Portal de Periódicos da CAPES

Can regional policies shape migration flows?

2022; Elsevier BV; Volume: 101; Issue: 3 Linguagem: Inglês

10.1111/pirs.12670

1435-5957

Augusto Cerqua, Guido Pellegrini, Ornella Tarola,

Regional Economics and Spatial Analysis

Our empirical analysis focuses on the effect of regional policies on migration attraction factors in Europe. We employ a regression discontinuity design to assess the causal relationship between the reception of large amounts of public funds and migration flows in the EU-15 regions. In highly-subsidised regions, we find a large increase in the share of foreign citizens from less-developed countries when compared to low-subsidised regions with similar pre-treatment characteristics. The analysis shows that such an increase is due to the positive impact of the European regional policy on job market opportunities as well as the improvement of public goods supply. El análisis empírico de este estudio se centra en el efecto de las políticas regionales sobre los factores de atracción de la migración en Europa. Se empleó un diseño de regresión discontinua para evaluar la relación causal entre la recepción de grandes cantidades de fondos públicos y los flujos migratorios en las regiones de la UE-15. En las regiones altamente subvencionadas, se encontró un gran aumento de la proporción de ciudadanos extranjeros procedentes de países menos desarrollados, en comparación con las regiones poco subvencionadas con características similares antes del tratamiento. El análisis muestra que dicho aumento se debe al impacto positivo de la política regional europea en las oportunidades del mercado laboral, así como a la mejora de la oferta de bienes públicos. 本研究では、欧州における移住の誘因に対する地域政策の影響に注目して実証分析を実施する。回帰不連続デザインを用いて、EU15の地域における移住フローと多額の公的資金の受け入れとの因果関係を評価する。多額の補助金の給付を受けている地域では、給付以前は特性が同様であった補助金が少ない地域と比較すると、後開発途上国出身の外国人の住民の割合が大幅に増加している。分析から、この増加は、雇用市場における機会に対する欧州地域政策の正の影響及び公共財供給の改善によるものであることが示された。 From an economic point of view, migrations are the empirical expression of factor mobility across international borders and, therefore, globally increase the efficiency of production processes. However, from a social perspective, they are complex events, often costly and wasteful, which require resources and flexibility both from those emigrating and from those who welcome immigrants. Faced with a dramatic increase in migratory flows to high-income countries (International Organization for Migration, 2019), and the strong heterogeneity of migrations across countries and regions, the economic literature has investigated what the mechanisms are that guide the choice to migrate and the final destination, to help developed countries to govern this complex phenomenon. The literature highlights that migration flows are driven mainly by economic and social opportunities. 11 In the seminal contribution by Todaro (1969), people compare the economic and social conditions of their native country with those possibly faced in the host country. This comparison leads to an income-maximizing behavior so that single economic agents, the migrant, take into account differentials in wages and employment conditions between countries. Whereas embracing this macroeconomic perspective emphasises the key features of the labour market in different geographical areas, in the corresponding microeconomic framework (Todaro, 1976), migration represents an investment in human capital. Thus, the location choice is driven by a cost–benefit analysis considering the returns of moving, net of the cost of traveling, assimilating a new culture and an unknown language, among other things. The welfare system (Borjas, 1999, see also Conte & Mazza, 2019 for a survey), and healthcare services (Borjas & Hilton, 1996; Preston, 2014) provide migrants with the opportunities to attain better living standards. Further dimensions of analysis are represented by natural amenities (Glaeser, Kolko, & Saiz, 2001; Graves, 1976), such as proximity to lakes and green landscape, and religious and ethnic differences. Our hypothesis is that some of these opportunities might be modified across European regions by the availability and use of European Structural and Cohesion Funds (EUF). The EUF foster structural and economic homogeneity across European regions in the context of infrastructure, education, and labour markets. Our theoretical entry point is based on Hatton and Williamson (2005), where an individual is more likely to migrate, the higher the compensating differential is, namely, the net wage in the destination country minus the net wage in the home country, and the lower the fixed migration cost is. An important addition to this model is the development of the so-called welfare magnet hypothesis, according to which migrants select the receiving country depending on the generous benefits they can get. Borjas (1999) shows that generous welfare programmes offered by many US states have become a magnet for immigrants. Concerning the