Immigration and Native Employment. Evidence from Italian Provinces in the Aftermath of the Great Recession∗

This study exploits the variability in the incidence of recent immigration inflows and the change in native employment in the Italian provinces to shed light on the impact of immigration on employment in rigid local labour markets. The study focuses on the period that followed the financial and sovereign debt crises, which strongly hit the labour markets of the Italian provinces. The results reveal a negligible overall impact of immigration on provincial employment which, however, hides differentiated impacts for different groups of natives. Employment responses to immigration shocks vary greatly depending on the skills and gender of the natives. JEL classification: J15, J61, R10, R23.


Introduction
Although immigration from poor to developed countries is a longstanding phenomenon (e.g. Peri 2016), the previously smooth pattern in the flow of immigrants has experienced a sharp rise in the most recent years. This has been particularly so in the case of Europe, that gained some 22 million international migrants between This unexpected and unprecedented influx of immigrants has resulted in an increase in foreign-born workers in the Italian labour force, fueling the argument that immigration has a pernicious impact on native employment. In this regard, theoretical arguments in the literature predict that this impact depends mainly on three aspects. First, whether and to which extent immigrants are complementary or substitute in production to native workers. Immigration is assumed to exert a positive effect on native's employment in the former case (Foged & Peri 2016), whereas natives may instead experience job losses in the latter (Borjas 2003). Second, the skill composition of the immigrant population would play a crucial role.
If immigrant inflows alter the skill distribution of the workforce (because, for example, immigrants are mainly low skilled, as is the case of Italy), the adjustment to restore the pre-immigration equilibrium will imply changes not only in wages, but also in the employment structure (Dustmann et al. 2005). Last, the employment effect of immigration would depend on the amount of rigidities in the labour and product markets, that is, on the institutions of the host country. In brief, wage adjustments to immigration shocks would be more intense in economies with flexible institutions whereas native job losses would be more frequent in countries with restrictive institutions (Angrist & Kugler 2003).
The evidence on the employment impact of immigration is inconclusive. Some studies for the U.S. estimate a detrimental impact of immigration on native employment (Altonji & Card 1991, Anastasopoulos et al. 2018, while others provide evidence in favor of a negligible and even positive impact (Card 2005, Ottaviano et al. 2013, Basso & Peri 2015. In the case of Europe, studies have focused on countries with a long tradition of immigration - Hunt (1992) and Edo (2015) for France; Dustmann et al. (2005) and Lemos & Portes (2014) for the U.K.; D'Amuri et al. (2010), Glitz (2012) and Dustmann et al. (2017) for Germany; Foged & Peri (2016) for Denmark; Basten & Siegenthaler (2018) for Switzerland -, whereas evidence for countries in which immigration is a more recent phenomenon is less abundant-e.g.  for Spain. 2 Overall, it can be said that the evidence from immigration in Europe is also ambiguous, although results in recent studies that exploit variability among European countries point to a more intense employment response in economies with less flexible institutions (Angrist & Kugler 2003, D'Amuri & Peri 2014, Moreno-Galbis & Tritah 2016. In the specific case of Italy, the literature on the employment effect of immigration is surprisingly scarce, particularly when it comes to the impact of the most recent immigration episodes. 3 Venturini (1999) analysed the impact of illegal immigrant workers in Italy on legal employment between 1980 and 1995, concluding that immigrants working without a regular contract crowd-out legal workers in the agricultural sector. On the other hand, she identifies the presence of a complementarity in production between legal workers and illegal foreign workers in the non-tradable sector. Along these lines, Venturini & Villosio (2006) concluded that the presence of regular immigrants in Italy in the period from 1993 to 1997 did not affect the probability of native workers to change their status from employed to unemployed and vice-versa. For a period closer to the one in this study, Labanca (2020) analysed the employment impact of the unexpected migration flows subsequent to the Arab Spring (from 2009 to 2012). His findings suggest that immigrants tended to displace native Italian workers in the short-run, particularly in sectors like mining, hotel and restaurant, and wholesale trade. The effect is instead positive in construction and educational services. 4 Therefore, according to theoretical predictions and existing evidence, the impact of recent immigration episodes in Italy may have been less straightforward than some actors preach and, at the same time, may have affected different population groups differently, depending on their work characteristics and the elasticity of their labour supply. This study tackles this issue by analysing the impact that the recent migratory waves have exerted on the native employment of the Italian provinces, in a context of economic recession and quite rigid labour market institutions.
