Migration and public finances in the EU

We provide novel and comprehensive evidence on the net ﬁscal contributions of natives and migrants to the governmental budgets of EU countries. We account for income taxes and cash beneﬁts, along with indirect taxes and in-kind beneﬁts, which are often missing in standard datasets. We ﬁnd that on average, migrants were net contributors to public ﬁnances over the period of 2014–2018 in the EU and, moreover, that they contribute approximately e 1.5 thousand more per capita each year than natives. We also show that this diﬀerence is partly due to selection on characteristics that make migrants net ﬁscal contributors, such as demographic factors and employment probability.


Introduction
Migration flows towards the European Union have been on the rise in recent years: between 2014 and 2018, the number of foreign-born residents in EU countries increased by about 7 million, so that by 2018, migrants accounted for 11.3% of the EU population, up from 9.8% in 2014. At the same time, the salience of migration in public opinion has also increased: the share of EU residents who think that migration is one of the two most important issues facing the EU increased between 2014 (24%) and 2018 (40%), according to Eurobarometer.
Concerns about migration are multifaceted, ranging from fears related to labour market competition, increased crime, or adverse effects on student achievement to more "culturerelated" concerns. However, one prominent source of concern is the impact that migration has on the tax and welfare system and thus its net fiscal consequences. Indeed, 43.8% of Europeans believe that immigrants take out more from the state in welfare and use of public services than they put in through taxes, according to the 2014 European Social Survey. Previous studies have shown that concerns about the burden of migration on the public budget dominate those about labour market competition and economic efficiency (Dustmann and Preston, 2007) and that fears of welfare abuse by migrants drive negative attitudes towards migration in Europe (Boeri, 2010). The impact that migration has on the tax and welfare system and the net fiscal consequences of migration are of major concern in the public debate over the pros and cons of migration. Boeri (2010), for instance, shows in an analysis of data for several European countries that there is no evidence that migrants -especially skilled migrants -receive more transfers from the state than they contribute to the public purse. Immigrants are often feared to be a burden on the receiving country's welfare state, as they may receive more in social transfers than what they pay in via their taxes. Migration, on the one hand, expands the population, bringing in new sources of public revenues, but on the other hand, it imposes possible new demands on public services.
One specific concern is also that the generosity of welfare provision in destination countries encourages migration of welfare-dependent migrants. Preston (2014) provides an excellent conceptual survey of the key issues in computing the fiscal effects of migration.
In this paper, we directly address these concerns by providing novel and comprehensive evidence on the net fiscal contributions of migrants across EU countries for five fiscal years over the period (2014)(2015)(2016)(2017)(2018) spanning the so-called refugee crisis. To do so, we first build a novel dataset for the microeconomic analysis of the net fiscal contribution of migrants across all EU countries. The dataset that we build allows us to take into account not only income taxes, social insurance and social security contributions paid and cash transfers received by migrant households but also the value added tax (VAT) paid 1 and their receipt of inkind benefits such as education, healthcare and social housing. We define the EU-14 as the member states of the European Union that joined prior to the 2004 enlargement, with 1 Please note that we do not account for excise taxes paid by households. the exclusion of the United Kingdom, which left the EU in 2020. We use this dataset to assess the average annual amount of fiscal revenues and government expenditures that can be traced back to natives and to migrants, and we focus on the EU-14, where the bulk of the migrant population is concentrated.
We show that these countries spent on average e9,600 per year for each of their native residents, which contrasts with an annual per capita average of e8,200 for their migrants.
Migrants therefore tend to be less expensive for the public purse than natives. However, we demonstrate that migrants' average annual per capita fiscal payments (e9,600) are in line with those of natives. Overall, while natives made net contributions to public coffers over this period of on average e32 per capita each year, migrants' net yearly contribution totals an average of e1,510 per capita. On average, across the EU-14, migrants make larger net fiscal contributions than natives, even when they are compared to natives in the same ventile of the national income distribution.
We investigate this aspect in more detail in the second part of our analysis, where we compare differences in net fiscal contributions between migrants and natives with similar characteristics. We show that controlling for demographic characteristics cancels the average gap in net fiscal contributions between migrants and natives, which indicates that migrants are favourably selected on characteristics that make them net fiscal contributors. However, when we condition on employment status as well, the migrant-native gap becomes negative, which indicates that immigrants' higher employment probability partially explains their higher net fiscal contributions. Finally, we show how the net fiscal contributions gap changes across migrants with different migration seniority.
Estimating the fiscal impacts of migration is not straightforward (Boeri and van Ours (2021), Chapter 9.) Preston (2014) neatly discusses the key challenges faced by researchers in this field. There are essentially two approaches to estimating the fiscal effects of migration.
The first stream of literature is heavily model based and aims to provide an assessment of the dynamic fiscal effects of migration over the life cycle. This is the approach taken in the studies of Storesletten (2000) for the US and Storesletten (2003) for Sweden. These papers account for the general equilibrium effects of migration and provide a longer-term estimate of its fiscal effects, but the analysis is strongly reliant on the structure imposed by their modelling assumptions. For instance, two recent papers have argued that some such studies underestimate the fiscal contribution of migrants because they ignore the indirect fiscal benefits of migration, operating through an increase in high-skilled native wages (Colas and Sachs, 2021), or because they abstract from the increase in revenues from capital taxes generated by higher returns to capital due to migration (Clemens, 2021). Other papers have taken a more data-based but also more static approach. The results of these studies are heavily context dependent. For instance, Dustmann and Frattini (2014) show that migrants who came to the UK after 2000, especially those from EU countries, have made a substantial positive fiscal contribution (see also Dustmann et al., 2010). Conversely, Ruist (2015) finds that refugees in Sweden in 2007 both contributed less and cost more to public finances than other residents. Likewise, National Academies of Sciences (2017) finds that low-skilled migration to the US has imposed a net fiscal cost on the country. The paper most closely related to our work is Liebig and Mo (2013), which assesses the net fiscal position of migrants relative to that of natives in European OECD countries and Australia, Canada and the US for 2007-2009, relying on several household surveys. They show that in most countries, the fiscal impact of migration is very small in terms of GDP and is approximately zero on average across the OECD countries considered. In countries where migrants' fiscal position is worse than that of natives, this impact is driven by lower tax payments rather than higher benefit receipts. Age is shown to be the key determinant of cross-country differences in migrants' net fiscal position.
