Don’t Stop Me Now: Cross-Border Commuting in the Aftermath of Schengen

1 Introduction

The economic integration of European countries was designed to create a unified economic area, which was achieved through the implementation of the ‘four freedoms’: capital, goods, services and labour. Fostering economic ties between its countries was one of the main objectives of the EU policy to create ‘an ever closer union’ (König and Ohr 2013). Despite this strong integration policy and in contrast to the goods and services markets being very integrated, the labour markets in Europe remain quite distinct (Baldwin and Wyplosz 2015), questioning the effectiveness of EU policies to promote internal mobility.

In this paper we aim to study the success of one of the most important EU policies implemented to promote labour mobility: the Schengen agreement. By abolishing internal border controls, the policy makes cross-country inter-regional travel-to-work journeys shorter and easier, thus creating strong incentives for workers to cross-border commute across countries. We use Switzerland, which joined the Schengen area in December 2008, as a case study to assess the effectiveness of such policy for two main reasons. First, the implementation of the Schengen agreement after the granting of the freedom of movement to all EU-15 and EFTA citizens in 2007 allows us to identify the specific effect of the implementation of Schengen in promoting labour mobility. Second, the number of workers employed in Switzerland but residing in a neighboring country have almost tripled since 2000, accounting nowadays for approximately 6.5 percent of the Swiss labor force and 5.1 percent of the Swiss working age population (Dorn and Zweimuller 2021; European Commission 2017; Edzes, van Dijk, and Broersma 2022). Figure 1 shows that the large majority of cross-border commuters to Switzerland reside in France, Germany and Italy.

Figure 1:

Cross-border commuters to Switzerland (2005–2015). The figure reports the total number of cross-border commuters to Switzerland in the period 2005–2015, independent of the country of origin (solid black line) and those coming from France, Germany and Italy (solid gray line). Source: Swiss Federal Statistical Office.

In the past decade, the combination of an increasing number of refugees, growing migratory pressure, security concerns and a rather weak economic recovery has led governments to re-introduce temporary internal border controls from time to time at certain crossings, with large economic costs for business and citizens (Ademmer et al. 2015). Similar measures have been implemented during the Covid-19 pandemic to limit the spread of the disease (Carrera and Luk 2020). The deeper ongoing discussion is about the possibility to permanently re-introduce border controls and terminate the Schengen agreement (Bertelsmann Foundation 2016). The cost of such measure for the EU economy has been estimated to amount to more than 100 billions euro (EPRS 2016; France Strategie 2013, 2016); to Switzerland alone, it could cost up to 1.5 billion Swiss Francs per year. For commuters, the cost would not only be in terms of time lost at the border, estimated to range between 1.3 and 5.2 billions euro (Eisele 2022), but also in terms of lost job opportunities. For France alone, this would imply the loss of up to 10,000 cross-border workers, amounting to a loss of 150–300 millions euro annually; for Switzerland, it could imply a decline in the rate of cross-border commuters between 27 % and 62 %.

Within a Difference-in-Difference (DiD) estimation framework, we choose our sample to include all individuals in the labour force who resided during the period of our analysis (2005–2015) in regions of Germany, Italy, France and Switzerland, which share the border either with Switzerland or with a Schengen country.^[1] Among those, we define as treated all individuals who gained the chance to commute cross-border to Switzerland without barriers due to Switzerland joining the Schengen area. Those are individuals who lived in regions of Italy, France and Germany, which share the border with Switzerland and individuals who lived in regions of Switzerland, which share the border with Italy, France and Germany. The control group is made by individuals who lived in regions of Italy, France and Germany, which share the border with countries already in the Schengen area, but not with Switzerland. Due to data availability, we exclude from our sample residents of Austria, which nevertheless account for a negligible percentage of the total number of cross-border commuters.^[2] We also exclude from our sample individuals living in Lichtenstein as the country joined the Schengen area in 2011, i.e. during the period of analysis.