The Italian case is of particular interest for several reasons. First, mostly because of its position in the centre of the Mediterranean Sea, Italy has become one of the most popular destinations for African migrants since the beginning of the 21st century, while it has also attracted citizens of Central and Eastern European countries since their accession to the European Union in 2004 and 2007 (Hanson & McIntosh 2016, Labanca 2020. Interestingly, part of this period of intense immigration in Italy overlaps with the severe downturn of the Italian economy caused by the global financial crisis and the European sovereign debt crisis. According to OECD (2017), Italy began to recover from the long and deep recession caused by these crises only in 2017. The figures indicate that the Italian real GDP per capita fell by about ten percentage points during the recessionary period. In this regard, 4 It should be noted that Labanca (2020) focused only on illegal immigrants from Egypt, Libya, Tunisia and Yemen, and was interested in the specific sectorial effects. In contrast, our study considers immigrants regardless of their country of origin and pays attention to the heterogeneity of the effect depending on the gender and skills of the native population. the extant literature indicates that immigrants usually have (i) a lower reservation wage; and (ii) different social norms (e.g. Edo 2015). This makes them more likely to be hired, as they are less costly compared to natives. The presence of a recession may further increase this mechanism as firms and capital owners often need to face budget cuts. Therefore, the potentially negative employment effect of immigration may be further exacerbated by the economic downturn. In this scenario of deep recession, one of the sectors more harshly affected was the labour market. To this extent, it is important to note that the Italian labour market has been characterised by a high degree of employment protection and downward wage rigidity. 5 In contrast to countries like the U.S. and U.K., wage flexibility in Italy is constrained by the system of wage bargaining, which is highly centralized at the national level (e.g. D' Amuri et al. 2010). This makes wages less sensitive to labour supply (or demand) shocks, provoking adjustments via changes in employment. This feature clearly plays a crucial role in the extent to which local labour markets are able to absorb an immigration-induced supply shock. Hence, it is motivating exploring how local employment reacted to sizeable immigration flows in a context of economic recession in a country characterized by a far from flexible labour market.
Second, the distribution of immigrants in the Italian geography is far from uniform. In fact, spatial disparities in the immigrant population resembles the ones frequently reported for other socio-economic variables in Italy (González 2011). This is consistent with the attraction of immigrants to places that offer greater economic opportunities. Immigrants were about 10% of the population in the northern and central parts of the country in 2017, whereas they were just about 4% in the South. In terms of the total number of immigrants in Italy that year, 58% of them concentrated in the northern regions, about a quarter in the central regions, while only 16% in the Southern ones. It is worth noting that there were also sizeable 5 Italy is among the countries with the highest values of the OECD strictness of the employment protection index (http://www.oecd.org/employment/emp/ oecdindicatorsofemploymentprotection.htm) and the ILO employment protection legislation index (https://www.ilo.org/dyn/eplex). Similarly, the value of the index of flexibility of wage determination produced by the World Economic Forum (https://tcdata360.worldbank.org) reveals that the degree of centralization of wage bargaining in Italy is comparable to that in France, Germany, and Spain, but far above that in Denmark, U.K. and the U.S. differences in these figures between provinces within regions. For example, in Lombardia, a region in the North, the proportion of immigrants in the population was 5.1% in the province of Sondrio and 13.9% in that of Milano, whereas in Puglia, in the South of Italy, the rate was 2.2% in Taranto and twice this figure in Foggia (4.5%). 6 The great spatial variation in economic performance and resilience to recessions (Faggian et al. 2018), jointly with the heterogeneous distribution of immigrants that characterizes the Italian provinces, provide an excellent framework to evaluate the reaction of native employment to a migration-induced shock to local labour markets during a period of severe economic downturn.
Finally, the issue of immigration has reached a central position in the political debate in Italy today (Mayda et al. 2018). 7 There is a fairly widespread belief that the labour market outcomes of the Italian-born have worsened as a result of increased labour competition brought by immigrants (Mayda 2006). In fact, the perception that immigration hinders the employment of natives has gained momentum, amplifying anti-immigration messages. To this extent, if in the past the Italian public opinion was split into two opposing factions, partisans and opponents (Gavosto et al. 1999), more recently the latter have somewhat "taken over" the former. 8 Against this background, this study tests whether the recent influx of immigrants to local labour markets in Italy has really had a negative impact on the employment of natives. By using data drawn from the Italian Labour Force Survey (LFS) and the Demographic Portal of the Italian Office for Statistics (DP-ISTAT) during the period 2009-2017, we apply the so-called "spatial correlation approach" (Borjas 2014) to estimate the effect of a change in the immigrant population on the change in native employment in the Italian provinces. In the first place, this effect is estimated for the overall native population. Then, differences between groups of 6 See Figure SM1 in the OSM for additional information on the distribution of immigrants in the Italian provinces. 7 In the political elections of March 2018, one of the parties that won most public support was the Northern League, well-known for its anti-immigration rhetoric.
8 As an example, the 2017 Special Eurobarometer (number 469) reveals that the Italian population greatly overestimates the proportion of immigrants in the country's population. Similarly, the evidence derived from the Standard Eurobarometer surveys indicates that immigration and unemployment are among the problems that most concern the Italian population. natives are explored. To be clear, the study hypothesizes that the complementarysubstitutability relationship between immigrant and Italian-born workers varies for high and low educated natives and by gender. In all cases, in order to identify the causal link that connects immigrants to native employment, we control for potential labour demand shocks and compensating adjustments through internal migration that could confound the estimate of the impact of immigration.