Our analysis is based on the European Union Statistics on Income and Living Conditions (EU-SILC) dataset, which provides a representative sample of all EU resident households, recording their income and their receipt of many cash benefits. We complement the EU-SILC data with EUROMOD (Sutherland and Figari, 2013), a microsimulation model that simulates contributory and noncontributory cash benefits and social security contributions paid by workers. We further use information from aggregate administrative data to add receipt of in-kind benefits to our dataset. Finally, we match households included in EU-SILC data to similar households in the EU Household Budget Survey (HBS) to impute a pattern of household consumption and measure individual households' contribution to aggregate demand and VAT revenues.
The paper is structured as follows: first, in section 2, we outline relevant facts necessary to understand migration flows in recent years in the EU and the nonnative population share, while in section 3, we describe input data and the data sources that we use to assess the net fiscal contribution of native and migrant populations. In section 4, we describe how we derive individual contributions to public expenditures and revenues with relevant sources, along with some aggregate statistics by type of revenue (section 4.2) or expenditure (section 4.1). Moreover, we define the net fiscal contribution of migrants and provide evidence for the difference between the two subpopulations, in aggregate terms (section 4.3), by position in the national income distribution (section 4.3), by years since migration (section 4.5), or by calendar years (section 4.4). Finally, in section 5, we compare differences in the nativemigrant net fiscal contribution with variables related to the characteristics of the migrant population.
2 Background: Migration in the EU From a historical perspective, migration from, towards and within Europe has changed significantly over time, as highlighted by Van Mol and De Valk (2016). After the fall of the Iron Curtain and the opening of borders to Eastern European countries, inflows of migrants increased significantly. In the 21st century, migration from outside the EU typically followed the extent of violent conflicts, leading to a substantial decline in migration from outside the EU at the beginning of the 2000s. Starting from 2006, with the rise of conflicts across the world (e.g., in Afghanistan and Iraq and following the Arab Spring movements), the EU again observed an increase in migration from outside Europe. Moreover, after some years of moderate migration in the 2010s, the so-called refugees crisis led again to a substantial increase in migration to EU member states, particularly Western European states. This crisis was different from other migration inflows in history in terms of not only the magnitude of the migration pattern but also the resistance to it among politicians and populations and the related public disputes. With respect to within-EU migration (see, e.g., Van Mol and De Valk, 2016), the free movement of EC citizens (article 45 of Treaty on the Functioning of the European Union -TEFU) related to the Maastricht Treaty in 1992 led to a continuous increase in free movement within the European Union and therefore also an increase in intra-EU migration and emigration. This phenomenon was only highly visible during the economic crisis of 2008, when Southern European countries experienced substantial outflows to other European countries.
We focus primarily on the EU-14 in aggregate and on five member states, namely, the four largest, with more than 45 million residents each (France, Germany, Italy and Spain) and one midsize country with a large share of migrants (Sweden). States that joined the EU after 2004 present a low proportion of migrants in their populations 2 ; therefore, we decided not to separately include them in our analysis.
According to EUROSTAT, in the years between 2009 and 2018, the share of migrants in the EU-14 increased from approximately 11.5% in 2009 to 11.8% in 2014 and then to 13.5% in 2018 3 . In total, this is an increase of approximately 2 p.p. In particular, strong increases in the migrant population share have been observed in countries such as Sweden and Germany but also in Austria, Belgium and Denmark, while in countries such as France, Spain and Italy, alongside Finland and Portugal, the share of migrants has increased only marginally.
Our paper focuses especially on the years 2014-2018, in which several countries saw an increase in the size of the non-EU population (see also Figure 1), with a relatively more stable presence of EU migrants. Relative changes in size of non-native population had been caused by as an increase in residence permits granted for work, family or education, but it is also as a consequence of recognition of asylum seekers as refugees. In our analysis, we are not able to observe asylum seekers due to the survey design of the EU-SILC, as those living in collective shelters are missing from the sampling frame in every hosting country, resulting in undercoverage of the actual migrant population. In addition, it would be particularly difficult to correctly establish the amount of benefits received by the asylum 2 See Figure 20 in Appendix B for a description of the relative size of the migrant population in new member states.
3 Source: Own calculation with EUROSTAT (migr_pop3ctb) data. seeker population. Conversely, asylum seekers who become recognized refugees and start to participate in the labour market are identified in our dataset. Figure 1 highlights that over the time period 2014-2018, the share of migrants in the total EU-14 population increased by 1.7 p.p., from approximately 11.8% to 13.5%. This increase was driven mainly by migrants from outside the EU, whose share in the overall population increased by approximately 1.1 p.p., from 7.7% to 8.7%. On the other hand, the share of migrants from within the EU-14 increased only slightly less strongly by 0.7 p.p., from 4.1% to 4.8%. Focusing on single countries, we observe a strong increase in the shares of both EU and non-EU migrants in Germany between 2014 and 2020. In the other countries, such as Spain, Sweden and France, the increase in the migrant population over this time period was driven mainly by an increase in the number of non-EU migrants. Despite this general trend, we also observe substantial differences in the size of the migrant population across EU countries. As highlighted in Figure 2, in Luxembourg, which is clearly an outlier due to its size and special characteristics within the EU-14, almost half of the population are migrants. In Sweden, almost 20% of the population was born outside of the country, substantially above the EU-14 average of 13.8%. This is also the case in Germany (17.3%). On the other hand, in countries such as Finland (6.7%) and Italy (10.2%), the share of migrants in the total population is substantially below the EU-14 average.
Additionally, the composition of migrants varies substantially across the EU-14 countries.
While, for example, the majority of migrants in Ireland and Luxembourg were born in another European country, the opposite holds in most of the other member states. In the Nordic countries Sweden and Denmark, a large majority of migrants are from outside the EU.
The same is true not only for many Southern European countries, such as Italy, Portugal, Spain and Greece, but also for France and the Netherlands. As we show in detail in section 3, migrants differ substantially from natives in their characteristics. These differences appear not only in household characteristics and age structure but also in education and employment rates and, as a result, in income. All of these differences play a crucial role in the assessment of the fiscal impact. Indeed, the differences in individual characteristics and in the legal constraints faced by the two populations on intra-EU mobility but also, for instance, recognition of foreign educational qualifications leads to stark differences in their respective labour market success and general integration process, which is in turn decisive for their fiscal impact (see, e.g., Hinte and Zimmermann (2014) or Hansen et al. (2017)).

