Not Just a Work Permit: EU Citizenship and the Consumption Behavior of Documented and Undocumented Immigrants

This paper explores the impact of the 2007 EU enlargement on the consumption behavior of immigrant households. Using data from a unique Italian survey and a diff-in-diff approach, we find that the enlargement induced a consumption increase for immigrants from new member states. This effect concerned both undocumented and documented immigrants, albeit through different channels. Detailed information on immigrants' legal status and sector of employment allows us to shed light on the exact mechanisms. Following the enlargement, previously undocumented immigrants experienced increases in labor income by moving away from the informal sector, whereas immigrants who were already working legally in Italy benefitted from an increased probability of getting permanent contracts. Enhanced employment stability in turn reduced uncertainty, leading to an increase in documented immigrants' consumption.


Introduction 1
Despite the large literature on the economic analysis of immigration, little is known about immigrant households'consumption behaviour in the host economy. In principle, a high level of uncertainty can depress the economic activity of households, including their consumption. When economic decisions are costly to revert, high uncertainty may induce individuals to postpone their decisions until uncertainty is su¢ ciently resolved and more information is available (Bernanke, 1983). Immigrants tend to face higher economic uncertainty than natives, which may a¤ect their consumption behavior. 2 On the one hand, undocumented immigrants are constantly at risk of being apprehended and subsequently deported, and when employed, they work in the informal economy and earn lower salaries (Dell' Aringa and Neri, 1987). Legalization procedures di¤er by country but are costly and burdensome in general. 3 On the other hand, documented immigrants are allowed to stay in the host country for a prespeci…ed period of time and are obliged to leave when their permit expires. Permits can be renewable but this is usually subject to ful…lling certain conditions such as earning high-enough income and/or not entailing in any criminal activity.
One of the fundamental principles of the European Union (EU) enables immigrants from the member states to live and work in the EU without the need of a work permit, and grants them the right to equal treatment with natives in employment, wages, and working conditions. 4 Thus, the EU accession plausibly implies an improvement in the employment opportunities of citizens from new member states, while reducing the degree of uncertainty and the precautionary savings motive. This could translate into higher income, and thus 1 We are indebted to Gian Carlo Blangiardo for kindly providing us with the data. Many thanks to Árpád Ábrahám, Gaetano Basso, Andrea Brandolini, Matteo Bugamelli, Francesca Carta, Tanika  (Project A03) is gratefully acknowledged. 2 Dustmann (1997) develops a model of return migration and shows that in fact immigrants may engage in more precautionary savings due to higher income uncertainty.
into an increase in consumption, in particular among the undocumented immigrants, who could now move into the formal sector. In the case of documented immigrants, the reduced labor market uncertainty coupled with a higher probability of getting a permanent contract may also boost household consumption expenditure. 5 As a result, extending citizenship rights might have an important impact on domestic demand. Despite its relevance, the link between citizenship and consumption has been largely overlooked empirically. Using data from a unique survey in Italy and employing a di¤erence-in-di¤erences approach, we study whether and through which channels the extension of EU citizenship a¤ected the consumption behavior of immigrant households following the 2007 enlargement. In our research design, we restrict our sample to immigrants who arrived in Italy before 2007 and compare the monthly consumption of households from new member states (Romania and Bulgaria) and candidate countries (Albania, Bosnia and Herzegovina, Croatia, Kosovo, Macedonia, Montenegro, Serbia and Turkey), before and after the enlargement. 6 ;7 We test the validity of our identi…cation strategy by addressing anticipation and composition e¤ects as well as spillover e¤ects by using immigrants from A8 countries as an alternative control group and by exploiting heterogeneity across regions and occupations. 8 Italy provides an ideal context to study the e¤ects of the 2007 enlargement as it has long been one of the main destinations for both Romanians and Bulgarians, even before 2007. Moreover, although the EU accession of Romania and Bulgaria was an expected event, its labor market consequences in Italy were not, and the EU accession implied for Romanians and Bulgarians full rights to work (see Section 3 for a detailed discussion).
Our unique dataset allows us to focus both on documented immigrants, for whom citizenship mainly implies that they do not need to renew their permits any more, and on undocumented immigrants, who bene…tted from legalization. Furthermore, detailed information on labor market outcomes, including the sector of employment (informal/formal), allows us to shed light on the exact mechanisms. We …nd that the EU accession signi…cantly increased average monthly consumption of immigrant households in the year of accession, 5 See Campos and Reggio (2015) for the relationship between labor market uncertainty and consumption and Barceló and Villanueva (2016) for the e¤ect of permanent contracts on household consumption and wealth accumulation. 6 A similar identi…cation strategy has been adopted by recent papers that study the labor market e¤ects of the 2004 enlargement (see Elsner, 2013a and2013b;Ruhs, 2017;and Ruhs and Wadsworth, 2018). 7 Although Iceland is among the candidate countries, their nationals can work in Italy as well as in other EU countries on the same footing as EU nationals, since they belong to the European Economic Area. Therefore, we do not consider Icelanders as part of our control group. 8 A8 countries are Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Slovakia and Slovenia, which joined the EU in 2004.
but also a few years later, in line with the presence of liquidity constraints. The increase in consumption involved both undocumented and documented immigrants albeit through di¤erent channels. Speci…cally, the former increased their expenditure on food, clothes, and other basic-need items due to increased labor income. We provide evidence that this was mainly achieved by moving from the informal towards the formal sector. Documented immigrants instead, increased mostly the consumption of durable goods. In their case, the underlying mechanism is a gradual increase in employment stability through permanent job contracts, which reduces the uncertainty about future labor income and thus, increases the propensity to consume. 9 At the same time, the probability of holding savings and remitting goes down, consistent with a reduced precautionary savings motive. 10 While legalization policies are usually di¢ cult to implement due to the high political cost entailed, our results imply that simplifying instead the bureaucratic procedure of work permits for documented immigrants may lead to similar consumption increases. The labor market channel that our analysis highlights is in line with Mastrobuoni and Pinotti (2015), who exploit the same natural experiment and …nd that immigrant crime decreases due to increased employment opportunities.