The empirical model also controls for unobserved province effects and provincespecific trends in native employment. While the former seeks to account for the large heterogeneity of labour markets between the Italian provinces in general, and their employment levels in particular, the latter aims to capture differences among provinces in the path followed by employment after the crisis. In addition, an IV estimator is implemented to account for the likely endogeneity caused by the non-random sorting of immigrants into local labour markets. To isolate the supply-driven shocks associated to immigration in each provincial labour market, the empirical analysis uses the so-called "shift-share" instrument, that combines information about the pre-sample settlements of immigrants in the Italian provinces and the evolution in the number of immigrants by country of origin in the whole of Italy.
In contrast with the idea that immigrants indiscriminately "take away jobs from natives", the evidence in this study points to an overall negligible impact of immigration on native employment in the representative Italian province. However, when considering the effect on specific groups of natives the results reveal a positive impact on high-educated and a negligible one on low-educated individuals. When using occupations instead of formal education to distinguish native workers with different labour market skills, the results identify a positive, albeit marginally significant, effect on skilled manual native workers, whereas a negligible impact is observed for workers in occupations that require lower skills and for white collar workers. These results partly deviate from the canonical theoretical model of immigration (Boeri & Van Ours 2008) according to which immigrants -that in Italy are mostly low-skilled (see Bratti & Conti 2018) -act as comple-mentary with high-skilled natives but compete for jobs with low-skilled ones. The evidence in this study suggests that in the Italian provinces, during the period under analysis, immigrants would have been less substitutes with low-skilled natives than estimated by Ottaviano & Peri (2012) for the U.S. and by Romiti (2011) for Italy in the period from 1987 to 2004. Interestingly, the results that distinguish by gender indicate that the employment of native males was not significantly affected by the immigration shocks of the 2009-2017 period. This is so regardless of their skills. In contrast, the employment of native females in the representative Italian province would have been stimulated by immigration, particularly in the case of women with high-education and working in skilled manual occupations.
The rest of the paper is structured as follows. Section 2 outlines the empirical model and discusses the identification strategy, while section 3 introduces the dataset and provides preliminary descriptive evidence on the relationship between immigration and native employment. The main results are presented and discussed in Section 4, distinguishing between the overall effect of immigration and the specific effects for groups of natives formed according to their skills and gender.

The Spatial Correlation Approach
Our interest in this study is the estimation of the short-run impact of immigration on the native employment of the representative (average) Italian province in the period under analysis. To this aim, we follow the so-called "Spatial Correlation Approach", pioneered by Altonji & Card (1991), which exploits the fact that different places generally experience non-homogeneous immigrants' inflows (in terms of the number of people entering each particular labour market). 9 The uneven spatial distribution of foreign-born individuals in Italy represents an interesting source of variation that can be exploited to estimate the impact of the recent immigration flows on native employment. As in Card (1990), Hunt (1992), Basso & Peri (2015), Foged & Peri (2016), Borjas (2017), we estimate the effect of interest from a specification where the employment of the total native population in a province is assumed to depend on the total number of immigrants in that province. Unlike the "National Skill-Cell Approach", the one based on regional variations allows to identify the "overall" impact of immigration rather than the "relative" effect that immigrants exert on the most similar natives, i.e. belonging to the same educationexperience group (Dustmann et al. 2016). In addition, in contrast to the approach based on the skill-cells, the spatial strategy does not impose the assumption that immigrants and natives are homogeneous in terms of their observable levels of education and experience. Indeed, the evidence indicates that immigrants experienced the so-called skill-downgrading (Dustmann & Preston 2012), leading to an incorrect classification of immigrants in the education-experience groups that, in turn, biases the estimated impact of immigration.
Based on Card & Peri (2016), the baseline specification used to estimate the overall impact of immigration on native employment in the Italian provinces is: 10 where ∆(N p,t ) = (N p,t − N p,t−1 ) and ∆(M p,t ) = (M p,t − M p,t−1 ) are the changes in, respectively, native employment and the number of working-age foreign-born individuals in province p between years t and t − 1. L p,t−1 , the working-age population of province p in t − 1, accounts for the size of the province labour market in t − 1. Therefore, the outcome variable is the yearly change in native employment of province p relative to the size of its labour market in t − 1, whereas the regressor proxies for the relative flow of immigrants in each province and year. The specification includes time dummies, ψ t , to account for country-wide year-specific 10 Borjas (2003) suggested an alternative specification to analyze the impact of immigration on native labour market outcomes, that has been used by several subsequent studies. However, we have not followed this approach to minimize the risk and consequences of spurious correlation between the variables of interest in this study (as pointed out by Card & Peri 2016).
shocks in employment, and province fixed-effects, µ p , that control for provincespecific trends. 11 Specific trends induced by provincial differences in the impact of the crisis on employment during the period analysed is a potential source of heterogeneity that must be taken into account to properly identify the impact of immigration on native's employment (Wooldridge 2002). 12 Finally, ν p,t is an i.i.d.
random term with zero mean and variance σ 2 ν .
The coefficient of interest, β, measures the short-run response of relative native employment in the representative Italian province associated to an increase of immigrants in the province of one percent point of its working-age population. As mentioned above, the aggregate spatial approach internalizes the possible spillover effects between different education-experience groups and therefore identifies the overall effect of immigration on native employment of the representative Italian province over the period analysed.