Data and baseline evidence
The key dataset used in this paper is the EU-SILC. This survey is released yearly by EU-ROSTAT and is run by the national statistical institutes of each EU member state using a common framework to harmonise variables, concepts, design, imputation, weighting and sampling error calculations across countries. This common framework ensures comparability across countries. The EU-SILC collects information on income and living conditions at both the individual and household levels. We use five consecutive waves of the cross-sectional EU-SILC from 2014 to 2018. As already mentioned, we primarily focus on the EU-14 in aggregate and on five member states, namely, France, Germany, Italy, Spain, and Sweden, the latter of which, despite being a midsize country, has a large share of migrants. Consistently with most of the literature, we group respondents of the survey by their country of birth 4 into natives and migrants, without distinguishing between EU migrants (i.e., nonnative residents born in another EU country) and non-EU migrants (i.e., nonnative residents born in a country not belonging to the EU). Unfortunately, for Germany among the others, the EU-SILC data does not allow us to distinguish between EU and non-EU migrants within the foreign born population.
We define the household head as the member who has the largest gross income or, in the case of equal income, the oldest. In the very few cases where this rule does not allow for identification of a household head, always involving couples, we classify the man as household head, following the prevalent social norm in the EU. We define a household as Germany and Sweden host a larger share of highly educated migrants than other countries, likely because of the characteristics of their labour markets. In contrast to those in other countries, migrants in France are older and less educated than natives, which could be a joint effect of France's generous welfare system and longer history of migration, especially from former colonies. In terms of employment probability, defined as the probability of being in employment conditional or unconditional on being of working age (25-64 years), large heterogeneity emerges between natives and migrants. In particular, in France and Sweden, working-age natives are more likely to be employed than migrants, whereas the difference is negligible in Germany, Italy and Spain and slim in the EU-14. The employment probability of natives drops in all countries when all ages are considered, most likely because there are fewer migrants above retirement age. Regarding years since migration (YSM), migration seems to be a more recent phenomenon in Italy, Spain and Sweden than in France and Germany. We also define income as the sum of income from employment, self-employment and other sources of income. For EU-14 countries that do not belong to the euro area (Denmark, Sweden), we use the conversion factor provided by EUROSTAT. Moreover, we use the EUROSTAT harmonised index of consumer prices (HICP) to adjust all monetary variables to 2018 euros. Migrants earn more than natives in the EU-14 aggregate but also in Germany and (with a smaller differential) in Italy, possibly because income is a relevant reason for migration among individuals. Migrants earn much less than natives in France and (with a smaller differential) in Spain and Sweden. Notes: Our calculations from EU-SILC data. Natives are respondents of the survey who live in their country of birth; migrants are nonnative residents born in another EU country or in a country not belonging to the EU. Education is defined as low if the respondent had no education or primary education, whereas it is defined as high if the respondent had at least postsecondary education. Employment probability is defined as the probability of being active in employment in the overall reference population, while employment probability in working age is the fraction of people in employment conditional on being of working age. YSM stands for years since migration. The migration status of a household is the migration status of the household head. Income is defined as the sum of income from employment, self-employment and other sources of income.
The number of observations refers to the number of respondents in the five waves considered.

Individual contributions to public expenditures and revenues
For our analysis of the fiscal cost of migrants, we extensively use EUROMOD, the tax-benefit microsimulation model for the European Union (Sutherland and Figari, 2013). EUROMOD was developed at the Institute for Social and Economic Research at the University of Essex as an EU-funded European academic project. Since 2017, the European Commission has taken over responsibilities for its annual updates and developments through the Joint Research Center in Seville. EUROMOD relies on EU-SILC microdata, which are representative at both the household and individual levels for each EU member state. It is a unique tool for comparative research on taxes and benefits in the EU. EUROMOD guarantees, through a standard set of protocols between developers and national teams, a common framework and therefore guarantees cross-country comparability.
As EUROMOD simulates benefit eligibility irrespective of citizenship and YSM, we extend it by taking into account differences in the tax-benefit system between native and migrant residents, as in many countries, eligibility criteria for specific benefit policies either are conditional on a minimum number of years of residence in the country or are subject to a minimum contribution history (e.g., unemployment benefits). EUROMOD policy changes and normative sources are specified in an annex file, where we explain the EUROMOD extensions according to national legislation and personal working history using information from the EU Commission website (https://tinyurl.com/3r3kuhav), integrated with country-specific sources of information 5 .
EUROMOD allows us to account for personal income taxes and social security contribution (SSC) payments at the individual and household levels. It also allows us to account for cash transfers received as pension payments or other contributory and noncontributory benefits (e.g., unemployment or family benefits). However, to have a full picture of the net fiscal contribution of migrants vis-a-vis that of natives, we also need to account for in-kind benefits and indirect taxes (Figari and Paulus, 2015;Christl et al., 2020).

