Even worse than the undocumented? Assessing the refugees’ integration in the labour market of Lombardy (Italy) in 2001– 2014

The paper aimed to assess possible gaps in labour market integration or refugees compared with other migrants, in cluding irregular migrants in the Italian region of Lombardy in 2001– 2014. Despite the Italian labour market's remarka ble capacity to integrate foreign workers, our findings sup port the results of studies carried out in other contexts on a ‘refugee gap’ in labour market integration. Our research shows that refugees have more considerable difficulties in tegrating into the labour market than other migrants and notably compared with their co- nationals, including those undocumented that may be in refugee- like situations. Refugees show higher unemployment levels, irregular em ployment and a higher incidence of very low monthly work-related incomes compared with other regular migrants. At the same time, they show higher levels of unemployment compared with irregular migrants.


The refugee gap
Immigrants' employment is one of the most researched areas in the field of integration (De Haas et al., 2019;Panichella, 2018;Scholten et al., 2015). Still, only a limited number of studies focus on the impact of migrants' entry channels on their labour market outcomes. Previous research in the United States, Australia, Canada and northern and continental Europe has repeatedly underlined that refugees face substantial difficulties in the process of labour market integration compared with other categories of immigrants (e.g. Mpofu et al., 2012;Waxman, 2001). Connor (2010) has defined such difficulties in terms of both entrance in the labour market and earning compared with other immigrants' categories as 'refugee gap'. Bakker et al. (2017) talk about a 'refugee entry effect', stressing, in particular, the initial disadvantage that refugees usually 1. face at the beginning of their working career in the host country that tends to decrease (but rarely disappear) over time. Both personal characteristics of refugees and host country-specific characteristics are responsible for refugees' penalization (Bevelander, 2016;Rengs et al., 2017).
From the refugees' point of view, the social and human capital theory has frequently been used to partially account for this result (de Vroome & van Tubergen, 2010). Refugees can hardly rely on economically advantageous social ties which would strengthen their employment prospects and knowledge of recruitment methods upon their arrival (Friedberg, 2000). Previous studies have found that refugees have, on average, lower education, language proficiency and employability skills compared with other categories of immigrants (Bevelander & Pendakur, 2014;Connor, 2010). Indeed, a large body of research suggests that refugees are less positively selected to be integrated into the host country labour market compared with other categories of immigrants (Chiswick, 2005;Constant & Zimmermann, 2005;Ruiz & Vargas-Silva, 2018). Not choosing to migrate upon their initiative, they did not invest in preparatory socialization, for instance by studying the language of the destination country before migration.
Additionally, having most probably fled their country abruptly, they may have interrupted their education, while not bringing with themselves certifications of their skills. Even in destination countries where refugees belong to selected networks characterized by high education, their qualifications are rarely recognized (Ager & Strang, 2008).
Several quantitative studies clarified that the lack of human and social capital does not fully explain the refugee gap in economic integration, at least in the medium-short term, suggesting the existence of additional relevant aspects often unaccounted for in standard labour force surveys (Connor, 2010). First, because of the traumatic experience they faced before and during migration, refugees have poorer mental and physical health than other migrants (e.g. Bratsberg et al., 2017;Connor, 2010;Hugo, 2011;Ortensi & Kingston, 2021;Ruiz & Vargas-Silva, 2018). Second, other characteristics, in many cases unobserved, such as the reason for migration, help to explain differences between different categories of immigrants. Refugees may lack personal motivation compared with other migrants, especially the economic ones (Campbell, 2014;Rengs et al., 2017). Third, in his labelling theory, Zetter (1991Zetter ( , 2007 ascribes to the mechanism of identity redefinition a role of its own in remoulding refugees as passive victims in need of assistance, perpetuating their dependency and disadvantage. Refugees hosted in reception centres before being granted international protection could feel less pressure to work immediately after the arrival than unassisted irregular migrants. Other factors may be ascribed to the host country-specific characteristics. The reception system often builds 'mechanisms of exclusion' for asylum applicants (including deportation, detention, dispersal, resettlement policies, regimentation of their daily routines and restriction in their mobility), which negatively impact on their integration and agency (Bloch & Schuster, 2005;Bevelander, 2016;Rengs et al., 2017;Schuster, 2004). The legislation also may deny refugees the authorization to work upon their arrival (Shisheva et al., 2010). At the same time, employers could be reluctant to hire them, uncertain about the asylum application results: their efforts to train asylum seekers could be vanished by a failure in the application. This approach goes hand in hand with the externalization of the European Union's borders, creating a parallel between migrants fragmented journeys to reach the EU and their fragmented integration trajectories (Collyer, 2010).
Finally, host countries' policies to attract labour migrants (e.g. quota systems, systems based on both quota and qualifications) are not designed to meet refugee needs. Ad hoc policies are needed to integrate refugees in the labour market, including language training and other forms of support, including mental health issues (Bevelander, 2016;Ruiz & Vargas-Silva, 2018).