Identification of the Effect of Immigration
In the specification in (1), the relative change in native employment is regressed on the relative change in province immigration to get rid of the unobserved timeinvariant differences between local labour markets that may confound the estimate of the impact of immigrant inflows. The specification in relative changes, therefore, controls for the correlation between the two variables of interest that may be due to permanent or persistent local economic conditions driving both the foreignborn population and the level of local employment. The analysis also accounts for province-specific trends in native employment (province fixed-effects in the specification in the changes of the variables) which is another source of province heterogeneity that could confound the estimate of the effect of immigration. But besides the unobserved local economic conditions, the empirical model must account for an important feature that characterizes the performance of local labour markets and, therefore, affects both natives and immigrants, namely the evolution 11 It should be noted that the province fixed-effects, µ p , result from the differentiation of the specification in levels that includes province-specific trends. 12 We thank an anonymous reviewer for raising this point.
of the industry in which they are employed (e.g. Acemoglu & Autor 2011, Basso & Peri 2015. The period analysed was characterized by the turbulences caused by the global financial crisis and the European sovereign debt crisis on the Italian economy. It is sensible thinking that the particular reaction of the Italian local economies in general and their labour markets in particular depended heavily on the productivity changes of the industries in which they are specialized. If so, the lack of control of the productivity changes that different industries experienced in the period analysed will lead to confounding the estimation of the effect of immigration. Therefore, in order to identify the specific impact induced by immigration flows on the change in native employment, we include in equation (1) the "quantity version" of the so-called Bartik shock (Bartik 1991), defined as: 13 and E j,p,to /E p,to is the employment share of industry j in province p in the initial year t 0 , E IT j,t is the employment of industry j in Italy in year t, and thus the second term in the left-hand side is the annual growth of employment in industry j in the country. The Bartik shock captures changes in province labour demand that are sector-driven, and could hinder the identification of the effect of immigration.
A well-known criticism of the spatial correlation approach is that local labour markets are not closed economies (e.g. Borjas 1999Borjas , 2006. This means that there may be compensatory flows if some natives move to other locations as a reaction to the changes in wages and employment opportunities induced by immigrant inflows. Under this scenario, an analysis conducted at the local level could indicate a weak (or even absent) correlation between immigrants and native labour market outcomes, not because foreign-born individuals are not actually harmful for the employment perspectives of the natives, but because internal migration diffuses the effect of the immigration shock to other local labour markets. To counter the concern of compensatory flows, studies using the spatial correlation approach have claimed that immigrants do not induce significant migratory responses by natives (e.g. Card & DiNardo 2000, Peri & Sparber 2011. In the specific case of Italy, Venturini & Villosio (2006) and Mocetti & Porello (2010) found that immigration has a trivial impact on overall native internal mobility, albeit there could be some compensatory responses of the low-educated and highly-educated natives that would alter the skill composition of the regional labour markets. 14 Although we believe that the annual changes considered in this study do not leave much room for labour market adjustments through compensatory population flows, we add to the baseline specification in (1) a control of internal migration, namely the net migration rate: where I p,t is the number of people immigrating into province p at year t, O p,t the number of people emigrating out of the same province in the same year, and L p,t−1 is as defined above. 15 As a result, the extended specification is as follows: As in the case of the baseline specification in (1), the coefficient of interest is β which captures the short-run impact of immigration on native employment, net of productivity-induced changes in local labour demand, internal compensatory flows, province unobserved heterogeneity, and province-specific trends.
However, it is well known that the identification of the causal effect of immigration based on the specification in equation (2) faces another problem, namely that immigrants' location decisions are not randomly taken. In brief, we will observe a spurious positive correlation between the change in native employment and the inflow of immigrants if the latter tended to settle in provinces with positive, or less negative, demand-driven shocks during the period analysed. A common way to solve this bias in the estimate of the causal effect of immigration is using an instrumental variable approach. Following the path set by Altonji & Card (1991) and Card (2001), we use an instrument that proxies the labour supply-driven shocks of the immigrants' inflow. The main rationale behind this instrument is that immigrants tend to settle in locations characterized by the presence of other individuals coming from the same country of origin (e.g. Bartel 1989). The number of foreigners from a country in province p at year t is assumed to be connected with the past number of immigrants from this country in the province but unrelated to the current shocks that affect the local economy. A shift-share type of instrument for and IT b p,t−1 refers to the Italian-born population in working-age in province p at year t − 1. 16 The subscript o denotes the immigrants' countries of origin and t 0 a baseline year that must be distant enough from the years in which the change in native employment is measured to guarantee the unrelatedness to current shocks.