Public expenditures
As highlighted, for instance, by Figari and Paulus (2015) and Christl et al. (2020), in-kind benefits play a crucial role not only in redistributive terms but also in the definition of the net fiscal impact of individuals. Hence, in addition to the standard transfer received (all types of cash benefits received at the individual or household level covered in EUROMOD, such as family benefits, unemployment benefits, and pensions), we include the monetary value of in-kind benefits that individuals receive for health, social housing and education, integrating them with EUROMOD individual and household cash benefit variables. We provide details on the imputation of the monetary value of in-kind benefits in the Appendix in section A.3 6 .
Native and migrant households could receive in-kind benefits differently because of their demographic and economic characteristics. On the other hand, where there is no reason to believe that migration status has a role in the distribution of in-kind benefits or we have no information to distribute the costs of benefits based on the migration status of households, we ignore these items in the measurement of the individual fiscal cost, as they could, at best, be only uniformly distributed. These refer to expenditures including other general functions of the public administration, infrastructure and other indivisible services provided to the whole community, defence, culture, public order and safety, environmental protection or public debt transactions, which we ignore here. Table 2 shows that our analysis accounts for approximately 70% of total public expenditures over the whole considered period in the focal countries and in the EU-14.  Overall, in EU-14 countries, expenditures in favour of natives are higher; i.e., natives tend to benefit more from public expenditures than migrants. This is particularly true for countries such as Italy and Spain, where we observe an enormous difference in the average benefit between native and migrant beneficiaries, while these differences are quite small in Germany and Sweden. Conversely, average public expenditures to the benefit of migrants are (slightly) larger than those to the benefit of natives in France.
Differences in average public expenditures could, however, be the result of differences in the composition of the native and migrant populations. We therefore disaggregate the total benefits into four subcategories: (i) old age and survivor benefits and pensions; (ii) 6 See also Verbist and Matsaganis (2014). benefits for education, health and housing; (iii) a broader group of social assistance benefits and pensions, such as family benefits (including maternity and parental leave benefits, child and child care benefits), disability benefits and pensions (including benefits and pensions for accidents, receiving care, caring benefits and health pensions for people with disabilities) and, finally, heating benefits and municipality benefits for housing; and (iv) unemployment and work-related benefits. Figure 4 displays how the total public expenditure directed towards migrants and natives is allocated across these four categories. The provision of in-kind goods such as social housing, health and education is the largest source of expenditures for both natives and migrants in every country and in the EU aggregate. Family benefits are the third highest kind of expenditure for both natives and migrants, whereas pensions are the second item for both groups. However, the relative shares of these expenditure categories for natives and migrants change remarkably: the family share for migrants is on average twice that of natives, whereas the pension expenditure of migrants is 0.7 that of natives, ranging from 0.4 in Italy to 0.89 in France. Public expenditures going to migrants are mostly health, housing and education benefits and family benefits and social assistance, whereas migrants profit less from old-age benefits. The expenditure share of old-age benefits is clearly larger for natives than for migrants, especially in Italy and Spain and to a lesser extent in Sweden, highlighting the unequal demographic composition of the two groups. In general, we also observe that a larger share of expenditures on migrants are work-related benefits in Italy and Spain than in other countries. (iii) social assistance benefits and pensions: family benefits (including maternity and parental leave benefits, child and child care benefits), disability benefits and pensions (including benefits and pensions for accidents, receiving care, caring benefits and health pensions for people with disabilities), heating benefits and municipality benefits for housing; and (iv) unemployment and work-related benefits.

Public revenues
Government revenues within our framework consist of three main categories. First, social security contributions (SSC) consist of both employer and employee contributions, as well as those of the self-employed. The second is labour and capital income taxes. Third, we include value-added taxes (VAT), which are typically not included in analyses based on microdata as these data lack precise information on indirect tax payments 7 . SSCs, income taxes and VAT account for almost 85% of total public revenues in EU-14 countries, as shown in Table   3. The share of these three sources in total public revenues is heterogeneous across countries, ranging from 70% in Sweden to more than 90% in Germany.  Figure 5 shows the average per capita fiscal revenues paid by natives and migrants: on average, across the EU-14, migrants generate slightly higher per capita revenues than natives. However, these findings differ substantially across the EU-14 member states. While in Germany, for example, migrants provide higher revenues for the public purse than natives, the opposite holds true in France, Italy, Spain and Sweden. Figure 6 splits the composition of public finance revenues into the three categories considered for natives and foreigners: at the EU-14 level, the composition of fiscal payments is similar for both immigrants and natives. Social security contributions, which are proportional to labour income in most EU countries, account for about half of total revenues, although their share in total revenues is higher for immigrants (54%) than for natives (48%).
Income tax revenue accounts for a slightly larger share of total fiscal revenues for natives (35%) than for migrants (32%), which reflects immigrants' generally lower incomes and progressive taxation in most EU countries. The share of total revenues accounted for by indirect taxes (VAT) is higher in countries such as Italy or Spain (21%), where VAT rates are higher than in the other countries (respectively, 22% and 21% versus 19% in Germany and 20% in France 8 . Differences between natives and migrants seem not to be particularly meaningful within each country and reflect income distributions and taxation schemes.