EU-15 countries, which also represent the focus of our empirical analysis, De Giorgi and Pellizzari (2009) estimate the extent to which welfare generosity affects the location decisions of migrants. The conclusions are that these welfare magnets are positive but relatively weak, compared to the role of labour market conditions, such as the unemployment rate and the level of wages. However, the paper suggests that the critical issue is to what extent the variation in the welfare institutions across the countries of the EU will generate changes in the spatial distribution of the migration flows. In a recent report by the World Economic Forum (Schwab, 2017), inadequate or limited urban services and infrastructure are indicated among the causes of migration (push factors), whereas affordable and accessible urban services (including health care, education, utilities, and transport) are among the pull factors. In such a context, there is a clear role for government intervention (see, for example, Cebula, 2002). Policies to increase welfare or stimulate growth, with positive effects on the labour market, affecting the factors that attract migrants, are however often constrained by the limits placed on public spending and also, for the regions, by the resources from state funding. The availability of EUF can relax these constraints and thus partially change the direction of migratory flows in the European space. There are two main channels through which EUF influence these flows, one direct and the other indirect. The direct channel concerns the financing by EUF of forms of welfare, such as health, education, funds for poverty and immigration. In the programming period 2000–2006, the share of EUF dedicated to these activities was almost one quarter. The indirect channel regards the positive effects of EUF on the growth and therefore on the economic well-being of the region, an important factor in attracting migrants. Several papers show that EU Cohesion Policy has had a positive impact on economic performance and on the convergence in income and wealth among European regions (see, among others, Basile, Castellani, & Zanfei, 2008; Becker, Egger, & Von Ehrlich, 2010; Pellegrini, Terribile, Tarola, Muccigrosso, & Busillo, 2013). For example, in Pellegrini et al. (2013) it is estimated that EUF had an average impact between 0.6% and 0.9% per year on the growth rate of the financed regions. Therefore, although EUF are certainly not a factor taken into account by migrants when deciding the region of destination, they do have a possible effect on the factors underlying this choice. Estimating this effect can be of crucial importance because it expands the range of information available to policy-makers on the overall economic and social impact of EUF. The identification of these effects is mainly an empirical question. The analysis is a very complex task, first because we need a causal identification, that separates the effects of EUF from those of the multiple factors affecting migrations, and second because the effects sought could be weak and statistically irrelevant. Our paper deals with the empirical identification of the direct and indirect impact of EU Cohesion Policy on migrations in the EU-15 regions, and the evaluation of their importance and significance, in a counterfactual framework. The impact we are looking for is on the margin (or discontinuity): given two ex ante economically similar regions, the presence or absence of substantial funding due to EUF can affect welfare and regional growth and therefore the differences in the degree of attractiveness for migratory flows. The discontinuity present in the allocation of EUF gives us the opportunity to use a causal model such as the regression discontinuity design (RDD), which is a quasi-experimental method with a very high internal validity (Lee & Lemieux, 2010). Comparing the economic scenario arising under the policy intervention with a counterfactual situation (what would have happened if the policy had not been implemented), this method isolates the impact of the EU Cohesion Policy from the confounding effects induced by other factors (see Mohl & Hagen, 2010), and then empirically evaluates the existence and significance of these effects in a statistically robust way. To this end, we exploit the allocation rule of regional EU transfers: less developed regions, with per capita GDP in purchasing power standards (PPS) below 75% of the EU average, qualify for Objective 1 (Ob.1) status, that is, they receive most of the EUF (called "treated" regions). We assume that non-eligible regions, with a per capita GDP level just above the 75% threshold (called "non-treated" regions), represent a valid counterfactual scenario to those just below the threshold (Cerqua & Pellegrini, 2018). We analyse the impact of EUF on the share of foreigners with or without European citizenship in the EU-15 regions. We also split the estimation by continent of origin. Our empirical