The validity of the instrument requires that, conditional to the controls and unobserved province heterogeneity considered in equation (2), the distribution of immigrants by country of origin in the Italian provinces in the baseline year does not correlate with province-specific demand changes in native employment in the period analysed. As in Bratti & Conti (2018) the baseline year is set to 1995, which is well before the onset of the financial and sovereign debt crises that strongly affected the Italian economy in the period under analysis. This year also predates the accession to the European Union of the member states from central and eastern Europe that spurred a substantial inflow of immigrants from these countries to Italy, as well as the migratory shock that followed the Arab spring. These two facts work in favour of the validity of the instrument, since it is sensible arguing that province-specific labour market shocks that affected the distribution of immigrants in 1995 do not strongly correlate with changes observed in employment from 2009 on (see Goldsmith-Pinkham et al. 2020). Even in the case of strong persistence in the province-specific shocks, their effect on employment changes about a decade and a half later should be largely captured by the controls (particularly the Bartik variable) and the elements associated to unobserved heterogeneity of the province included in equation (2). On the other hand, the validity of the instruments also requires that the overall inflow of immigrants from each country of origin to Italy does not correlate with shocks exerting an impact on employment changes in the province. Considering the prevalence of immigrants from the above-mentioned origins, push factors associated to the internal situation of the places of birth probably had more influence on migration decisions that pull factors motivated by the economic performance of the Italian provinces.
In any case, some evidence will be provided in section 4 to mitigate concerns about the exogeneity of the instrument.

Data and Descriptive Analysis
In this section we provide information on the data sources used to construct the variables introduced in the previous section. It also presents the results of a descriptive analysis that sheds some preliminary light on the relationship between the flow of immigrants and the evolution of native employment in the Italian provinces over the period 2009-2017.

Data Source
Population censuses are the data sources commonly used in the existing literature on the economics of immigration. However, such information is not available It is worth mentioning that, since the objective of the paper is to assess the effect of immigration on native employment, we consider only the working-age population, for both natives and immigrants. In Italy, the minimum legal workingage is 15 years, so we include in the analysis individuals from 15 to 64 years of age. On the other hand, it should be said that the main results in the paper are obtained for the Italian provinces (NUTS 3 regions in Italy), which is the territorial breakdown closer to the concept of local labour markets for which the required data for the analysis can be computed. Due to some changes in the configuration of the set of provinces in the period analysed, we had to make some adjustments that 17 See http://demo.istat.it/index_e.html. Suitable data on all foreign-born individuals, either non-citizens or naturalized, is not available for the Italian provinces. As pointed out by an anonymous reviewer, this may rise concerns due to the endogeneity of immigrants' naturalization. However, it should be taken into account that our empirical exercise focuses on annual changes of the variables of interest. In this case, as emphasized by Angrist & Kugler (2003), the group of non-nationals largely overlaps the group of recently-arrived foreign born.

Results
This section summarizes the results of the estimation of the effect of immigration on native employment from the specifications sketched in section 2 using the data for the Italian provinces described above. The impact on the overall native population is discussed first. Next, we explore differences in the effect of immigration on different groups of native workers, defined based on their skills and gender.
Weighted regressions, using as weights the total number of working-age individuals in the province at the beginning of the period, are used in all cases, while standard errors are clustered at the province level.

Overall impact of immigration
The ordinary least squares estimation (OLS) of the impact of immigration on native employment from the baseline specification in equation (1)  This positive response of native employment was also derived from a group of EU countries that includes Italy for the period 1998-2004 by Moreno-Galbis & Tritah (2016). In addition to the complementarity between immigrants and natives, these authors argued that immigrants could be exerting a positive exter-nality since they are more profitable workers (the productivity-wage gap is wider for immigrants than for natives). Increase in profits would lead firms to open more vacancies, which would improve the employment prospects of the natives.
However, this type of adjustment is more likely in host countries with flexible institutions, which is not the case of Italy in the period analysed. In fact, the positive estimate of the effect of immigration on native employment could be due to the bias of the OLS estimator if the location decisions of immigrants in the period analysed were not random, and immigrants moved to those provinces less affected by the recession. The inclusion of the Bartik variable aims to control at least part of this problem, particularly with regard to industry-specific shocks that affected labour demand in each province depending on its sectoral composition. Still, to address this potential source of bias we use the immigrants shift-share instrument to implement an IV estimator for the extended specification.
Before presenting the results obtained with this estimator, it is worthwhile to show some evidence supporting the validity of the instrument. As mentioned in section 2, it can be argued that highly persistent province-specific shocks induce correlation between the provincial distribution of immigrants in 1995 (year used to compute the share component of the predicted amount of immigrants in each province-year) and shocks to changes in provincial employment during the period 2009 to 2017. In that case, the instrument and changes in native employment will be spuriously correlated, leading to violation of the exclusion restriction. To rule this concern out, in the spirit of provinces before 2009 prevents us to carry out such type of test. As an alternative, we can test the validity of the instrument computed for the second part of the period only (2013-2017) relative to changes in native employment for the first part (2009)(2010)(2011)(2012)(2013). A significant correlation between them would point to strong serial correlation between earlier demand shocks and later values of the instrument (changes in predicted share of immigrants), casting doubt on its validity. The result clearly suggests that this is not the case, since the correlation between the change in employment in the first half of the period and the instrument in the second is not statistically different from zero. 25 Therefore, these results, jointly with the arguments provided in section 2, support the validity of the instrument based on the immigration enclaves in 1995.