The case of Italy
Because of its geographical position and ties with countries situated along the main migration routes towards Europe, Italy received an unprecedented number of 447,600 asylum applications between 2014 and 2018 (Eurostat, 2020a). Despite being considered a 'new country of asylum', Italy has received a not negligible number of asylum applications in recent decades, especially during the Yugoslav wars from the 1990s to the early 2000s (Bona, 2016), and the unrest in Northern Africa in 2011. Around 400,000 applicants claimed asylum in Italy between 1985 and 2013 (Eurostat, 2020a). However, the overall perception of Italy as a country of transit rather than a destination has pushed the topic of refugees' integration into the Italian labour market into the background (European Parliament, 2017). Refugees' economic integration in the Italian context and in general in southern European countries, characterized by low penalization in employment and concentration in low-skilled jobs (Fellini, 2018;King et al., 2000), has been less investigated compared with northern and continental Europe, leaving several open research questions. Most of the studies on migrants' labour market integration show that, especially in the first years after migration, refugees have lower employment rates compared with migrants admitted through other entry channels (e.g. Bevelander & Pendakur, 2014;Martín et al., 2016;OECD, 2015). Refugees who settled in Italy in the past may have experienced the same difficulties in economic integration compared with other migrants documented in northern and continental Europe (Korac, 2003). However, there is the possibility that they instead 'benefited' from the distinctive Italian pattern of segmented early incorporation into the labour market, achieving a level of economic integration similar to migrants admitted according to other entry channels. Another open question is how refugees' economic integration compares to that of undocumented economic migrants from the same country of origin that may themselves be de facto in refugee-like situations. These two subpopulations share a common experience of factors such as poverty, inequality, discrimination, human rights violations and conflicts in the origin country. In such contexts, there are great difficulties in distinguishing economic migrants and refugees (Pastore & Roman, 2014;Van Hear, 2011). However, it should be stressed that after arrival, their treatment is different: asylum seekers receive some assistance, while irregular immigrants do not.
A reflection on and an assessment of refugees' labour market integration in Italy is timely and needed to inform policies aimed at the management of the high number of refugees who have applied for asylum in Italy in the last years and in the light of the expected economic crisis driven by the COVID-19 pandemic.
The current study use data collected by the Regional Observatory for Integration and Multiethnicity (ORIM) of the region of Lombardy in Italy. This study contributes to the existing literature by analysing data on refugees' position and wages in the Italian labour market from 2001 to 2014-a period that mostly precedes the recent refugee crisis -to assess possible gaps in labour market integration compared with other migrants. Moreover, thanks to the unique nature of our data, we compare refugees to both regular and undocumented migrants for each year covered by the surveys.
Do differences by entry channels fade in the Italian labour market?
In Italy, the endemic difficulties in distinguishing refugee from other migrants have been favoured by immigration policies for many years (Ambrosini, 2014) for asylum seekers and refugees is evident in the earlier studies on integration of such population: especially in the first years after their arrival in Italy, refugees and undocumented migrants from countries characterized by political instability, poverty, inequality and widespread human rights abuse were vulnerable, fuzzy subpopulations (Ambrosini, 2014;Bolzoni et al., 2015;D'Angelo, 2019;Schuster, 2005a). Asylum applicants whose application was rejected, in most cases, followed the typical southern European pattern, facing an initial phase in a situation of undocumented migratory status and illegal employment (Fullin & Reyneri, 2011). Recognized refugees followed a similar pattern, with the legal status granted according to the protection channel as the only difference. One of the main features of the Italian economic integration pattern is that employment has legitimized refugees' settlement more than protection has. Scholars pointed out that back in the past it was easier to obtain legal status through the channel of economic migration than through the asylum route (Ambrosini, 2014;Korac, 2003). Being considered migrants instead of refugees stimulated agency and indirectly helped them achieve sufficient integration, albeit with great suffering due to very harsh initial conditions (Ambrosini, 2014). The significant capacity to integrate workers in the regular and irregular Italian job market and the little de facto distinction between different entry channels-that generally also characterize the job markets of southern European countries (Fellini, 2018)-should, in theory, represent a 'favourable situation' at least for the early economic inclusion of refugees in the Italian labour market. The economic inclusion of refugees was driven by the Italian labour market's broad capability to incorporate foreign-born workers although, even more than in the average of European countries, in segmented and ethnicized low-income and low-profile jobs (Zanfrini, 2014).