Net fiscal contribution of migrants
In the previous subsections, we separately analysed the expenditures and revenues side of public finances. We can now put together these two determinants of the public budget and compute the net fiscal contribution made by migrants and natives. Specifically, for each individual i who currently resides in country c of migration status j ∈ {native, migrant} and observed in survey year t ∈ {2014, ..., 2018}, we define the net fiscal contribution (NFC) as the difference between public revenues generated (i.e., taxes paid) R and public expenditures occupied (i.e., benefits received) E:  Such heterogeneity reflects the underlying heterogeneity in the characteristics of the migrant population across countries, as we investigate below. However, one striking regularity highlighted by Table 4 is that the countries where migrants' NFCs are larger than those of natives are also those where migrants' average taxable income is higher than natives' and vice versa. However, the differences in per capita NFCs between migrants and natives insurance contributions and other social insurance contributions -contributions paid by individuals but not directly linked to employment or self-employment (e.g., due to benefits and paid by the benefit recipients, health contributions paid by the general population). We then compute the group-specific average NFC for each migrant and native in each ventile and take the difference between the two groups. We thus obtain the ventile-specific difference in the net fiscal contributions of migrants and natives, namely: where In Figures 7 -12, we report ∆N F C v,c for each ventile v, and we fit a polynomial function of degree nine across these 20 points to facilitate readability for the EU-14 as a whole (7) and for each of our five selected countries separately (Figures 8-12). In the bottom part of each figure, we also show the histogram of the distribution of migrants across ventiles of the native distribution of taxable income (i.e., t N v,mig,t , t N mig,t ).
On average, across the EU, migrants contribute to public finances significantly more than natives at all ventiles of the income distribution, with the exception of the bottom ventile, which is also, however, the one into which migrants are disproportionately likely to fall (Figure 7). In four of the five countries that we focus on, we observe the same pattern of higher migrant relative to native NFCs in the same position of the national income distribution, with the difference increasing along ventiles. In France, in contrast, migrants' NFC is lower than natives' at all points of the distribution except in the top ventiles ( Figure   8).
Although everywhere immigrants are more easily found in the bottom ventiles than natives, the degree of concentration at the bottom of the distribution varies considerably across countries. In France, Spain and, especially, Sweden, migrants are disproportionately concentrated in the very bottom part of the distribution. In Italy, on the other hand, the distribution is slightly less skewed, while in Germany, the distribution of migrants is more similar to that of natives than anywhere else, even though migrants are still more likely than natives to be in the bottom income ventiles. Because of the progressive tax and benefit system of EU countries, a relatively larger share of low-income migrants increases the likelihood that they receive more in welfare benefits than what they paid in through taxes and SSCs. The interplay of differences in the income distribution and NFCs conditional on disposable income determines the aggregate results displayed in Table 4.  Source: Our calculations from the EU-SILC dataset and EUROMOD. In the upper panel, we report the ventile-specific ∆N F C v,EU from equation 2 (dotted line) and a polynomial approximation of degree 9 for the distribution of ∆N F C v,EU (dashed line), while in the lower panel, we report the histogram of the distribution of migrants across ventiles of the native distribution of taxable income.  Source: Our calculations from the EU-SILC dataset and EUROMOD. In the upper panel, we report the ventile-specific ∆N F C v,F R from equation 2 (dotted line) and a polynomial approximation of degree 9 for the distribution of ∆N F C v,F R (dashed line), while in the lower panel, we report the histogram of the distribution of migrants across ventiles of the native distribution of taxable income.   Source: Our calculations using the EU-SILC dataset and EUROMOD. In the upper panel, we report the ventile-specific ∆N F C v,IT from equation 2 (dotted line) and a polynomial approximation of degree 9 for the distribution of ∆N F C v,IT (dashed line), while in the lower panel, we report the histogram of the distribution of migrants across ventiles of the native distribution of taxable income.  Source: Our calculations from the EU-SILC dataset and EUROMOD. In the upper panel, we report the ventile-specific ∆N F C v,ES from equation 2 (dotted line) and a polynomial approximation of degree 9 for the distribution of ∆N F C v,ES (dashed line), while in the lower panel, we report the histogram of the distribution of migrants across ventiles of the native distribution of taxable income.  Source: Our calculations from the EU-SILC dataset and EUROMOD. In the upper panel, we report the ventile-specific ∆N F C v,SE from equation 2 (dotted line) and a polynomial approximation of degree 9 for the distribution of ∆N F C v,SE (dashed line), while in the lower panel, we report the histogram of the distribution of migrants across ventiles of the native distribution of taxable income.