The results of the first-stage regression are summarised in Table A2 Table 1. It can be observed that after controlling for the endogeneity of immigration shocks, the estimated effect on native employment is still positive although not statistically different from zero. It is important to notice that although the IV estimate of the effect of immigration is slightly higher than that obtained by the OLS estimator, it is estimated with less precision, leading to a non-significant effect. Therefore, this result points to This conclusion about the overall impact of immigration on native employment in the Italian provinces in the period analysed is robust to a set of alternative specifications and samples, as reported in columns (4) to (11) of Table 1. 26 In particular, 25 The value of the parameter estimated in the simple regression between the two variables is 0.011, with s.e.=0.026. 26 We thank three anonymous reviewers for suggesting several of the robustness checks.
column (4) shows that the estimated effect is not driven by the use of province population weights, since similar results are obtained with the unweighted IV estimator. In turn, results in column (5) correspond to the specification that adds two lags of the change of the immigrants indicator. Following Jaeger et al. (2019), in this way we aim to capture the dynamic response of employment to immigration, due to adjustments to past immigration shocks. In brief, these authors argue that the instrument will be correlated with ongoing responses to previous shocks if the provincial distribution of the inflow of immigrants remains stable during the analysed period. As a result, the IV estimator based on the shift-share instrument will not identify the short-run effect of immigration on employment but a mixture of the short-and long-run effects. We also check the robustness of the estimated short-run impact of immigration to the exclusion of internal migration. This is important inasmuch as it can be argued that this variable is a bad control given that, as long as it is determined by immigration shocks, it is an outcome variable rather than a valid control. The results in column (6)  The result of the first-stage F-statistic in the dynamic specification clearly rejects that the instrument and its lags are jointly weak. As indicated in Jaeger et al. (2019), evidence on weak instruments is obtained when there are no substantial changes over the period analysed in the composition by country of origin of national inflows, meaning that the instruments will be highly correlated. In this regard, it is worthwhile noting that there is no significant serial correlation neither in the instrument nor in the immigration regressor used in this study. On the other hand, it should be mentioned that similar results were obtained with a different (reasonable) number of lags. low (∆Π IL ) educated natives. These controls aim to capture differences in the propensity to migrate of different population groups that affect the composition of the population. As shown in column (7), the estimated effect of immigration in this alternative specification is somewhat lower and remains not statistically significant. The main conclusion on this effect is also not affected by the inclusion of the proportion of workers in each skill group. Although results in column (8) confirm that the shares of skilled manual (E skm ) and white collar (E white ) workers affect positively changes in native employment, their inclusion as controls does not modify the conclusion about the impact of immigration. Furthermore, its impact is also not affected by the inclusion of an additional control that aims to account for the effect of agglomeration economies. To be clear, the results in column (9) correspond to the estimated coefficients in a specification that includes the lagged annual growth of population density (i.e. ∆ϑ p,t−1 ) interacted with year dummies to allow for the effect of agglomeration to vary across years. 28 Finally, we check the robustness of the estimated impact of immigration to the exclusion from the analysis of the largest provinces (column 10) and immigrants from EU 15 countries others than Italy (column 11). In the first case, removing provinces with more than 2 million inhabitants (Milano, Rome and Naples) leads to an increase in the estimated effect of immigration. However, it is not statistically significant as there is also a decrease in the precision with which the parameter is estimated. Therefore, there is no evidence that the estimated effect reported in column (3)  Overall, these robustness checks confirm that there was a positive although not significant effect of immigration on native employment in the Italian provinces in 28 We thank an anonymous reviewer for suggesting the inclusion of the interaction between a measure of agglomeration and year dummies as a robustness check. The results in the next subsection are obtained without the agglomeration controls since their inclusion does not modify the estimated effect of immigration and due to our concern about the endogeneity of population density, the treatment of which is beyond the scope of the current study. Undoubtedly, this is an issue that deserves further attention in a specific study.

Heterogeneity in the impact of immigration
So far, we have considered all native-born workers in a single group, as if they were homogeneous workers and were similarly affected by immigration shocks, regardless of their job characteristics. Nevertheless, both theoretical arguments and empirical evidence seem to contradict this hypothesis (Kerr & Kerr 2011, Borjas 2014, Dustmann et al. 2016. In brief, immigrants can act as complementary for a part of the native population, specifically the highly educated (Chassamboulli & Palivos 2013, Dustmann et al. 2017) and as substitute for natives with low levels of education (Altonji & Card 1991, Dustmann et al. 2017. If so, the negligible overall estimated effect could be masking significant opposite effects for native workers with different skills, which cancel out in the aggregate. Therefore, following the advice of Dustmann et al. (2016), we investigate the effect of immigration shocks on the employment of several native groups. Specifically, to assess the impact of immigration on natives with different skills, we classify them into two groups based on their level of education: one formed by native workers with primary and secondary education and another composed of natives with a university degree and higher stages of tertiary education. According to the low level of education of immigrants in Italy during the period analysed (e.g. Fullin & Reyneri 2011, Bratti & Conti 2018 and the imperfect transferability of the education that they acquired in their countries of origin, i.e. skill-downgrading (e.g. Dustmann & Preston 2012), the former group of natives would have been more exposed to immigrants' competition. Also, the elasticity of labour supply is likely to differ between the two groups, leading to different employment responses to the immigration shocks (Dustmann et al. 2016). Finally, natives with different levels of education could be subject to downward wage rigidities with different intensity (e.g. depending on the type of contract -Edo 2015). Therefore, the extended specification in equation (2) is estimated by IV for the samples defined by these two categories of workers to identify the specific effect of immigration on the employment of native workers of high and low education. 29 The results are summarized in the first two columns in Table 2 is consistent with a situation in which immigrants are employed in occupations different from the ones undertaken by natives (even if similarly skilled, see Ottaviano & Peri 2012), that are typically manual intensive (Peri & Sparber 2009, Foged & Peri 2016. Therefore, in this scenario immigrants (i) do not directly compete with natives and (ii) induce natives to upgrade their jobs, moving to more communication intensive tasks, for which they have a comparative advantage vis-à-vis immigrants (Peri & Sparber 2009, Giuntella 2012).