Research has shown that during and after the crisis, labour market outcomes worsened, with immigrants hit harder than natives by the crisis. However, if the increase in the penalization in unemployment rates was relatively limited, occupational segregation in the Italian labour market became stronger than before the crisis (Fellini, 2018).
Legal barriers to labour market participation have been gradually reduced over time. Until 2005, asylum seekers waiting for a decision about their application could not work. Decree 140, 30/05/2005, article 11 gave asylum seekers whose application was still pending after 6 months the possibility to work. More recently, decree No. 142 of 15/07/2015 article 22 further reduced the period to two months since the application and gave them access to Employment Offices on equal terms with natives and labour migrants.
Migration policies and the labour market structure actively reduced differences by entry channels in Italy.
Still, we also need to briefly remark the role of key structural differences between refugees, irregular and regular migrants in shaping their labour market outcomes. Refugees in Italy are mainly childless young men. Gender segmentation helps immigrant women, even if they are irregular, to find employment in the domestic and healthcare sectors. These sectors were also less affected by the 2008 crisis than the manufacturing and construction industries, where the male component is concentrated (Fellini, 2018). Another essential feature is the presence of the family. Usually, irregular migrants and refugees are alone in Italy, whereas regular ones increasingly live with their reunited or newly formed families. Family migration could hamper women's participation in the labour market, especially when considering migrants coming from countries and regions where female employment rates are particularly low (e.g. the Middle East and North Africa). For female irregular immigrants and more generally for female migrating alone, having a family in the homeland is a strong incentive to work (Cvajner & Sciortino, 2010).
Indeed, results from the 2014 labour force survey show that migrants who entered Italy by the international protection or asylum channel have a lower employment rate than migrants who arrived for work reasons. Still, their employment rate is higher compared with those observed among the refugees in many other EU member states, including notably Sweden, Finland and Belgium (Martín et al., 2016).
Given this framework, we aim to answer the following research questions: (Q1) Were refugees in Italy more at risk of unemployment than other migrants, including undocumented migrants and/or co-nationals (i.e. sharing the same country of birth) between 2001 and 2014?
By fitting models on the refugees' co-nationals, we can control unobserved effects restricting the focus on individuals that share the same background and may, therefore, be in a refugee-like situation. Group specificities and negative cultural norm images that impact job market integration (Confurius et al., 2019) may, for example, disproportionally affect communities characterized by a large number of refugees and undocumented migrants.
(Q2) Was there a similar regular employment occurrence among refugees compared with all other regular migrants and regular co-nationals between 2001 and 2014?
With this analysis, we want to assess whether refugees are penalized in the access to the regular labour market compared with other migrants or, on the contrary, their status is beneficial to the process of finding a legal job.
(Q3) Are refugees more likely to receive the lowest average job-related income compared with other migrants between 2001 and 2014?
With this analysis, we want to deepen the reflection beyond the dimension of mere inclusion, analysing the risk of being trapped into lowest-wage jobs. Especially in the Italian context, regular employment may be a weak indicator of a good job. Irregularly employed migrants predominantly work in the same sectors of activity as those regularly employed. The borders separating formal and informal economy are not easy to trace in a landscape marked by the deregulation of the labour market and the growing precariousness of work relations (Zanfrini, 2019a). For this reason, we also included in our analysis the average monthly wage. Despite not being a clear indication of a 'bad job', adaptability to extremely low wages, apparently advantageous in the short-term, especially for newcomers, may be problematic in the long run.