Evolution over time
Our data span a period of five years characterized by intensive migration waves (see section 2). A crucial question is therefore how the net fiscal contribution of migrants vis-a-vis that of natives evolved over time. Hence, in this section, we do not pool the data over all available years but rather study how per capita N F C changed over time. Figure 13 reports the annual migrant-native difference in per capita revenues (∆R, in yellow) and per capita expenditures (∆E, in green) across all EU-14 countries pooled and for each of our five selected countries. Across all EU-14 countries, migrant-generated fiscal revenues were very similar to those of natives, although they decreased slightly over time.
Likewise, while public expenditures for migrants were significantly lower in all years than expenditures for natives, in the more recent years, the difference between public expenditures for migrants and natives decreased. These two trends therefore show a tendency towards a reduction in migrants' net fiscal contributions over time, even though as of 2018 they were still contributing significantly more to public coffers than natives. As we already mentioned in the previous sections, differences in migrant-native expenditures are slightly positive in France and close to zero in Sweden, and they are quite constant for the time span considered.
Conversely, migrant-native differences in revenue and expenditure are both converging to zero in Germany, while differences in per capita expenditures between migrants and natives are increasing over time.
What role do individual characteristics play in explaining the level and evolution of migrant-native differences in net fiscal contributions? We have seen above that migrants' NFCs tend to be higher than those of natives with similar income; what would happen if migrants and natives had similar observable characteristics?
We address this question by estimating a regression of the type: where i indexes individuals, c the country of residence and t the year of observation.
I is a dummy variable that identifies migrants vs natives, and X is a vector of individual characteristics. Therefore, γ c,t measures the average difference in NFCs between migrants and natives living in country c in year t, conditional on the variables included in X.
We estimate equation (3) separately for each country c and alternatively pool all EU-14 countries and report in Figure 14 the estimated γ c,t and the associated 95% confidence intervals. In all cases, we report estimates from three different versions of equation (3): first, one in which we do not include X (unconditional ); second, one in which we include in X individual characteristics such as age, gender, education and household size (cond. demographics); and third, one in which we additionally include employment status (cond. As we have seen in Figure 13, at the EU-14 level, the differential between migrants' and natives' per capita NFCs is positive in all years, although its size decreases over time.
When we condition out individual characteristics and essentially compare migrants and natives who look alike in terms of age, gender, education and household size, the estimated differential decreases in all years, which indicates that part of the fiscal advantage of migrants is explained by migrants having demographic characteristics that make them more likely to be net fiscal contributors. In 2017 and 2018, migrants and natives with similar characteristics were indeed making exactly the same net fiscal per capita contribution, indicating that migrants' superior fiscal stance was -in those two years -entirely driven by their favourable selection. Last, migrants are more likely to be employed than natives across the EU-14 as a whole: once we condition on both individual characteristics and employment status, the net fiscal contributions of natives are larger than those of immigrants in all years. Thus, the higher employment probability of natives is what ultimately drives migrants' positive NFC.
Interestingly, the trend of NFC over time is exactly the same regardless of what variables are conditioned out.
Germany and Sweden follow a trend that is very similar to that of the EU-14 as a whole, although the migrant-native NFC differential is always positive in Germany and always negative (except in 2008) in Sweden. Italy and Spain are the only two countries where the unconditional migrant-native difference in net fiscal contributions is increasing over time.
Additionally, however, the "conditional" differences are smaller in this case, which indicates that migrants are overrepresented with respect to natives in demographic groups that are net contributors. Were the demographic composition of migrants and natives the same, the NFC of migrants in Spain would not be significantly different from that of natives in any year. In Italy, in contrast, it would still be higher than natives', albeit to a lesser extent and increasingly so over time. Adding labour market status as an additional conditioning variable also makes the net fiscal contribution of migrants in both Italy and Spain become flat and negative at approximately e-2.000. Were the labour force participation of natives and migrants the same, migrants' NFC would be negative everywhere.
France stands out among the countries for its consistently negative difference in migrants' NFC relative to natives' that is increasing over time. However, the difference vanishes once differences in demographic characteristics are accounted for, which indicates that unlike those in the other countries, migrants in France have age-gender-education profiles that make them more likely to be net fiscal recipients. However, migrants in France also have higher labour market participation than comparable natives: were their employment status similar to that of natives, their NFC would have been increasingly negative.
mation coming from EU-SILC's Self-defined current economic status (pl031) and Status in employment (pl040) Figure 14: Migrant-native differences in NFCs, conditional on different sets of variables and over time, 2014-2018.

Years since migration in the country
Finally, the net fiscal contribution of migrants may change with time spent in the host country, as highlighted by Brell et al. (2020). In this section, we investigate whether years since migration positively or negatively affects the NFC of migrants. To this end, denoting years since migration as YSM, we estimate the following model: where, in addition to the notation introduced before, τ t are year dummies and YSM is included with a polynomial of degree two to account for nonlinearities. As before, we estimate this model first using an empty X matrix and then including in X only demographic characteristics and finally including employment status as well. Hence, γ c measures the average difference in NFCs between migrants who have just arrived in country c (i.e., when YSM=0) and natives net of the inclusion of covariates, whereas when YSM=0, β c and η c capture how the differential evolves nonlinearly over time. However, when we also control for employment status, the effect of YSM declines, although it remains positive.
The expected evolution of the migrant-native NFC differential E[∆N F C c ] over years since migration can therefore be described by the following equation 10 : The relationships between ∆N F C i,c and Y SM are also plotted in Figures 15-17, depicting the relationships with YSM, which is increasing at a decreasing rate (5) as in panels A, B and C of Table 5, respectively. Figure 15 shows that, on average across the EU-14, the net fiscal contributions of migrants who have been in the country for at least five years are larger than those of natives.
There is, however, heterogeneity across countries. For instance, in Italy and Spain, immigrants' net fiscal contributions are higher than those of natives regardless of years since migration. Conversely, in Germany, only immigrants with at least sixteen years of residence are stronger fiscal contributors than natives. In all cases, the differential in fiscal contribu-10 Note that our estimates are based on pooled cross-sectional data for 2014-2018; hence, we are comparing immigrants who have been in the country for a few years with other immigrants who have been in the country for a longer period. Therefore, these estimates do not precisely allow us to distinguish the so-called cohort effect (i.e., differences in potentially unobserved characteristics between different migrant cohorts) and residence effect (i.e., the effect of spending more time in the host country).  Notes: The dependent variable is net fiscal contribution, expressed in 2018 euros. With reference to equation (4), we report only γc, which is the estimated coefficient of nonnatives; βc is the estimated coefficient of the linear effect of YSM, and ηc is the estimated coefficient of the quadratic effect of YSM. In specification (a), we include country and year fixed effects and their interactions as well as years since migration and its quadratic term; in panel (b), we also include age, gender, highest personal education status and the household dimension; and in panel (c), we also include employment status. Robust standard errors are calculated in all specifications. Significance levels: * ≤ .1, * * ≤ .05, * * * ≤ .01.
tions between immigrants and natives increases until at least 22 years since migration (for the aggregate EU-14). Figure 16 shows that if the distribution of demographic characteristics were the same for natives and migrants, in all countries, newly arrived migrants would have lower NFCs than natives. The effect of length of residence operates mainly through an increase in the probability of being in employment: Figure 17 shows that were the labour force participation of migrants the same as that of natives, immigrants' NFCs would always be lower than natives', and the effect of years since migration would be largely reduced.    Source: Our calculations from EU-SILC and EUROMOD data.