Information about the actual changes at the frontiers between Switzerland and Italy, France and Germany due to the Schengen implementation is lacking. Anecdotal evidence provides quite a contradictory picture. On one hand, systematic Swiss-German and Swiss-French border checks were described as the norm up until 2008, which led to long queues of cross-border commuters at the frontier.^[3] According to other sources, however, random checks were happening even before the implementation of Schengen, making the official removal of border checks a less critical change (Confoederatio Helvetica 2004). Nevertheless, we envision overall a positive impact of the legal removal of border checks for cross-border commuters, as generally their “life got easier” and “crossing the border became much faster”.^[4] In fact, according to the European Parliament (2016), depending on the intensity of the checks, the time lost at the border could have been up to 10–20 min for passenger cars.^[5] Moreover, the costs linked to time delays for in-commuting and out-commuting workers could have amounted to a range between 17 and 37 min.

After controlling for several individual and regional features, we find that among individuals in the labour force the probability of crossing the border for work is approximately 0.5 percentage points higher after the implementation of Schengen. This result is shown to be consistent to multiple alternative specifications and several robustness checks. We also disentangle the effect according to the direction of traveling: we estimate the effect from Switzerland to be smaller and equal to 0.3 percentage points, while the effect to Switzerland to be larger and equal to 0.5 percentage points. When we restrict the sample to inter-regional commuters the probability to cross-border commute is higher and approximately equal to 4 percentage points. The estimate of a triple difference model informs us about older tertiary educated males being more likely to commute to Switzerland after the implementation of Schengen. We also explore the potential channels through which the Schengen effect operated: we show evidence of a redirection of inter-regional commuters from internal and other foreign regions to Swiss ones, while we do not find evidence of geographical migratory flows to bordering regions nor of sectoral transitions with the purpose of commuting. Finally, we provide evidence of no significant labour market changes in local labour markets of treated regions after the policy change.

The rest of the paper is organized as follows. In the next section, we review the related literature, while in Section 3 we describe the institutional background. Section 5 discusses the identification and the empirical strategy, while Section 4 presents the data. We illustrate the main results and several robustness checks in Section 6, while in Section 7 we explore the potential channels which led to higher commuting rates. In Section 8 we investigate local labour market effects and finally, Section 9 concludes the paper.

2 Literature Review

This paper fits into the literature which studies the role of borders as an obstacle to labour market integration. Evidence has been provided of a spatial segmentation of labour markets along national borders in EU-15 countries due to significant border impediments on both sides of national frontiers (Niebuhr and Stiller 2004). Even within countries, regional borders could represent a strong impediment to commuters and impose strong spatial imperfection in the labour market (Capello, Caragliu, and Fratesi 2018; Persyn and Torfs 2016). Policy measures implemented to lower existing barriers have been proven to be effective in improving labour market integration. The harmonization of accounting and auditing standards has increased significantly international labour migration in the accounting profession compared to other professions (Bloomfield et al. 2017). Policies such as the abolition of the border controls in the Schengen area and the introduction of the Euro currency have led to an improvement in cross-border integration, although had no effect on labour market integration due to strong and persistent language barriers (Bartz and Fuchs-Schündeln 2012). Our paper adds to this literature by directly identifying and quantifying the effect of the implementation of the Schengen agreement in Switzerland on cross-border commuting.