The literature on the consumption behavior of both documented and undocumented immigrants is scarce mainly due to data limitations. Two recent exceptions, exploiting the same dataset that we use, are Dustmann et al. (2017) and Barigozzi and Speciale (2011). Dustmann et al. (2017) use amnesty quotas to analyze the e¤ect of immigrants' legal status on their consumption behavior in Italy and …nd that undocumented immigrants consume about 40% less than documented immigrants and that this is partly due to their lower income. In our analysis we highlight an additional channel, that is the increased probability of getting a permanent contract for immigrants that were documented even before their home country accessed the EU. In the new legal framework, work permits of citizens from new member countries were no longer of limited duration, which plausibly made …rms more willing to o¤er them permanent contracts. Enhanced labor stability in turn decreases the uncertainty about future labor income, and thus increases the propensity to consume. Barigozzi and Speciale (2011) also focus on Italy and study the di¤erences in 9 We complement our …ndings with additional evidence from Social Security records that point to the same direction.
1 0 According to the literature, one of the reason why immigrants remit is to insure themselves against risk. Amuedo-Dorantes and Pozo (2006) show that remittances act as a form of family-provided insurance and self insurance. Delpierre and Verheyden (2014) develop a model with endogenous remittances, savings, and return decisions under uncertainty and show that when migrants face relatively low wage risks in the host country, they tend to remit less. the consumption behavior of natives, documented and undocumented immigrants. They …nd that the permanence in the host country plays an important role in attenuating these di¤erences. In our empirical exercise we control for years of residence and we show that the immigrants that bene…tted most from the EU enlargement were those with less than 5 years of residence in Italy, who were not eligible for permanent residence permits. Moreover, in a placebo exercise we …nd no e¤ect among immigrants who held or were eligible to apply for the Italian citizenship.
Our …ndings also contribute to a very recent literature that studies the labor market e¤ects of faster access to citizenship as well as of di¤erent asylum policies. This strand of the literature suggests that faster access to citizenship improves the labor market attachment of female immigrants and their investment in host country-speci…c skills (Gathmann and Keller, 2018). Similarly, higher recognition and decision rates boost the employment prospects and the economic integration of refugees (Fasani et al., 2018), while a lengthy period before obtaining the right to work seem to hamper them (Ballatore et al., 2017). We also explore alternative explanations, such as easier access to credit, but the labor market channel remains the most plausible underlying mechanism.
The paper is organized as follows. The next section provides a brief description of the Italian labor market in the context of immigration and of the legal framework regarding immigrants. Section 3 describes the natural experiment we exploit in our analysis. Section 4 presents the data, sample and identi…cation strategy and Section 5 outlines the empirical approach and presents the main results. Section 6 explores possible underlying mechanisms while Section 7 performs a series of robustness checks and discusses alternative channels.

Background
Immigration is considered to be a structural characteristic of the Italian society and labor market (Quassoli, 1999). The empirical evidence shows that the demand for immigrant workers in Italian labor market is mainly concentrated on unskilled jobs (Fullin and Reyneri, 2011) and compared to employed Italians, immigrant workers are more likely to be employed in sectors with low-pay, high job instability, and weak employment protection (Ambrosini and Barone, 2007). Immigrant workers are also more exposed than natives to temporary employment contracts (Barone, 2009), which are consistently found to be associated with lower job satisfaction and greater di¢ culty in balancing work and family, and provide less opportunities for work-related training and career advancement opportunities compared to permanent contracts (see, for example, Bentolila and Dolado, 1994;Blanchard and Landier, 2002;Bonet et al., 2013;Booth et al., 2002;Dolado et al., 2002). 11 In relation to the legal framework for immigration, Italy o¤ers various types of residence permits, including those granted for work reasons, which can be either temporary and need to be renewed in certain intervals (permesso di soggiorno) or permanent (carta di soggiorno). 12 Residence permits for work reasons are subject to quotas set by the government each year for di¤erent categories of immigrant workers (see, for a discussion, Pinotti, 2017). 13 The type of employment contract has a direct e¤ect on the frequency that residence permits need to be renewed. The temporary residence permit for work reasons has a validity of two years for immigrants working under an open-ended (permanent) contract and a validity of one year for those with …xed-term (temporary) contracts. Immigrants become eligible for a permanent permit of unlimited duration after …ve years of legal residence in Italy and the successful completion of an Italian language test.
Despite its comprehensive legal framework, as of January 2018, there were an estimated 533,000 undocumented immigrants in Italy (ISMU, 2018). Undocumented immigrants either enter the country without a permit or they enter on short-term visas (e.g. tourist or student visa) and then overstay despite having their documents expired (Fasani, 2015;Fullin and Reyneri, 2011). They tend to choose countries where it is easy for them to work for a period, even without a work permit, which they might obtain subsequently through regularization programmes or by …nding an employer in the formal sector to sponsor them (Levinson, 2005). Although the immigrants'legal status (documented/undocumented) and sector of employment (formal/informal) are not necessarily reciprocal, the relatively large informal economy in Italy has been a major factor in promoting undocumented immigration 1 1 In the late 1980s, Italy was considered to have one of the strictest labor markets in terms of employment protection legislation (OECD, 2004). To provide more ‡exibility to employers, Italy relaxed the rules for the use of temporary contracts in 1987, which, prior to this year, could only be used for seasonal work, speci…c projects or for replacement of absent workers temporarily (Kugler and Pica, 2008). Since then, temporary contracts steadily increased as a share of total employment (Cappellari et al., 2012). In the period of our analysis (2001-2012) around 13% of the workforce was under temporary employment contracts (Istat, Labour Force Survey). While the extended use of temporary contracts allowed for more ‡exibility in the labor market, large di¤erences in terms of employment protection legislation between permanent and temporary contracts have been a concern (Garibaldi and Taddei, 2013). 1 2 Other types include those granted for family reasons (e.g. spouse or dependent child of a legal resident) and special permits for study purposes and permits for asylum seekers/humanitarian reasons. 1 3 Immigrants can be sponsored for two main types of permits: type-A permits for domestic and care workers employed by families and type-B permits for workers employed by …rms. The latter are further distinguished between B1 and B2 permits for construction and non-construction workers. (Reyneri, 1998). 14 The …ndings of recent studies suggest that once undocumented immigrants are regularized, the majority move to the formal sector (Fullin and Reyneri, 2011) and stay in it (Di Porto et al., 2018

