***
    - 2010s cohort: -0.958 [0.068]*** ; with cohort effects: -0.765 [0.029]***
- Fixed effects and sample:
  - Country-Year fixed effects: Y / Y Y (table shows Country-Year fixed effects indicated as Y/Y/Y)
  - Region fixed effects: Y / Y
  - Observations:
    - Female columns: 4,993,587
    - Male columns: 4,698,090
  - Pseudo R-squared:
    - Female baseline and with cohort effects: 0.148 and 0.148
    - Male baseline and with cohort effects: 0.163 and 0.163

### Quality of employment (Table A.2)
- Dependent variable: probability of having a full-time & permanent job. Robust standard errors clustered at the regional level.
- Regression split by gender (Female column (1); Male column (2)).
- Key coefficient estimates (with clustered standard errors in brackets):
  - Being immigrant:
    - Female: -0.380 [0.046]***
    - Male: -0.530 [0.046]***
  - Years since migration:
    - Female: 0.019 [0.005]***
    - Male: 0.015 [0.003]***
  - Years since migration squared/100:
    - Female: -0.023 [0.007]***
    - Male: -0.020 [0.005]***
  - Education:
    - Female: 0.084 [0.004]***
    - Male: 0.045 [0.005]***
  - Age:
    - Female: 0.119 [0.008]***
    - Male: 0.160 [0.003]***
  - Age squared:
    - Female: -0.001 [0.000]***
    - Male: -0.002 [0.000]***
  - Being married:
    - Female: -0.234 [0.021]***
    - Male: 0.297 [0.005]***
- Fixed effects and sample:
  - Country-Year fixed effects: Y / Y Y (table indicates country-year fixed effects present)
  - Region fixed effects: Y / Y
  - Observations:
    - Female: 4,993,587
    - Male: 4,698,090
  - Pseudo R-squared:
    - Female: 0.0988
    - Male: 0.0826

### Simulation of Employment Paths for Migrants from Individual Host Countries (Figure A.1)
- Simulated employment probability paths shown for multiple host countries and separately by gender: Austria (AT-Female, AT-Male), Belgium (BE-Female, BE-Male), Denmark (DK-Female, DK-Male), Spain (ES-Female, ES-Male), Finland (FI-Female, FI-Male), France (FR-Female, FR-Male), Ireland (IE-Female, IE-Male), Italy (IT-Female, IT-Male), Netherlands (NL-Female, NL-Male), Norway (NO-Female, NO-Male), Portugal (PT-Female, PT-Male), Sweden (SE-Female, SE-Male), United Kingdom (UK-Female, UK-Male).
- Legend categories displayed in each panel:
  - All immigrants
  - EUR-NAMENA
  - Asia
  - Lat.Am.
  - Natives
- X-axis in each panel: Years since migration (tick marks shown at 0, 5, 10, 15, 20 in many panels).
- Y-axis in each panel: probability values shown from .2 to 1 in increments (.2, .4, .6, .8, 1).

### Effects of Initial Conditions, Alternative Measures (Figure A.2)
- Panel A: OECD unemployment gap
  - Plots marginal effects of years since migration on Pr(emp) by the unemployment gap at arrival.
  - Female panel y-axis scale shows effects from .002 to .01; x-axis unemployment gap at arrival from -8 to 8.
  - Male panel y-axis scale shows effects from 0 to .004; x-axis unemployment gap at arrival from -8 to 8.
- Panel B: OECD output gap
  - Plots marginal effects of years since migration on Pr(emp) by the output gap at arrival.
  - Female panel y-axis scale shows effects from .0 02 to .0 1 (displayed as .0 02, .0 04, .0 06, .0 08, .0 1); x-axis output gap at arrival from -8 to 8.
  - Male panel y-axis scale shows effects from .001 to .0035; x-axis output gap at arrival from -8 to 8.
- Panel C: OECD labor force participation gap
  - Plots marginal effects of years since migration on Pr(emp) by the LFP gap at arrival.
  - Female panel y-axis scale shows effects from 0.00 to .008; x-axis LFP gap at arrival from -4 to 4.
  - Male panel y-axis scale shows effects from .001 to .0035; x-axis LFP gap at arrival from -4 to 4.

*Source: wp18232 - Appendix Tables and Figures*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp18232.pdf_