Lombardy as the context of our study
Our study is based on regional data collected in the northern region of Lombardy. There is agreement on the fact that attention should be paid to the geographical diversification of migrants' inclusion models to frame analyses correctly (Zanfrini, 2014). Beyond national trends, the local perspective is crucial in the interpretation of migrant integration into the labour market. Lombardy, especially regarding the area of Milan, is one of the wealthiest regions of Europe (Eurostat, 2020b). It is the most populated Italian region and hosts about a quarter of Italy's foreign population. Indeed, Italy's complex economic geography, characterized by marked regional and occupational concentration, has been fundamental for immigrant workforce placement. Several occupational and integration models are present at the same time in the country. The focus on Lombardy offers the opportunity to analyse refugee integration into different territorial settings (Ambrosini, 2013): the areas whose economy is based on small and mid-size enterprises, the metropolitan economies model (Milan) and the seasonal activities model in the central and northern areas where immigrants work in the tourism and the services.

MATERIAL S AND ME THODS Data
This analysis is carried out on the pooled data set resulting from 14 rounds of the Italian survey on immigrants carried out by the Regional Observatory for Integration and Multiethnicity of Lombardy (ORIM; Polis, 2019).
ORIM is an annual cross-sectional survey carried out using the centre of aggregation sampling method (Baio et al., 2011). Studies based on centres of aggregation are specifically designed to collect information on a representative sample of migrants, which also includes undocumented migrants. The interviews are performed face to face in Italian or a foreign language by skilled foreign interviewers who have undertaken specific training. The target population includes men and women with a migrant background 3. aged 14 and over (including undocumented, naturalized and second-generation migrants). Pooling data from different years help mitigate the problem of the small number of undocumented migrants and refugees in annual surveys. Table S1 in the Appendix S1 provides details about each survey included in the pooled data set.
We are aware that the data have some limitations. First, ORIM surveys do not specifically target the refugee population. For this reason, we cannot refine our analysis by taking into consideration differences according to the type of asylum-related permit and between recognized refugees and asylum seekers (Bakker et al., 2017).
Second, by using ORIM data, we can only account for the experience of migrants holding a valid asylum application or a permit for humanitarian reasons, therefore missing out on information about those who have converted their refugee permit into other typologies. However, we can partially control this bias by controlling for the number of years since migration in the statistical model. Finally, while on the positive side, the use of a pooled dataset provides a large sample size and allows comparison between different years, on the other, it restricts the availability of information to that collected in each of the 14 surveys.
Despite these limitations, this data set provides comparable information on a large number of years. It is unique in the southern European context: it includes information on refugees living outside and inside accommodation centres and facilitates comparisons with other migrants, including undocumented migrants.

Method
The analysis of migrants' and refugees' labour market integration implies some complications. First, being employed and unemployed are conditions that depend on participation choices; in other words, leaving inactivity is the result of self-selection. Second, the possibility of being legally hired depends on legal status. Another crucial point is related to refugee status. The refugee status is assigned following rules and judgement by the territorial commission, and it implies a different condition compared with other migrants. This status is available only to a selected subpopulation with characteristics that may impact job integration. In other words, being granted the refugee status may be an endogenous condition for job market participation and integration.
To deal with these issues, we analyse unemployment conditional to endogenous selection into labour market participation. Consistently, we analyse regular employment and monthly wage conditional to endogenous selection into employment status (regardless of the regular or irregular job position). We operationalize our models as follow: we fit eight probit models with endogenous sample selection and treatment. Tables S2 and S3 in the Appendix S1 contain details on the eight models.
We tested the endogeneity of the treatment in every model, using the level of the country of origin's Human Development Index in the year of the migrant's arrival in Italy (HDI, 2020) as an instrumental variable. The refugee status is endogenous for the models that analyse unemployment and average monthly wages of refugees compared with all regular migrants (Models 1 and 6) and regular co-nationals (Models 2 and 7). In the models that examine refugees compared with undocumented co-nationals (Models 3 and 8) and in models analysing regular employment (Models 4 and 5), no endogeneity was detected. This is an interesting result in itself. The differences between the subpopulations of refugees and undocumented co-nationals are so limited that we can model the refugee status as an exogenous treatment (i.e. assigned randomly). The same occurs for the model on regular employment fit on employed migrants. Here, the effect of the selection into employment for refugees is so strong that employed refugees are no longer significantly different from other migrants and, again, the treatment can be considered exogenous. However, the treatment is endogenous for Models 6 and 7 that analyse work-related income suggesting that refugee characteristics interact with the typology of jobs that refugees hold.