The Natural Experiment
Bulgaria and Romania joined the EU on January 1, 2007. 15 In fact, the EU accession of Romania and Bulgaria was an expected event as the accession negotiations were successfully concluded in 2004. However, the accession treaties allowed member states to impose temporal labor market restrictions on Bulgarian and Romanian workers for up to seven years after accession. All EU states were required to open their labor markets to the citizens of the two newest members by the end of 2013, but they had to give justi…cation if they wished to restrict access beyond 2011. 16 The majority of member states, including Italy, announced that would impose interim restrictions to protect their labor markets from a large ‡ow of immigrants from the new member states and therefore Romanian and Bulgarian immigrants would still be required to have a permit in order to work. 17 However, just three days prior to the EU accession, on December 28, 2006, the newlyelected, center-left government in Italy lifted the restrictions for high skilled employment as well as in sectors where the vast majority of Romanian and Bulgarian immigrants used to work, such as construction, hotel and tourism, domestic work, care services, agriculture, engineering and seasonal work. In these sectors, employers of Bulgarian and Romanian 1 4 A unique feature of our data is that we can observe both undocumented/documented immigrants working in the informal/formal sector. In our sample, all undocumented immigrants but also 14% of the documented immigrants work in the informal sector. 1 5 Following the EU enlargement in 2007, Romanian and Bulgarian immigrants in Italy were instantly granted with the EU citizenship and became documented without the need of obtaining/renewing any residence permit. 1 6 Some member states, that had imposed interim labour market restrictions in 2007, lifted them in advance of this deadline. Denmark, Greece, Portugal, Ireland and Hungary opened their labor markets to arrivals from Bulgaria and Romania before the end of the transitory period, while the UK, Germany, Austria, France, Belgium, the Netherlands, Malta and Luxembourg maintained restrictions for the maximum period. 1 7 Note that work permits were not transferable across member states. Therefore, Romanian and Bulgarians citizens, who were legally working in one of the member states at the date of accession and had a work permit for 12 months or longer, would have direct access to the labor market of that member state but not automatically to the labor markets of other member states. If they voluntarily left the labour market of the host member state, they would lose the right of access to the labor market of that state. These interim restrictions only applied to access to the labour market by workers and not to the free provision of services nor to the freedom of establishment, students, tourists, pensioners, etc. workers simply needed to submit a copy of the employment contract to the local labor o¢ ce. Migration quotas were maintained only in the manufacturing sector but were eased so as to accommodate a larger number of workers from the new member states. 18 As a result, in 2007, Italy was the only major economy in Europe to lift restrictions on workers from Romania and Bulgaria, granting them in practise full rights to work in Italy. 19   and over and reside in Italy at the time of the interview. A unique feature of the ISMU survey is that its sampling scheme was speci…cally designed to collect information on a representative sample of both documented and undocumented immigrants (see Data Section in the Appendix). In order to obtain truthful answers from the respondents on legal status and informal employment, no sensitive information is asked (e.g. name and address) and the data are collected in public spaces by interviewers with a foreign background, who have 1 8 Italy fully liberalised its labor market for citizens of Romania and Bulgaria as of 1 January 2012. 1 9 See Migration Advisory Committee Report (2008) and House of Commons Home A¤airs Committee Report (2007). See also Mastrobuoni and Pinotti (2015) for a similar discussion. The ISMU data include rich information on personal characteristics such as age, gender, education, marital status, country of origin, years of residence in Italy, residence permit as well as employment status and labor income. 21 Information on the residence permits allows us to identify the legal status of the respondents. In particular, we consider immigrants as documented if they reported to have a valid residence permit (permanent or temporary) at the time of the interview. Employed respondents were also asked about their labor income, type of employment contract and occupation, and importantly for our analysis, whether they work in the formal or in the informal sector. (ii) housing such as rent, mortgage, maintenance, bills; and (iii) other items such as transportation, leisure, installment purchases and debt. 22 Our main dependent variable is the total consumption of immigrant households in the host country, i.e. the sum of these three types of consumption expenditure, but we also explore each of the disaggregated consumption categories separately. In our benchmark estimates, we use the average monthly household consumption controlling for the number of household members and cohabiting children as well as the total number of children (cohabiting or not) and whether the spouse lives with the respondent or abroad. We also check the sensitivity of our main results using equivalized consumption (see Section 7). Respondents were also asked to report in euros 2 1 The country of origin refers to the individual respondent rather than the whole household. We check the sensitivity of our results to the de…nition of immigrant households by restricting the sample to immigrants who are living with a partner from the same country of origin or singles/not living with a partner (see Section 7). 2 2 Throughout the paper, we use the term 'durables' to refer to the latter as it is likely to include large and long-term purchases such as cars or home appliances that are usually bought in installments. their average monthly remittances as well as their monthly savings in Italy. Information on monthly remittances is likely to be subject to measurement error while there is no information on savings held in the home country (see Dustmann et al., 2017 for a further discussion). Moreover, many respondents report zero monthly remittances and/or savings. Therefore, in Section 6 we adopt a linear probability model as well as a model that takes censoring at zero into account to study the e¤ect both on the extensive and the intensive margin of savings and remittances. Nevertheless, these results should be interpreted with caution due to the measurement of these variables in the ISMU survey.
Two factors should be mentioned before going ahead. First, the ISMU survey concerns only the Lombardy region of Italy. However, Lombardy can be considered as a good approximation of the whole country as it is the most populated and one of the largest and wealthiest regions of Italy, and has the largest migrant population in the country accounting for 25% of the total (IReR, 2010). 23 Second, due to the cross-sectional nature of the survey, we are not able to trace the same individuals over time. Still, we are able to recover some retrospective information on whether respondents in our sample were documented and working in the formal sector before the EU enlargement. In particular, given that there was no need of obtaining/renewing the work permit in 2007 among the treated, we can infer that those with a valid work permit in 2007 had obtained it beforehand. Moreover, we use the Social Security records, that allow us to follow individuals over time, obtaining some additional evidence regarding the labor market outcomes (wages and type of contract).