results show a statistically significant increase in the share of foreign citizens in Ob.1 regions due to EUF. This impact on the spatial distribution of migration is mostly driven by the increase in the number of non-European immigrants coming from less developed countries. The results are robust to different specifications of the RDD and hold even when we take into account that regions with direct access to the sea or bordering with countries not belonging to the EU-15 might be more vulnerable to the arrival of legal and illegal migrants. Our paper makes a twofold contribution. First, although there is a vast empirical literature on the economic impact of EUF (see the meta-analysis by Dall'Erba & Fang, 2017), the link between EU regional policy and migration is arguably an under-researched topic, and regional data on migration are scarce. The two most relevant exceptions are Kessler, Hansen, and Lessmann (2011), who adopt a model of residential and political choice, and Egger, Eggert, and Larch (2014), whose analysis is based on a new economic geography model. However, the empirical analyses of these papers rely on country-level data, which — considering the regional nature of the policy and the large regional inequalities within most EU countries — lead to a limited explanatory power of the empirical models used on the spatial distribution of migrations. In our empirical analysis, we have created a new dataset containing migration flows among European regions at the NUTS 2 level for the years 2001 and 2011, considering intra- and extra-European migration flows. To the best of our knowledge, this is the first time that a similar dataset has been set up. Second, although the RDD has already been used to investigate the impact of EUF on GDP and employment growth (see, Becker et al., 2010; Pellegrini et al., 2013), as far as we are aware such an empirical approach has never been adopted to assess the causal relationship between EUF and migration flows. Given the increasing share of the EU budget devoted to the Cohesion Policy since the mid-1970s and the dramatic increase in migratory flows to Europe, the policy's contribution to the attraction of migrants is crucial information to EU policy-makers in shaping the regional distribution of EUF. The paper is structured as follows. The EU Cohesion Policy and its potential relationship with migratory flows is presented in Section 2. The data and the evaluation methods are discussed in Section 3. The results of the empirical analysis are presented in Section 4, while Section 5 assesses the robustness of the estimates. The final section concludes the paper. The EU has invested locally through its regional policy since 1975 to improve the competitiveness of slow-growth regions, and correct regional unbalance. The relevance of the EU Cohesion Policy widely increased with the 1988 landmark reform, which followed the accession of Greece, Spain, and Portugal to the EU and the widening of regional disparities within the EU over the previous 15 years. Since then, the importance of the EU regional policy has not ceased to increase, and the reduction of regional disparities in development has become the key goal of Cohesion Policy. Indeed, EUF have been primarily assigned to less developed regions of the EU to compensate them for the absence of some preconditions for growth—infrastructure, accessibility, education, and health care (Camagni & Capello, 2015). Nonetheless, the targets of regional policy have changed over time. The initial priority was given to unemployment, industrial reconversion, and modernisation of agriculture. The priorities have then been extended to include disparities in terms of innovation, level of education, environmental quality, and poverty, as shown by the funding composition under different types of expenditure. From the 1989–1993 to the 2000–2006 programming period, the share of funding allocated to the less developed regions for supporting human capital development (e.g., training, education, and social inclusion) has passed from 20.6% to 24.5%. The share of funding for environmental infrastructures has also increased (from 1.6% to 14%), while that directed to non-environmental infrastructures has decreased (from 36.3% to 30.9%) (European Commission, 2014). Interestingly, the focus on regional disparities has contributed to the development of welfare programmes, which, in turn, might have raised the capacity to attract migratory flows. The focus of the EU regional policy is on the least economically developed regions, the so-called Ob.1 regions (named convergence regions in the 2007–2013 programming period), defined as regions, at the NUTS 2 level, with per capita GDP in PPS below 75% of the EU average. The eligibility threshold is computed years before the beginning of the actual policy interventions. For the programming period 2000–2006, the European Commission computed the eligibility threshold on the basis of the average values observed for the period 1994–1996. The fact that the forcing variable is measured years before the beginning of the actual policy interventions guarantees that the treatment assignment mimics a randomisation process around the threshold. The funding allocated to these regions consisted of almost 70% of total allocations for EUF for the period 2000–2006, amounting to some €136 billion. 