This paper also fits into the literature that with a DiD estimation strategy evaluates the effects of newly implemented policy interventions on cross-border commuting and, ultimately on wages and labour supply (Bigotta, Oscar, and Losa 2012; Dustmann, Schönberg, and Stuhler 2016). Recently, Hafner (2022) shows that the gradual implementation of the freedom of movement to cross-border commuters between EU and Switzerland increased cross-border commuting, with a small positive impact on the wages of workers in France. However, their analysis stops in 2007, just before the implementation of Schengen. From a different perspective, Dicarlo (2022) studies the impact of the same exogenous event on the labour demand in one of the sending country, i.e. Italy. On top of narrowing the focus to a single country, they also do not disentangle the Schengen impact from the freedom of movement effect. With a similar approach, Anelli et al. (2023) analyze the adjustments of Italian firms in reaction to the large outflow of workers following the Great Recession. Other recent papers study the effect of the implementation of the Swiss-EU Agreement on the Free Movement of Persons on the quality of products, trade, and effective global value chains Ariu (2022) and on the number of patents and inventing teams (Cristelli and Lissoni 2020). The most closely related papers are the ones by Beerli et al. (2021) and Bello (2020). Beerli et al. (2021) investigate the labour market effect of the gradual implementation of the freedom of movement to cross-border commuters in bordering regions of Switzerland. They analyse three phases of the implementation: a pre-phase before 1999, a first phase between 1999 and 2004 and a second phase between 2004 and 2010. They find no significant effect of the treatment in the first two periods but a strong positive effect on Swiss labour supply and demand in phase two. Our paper differs in three main aspects. First, we study the effect of the implementation of the Schengen agreement using a different identification strategy and different data on multiple countries, while they focus on Switzerland only. Second, we are able to identify the Schengen effect, while by pulling together the years between 2004 and 2010, Beerli et al. (2021) capture multiple events, such as the changes in the freedom of movement to cross-border commuters in border regions (in 2004), the liberalization for all EU-15/EFTA workers in the whole country (in 2007), and the implementation of the Schengen agreement (in 2008). Finally, while they focus on the labour market effects in Switzerland, our focal point is the cross-border commuting phenomenon in both directions. Bello (2020) analyses the responsiveness of cross-border shopping and commuting between Italy and Switzerland to changes in the exchange rate in the period 2005–2015. Her results show that a 10 % appreciation in the Swiss franc lead to an increase in the number of vehicles crossing the Swiss-Italian border by approximately 2 % more than in the rest of the Ticino canton and to a 2.5 % increase in the number of cross-border workers in municipalities within a driving distance and in their labour supply. Although the outcome of interest is similar,^[6] in her analysis Bello (2020) discards the implementation of the Schengen agreement in December 2008, which could have strongly affected the commuting flows. Her analysis is also restricted to the Italian border with Switzerland, limiting the scope of the paper. Using a larger sample which includes three out of the four Swiss borders and considering bilateral flows to and from Switzerland, our paper complements this work by identifying the specific effect of the implementation of the Schengen agreement on cross-border commuting for work, after controlling for the exchange rate fluctuations.

3 Institutional Background

The Schengen agreement allows access to the Schengen Area, a zone in which internal border checks are abolished, vehicles are allowed to cross borders without stopping, and residents in border areas are granted freedom to cross borders away from fixed checkpoints. Swiss citizens voted by a 55 % majority to join the Schengen area in 2005 and on 27 November 2008, the interior and justice ministers of the EU in Brussels announced Switzerland’s accession to the Schengen passport-free zone from 12 December 2008. Since then, the land border checkpoints have officially remained in place only for goods movements, and systematic checks on individuals at the frontier have been completely abolished. In practice, there is no clear understanding about the actual changes, compared to was already happening at the borders. The BBC documented that “in Geneva and Basel in particular, where tens of thousands of people live in France or Germany, but work in Switzerland, queues at the border have long been a source of irritation”, before the implementation of Schengen.^[7] Other sources reported that “Swiss-German border checks used to be the norm up until 2008, when Switzerland joined the Schengen Area”.^[8] On the other hand, the Swiss press reported that although from December 2008 the Swiss Border Guard increasingly took part in Schengen substitute measures at the EU external borders, only 150 guards were redeployed to replace international police officers at Swiss airports to carry out controls on passengers crossing external Schengen borders (Confoederatio Helvetica 2004). Also, the number of guards at the borders appeared to be reduced only by approximately 70 units (Il Parlamento svizzero 2011). Moreover, at some borders, checks were already performed “by sampling” a small number of vehicles crossing the border to Switzerland on a daily basis. The chief of the Swiss border police also admitted to the local radio that most systematic checks had already been removed before Switzerland joined the Schengen area and people would have felt the difference, but only slightly.^[9] Overall, the anecdotal evidence on border checks seems to provide contradictory pictures and seems to suggest that the approach to border checks was quite heterogeneous across different borders. Nevertheless, it is likely that cross-border commuters benefited from the official removal of border checks for whom “crossing the border became much faster”^[10] and made “their lives easier”.^[11]