Sample and identi…cation
In our analysis we use all nine waves (2004-2012) of the ISMU data that include information on average monthly household expenditure to explore the impact of the EU enlargement on the consumption of immigrant households from new member states. Our treatment group consists of Romanians and Bulgarians. A natural control group for new EU member countries is the EU candidate member countries as they should be comparable on the basis of the political and economic conditions (Mastrobuoni and Pinotti, 2015). Moreover, their attitudes towards risk should be similar to those of Romanians and Bulgarians given their common migration choices. 24 Therefore, immigrants from Albania, Bosnia and Herzegovina, Croatia, Kosovo, Macedonia, Montenegro, Serbia, and Turkey constitute our control group. 25 Since Italy experienced an expansion of migration from Romania and Bulgaria following their accession to the EU, the causal e¤ect of the EU enlargement on the consumption of immigrant households would be contaminated by the di¤erent selection of new immigrants following the EU accession. To address this issue, we restrict our sample to immigrants who arrived in Italy before 2007, i.e., before Romania and Bulgaria joined the EU. 26 We also restrict our sample to immigrants who do not hold Italian citizenship by the time of the interview and with no more than ten years of residence in Italy by the time of the EU accession since a non-EU citizen, having legally resided in Italy for ten years, is eligible to apply for the Italian citizenship. 27 In Figure 1, we present the average monthly consumption of immigrant households in the host country for the treatment and the control groups, before and after the enlargement for each year. As shown in panel A, Romanians and Bulgarians living in Italy had lower average monthly consumption than immigrants from EU candidate countries. The di¤erence remained fairly constant until 2007, suggesting that the consumption expenditure of treatment and control groups were following parallel trends prior to the EU enlargement.
In 2007, with the EU accession, the average monthly consumption of the treated group increased substantially, while the one of the control group continued to grow at approximately the same rate as in the previous years (panel A). This increase in total consumption was mainly driven by the increase in the expenditure on food, clothing and other basic needs and on transportation, leisure, installment purchases and debt (panels B and D) and is evident not only immediately after the EU accession but also in the following years. By contrast, the immigrant households'housing expenditure continued to grow in the year of EU accession at approximately the same rate as in the previous years, both for the treated and the control group (panel C). Table 1 presents the means of all the variables included in our analysis for the treated and control groups in our sample prior to (2004)(2005)(2006) and after (2007-2012) the EU enlargement (see Appendix Table A2 for a description of these variables). Focusing on the individual and 2 5 As a robustness check, we repeated the analysis using an alternative control group that consists of the nationals of Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Slovakia and Slovenia, which joined the EU in 2004 (see Section 7). 2 6 As a further robustness check, we repeated the analysis by restricting our sample to Romanian and Bulgarian migrants who moved to Italy at least one year, then two years and …nally three years before the EU enlargement (see Section 7). Note that the latter corresponds to immigrants who arrived in Italy before the end of the accession negotiations of Romania and Bulgaria in 2004. 2 7 We use this excluded group to perform a placebo exercise in Section 7.
household characteristics, the treatment and control groups are similar to each other before the EU enlargement in terms of age composition. However, there are notable di¤erences in other characteristics. Immigrants from Bulgaria and Romania are more likely to be female and to be more educated than immigrants from EU candidate countries. They are less likely to have a valid residence permit (being documented) and they reside in Italy for a smaller average number of years. Moreover, they tend to live in smaller households with fewer children. In terms of employment outcomes, Bulgarian and Romanian immigrants have lower labor income than immigrants from EU candidate countries and are more likely to work under temporary contracts and in the informal sector.
Focusing on the before and after trends, the treated group experiences higher consumption and labor income increases than the control group after the enlargement. The share of females and the average years of residence evolve in a similar way among the two groups while the composition by education remains pretty stable over time. There is an increase in the fraction of immigrants with at least one child and a decrease in the fraction of those with spouses living abroad, especially in the treated group. These di¤erences could a¤ect our analysis, as other things being equal, they would lead to an increase in the number of household members that would translate mechanically into an increase in household consumption expenditure. Therefore, in our analysis we always control for the changing household structure and perform a series of robustness checks on this issue.
In the next section, we account for compositional di¤erences between the treated and control groups and test the validity of the parallel trends assumption using a regression framework, which reinforces the causal interpretation of the e¤ect of EU accession on the monthly consumption expenditure of immigrants from new EU member countries. In what follows, we also examine whether any trends in the observable characteristics of immigrants are the same across the treatment and control groups. In particular, we test the validity of our approach by analyzing whether the composition of Romanians and Bulgarians changed in some systematic way following the EU enlargement.

Short-term analysis
Since the observed di¤erences in consumption presented in Figure 1 may re ‡ect the underlying di¤erences between the treatment and the control groups rather than a treatment e¤ect, it is important to control for individual and household characteristics. For this purpose, we …rst focus on the short-term impact of the EU accession (i.e. from year 2006 to 2007) and set our empirical model as follows: where i is an index for the households, c is the country of origin, p is the Italian province of home ownership in Italy. In our full speci…cation, we also include the respondent's average monthly labor income in addition to individual and household controls as a proxy for the household income. Finally, province of residence in Italy is denoted as p and icpt is an error term. As immigrants of the same nationality are likely to reside in the same province, the consumption expenditure may be correlated within country of origin groups but also within provinces. We thus cluster standard errors by Italian province of residence and country of origin using the two-way method proposed by Cameron et al. (2011). 28 In equation 1, the coe¢ cient is the shared e¤ect of the EU enlargement. The main coe¢ cient of interest is which is the di¤erence-in-di¤erences coe¢ cient, comparing monthly consumption of immigrant households from new member states and EU candidate countries in the host country, before and after the EU enlargement. Table 2 presents the shortterm estimates in separate panels for total consumption and for the broad categories of consumption expenditure. In each panel, we include country of origin and Italian residence of province …xed e¤ects, and gradually add individual and household controls. In the last column of each panel, we also control for the respondent's average monthly labor income net of taxes. 29 As shown in panel A of Table 2  consumption, and other basic needs of around 7.7%, which is similar in magnitude to the 2 8 This results in 88 clusters in the short-term analysis. Nevertheless, we also check sensitivity of our results clustering standard errors solely at the country of origin level (11 clusters) using the wild bootstrap method (Cameron et al., 2008) with 1000 replications to account for the small number of clusters (see Section 7). 2 9 One drawback is that household income information is not available for survey years earlier than 2007. Therefore, we use the respondent's labor income as a proxy of the household income (the correlation coef-…cient between household income and respondent's labor income in our sample is around 0.27 for the years 2007-2012). Dropping the respondent's labor income from our preferred speci…cation, leaves our main results unchanged. e¤ect on total consumption, and their expenditure on less basic needs and durable goods around 14%.