However, the interpretation of these results should take into account that "overeducation" is a characteristic of labour markets in different countries in southern Europe, including Italy (e.g. Flisi et al. 2014). In short, the considerable proportion of native workers in the Italian provinces employed in occupations that required less education than they had can somehow affect the conclusions derived for the groups of workers with different levels of educational attainment. To overcome this drawback, under the usual assumption that occupations differ in term of the required skills, we complement the analysis with the results of the grouping of the native population according to occupations. In particular, we classify native workers into three categories of occupations, from more to less skilled: white collar, 29 The results of this section using the OLS estimator are reported in Tables SM4 and SM5 of  the OSM. 30 The canonical theoretical model of immigration predicts a negative effect of low-skilled immigrants on the employment prospects of their natives counterparts (Boeri & Van Ours 2008). skilled manual, and blue collar. The results are shown in the last block of columns in Table 2. Although the estimated effect of immigration is not significant in any of the skill groups, the point estimate for skilled manual natives is much higher than that of the other two groups, being almost significant (at the 10% level). Therefore, these results suggest that, in the aftermath of the Great Recession, immigration may have stimulated, on average, the employment of skilled manual natives, while having no effect on white collar and low skilled manual workers. However, there could have been a wide dispersion in impact even for skilled workers which, as shown below, can be explained by different responses of female and male natives.
As a final stage in the analysis, we explore heterogeneous responses in native employment by gender. The reason is that some studies have suggested that the impact that immigrants exert when entering the host country's labour market might affect in a different way male and female natives (Barone & Mocetti 2011, Farré et al. 2011, Forlani et al. 2015. This could be particularly important in the case of the Italian labour market, characterized by striking gender disparities. For example, the employment rate of the working-age population in Italy in 2017 was 67.1% for males but only 48.9% for females. However, the gender gap narrowed in the case of workers with tertiary education. In this case the employment rates were 83.1% and 74.7%, respectively. 31 In this context, the complementarity/substitutability mechanisms may have worked differently for native male and female workers. For example, Barone & Mocetti (2011) argued that the high presence of immigrants providing household services is associated with an increase of the hours worked by the high-skilled native females. A gender heterogeneous response could indeed be behind the low precision with which the effect of immigration is estimated for the overall population of natives and, particularly, for the group of skilled manual workers.
For these reasons, in Table 3 we report not only the estimated effect of immigration on the employment of female and male natives, but also that obtained by distinguishing between female and male workers of different levels of education and occupations. First, it is observed that the estimate of the overall effect for female natives is positive and significant, whereas it is not statistically significant and, indeed, very close to zero in the case of their male counterparts. Second, the distinction by skills reveals interesting differences in the reaction of female and male employment to immigration shocks. The results confirm complementarity with the highly educated natives. More precisely, they suggest a positive and significant effect for the high-educated females of around 0.20. For males with high education the estimated effect is somewhat smaller in magnitude (0.18), being estimated with much less precision. In fact, for this group of native workers the estimated effect is not significant from a statistical point of view. Interestingly, the results do not support the claims that immigrants hinder employment of the low-educated Italians. The effect estimated for the native males with low education is negative, which is consistent with certain degree of substitutability between low-skilled immigrants and their native males counterparts. However, the coefficient for this group of workers is not statistically significant. The effect of immigration is also not significant for low educated native females, although in this case the point estimate is positive.
The positive impact on employment for females (both high-and low-skilled) may be explained by the fact that immigrants, particularly females (whose share over the total immigrant population in Italy is higher than that of males in the period under analysis 32 ) tended to substitute native females in the household production services (Cortés & Tessada 2011, Farré et al. 2011) therefore allowing the latter to increase their labor force participation. Consistent with the evidence in Forlani et al. (2015) from a group of developed countries, the evidence from the Italian provinces during the period analysed in this study supports a positive impact of immigration on the employment of native females, particularly for highly skilled women.