More recent developments put the Schengen agreement in Switzerland at risk. In February 2014, the popular campaign “Against mass immigration” pushed for the re-introduction of immigration quotas (Abu-Hayyeh, Murray, and Fekete 2014), leading Switzerland and the EU to sign on December 2016 an agreement requiring employers to prioritize Swiss-based job seekers, penalizing cross-border commuters. Moreover, the terrorist attacks in Paris (November 2015) and Brussels (March 2016) and the consequent geopolitical turmoil also led several EU Member States to re-establish border controls, disrupting free movement in the Schengen area, mainly in border regions (Evrard, Nienaber, and Sommaribas 2020). Finally, as a result of the COVID-19 outbreak in 2020, Switzerland closed dozens of its border crossings and reduced others to rush-hour opening times, with evidence of ‘monstrous traffic jams’ at the borders and waiting times extended to up to 3 h (The Local 2020).

4 Data and Descriptive Statistics

We use data from the European Labour Force Survey (ELFS) for the period 2005–2015.^[12] We define cross-border commuting based on place of residence at the time of the interview and working place at the time of the interview being located in two different NUTS2 regions in two different countries. We have information at NUTS2 level for Italy, France and Switzerland, which corresponds to the first-level administrative division of the country (regions), while only information at NUTS1 level (macro-regions) for Germany and Austria. We exclude from our sample individuals living in Lichtenstein as the country joined the Schengen area in 2011, i.e. during our period of analysis.^[13] We exclude individuals living in the Italian region ITH4 and the German regions DE2, DE3, DE4, DE8, DED which share the border countries which joined the Schengen area during the period of observation.^[14] Finally, we exclude from our analysis individuals residing in Austria, as there is only one Austrian region which shares the border with Switzerland for a very limited number of kilometers, while sharing also the border with Germany and Italy. Nevertheless, according to the Swiss Federal Statistical Office, in 2014 among the 290,000 Europeans who commuted across the border to work in Switzerland, more than 97 % traveled from Germany, France and Italy. In the same year, according to the Swiss Labour Force Survey more than 99 % of Swiss residents commuting across the border traveled to Italy, France and Germany. We complement this dataset with Eurostat, Cambridge Econometrics, Bank of International Settlements (BIS) and OECD data to gather additional information at regional level. We construct all regional variables as the ratio between the value in the origin region and the average value among all potential destination regions abroad. These variables should capture the push-pull factors determining the decision of individuals to commute across the border. When deciding whether to work in the region of residence or abroad, we expect individuals to compare the economic conditions in the origin region with the economic conditions in potential destinations. For example, we expect that the higher the value of the unemployment rate in the origin region with respect to the average value in potential foreign destination regions, the higher the probability to commute across the border, i.e. a positive coefficient of the unemployment rate variable.^[15] We use OECD data on unemployment and youth unemployment to construct measures of unemployment at regional level for the years 2005–2015 for specific sub-categories of individuals. We use Cambridge Econometrics data on the share of employment by sector at regional level to capture the way the structure of regional economies has changed over time during different phases of the business cycle. In addition, we take into consideration the quality of the infrastructures by including a measure of road length between two regions (in kilometers), as provided by Eurostat. To take into account differences in real estate prices across different countries we use national data on house prices from the Bank of International Settlements (BIS). From Eurostat we also gather information on the exchange rate between all the countries considered and Switzerland to capture the effect of the exchange rate volatility on the decision to cross-border commute. Finally, to account for the language barrier, which in the literature has been mentioned as a rather important deterrent to mobility in Europe (Bartz and Fuchs-Schündeln 2012), we also use a measure of closeness of languages (Dyen, Kruskal, and Black 1992).^[16]

We focus on the period 2005–2015, during which the survey data collection was unchanged both over time and across countries and no suspension of the Schengen agreement in any of the countries considered was implemented (Section 3).