Pre-trends and persistence
In order to test the validity of our analysis, we now adopt a more generalized framework like in Autor (2003) that allows us to test for parallel trends but also to examine the persistence of the e¤ect. For this purpose, we use data for the period 2004-2012 and we augment equation (1) with lags and leads of the treatment as given by equation (2) ln(cicpt) = leads & lags of the treatm ent including year …xed e¤ects ( t ) in our speci…cation to capture the common time trends in the monthly consumption expenditure of the treatment and the control groups, and the changes in macroeconomic variables (e.g. in ‡ation). The term new EU c is not shown because the coe¢ cient is absorbed by the country of origin …xed e¤ects ( c ).
In equation (

Mechanisms
One of the most important bene…t for the immigrants of the new EU member countries is the right to work in all EU countries without the need of a work permit. As discussed in Section 3, Italy had initially announced that it would impose interim restrictions to protect its labor market just like other EU countries did. However, just a few days before the accession Romanians and Bulgarians acquired full rights to work in Italy. This could have direct e¤ects on the employment probability and the labor income of our treated group which may explain the increase in the immigrant household consumption that we documented in the previous section. In fact, after the EU enlargement the labor income and the percentage of those with permanent contracts increased among immigrants in the treated group (by 6% and 8 percentage points, respectively) while they remained fairly constant among immigrants in the control group (see Table 1). Moreover, before the EU enlargement 20% of immigrants in our treated group were undocumented and 31% were working informally. After the EU enlargement, they all became documented (as they gained EU citizenship) and the percentage of informality decreased sharply -yet did not disappearto 18%. At the same time the percentage of undocumented and of those working informally decreased only slightly among immigrants in the control group. Table 4 presents the results of regressions for di¤erent labor market aspects. 30 We observe a positive labor force participation e¤ect after the accession (column 1) and a positive, though not statistically signi…cant, employment e¤ect (column 2). Indeed, most immigrants who were legal residents before the accession were already employed since obtaining a work permit is the most common way of becoming documented in Italy (Mastrobuoni and Pinotti, 2015). 31 Moreover, even undocumented immigrants tend to work but in the shadow economy. 32 Note that the ISMU data contain information both for the formal and the informal employment, and thus it is not puzzling that the probability of employment did not increase signi…cantly. What did increase after the EU enlargement is the labor income (column 3).
The increase in the labor income is in line with Ruhs (2017) Romanian and Bulgarian immigrants in Italy were documented after 2007. Although the ISMU data provide information on current legal status, there is no information on former legal status (before the EU enlargement). Due to their cross-sectional nature, it is also not possible to distinguish between those immigrants from the newly accessed countries that were legalized by the EU enlargement, and those that were legally residing in Italy already before. Likewise, there is no retrospective information on informal/formal work. Still, we are able to single out a particular set of immigrants, for whom we can infer that they were both residing legally and working in the formal sector before the accession of their home country in the EU. We do so by focusing on a subsample of documented immigrants who reported in 2007 to have a valid residence permit for (dependent) work. The rationale 3 1 In our sample, 65% of all documented immigrants have a residence permit for work reasons. 3 2 In our sample, 74% of all undocumented immigrants work and all do so in the informal sector.
behind our strategy is that the respondents in our treatment group should have obtained the permit (i.e. legally residing and working in the formal sector in Italy as an employee) before the EU enlargement, since there was no need for them to obtain or renew it in 2007 after the EU accession. 33 ;34 This strategy has the advantage of identifying the set of immigrants, for whom the EU enlargement basically implied that they did not need to renew their permits any more as opposed to undocumented immigrants or those working informally, who derived more tangible bene…ts from becoming documented and in many cases moving to the formal sector. We thus replicate our short-term analysis using this particular group of immigrants. These results are reported in Table 5. We observe that the estimated e¤ect on food, clothing and other basic needs turns to be insigni…cant for immigrants who were legal already before the enlargement. On the other hand, there is an increase in their total consumption expenditure driven mainly by the increase in household consumption of durable goods such as transportation, installment purchases and debt. Table 6 focuses on the treatment e¤ect on labor market outcomes in the short-run for this subgroup of formerly legal immigrants in order to explore the underlying mechanism behind the response in consumption. The small and insigni…cant estimates on labor income (column 1) and on the probability of having a permanent contract (column 2) suggest that the labor market outcomes of immigrants from newly accessed countries, who had a valid permit and were working in the formal sector even before 2007, were not immediately a¤ected by the EU enlargement. This is not surprising since transitions into permanent contracts usually take time. Given that the ISMU data do not allow us to explore the long-term e¤ects of the EU enlargement on the labor market outcomes of this subgroup of immigrants, we provide further evidence using data from the Italian Social Security (INPS) records. The Social Security data contain information for a 6,5% random sample of all private sector employees in Italy (see Data Section in the Appendix for further details). Due to their administrative nature, these data include only immigrants that are working in the formal labor market as employees who in principle correspond to the ISMU subsample of documented immigrants with a valid work permit. 35 ;36 To have comparable results with the ISMU data, we restrict the sample to immigrants that work in a …rm located in Lombardy and appear at least once in the Social Security data before 2007 with less than 10 years of experience. 37 An advantage of the Social Security data is that we can also observe daily wages, which di¤erently from monthly wages in the ISMU data, do not re ‡ect changes in the labor supply. Appendix Table   A3 reports the descriptive statistics of our sample. We see that the treated and the control groups experience similar increases in daily and monthly wages after the EU enlargement.