22 The remaining part of EUF in the 2000-2006 programming period targeted EU regions that did not comply with the "75 percent eligibility rule", but were suffering from problems of economic backwardness (e.g., industrial decline or structural difficulties, long-term unemployment, rural underdevelopment, and low population density). However, these regions received a much lower regional aid intensity per head with respect to Ob.1 regions (Pellegrini et al., 2013). The subsequent programming periods registered relatively small changes in these allocation rules, which means that this allocation system is still intact (Cerqua & Pellegrini, 2020; Ferrara, McCann, Pellegrini, Stelder, & Terribile, 2017). It is worth noting that the "75 percent eligibility rule" allows the RDD approach to be used as an identification procedure of the policy impact on migration flows. Therefore, regions located just above or below the cut-off point are those suitable for our evaluation work as they share the same characteristics but the treatment. These regions are the focus of our analysis in the following sections. The territorial level of interest in our analysis is the one defined by the EU 2010 nomenclature NUTS 2. To account for net migration, we use census data provided by Eurostat on the composition of the population in 2001 and 2011 in the EU-15 NUTS 2 regions. 33 We do not consider the accession countries (EU-12) of 2004 that did not receive EUF before 2004. This is due to the peculiarities of our evaluation strategy as well as the lack of citizenship data for most EU-12 countries in 2001. This choice allows us to glean significant differences in the composition of the population that can be ascribed, at least in part, to the effect of local policies financed by the EUF, for example, to the increased public facilities supply in the highly-financed regions. Eurostat data provide accurate information on the number of citizens from the reporting country, other EU-countries, other European countries (those that do not belong to the EU), and non-European countries. Using census data allows us to escape from problems regarding the fragmented definition of migrants across EU countries that often constitutes a limiting factor for this kind of analysis (see Westoff & Frejka, 2007). 44 In this study, we use the change in the share of foreign citizenship as a proxy for migration. Although there are other potential proxies for migration (e.g., foreign-born data), in Europe, data on foreign citizenship are more usually employed, and more widely available, when immigration is considered (Coleman, 2003). While data concerning the year 2011 were fully available, for the year 2001 data regarding Greece, Belgium, Germany, and French overseas-departments were not available: for the first two countries, missing data have been retrieved using the figures provided by the Hellenic Statistical Authority (https://panorama.statistics.gr/en/) and Statistics Belgium (http://www.statbel.fgov.be), respectively. Regarding Germany, although the collection of census data in 2001 was not implemented, the German Federal Statistics Office collected data on a micro-census covering 800,000 persons, that is, a sample size of 1% of the German population 55 Data from the micro-census of 2001 allows us to trace out information in the time interval between two population censuses. Due to the reunification of Germany that occurred in 1989, the most recent census before 2011 was taken in 1987. (see Schwarz, 2001). For our analysis, the 2001 micro-census enables us to retrieve an accurate estimate of German population data by nationality at the NUTS 2 level and thus fill in the missing census data. Lastly, data on the four French overseas territories (Guadeloupe, Martinique, Reunion, and Guyane) have been gathered from the Atlas National des Populations Immigrées of each region published by the Institut National de la Statistique et des Etudes Economiques. The availability of data relative to two censuses allows us to calculate the percentage point difference in the share of citizens coming from the reporting country, other European countries and non-European countries as well as the absolute differences for each of the above groups. Such variables represent our main outcome variables. Unfortunately, it is not possible to directly calculate the change in the share of people with other EU citizenship as data in 2001 refer to EU-15, while data in 2011 refer to EU-27. However, we use the share of EU-27 citizens and the absolute number of EU-27 citizens in 2011 as additional outcome