Discussion
Although our analysis showed that on average migrants' net fiscal contributions are higher than natives', it also highlighted substantial heterogeneity across countries. Such crosscountry differences may originate partly from differences in the tax and benefits systems and partly from differences in the relative characteristics of the native and migrant populations.
Our results in section 4.4 showed that, indeed, migrant-native differences in NFCs are reduced when migrants are compared to natives with similar age-gender-education profiles and that compared to natives who have the same employment status, immigrants have lower net fiscal contributions.
These findings generalize beyond the five countries that are the most interesting cases for our purposes and to which we devoted our attention. We extend our analysis considering all EU member states in Figures 18 and 19. In the left panel of Figure 18, we plot the migrant-native differences in per capita NFCs for each country and year included in our analysis against the migrant-native differential in employment probability. The figure shows a strongly positive correlation between these two gaps: the higher migrants' NFCs is relative to natives', the higher is their employment probability relative to natives'.On average, across all countries and years, a 10 p.p. increase in the migrant-native employment probability gap is associated with an increase of approximately e2,630 in the migrant-native gap in net fiscal contributions.  Likewise, the right panel of Figure 18 plots migrant-native gaps in net fiscal contributions against the difference between the shares of individuals with low education in the migrant and native populations. In this case, the correlation is clearly negative: a 10 p.p. decrease in the migrant-native differential in the probability of having low education (i.e., higher education of migrants relative to that of natives) is associated with a e1,170 euro increase in the migrant-native gap in NFCs.
Hence, immigrants' (and natives') characteristics seem to matter a great deal in explaining their relative fiscal contributions: a more educated and more economically integrated migrant population leads to higher net contributions by migrants to the public purse. An obvious implication is that policies that foster migrants' labour market integration are desirable not only to facilitate their lives in the host countries but also to help those countries' public finances.
However, while migrants' characteristics matter, the size of the migrant population in a country does not seem to have any major effect on the amount of migrants' per capita net fiscal contributions. We show this in Figure 19.  The left panel plots the migrant-native difference in per capita NFCs for each country- year against the corresponding share of migrants and shows that the two are only moderately negatively correlated: the line fitting the scatter plot has a negative slope that is not statistically significant at conventional levels. However, the weak negative correlation between migrant concentration and the migrant-native NFC differential is driven largely by crosssectional differences across countries. While one might think that what is truly potentially costly for public finances is not the overall number of migrants but the number of recent migrants, the right panel of Figure 19 suggests instead that this is not the case. The figure plots the 2018 migrant-native NFC differential against the change in the share of migrants in the total population of each country between 2014 and 2018: the line fitted through the plot is almost perfectly flat, indicating the lack of any correlation between the relative fiscal stance of migrants and the size of migrant inflows in each country.

Concluding remarks
The general public discussion over whether the generosity of welfare provision in destination countries might encourage migration of welfare-dependent migrants is ongoing in Europe.
In this paper, we try to contribute to this discussion by providing novel and comprehensive evidence on the net fiscal contributions of migrants across EU-14 countries for the period 2014 to 2018, including the years of the so-called refugee crisis. We focus on Western European countries , where the bulk of the migrant population is concentrated.
To obtain a comprehensive overview of the contributions made and benefits received by each population group, we first build a novel dataset for the microeconomic analysis of the net fiscal contribution of migrants across all 14 EU countries. We take into account not only income taxes paid and cash transfers received by migrant and native households but also the VAT paid and the receipt of in-kind benefits such as education and healthcare at the individual level. We then assess the average annual amount of fiscal revenues and government expenditures that can be traced back to natives and to migrants.
Overall, we find that over the observed period, natives contributed on average e32 per year to public coffers while migrants contributed (net) an average of e1,510 per capita to the public finances of "host" countries every year. These results, however, differ across countries: we see that in France and Sweden natives contribute on average more than migrants to public finances, while in Germany, Spain, and Italy, the opposite holds true.
We also compare the differences in net fiscal contributions between migrants and natives while accounting for differences in characteristics. Our analysis highlights that controlling for demographic characteristics cancels the average difference in net fiscal contributions between migrants and natives, which indicates that migrants are favourably selected on characteristics that make them net fiscal contributors. However, when we condition on employment status as well, the migrant-native gap becomes negative, which indicates that the higher employment probability of migrants explains in part their higher net fiscal contributions.

A.1 SILC and EM input data
We were able to combine the set of information contained in EU-SILC data with the EU-ROMOD input dataset for all EU countries in the five cross-sectional waves considered. The combination was needed because the EUROMOD input file, which is derived from SILC, was missing some key variables for this project, namely, year of migration in the country (SILC variable rb031) and tenure status of the household (SILC hh021). The former variable is necessary to assess the relevant income period for certain policy simulations in EUROMOD, such as unemployment or old-age pensions, which require migrants to have spent a certain spell of time in the country before they become eligible for the benefit. The latter variable is needed to identify households in social housing and to correctly estimate the individual cost for social housing. The main variable to differentiate nationals from migrants is dcb, which classifies each individual in the dataset as either a native, a citizen born in another EU country or a non-EU-born citizen. The match between EUROMOD input data and UDB-SILC data was based on personal and household IDs. However, for some countries and years, this matching procedure did not work, as the household and individual IDs are not the same in the two datasets despite the observations being the same; therefore, we used exact matching on observables such as age (EU-SILC variable rx020), gender (EU-SILC variable rb090) and sampling weights (SILC variable rb050 and db090), household dimension and ranking in annual gross income within the household. Last, for those who were still unmatched, we performed probabilistic matching with the same observables.