However, there is an increase in the percentage of workers with a permanent contract only among the treated.
The panel nature of the administrative data allows us to follow individuals over time and to perform a regression analysis with worker and …rm …xed e¤ects. 38 In this way we are able to account for unobserved heterogeneity without the extensive list of controls that were available in the ISMU data and were important to include in a repeated cross-sectional setting. Table 7 reports the results of this analysis. There is no statistically signi…cant e¤ect on monthly (column 1) or daily (column 2) wages, while the probability of having a permanent contract increases from 2008 onwards (column 3). Moreover, pre-trends seem to be parallel as the coe¢ cients are not statistically signi…cant in the period before the enlargement. The regression estimates con…rm the cross sectional results from the ISMU survey (Table 4, columns 5 and 6) and suggest that although legalization is not the reason behind it, employers reacted positively to the fact that Romanians and Bulgarians did not need to pass anymore through the tedious bureaucratic procedure of renewing their work permit. In other words, the new legal framework after the EU accession acted as a 'permanent'work permit. Increased employment stability reduced the uncertainty for future labor income, which in turn increased their consumption expenditure. This result is in line with Gathmann and Keller (2018) who …nd that faster access to citizenship for immigrants in Germany has improved their labor market attachment.
The above labor market story is broadly consistent with the pattern of our consumption results. We observe an immediate increase in consumption in 2007 and then again a few years later. The former is due to the increase in food, clothing and other basic need expenditures 3 5 Hotchkiss et al. (2015) show that administrative data in the US may actually include a small number of undocumented immigrants with 'fake'…scal code. 3 6 As it is often common with administrative data, we are not able to distinguish unemployment from non-participation in the Social Security Records. 3 7 Since we lack information on the year of arrival in Italy, we use the date of entry in the labor market as a proxy of the arrival date. 3 8 We cluster standards errors at the worker and year level. by the previously undocumented immigrants whose labor income rises but also due to the increase in durable goods by the previously documented immigrants working in the formal sector. Although the latter did not experience any increase in labor income, they might have anticipated that they would be able to access a permanent job in the future. Our analysis supports this hypothesis as consumption increases again after having obtained the permanent contract.
It is also worth noting that these results are suggestive of a reduced precautionary savings motive. To verify this, we use the ISMU data that include some information on average monthly savings in Italy as well as on average monthly remittances. Unlike consumption expenditure, the information on remittances and savings in the ISMU data is imperfect (Dustmann et al., 2017). Regarding savings, we only have information on savings held in the host country but no information on savings held in the home country. Ideally, we would like to have a measure of total savings (both in Italy and in the home country) in order to be able to analyze precautionary savings. 39 In relation to remittances, the ISMU survey asks respondents to report the average amount they send home each month, which is subject to measurement error, especially if transfers take alternative forms than sending money or are less frequent. Moreover, remittances may either end up as savings or investment in the home country or …nance the consumption expenditures of family members who do not live in Italy. 40 This is why we study the two variables (savings in Italy and remittances) both separately and jointly as a composite measure of total savings. As many immigrant households in our sample report zero savings and/or remittances (42% and 47% of all cases, respectively), we …rst follow Dustmann and Mestres (2010b) and in our OLS estimates we set zero savings and/or remittances to 1 and use log(y+1) as our dependent variable, where y is savings, remittances or the sum of the two. Then, we adopt a linear probability model in order to study the extensive margin of savings and remittances. Table 8 reports the results of both models. 41 There is a negative statistically signi…cant e¤ect on remittances in 2010 and on savings in 2012. These are years for which we …nd a positive e¤ect on housing expenditures (Table 3, panel C) and a higher probability of having a permanent employment contract (Table 7, column 3). In line with Amuedo-Dorantes and Pozo (2006), 3 9 Dustmann and Mestres (2010a) show that not accounting for savings in the home country may result in distorted conclusions regarding immigrants'saving behavior. 4 0 Immigrants may remit for a variety of reasons, ranging from altruism, exchange, inheritance, or strategic motives to family insurance and investment motives (see, for an excellent review, Rapoport and Docquier, 2006). 4 1 These results should be interpreted with caution for the reasons we described above regarding the measurement of remittances and savings in the ISMU survey. who show that undocumented/risky-income immigrants tend to remit more, and Dustmann and Mestres (2010a and 2010b), who show that temporary immigrants are likely to remit and save more, we …nd that immigrants, by getting legalized and accessing permanent employment contracts, reduce savings and remittances after the EU enlargement.

Mechanisms
In this section we conduct various additional exercises. First, we address anticipation and spillover e¤ects that are common threats to identi…cation in a di¤erence-in-di¤erence framework. Second, we examine whether our estimates are sensitive to the measure of consumption, to the de…nition of immigrant households or to the way of clustering, and explore whether the business cycle or changes in household structure over time drive our results. Third, we consider selective out-migration (composition e¤ects) and omitted variables/unobserved heterogeneity. Lastly, we perform a placebo exercise and discuss alternative mechanisms that may lie behind our …ndings.

Robustness checks
We start by performing a series of exercises to examine the robustness of our benchmark estimates (Table 3, panel A, column 4) and present these results in Table 9. First, we proceed by looking at anticipation e¤ects as treated households that moved in Italy prior to the EU enlargement may have somehow anticipated that the labor market restrictions would have not been implemented. We thus restrict our sample to those that had migrated in Italy at least one year, then two years and …nally three years before December 27, 2007. Note that the latter corresponds to immigrants who arrived in Italy even before the end of EU accession negotiations of Bulgaria and Romania in 2004. These results are presented in columns 1-3.
We …nd that both the short-and the medium-run e¤ects of the EU enlargement on the total household consumption of the Romanian and Bulgarian immigrants remain positive and signi…cant, even for those who arrived in Italy well before the announcement of the policy, suggesting that our results are not driven by anticipation.