Measures
The variables used in the models were selected for their relevance according to the literature on the topic. They were defined as follows:

Dependent variables
The dependent variable for Models 1-3 is unemployment status (yes, no), for Models 4 and 5 is regular employment (yes, no), and for Model 6-8 is receiving an average monthly work-related income equal to or below the lowest quartile (yes, no). The lowest quartile was calculated on the whole migrant population surveyed on each annual survey to account for annual differences in the cost of living.
Data about monthly work-related income have the limitation that no information is available about the average number of hours effectively worked. For this reason, we fit additional models using quartiles according to the working condition (i.e. the type of contract for regular work or the self-declared stability of work for irregular work) and quartiles according to the type of job (profession) as consistency checks. These additional models brought to results very similar to Models 6-8; for this reason, they are not presented or discussed in the paper but are available upon request.

Endogenous sample selection
The variable accounting for endogenous selection is labour market participation (yes, no: homemaker, student, unable to work, retired) for Models 1, 2 and 3. Models 4 and 5, instead, are restricted only to labour market participants, and the variable accounting for endogenous selection is the employment status (yes: regularly or irregularly employed, no: unemployed).

Refugee status (yes, no)
This dummy variable accounts for being a refugee. We considered as 'refugees' those who have applied for asylum, have been recognized as refugees or have a residency permit for humanitarian, subsidiary or temporary protection.

Other variables
Gender (male, female), age (in single years, with a squared term) and years since migration (in single years, with a squared term) Many studies have shown that the employment outcomes of migrants improve proportionally to their age and length of residence. Women's labour market outcomes are usually worse than men's, especially when they were born in non-EU countries (Huddleston et al., 2013). Education (primary or none, lower secondary school, upper secondary school, university degree) Even though education is a crucial element in employment integration, immigrants' qualifications are often not recognized. Moreover, employers do not seem to place much trust in foreign educational qualifications and work experience. Skills and qualifications often do not fit the labour demand in the country of immigration (Huddleston et al., 2013;Lemaître, 2007). However, better-educated immigrants are usually more proactive in using their resources. Education can be regarded as a proxy for greater personal resources, which enable migrants to achieve their migratory goals more easily, to learn the language beforehand, and to manage themselves more effectively in the labour market. Better-educated migrants are therefore expected to have lower levels of unemployment and higher participation in the regular market job even if their qualifications may not match the requested skills.
Year of arrival (<1991, 1991-2000, 2001-2005, 2006-2008, after 2008) The year of arrival helps to roughly check for progress made by Italy in refugee reception and policies. It also indirectly takes into account changes in economic conditions (e.g. the beginning of the economic crisis from 2008) and different waves of refugees arriving in Italy.

Religion (Muslim, Christian, other or none)
The perception of higher cultural proximity or local preferences based on race, religion or ethnic origin may affect the chances of finding a job. Christian migrants may take advantage of the mediation of the Catholic Church and religious nongovernmental organizations to find a job (Ambrosini, 2014;Korac, 2003;Lemaître, 2007).

Presence of a partner (yes, no) and number of cohabitant children (continuous)
The presence of the family usually has positive effects on social integration. However, cohabitant children can discourage the labour market participation of foreign-born immigrant women (Huddleston et al., 2013).

Descriptive analysis
Asylum is a global phenomenon, but it is mostly males who make the journey to Europe (Wanner, 2016), and Italy is no exception. A high presence of young, unmarried, childless men has always characterized refugees in Italy and is consistently found in our sample ( Table 1).
The main sending countries of refugees to Italy have been changing over time due to the arising local political crises and unrest. After Italy became a consolidated country of immigration in the first decade of the 2000s, the difference in the median year of arrival between refugees and other migrants widened ( Table 2). For many years, Italy was mostly a transit country for refugees migrating onward after arrival (Schuster, 2005b). Moreover, the possibility provided by the law of converting residence permits for asylum-related reasons into permits for work further explains this finding.  The proportion of the inactive population is 7.0 per cent among refugees compared with 14.9 per cent of other migrants.