variables. In a subsequent analysis, we will then investigate the continent of origin of non-European citizens splitting between Africa, Asia, Central and South America, North America, and Oceania. We have also collected data on several demographic and economic pre-treatment covariates from the regional databases of Cambridge Econometrics and Eurostat, which might be linked to migration decisions. These variables are: the population density, the share of the population over 65, the total employment divided by the active population, 66 This variable can take on values above 1 as it represents the number of individuals working in the region divided by the number of active individuals (in employment or looking for employment) residing in the region. labour productivity (gross value added (GVA) per hour worked), the number of hours worked per employee, the share of employment in the primary sector, and the share of employment in the secondary sector. As for the EUF, we split the EU-15 regions into "treated" and "untreated" following the Ob.1 status assignment process relative to the 2000–2006 programming period. 77 Although our data starts in 2001, i.e., one year after the beginning of the programming period under analysis, this is not a concern as the bulk of EUF are spent in the final years of the programming period, including up to two years after the end of the programming period. We focus the empirical analysis on the split between Ob.1 and non-Ob.1 regions because Ob.1 expenditures account for more than two-thirds of the Cohesion Policy budget. Furthermore, we collected NUTS 2 regional data on EUF payments from 2001 to 2010 (see Roemisch, 2016). This continuous measure of treatment allows us to test the sensitivity of the binary analysis and gain more insights into the causal relationship between EUF and migration flows. Before carrying out the main analysis, we map and plot the population variables to start investigating possible patterns. Figure 1 illustrates the geographic distribution of the regional foreigner share deciles in 2001 (Panel A) and 2011 (Panel B). For instance, in 2001, the first decile of regions consists of the regions with foreigner shares between 0.6% and 1.3%. The regions in the first decile are indicated in light blue. The tenth decile consists of regions with shares between 9.2% and 26.6%. They are shown in dark navy. The intermediate deciles are indicated by intermediate shades of blue. In 2001 the highest shares of foreigners were localised at the core of the EU, in the UK, and Ireland. Ten years later, we observe a somewhat different pattern with a steep increase in the share of foreigners in many Italian, Portuguese, and Spanish regions. Panel C of Figure 1 depicts the percentage point difference between the percentage of foreigners in 2001 and the same variable in 2011, allowing us to identify the differences in migration trends among regions. A striking feature of this figure is the extent to which the top decile regions (the ones marked in dark orange) are concentrated in the periphery of the EU-15. Figure 2 reports the Ob.1 status assignment for the 2000–2006 programming period (Panel A) and the amount of EUF per capita allocated during the 2001–2010 period (Panel B). Both panels show a pattern very similar to the one arising in Figure 1, suggesting the presence of a positive relationship between EUF and the change in the share of foreigners. Indeed, peripheral regions tend to have both a higher share of foreigners and a higher EUF per capita for the period under consideration. This positive relationship is even more evident in Figure 3, where we plot the relationship between the change in the share of foreigners from 2001 to 2011 and the amount of EUF per capita allocated for the 2001–2010 period using a kernel-weighted local polynomial smoothing regression. The green line shows that there is a positive relationship between EUF and the percentage points change in the share of foreigners, which flattens out only at the right-hand side of the fund distributions. Nevertheless, only a rigorous econometric causal model can determine whether this relationship is causal or due to a spurious correlation. With this intent, in the rest of the paper, we compare regions with similar characteristics but which received a much larger share of EUF due to an exogenous assignment rule. We perform an empirical evaluation of the causal relationship between the EU regional policy and migration flows in the EU-15 regions. Our identification strategy exploits the allocation rule of regional EU transfers: only regions with a per capita GDP in PPS below 75% of the EU average (calculated as an average of three years before the beginning of the programming period) are qualified for Ob.1 funds, that is, receive a considerable amount of the EUF. Therefore, regions with a per capita GDP level just above 75% of the EU average (which did not receive the Ob.1 funds) can be considered valid counterfactual