A.2 Reweighting
The EU-SILC is supposed to provide a representative picture of resident households in each EU member state. However, these data tend to provide a biased representation of the population of migrant residents, partly due to their relatively small share and to greater difficulties in finding them present at home or willing to take part in an interview. In fact, the weighted frequency with the sampling weights included in the EUROMOD data provides a limited picture of the population of residents by citizenship and age groups, with nationals tending to be overrepresented and foreign residents conversely underrepresented. Hence, we validated the data with frequencies from aggregated external sources by country and country of birth. These external statistics are provided by EUROSTAT, which collects information mostly from administrative sources, including interior ministries or registry offices. To provide a picture of foreign-born residents closer to thier actual frequencies, we poststratified the SILC/EUROMOD input data using original weights and EUROSTAT information (contained in migr-pop3ctb dataset) on the number of natives and EU and non-EU foreign-born residents by age group, gender and country of birth in each EU country at the beginning of any year and at the beginning of the following year, which we averaged to obtain the mean value of foreign-born residents in any year of our analysis. We then adjusted sampling weights in each age-gender-birth country group by a multiplicative factor that minimized the distance of the sum of the weights from the control total for each group.
In general, we used 5-year age groups for nationals and 10-year age intervals for EU and non-EU citizens. However, in some countries, the sample size for nonnationals was so small that we had to increase the width of the age groups. Reweighting was performed with the Stata program survwgt (Winter, 2002). As already mentioned, we also considered that the EU-SILC data do not allow us to distinguish between country nationals and EU citizens in Germany. For these countries, we therefore include EU-born migrants in the non-EU-born migrants group.

A.3 In-kind and cash expenditures and expenditure validation
We obtained aggregate country-level public expenditures for health services for all EU-14 countries from EUROSTAT's database on general government expenditures by function (COFOG, gov-10a-exp, variable GF07). Such expenditures include spending on medical products, appliances and equipment (GF07-01), outpatient services (GF07-02), hospital services (GF07-03), public health services (GF07-04), R&D in health (GF07-05), and health expenditures not specified in previous categories (GF07-06). We then apportioned expenditures by age groups based on the OECD statistics on the distribution of the total cost of public healthcare by age class for the Netherlands in 2011 (OECD, 2017), assuming that this distribution is representative of the distribution of healthcare expenditure across the EU-14 by five-year age classes and that it did not change over the time span that we considered. Last, we divided the aggregate expenditures by age class by the number of residents within each age class to estimate per-capita expenditures. To the best of our knowledge, there is no alternative and similarly reliable source estimating the distribution of healthcare costs by individual characteristics. For the cost of social housing provision, we obtained from the COFOG database the total amount spent by public authorities or public institutions for housing and community amenities (COFOG, gov-10a-exp, variable GF06), which includes housing development (GF06-01), community development (GF06-02), water supply (GF06-03), street lighting (GF06-04), R&D housing and communities amenities (GF06-05) and other, i.e., residual, expenditures (GF06-06). We then used EU-SILC-provided information regarding the tenure status of households to obtain an estimate of the number of individuals living in social housing along with their sampling weight, which was used to compute an estimate of the average cost of social housing in each EU-14 country for every included year. Finally, we also included in-kind expenditures for education. We used the total public expenditure on education per full-time-equivalent student by education level and type of program (EUROSTAT educ-uoe-fine09 database), which is available for EU countries for any given year. Within each combination of country and year, we calculated the relative cost for each educational level with respect to the highest educational grade, namely, tertiary education. We then estimated, using the information retrieved in EUROMOD dec variable (present educational status) and sampling weights, the number of students for each level and then the relative cost of each educational level within a certain country, year and educational attainment combination. Next, we divided the total amount spent by public bodies retrieved by COFOG (gov-10a-exp, variable GF09 -Education), assigning to each student in the EU-SILC/EUROMOD the relative cost faced by public bodies for his or her education. This procedure enabled us to account for the different structure of educational patterns within each country in a more flexible and precise way we could do by following COFOG criteria (01: Preprimary and primary education; 02: Secondary education; 03: Postsecondary nontertiary education; 04: Tertiary education; 05: Education not definable by level; 06: Subsidiary services to education; 07: R&D Education; 08: Education not elsewhere classified). While in-kind benefits for education could be directly attributed to the child (and then to the household) that is receiving education, we imputed in-kind housing benefits by assigning the same amount of per capita expenditure for each person. We imputed in-kind health benefits prc-hicp-aind database), which gives comparable measures of inflation for the EU-14 countries over different years and provides the official measure of consumer price inflation in the euro area for the purposes of monetary policymaking. We retrieved the VAT rate for each subitem, and we finally calculated the amount of indirect taxes paid by each household in the EU-HBS database. Then, using income variables (when available) and demographic characteristics (such as age, gender, educational attainment and country of birth of the household head, household dimension and composition), we estimated via OLS for each household in the EU-SILC data the expected amount of VAT paid. As shown by Christl et al. (2022), there are substantial differences across effective VAT rates across migration statuses due to different consumption behaviour. In Table 7, we provide external validation of our estimated public revenues by comparing the aggregates with official statistics from EUROSTAT.

In person
All over the European Union there are hundreds of Europe Direct information centres. You can find the address of the centre nearest you at: https://europa.eu/european-union/contact_en

On the phone or by email
Europe Direct is a service that answers your questions about the European Union. You can contact this service: -by freephone: 00 800 6 7 8 9 10 11 (certain operators may charge for these calls), -at the following standard number: +32 22999696, or -by electronic mail via: https://europa.eu/european-union/contact_en

Online
Information about the European Union in all the official languages of the EU is available on the Europa website at: https://europa.eu/european-union/index_en

EU publications
You can download or order free and priced EU publications from EU Bookshop at: https://publications.europa.eu/en/publications. Multiple copies of free publications may be obtained by contacting Europe Direct or your local information centre (see https://europa.eu/european-union/contact_en).