Multivariate analysis
The first three models analyse the transition from unemployment to employment. The first model aims at assessing the effect of the refugee status on the incidence of unemployment. The model is fit on all regular migrants while controlling for endogenous self-selection into the labour market and the endogeneity of the refugee status. Readers interested in factors driving selection and refugee status will find models in the Appendix S1 (Tables S4-S6).
The coefficients for models with endogenous selection and treatment are difficult to interpret. To produce better understandable results, we calculated predictive margins for each model ( Table 3).
As we considered the refugee status as a treatment, we are interested in analysing the average treatment effect on the treated (ATET). In the case of our study, this is the average effect of the refugee status on unemployment among those individuals who got the treatment (i.e. were granted the refugee status).
Results from Models 1 and 2 show that regardless of the population we considered (all migrants or just conationals), being a refugee increases unemployed by 0.38/0.40.
Results from Model 3 fitted on refugees and their co-national undocumented peers allow further considerations. Even when we compare job market integration outcomes of refugees to the most vulnerable members of the same communities (with similar socio-demographic characteristics and same country of origin background), the refugees have an occurrence of unemployment increased by 0.189 compared with undocumented co-nationals.
In the case of model 3, it makes sense to consider other treatment effect estimation measures. First, we may consider the expected effect on unemployment in case each asylum seekers' co-national undocumented peer had been granted the refugee status (i.e. estimating the average treatment effect in the population, ATE). In this case, the expected proportion of unemployed would rise by 0.194 in the whole population (refugees + undocumented migrants). Model 3 also suggests that the expected percentage of unemployed in the population considered would be 19.7 per cent in case no one was granted asylum, while it would reach 39.9 per cent in case everyone was.
Models 4 and 5 analyse the subsequent step (formal inclusion) of labour market integration by evaluating the impact of the refugee status on having a regular job among migrants who are employed. The model is fit on all regular migrants that participate in the labour market while controlling for endogenous self-selection into employment. As in the case of Model 3, the effect of endogenous selection is so strong, especially among refugees that treatment of having a refugee status is not endogenous.
Models 4 and 5 show that the refugees have a lower occurrence of regular employment compared with both all regular migrants and the subsample of regular co-national peers. Being a refugee decreases the chance of being regularly employed by a significant 0.168 for those who have been granted the refugee status compared with all regular migrants, and by 0.174 compared with co-national peers.
Models 6, 7 and 8 account for low average monthly work-related income defined as an income equal or below the lowest quartile calculated on a yearly basis on the whole population surveyed. If we consider all regular migrants or only their regular co-nationals, being a refugee increases the occurrence of average monthly workrelated income below the lowest quartile by 0.36. On the contrary, no differences exist if we compare refugees to their undocumented co-nationals. As happens for Model 3, when dealing with undocumented co-nationals, it makes sense to consider the average treatment effect in the population (i.e. the expected impact on average monthly work-related income in case each asylum seekers' co-national undocumented peer was granted the refugee status). In that case, and consistently to ATET, the incidence of average monthly work-related incomes below or equal to the lowest quartile would rise by a non-statistically significant 0.081. Model 8 also implies that

TA B L E 3
Average treatment effects and potential-outcome means for Models from 1 to 8 The incidence of lowest wages is higher among undocumented migrants and refugees compared with regular migrants, with no significant differences between undocumented migrants and refugees ( Figure 3). Besides, while although significantly higher among refugees, the incidence of low wages was slightly decreasing from 2001 to 2014 (Q3).