comparisons to those just below the cut-off (which did receive the Ob.1 funds). Indeed, economies with similar per capita income levels share many structural attributes, including their levels of education, science and technology endowments, infrastructure quality, and institutional quality (Iammarino, Rodríguez-Pose, & Storper, 2019). This sharp discontinuity can be exploited via the RDD (see, inter alia, Hahn, Todd, & van der Klaauw, 2001; Lee & Lemieux, 2010), an econometric method which compares regions lying closely on either side of the threshold and delivers an estimation of the local average treatment effect (LATE), namely, the average treatment effect around the threshold. For those regions in the interval just above and below this threshold, the treatment assignment (i.e., Ob.1 funds) is to be considered as good as randomised. In other words, the RDD is equivalent to a local random assignment around the cut-off point. Lastly, the use of the RDD overcomes the inverse causality problem, which implicitly affects this kind of analysis. The estimation procedure begins with some graphical evidence. A simple way to evaluate the effect of EUF on migration is to plot the relationship between each dependent variable and the forcing variable (pre-treatment GDP per capita) for regions on either side of the 75% cut-off point. Figure 4 plots each of the primary dependent variables for the period from 2001 to 2011 for Ob.1 regions against non-Ob.1 regions. In each graph, the cut-off line sharply separates the treated and untreated regions. Each figure superimposes the fit of a quadratic regression model (estimated separately on each side of the cut-off point). Figure 4 clearly shows that, on average, the share of foreigners in Ob.1 regions increased more than in non-Ob.1 regions. This change seems to be due to both European and non-European citizens. Although the graphical evidence is important in showing possible differences between treated and untreated regions around the Ob.1 assignment threshold, a formal RDD regression allows us to calculate the extent of the observed differences and whether they are statistically different from zero. The results obtained using the RDD specification detailed in subsection 3.2 are presented in Table 1. Columns (1) and (2) report the EUF impact on the share of citizens with host country citizenship, with and without pre-treatment covariates X. This impact is negative and statistically significant at the 5% level in both specifications. The extent of the difference in the share of citizens with host country citizenship over the period from 2001 to 2011 is −2.23 percentage points in Ob.1 regions. This corresponds to a substantial increase in the share of foreigners as in EU-15 NUTS 2 regions considering that the average share of foreigners was 4.84% in 2001 and 6.45% in 2011. In the remaining columns, we investigate whether this increase is mostly due to foreigners with European citizenship (columns (3) and (4)) or to extra-Europeans (columns (5) and (6)). Although all coefficients are positive, only the one concerning the increase in the share of non-Europeans citizens is statistically significant from zero at the 5% level in both specifications. This means that in the analysed period, EUF brought about a wide expansion in the share of foreign citizens in Ob.1 regions, mostly driven by the increase in non-Europeans. Lastly, although we do not have data on the pre-treatment value for the share of people with other EU citizenship referring to EU-27, in columns (7) and (8), we check whether there are significant differences in the share of individuals with EU-27 citizenship in 2011. We find no statistically significant effect confirming that most of the immigrants come from non-Europeans countries. So far, we have looked at the difference in shares of groups of citizens with different citizenships, that is, host country, European and non-European. Although this analysis is informative in showing the migration trends among regions, it might conceal other patterns. For instance, an increase in the share of foreigners might derive from the migration of citizens from the host country rather than from an actual increase in the number of migrants from other countries. In addition, internal migration of natives can reduce foreign employment shares and thus lead to underestimating the effects of EUF. For these reasons, we complement the empirical analysis by looking at the net migration changes between 2001 and 2011 for each group of individuals under analysis. The estimates are reported in Table 2. The estimates reported in Table 2 show that our findings are mainly driven by the arrivals of foreign migrants—especially those coming from non-European countries—rather than by a reduction of individuals with a host country citizenship. The average increase of over 35,000 individuals from non-European countries for Ob.1 regions is sizable, especially considering that the averag