The evidence from the occupational groups that distinguish between male and female natives (columns 4 to 6 of Table 3) confirms the lack of a substitution effect of immigration on native employment, regardless of the skills of the lat-32 For more details, refer to Table SM1 of the OSM. ter. In fact, the results point to the complementarity between immigration and the employment of native females. To be clear, the point estimate of the impact of immigration on native male employment is quite modest (and statistically insignificant) in all occupation groups. By contrast, the impact on females in skilled manual occupations would have been significant and sizeable.
Summing up, the evidence from Italian provinces in the period that followed the financial and sovereign debt crises confirms that native employment reacted differently to immigration shocks depending on skills and gender. On the one hand, there would have been a positive response in the employment of highlyskilled natives, particularly in the case of females. On the other, recent immigrant inflows to the Italian provinces would have not hinder significantly the employment of low-skilled natives.

Conclusions
This study has provided evidence on the short-run effect of recent immigration shocks on native employment in the Italian provinces. The results are particularly interesting because they have been obtained for a country where immigration is a relatively recent phenomenon, the inflow of immigrants in recent years has been particularly intense, and their geographical distribution has been far from uniform.
Besides, in contrast to most previous studies, this one has considered a period of economic recession that strongly hit the labour markets of Italian provinces and, in particular, the employment prospects of their native populations. The overlapping of large immigration inflows and job losses would have contributed to fuelling a passionate anti-migratory rhetoric. Interestingly, the study has estimated differentiated employment responses for separate groups of natives to account for heterogenous impacts of the immigration shocks depending on characteristics of the natives and their labour supply elasticities. All of this for an economy with intense downward wage rigidity; a feature that has been shown to favour employment adjustments to immigration shocks.
Once local labour demand shocks, internal compensatory flows, province-specific trends in employment and, especially, the non-random spatial distribution of immigrants are controlled for, the results point to a negligible adjustment of provincial native employment to the immigration shocks over the period analysed. However, the study shows that this overall impact of immigration on employment hides interesting heterogeneous responses of different groups of natives. To be clear, the evidence from the Italian provinces since the onset of the Great Recession confirms that the employment response varied according to the skills of the natives. While immigration inflows would have stimulated employment of the highly-educated natives, the impact on the low-educated would have been negligible. Therefore, although most immigrants that arrived in Italy in the period under analysed were low-educated and that they probably experienced to some extent the skill-downgrading, it does not seem that immigrants substitute native of similar skills. Labour market rigidities that make difficult and costly to fire workers to replace them with newcomers could partly explain this result.
Interestingly, the study has revealed that the distinction by gender is crucial.
There are no signs that clearly indicate a pernicious effect of immigration inflows on the employment of native males in the aftermath of the Great Recession. In sharp contrast, the evidence points to a positive response of female employment, that would be particularly significant in the case of native women with high education and employed in skilled manual occupations. This result is consistent with immigrants substituting native females in housekeeping and child and elderly care, which leads them to increase participation in the labour market. In this regard, the results in this study are particularly important from a policy perspective due to the still substantial gender disparities in the participation rate that characterize the Italian labour market and the differences in female participation between the Italian local economies.
Finally, we must admit some shortcomings of this empirical exercise. For example, the analysis focused only on the partial (short-run) employment effect of immigration, although responses in the longer-term involving different mechanisms (impact on productivity, investments in education of the natives, innovation, etc.) can also be of great importance. On the other hand, it could be argued that the annual changes considered in this study prevent controlling for highly persistent dynamic effects even in the case of adopting the approach suggested by Jaeger et al. (2019). Last, it should be noted that the study have just considered the employment effect of immigration at the extensive margin, while responses at the intensive margin could also be relevant. In any case, based on arguments and evidence in the extant literature, we can speculate that these additional sources of influence of the immigration shocks in Italy in the aftermath of the Great Recession would have probably contributed to enhance the employment prospects of the Italian-born beyond the short-run effect estimated in this study. This, therefore, would contradict one of the most powerful arguments of rhetoric against immigration in Italy and in other European countries that have experienced similar immigration episodes in the recent past. Note: The dependent variable is the change in native employment as share of the initial working-age population, while the main independent one is the change in immigrant population as share of the initial working-age population. Columns (3) to (11) refer to the 2SLS using as instrument the change in the shift-share variable based on the residence permits issued in 1995. All regressions, except the one of column (4), are weighted by the total number of working-age individuals in the province at the beginning of the period. The R-squared reported in columns (3) to (11)     Note: Variables are expressed in changes over the entire period and are cleaned from the time average. The size of the circle is proportional to the initial working-age population in the province.    iv   Note: OLS estimates of Table 3. * * * p < 0.01, * * p < 0.05, * p < 0.1 vi Maps Figure SM1: Immigration in the Italian provinces as percentage of the province population (2017).  viii

Provinces
In order to have a homogeneous dataset over the period analyzed and due to changes in the definition of some provinces, we have merged together the following provinces: • Monza e della Brianza with Milano.

Industries
In order to homogenize the data relative to the industries classification, we have constructed 46 new industries that are defined as follows: 33 In the LFS of the first quarter of 2017 these two provinces are merged together under the name "Sud Sardegna".
ix x