Second, we try to understand whether there are any spillover e¤ects between the treated and the control group (the so-called SUTVA-see Rubin, 1977). In particular, if the treated and the control group competed for the same jobs, the EU accession could not only bene…t the treated but also negatively a¤ect the control, undermining our di¤erence-in-di¤erence strategy. To address this issue, we …rst compare provinces where the treated and the control group were of similar size before 2007 (Figure 3 panel a) to provinces where the treated group was the minority (Figure 3 panel b). 42 The idea behind our strategy is that spillover e¤ects should be stronger in provinces where the treated and the control groups are of similar sizes (potentially through the competition in the labor market) than in provinces where the treatment group was a minority. The e¤ect of the EU enlargement on consumption is not di¤erent between the two sets suggesting that SUTVA is likely to be satis…ed in our setting.
A similar picture emerges when we compare occupations that experienced an increase in the fraction of the treated group after the enlargement with industries/occupations that did not (Figures 4a and 4b). 43 We then check the sensitivity of our analysis to the alternative measures of consumption. In particular, we use individual consumption calculated as the ratio between household consumption and the number of members of the household residing in Italy, converted into equalized adults using three alternative equivalence scales. These results are presented in  Table A4 for the relative sizes of the treatment and control groups by Italian province of residence prior to the EU enlargement. 4 3 See Appendix Table A5 for the percentage point change in the fraction of Romanian and Bulgarian immigrants following the EU enlargement by occupation. 4 4 Standard OECD equivalence scale assigns a weighting of 1 to the …rst adult, 0.7 to the second and each subsequent adult, and 0.5 to each child, whereas modi…ed OECD equivalence scale assigns a weighting of 1 to the …rst adult, 0.5 to the second and each subsequent adult, and 0.3 to each child in the household. The equivalence scale used by ISEE, on the other hand, asssigns a weighting of 1 to the …rst person in the household, 0.57 to the second person, 0.47 to the third person, 0.42 to the fourth person, 0.39 to the …fth person, and 0.35 to the subsequent persons. 4 5 In these speci…cations we do not control for the number of household members and the number of cohabiting children as this is already taken into account by the equivalence scale. We still include though controls for the total number of chidren and for whether the spouse of the respondent lives abroad in order to account for non cohabiting household members. and Slovenia (together referred to as the A8 countries) that accessed the EU in 2004. 46 The advantage of using the nationals of A8 countries as a control group is that they are unlikely to be a¤ected by the EU enlargement since they acquired the EU citizenship already in 2004. Thus, possible spillover e¤ects are not a concern in this setting. These results are presented in Table 9, column 7. Comparison of these results to our benchmark estimates in Table 3 shows that they are quite similar in magnitude, although the coe¢ cients are less precisely estimated due to the smaller sample size.
Next, we check the sensitivity of our analysis to the de…nition of the immigrant households' country of origin. Throughout our analysis, we consider a household being from a particular country if the respondent is from that country. Although the ISMU data do As discussed in Section 4, consumption expenditure might be correlated within the country of origin groups, but also within provinces as immigrants from the same nationality tend to concentrate/live close to each other. Thus, throughout our analysis, we use twoway clustered (at Italian province of residence and at country of origin level), but we also check how robust our estimates are to di¤erent ways of clustering. In particular, we cluster standard errors only at the country of origin level (11 clusters) and use the wild bootstrap method proposed by Cameron et al. (2008) with 1000 replications to account for the small number of clusters. These results (column 9) are in line with the benchmark.
During the period of our analysis Italy experienced the Great Recession, which led to severe job losses. Immigrants were particularly a¤ected as they tend to be more susceptible to the economic cycle than natives (Dustmann et al., 2010). Although in our analysis both the treated and the control group comprise of immigrants and are therefore exposed to the recession in a similar way, it may be the case that the two groups were concentrated in provinces or occupations that were di¤erently a¤ected by the recession. Therefore, we add to our speci…cation province-year, and occupation-year …xed e¤ects in order to explore whether our results are driven by the business cycle at the local and occupational level. 47 To do so, we restrict the sample to employed individuals and use available information on the occupation of the immigrant (e.g. domestic worker, artisan, intellectual, employee in hotels/restaurants, construction worker, salesperson -see Appendix Table A5). The results after the inclusion of these new set of dummies (column 10) are similar to the benchmark.
Moreover, given that the EU enlargement may have a¤ected the fertility or family re-uni…cation decisions of the immigrants, we include in our speci…cation, in addition to the controls for the presence of a spouse living with the respondent in Italy and for the total number of children, their interaction with year dummies so as to invistigate whether changes in the household structure over time drive the results on consumption (column 11). The estimates of this further robustness check show that our results are not sensitive to the changes in the household structure over time.
Another possible threat to our identi…cation strategy is selective out-migration. It is possible that the composition of our sample changes after the EU accession given that the treated group acquired the right to move freely to other countries within the EU or due to return migration. In particular, mobility may be non-random and treated households that did not prosper in Italy may decide to leave the country in search of better opportunities elsewhere in the EU. If the composition of immigrants changed in some systematic way following the EU enlargement, then we need to take account of this selection when assessing the e¤ects of EU enlargement on household consumption expenditure. For this purpose, we estimate a version of equation (2), where the dependent variable is the immigrants'characteristics (female, young, low educated, number of household members). These results are presented in Appendix Table A6. We do not …nd any signi…cant change in the composition of our sample following the EU enlargement, con…rming the robustness of our results. Moreover, the ISMU data include direct information on the intentions of immigrants to leave Italy (to return to the home country or move to a di¤erent country). This information is available only in the period 2010-2012, so we cannot study the e¤ect of the EU enlargement. Still, we can check whether the intentions to leave Italy were systematically di¤erent between the treated and the control group in the period 2010-2012. Results in Appendix Table A7 show that Romanians and Bulgarians are not more likely to select into return migration/migration towards a di¤erent country than immigrants from EU candidate countries. This is true also when we restrict the sample to immigrants who arrived in Italy one or two years before the enlargement (columns 2 and 3).