DISCUSS ION
Research has repeatedly documented the existence of a gap in the labour market integration of refugees compared with other migrants (Bakker et al., 2017;Bauer et al., 2017;Bevelander, 2011;Cangiano, 2014;Connor, 2010;Delaporte & Piracha, 2018;Ruiz & Vargas-Silva, 2018). However, the Italian context, and particularly Lombardy, where migrants usually experience a more swift inclusion in the labour market, although in low-qualified and poorly paid jobs, was less investigated compared with other long-established migration countries in the centrenorth of Europe. We focused on the years from 2001 to 2014 to assess the integration of refugees that mostly came to Italy before the 2014-2017 refugee crisis. Our research shows that even in the Italian framework and in F I G U R E 3 Adjusted predictions at the means for the probability of having an average monthly wage equal to or below the lowest quartile among regular migrants participating in the job market by year of the survey according to Model 6 particular in Lombardy, one of the wealthiest regions of Europe, refugees show more considerable difficulties in integrating into the labour market compared with other migrants and notably compared with their co-nationals, including those undocumented.
The gap in labour market inclusion is the primary and most serious problem in the years we considered (Q1).
We also show that even though refugees and undocumented migrants are two similar and fuzzy subpopulations, the refugees' risk of unemployment is increased by 19.4 per cent compared with their undocumented co-nationals.
Our research shows that a slight penalization exists for refugees also regarding the occurrence of regular employment among the employed population with a residency permit (Q2). The incidence of average monthly wages below the first quartile is also higher among refugees compared with other migrants. Notably, the occurrence of very low average monthly wages is similar among refugees and undocumented migrants (Q3).
By providing evidence of higher unemployment, lower regular employment and higher occurrence of extremely low wages, our findings confirm the results of previous studies on a gap in labour market integration for refugees in Italy in the years 2001-2014.
We also contribute to the literature showing that, far from taking advantage of their legal status, refugees have a higher risk of unemployment than undocumented migrants from the same countries of origin and share a similar occurrence of extremely low wages.
However, the picture is not all negative if we consider the trend between 2011 and 2014: our crosssectional data show improvements over the period considered. Despite an overall increase in unemployment, the gap with other migrants in employment and regular employment among the employed is reducing in recent years.
What are the implications of this research for the management of post-2014 refugees flows and in light of the COVID-19 pandemic? It is too early to assess the labour market integration of refugees who arrived in recent years or the incoming crisis's impact. Moreover, the COVID-19 pandemic is expected to have a very harsh impact on the economy of Lombardy, one of the worst affected regions of Europe. However, we can provide some valuable evidence from the experience of the recent past to address future policies. The broad capacity to include foreign workers that characterized the Italian labour market in the past years is not enough to overcome the difficulties in labour market integration that typically affect refugees. Our data show that we cannot merely rely on the Italian job market features of 'low unemployment risk-no access to skilled jobs' pattern (Fellini, 2018), expecting refugees to overcome the gap induced by their unobserved vulnerabilities by themselves and show performances similar to other migrants. Integration policies addressed to refugees should be tailor-made thus taking into account their migrant trajectory and their specific needs and vulnerabilities. States should, therefore, provide ad hoc policies for labour market integration of refugees to avoid the trade-off between the aim of supporting a rapid transition towards autonomy and a substantial risk of permanent segmentation into worst and lowest-paid jobs (Zanfrini, 2019b). Finally, a new amnesty for irregular migrants has been issued by the Italian Government during 2020 (art. 103 DL 34/2020 and Law 77 of July 17th 2020). The amnesty was devoted only to specific sectors, particularly touched by workforce shortage during the pandemic (i.e. agriculture, breeding and fishery, domestic work and care work) and received 207 thousand applications between July and Mid-August. The regularization was undoubtedly crucial, especially in a period of sanitarian and economic crisis. Those who had applied for international protection could also apply for regularization to have a greater chance to get a permit of stay. For rejected or pending asylum seekers obtaining a permit of stay through a regularization means following a well-known path in the Italian context. Ambrosini (2014) has well explained that refugees have always been considered as migrants and therefore legitimated by their complementary economic role in a national economy that suffers from a shortage of low-cost labour. Regularizations cannot substitute a harmonized immigration and integration policy devoted to asylum seekers and refugees.
As we stressed, Italy is a country that lacks a real 'culture of asylum': the 2008 crisis has impacted heavily on this mechanism, fuelling fears of competitiveness with unemployed, low-educated and low-skilled natives (Zanfrini, 2014). Therefore, the perception of immigration as a security and public order problem has recently prevailed over economic acceptance. The COVID-19 crisis is expected to have a far worst impact. Given the recognized fragility of migrants in labour market integration, early labour market access policies-some of them successfully experimented in Italy (European Parliament, 2017)-should be prioritized for refugees to be legitimated by their status and by their work. At the same time, we hope that reversing current trends, efforts will also be made to spread the 'culture of asylum' (Korac, 2003;Zappacosta, 2019) in Italy as a fundamental human right.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/imig.12884.