Although in our analysis we include a comprehensive set of individual and household variables, a possible concern is that they may not fully control for all the relevant characteristics and thus equation (2) could su¤er from the omitted variables problem. To assess the in ‡uence of omitted variables relative to the one of observed characteristics, we use a method proposed by Altonji et al. (2005) and calculate the ratio of the in ‡uence of unobserved characteristics relative to the one of observed control variables that would be required so as to fully explain away our result. The intuition behind this approach is that if the inclusion of observed control variables substantially weakens the impact of the EU enlargement, then one would expect that the inclusion of additional controls (observed or unobserved) would reduce the estimated e¤ect even further. Conversely, if the inclusion of additional controls has no substantial e¤ect on the magnitude of the coe¢ cient estimate, then this will support the causal interpretation. Thus, a large ratio would imply that the unobserved heterogeneity cannot fully explain away the estimated e¤ect of the EU enlargement. In Appendix Table   A8, we present this ratio based on our main results on total consumption ( Table 3). The reported ratios are between three and seven, suggesting that in order to attribute the entire estimated e¤ect of the EU enlargement on the total consumption to selection e¤ects, the in ‡uence of unobservable factors would have to be between three to seven times greater than the one of the observable characteristics. These values are considered to be high (see, Therefore, we conclude that our estimates cannot be attributed to unobserved heterogeneity.

Placebo exercise and alternative mechanisms
We have seen so far that the EU accession increased the household consumption of the treated with respect to the control and provided evidence that the improved labor market conditions is a possible underlying mechanism. To provide additional supporting evidence, we perform a placebo exercise on a group of immigrants who were unlikely to bene…t from the EU accession and were excluded from our analysis so far. In particular, we focus on immigrants who either held the Italian citizenship or were eligible to apply for it by 2007 (having resided in Italy for more than 10 years). We compare this group of immigrants with immigrants in the sample we used for our benchmark estimates, that we further split between those with less than 5 years and those with 5-10 years of residence by 2007. The latter were eligible for permanent residence permits and therefore the expected bene…ts from the EU accession would be lower for them. Table 10 reports the results for these three separate groups. 48 We verify a positive and statistically signi…cant e¤ect among immigrants with less than 5 years of residence by 2007 (column 1), while the coe¢ cient of the interaction term is still positive but half in size and not statistically di¤erent from zero among immigrants with 5-10 years of residence (column 2). The e¤ect completely vanishes (and turns even negative) when we focus instead on immigrants who held or were eligible to apply for the Italian citizenship by 2007 (column 3). This placebo exercise is consistent with the notion that immigrants who bene…tted from the EU accession most were those not close to acquiring permanent residence or citizenship rights in Italy.
Still, the labor market mechanism does not exclude other channels that may have also contributed. More speci…cally, the EU accession may have also facilitated the access to credit for the treated households although in our benchmark speci…cation we do not …nd any evidence of increased expenditures regarding housing, at least in the short run ( Table   2, panel C). A possible reason is that mortgage payments and rent enter in the same way in the expenditures for housing. As the ISMU data do not contain any information on mortgages, we utilize a di¤erent data source, the Survey on Income and Life Conditions of Households with Foreigners conducted by the Italian National Institute of Statistics in 2009. The survey has been conducted only once and followed closely the design of the European Union Statistics on Income and Living Conditions (EU-SILC) but with a sample exclusively composed of households with at least one foreigner. Moreover, it has been speci…cally designed to be representative by nationality (see Data Section in the Appendix).
More importantly, the survey contains unique information on whether households have a mortgage and the year that they obtained it. We de…ne the treated and the control group in the same way as in the benchmark exercise and apply the same sample restrictions and identi…cation strategy but we see no clear di¤erence in the fraction of the treated and control with a mortgage issued immediately before or after the EU enlargement (see Appendix Table   A9). 49 Another possible mechanism is access to social bene…ts such as welfare or unemployment bene…ts. This mechanism could be relevant only for previously undocumented immigrants since documented immigrants, as long as they satisfy the eligibility conditions in terms of income thresholds and previous work experience, have access to social bene…ts even before  50 In relation to our …ndings, if all prices went down due to the EU enlargement, this would a¤ect both the treated and the control group, and hence this would not a¤ect our results. If we assume that only prices related to certain Romanian and Bulgarian products went down and that the demand for these products is generally higher among the treated, the e¤ect on total consumption expenditure would depend on the elasticity of these products and on their relative weight in the consumer basket. Although we cannot exclude this mechanism, we do not expect it to be the main driver of our results given the range of Romanian and Bulgarian products available in Italy.
Therefore, the improved labor market conditions continues to be the most plausible underlying mechanism behind the increases in the household consumption of the treated.
Moving out of the shadow economy for previously undocumented immigrants as well as the increased probability of getting a permanent job for previously documented mmigrants, whose work permit was no longer of limited duration, are the two main channels.

Conclusions
In this paper we focus on Romanian and Bulgarian households that had migrated to Italy before 2007 and study whether the accession of their home country in 2007 a¤ected their consumption behavior. We …nd that their average monthly consumption expenditure increased signi…cantly as soon as their home country accessed the EU. This increase is not just temporary and it cannot be attributed to the mere legalization.
On the one hand, immigrants from the new member countries who were working informally in Italy before the EU accession experience increases in labor income after the accession by moving away from the shadow economy. On the other hand, documented immigrants from the new member countries who were working formally in Italy even before accessing the EU do not experience wage increases but have an increased probability of getting a permanent contract after the accession. We conjecture that the resolution of uncertainty regarding the renewal of work permits has contributed to this e¤ect. In the new legal framework work permits did not have to be renewed for the citizens of the new member countries making …rms more willing to o¤er them permanent contracts. Enhanced labor market stability decreases the uncertainty regarding future labor income and it consequently increases household expenditures-particularly those on durables. Our results are robust to a series of robustness checks addressing anticipation and composition e¤ects as well as spillovers. We also discuss alternative possible channels, such as improved access to credit, and we conclude that improved labor market conditions is the most plausible underlying mechanism.
Our results have important policy implications in a period of increased legal uncertainty, following the decision of the United Kingdom to exit the EU, which is expected to increase (3) (1) (3) (1) (3) (3)        (10) includes the interaction between province and year and occupation (only available for employed individuals) and year …xed e¤ects and column (11) includes the interaction between having a spouse abroad and year and number of cohabiting children and year …xed e¤ects. For the remaining de…nitions, see Table 3 notes.              Table 3 Column 2  vs  Table 3 Column 4 3.19 Table 3 Column 3  vs  Table 3 Table 3 for the description of the full set of controls included in each speci…cation.