Our sample includes almost 3.1 million individuals, among whom we observe an increasing number of cross-border commuters. Specifically, the total number of commuters across the border went up from approximately 2000 individuals before 2009 to almost 4000 individuals in the years after 2010 (Figure 2(a)). When we confine our analysis to workers who commute towards Switzerland, we observe a spike starting from 2009, in line with the statistics provided by the Swiss Federal Statistics Office, which show that in the period 2009–2014 the number of people crossing the border to Switzerland for work has risen by almost 30 %. Looking at the flow from Switzerland to the bordering countries, although the numbers are much smaller, we observe an increasing trend from 2005, with a spike in 2011.

Figure 2:

Number of cross-border commuters to and from Switzerland (2005–2015). The solid grey line in Figure (a) reports the total number of cross-border commuters to Switzerland from France, Germany and Italy, while the solid black line reports the total number of cross-border commuters to other countries (not Switzerland). The solid black line in Figure (b) reports the total number of cross-border commuters from Switzerland to France, Germany and Italy. Source: ELFS data.

5 Identification and Empirical Strategy

The enlargement of the Schengen area to Switzerland in December 2008 represents an exogenous event which we exploit within a DiD framework. We select the labour force population (employed and unemployed individuals) as the sample on which to measure the policy-relevant treatment effect of removing the border controls on cross-border commuting. In fact, the Schengen implementation could have led not only commuters, but also non-commuting employed people and unemployed individuals to start commuting across the border. Being the removal of border checks bilateral, the policy is expected to have had an impact on commuting flows in both directions across the Swiss border. We define as treated all employed and unemployed individuals who live in regions of Italy, France and Germany, which share the border with Switzerland and all employed and unemployed individuals who live in regions of Switzerland, which share the border with Austria, Italy, France or Germany.^[17] The control group is made of labour force individuals (employed and unemployed) who live in regions of Italy, France and Germany, which share the border with another Schengen country, but not Switzerland (Figure 3).^[18] We exclude from our sample individuals living in Lichtenstein as the country joined the Schengen area in 2011, i.e. during our period of analysis. We also exclude individuals residing in Austria, as there is only one Austrian region which shares the border with Switzerland for a very limited number of kilometers, while sharing also the border with Germany and Italy. Finally, we exclude individuals living in the Italian region ITH4 and the German regions DE2, DE3, DE4, DE8, DED which share the border with countries which joined the Schengen area during the period of observation.^[19] The key aspect of this setting is that the control group is never observed to be exposed to the treatment, thus ruling out the possibility that a misclassification affects our sample split.

Figure 3:

Regions where treated and control workers reside. The yellow areas identify the Western European countries which belonged to the Schengen area in 2008. The grey areas identify the Western European countries which did not belong to the Schengen area in 2008. The red and blue areas identify the regions in France, Germany, Switzerland and Italy were treated and control individuals resided in 2008, respectively.

In compliance with the assumptions of our approach, Figure 4 reports the absolute number and the percentage of cross-border commuters in treated and control regions. Although in the years pre-Schengen, the number and the percentage of cross-border commuters were higher in treated regions, the trend was similar across the two groups (Abadie 2005; Bertrand, Duflo, and Mullainathan 2004).

Figure 4:

Cross-border commuters in treated and control regions. In Figure (a), the solid grey and black lines report the absolute numbers of cross-border commuters residing in treated and control regions, respectively. In Figure (b), the solid grey and black lines report the percentage of cross-border commuters (on total labour force) residing in treated and control regions, respectively. Source: Our own calculations using ELFS data.

To comply with the stable unit treatment value assumption (SUTVA), we report in Table 1 the demographic and work characteristics of labour force individuals in treated and control groups before and after the implementation of Schengen. The statistics show a close similarity between the features of individuals in the two groups and a good stability of such characteristics before and after the treatment.

We pool together data before and after the entrance of Switzerland in the Schengen area^[20] and we estimate the following equation:

where i identifies the individual, c the country of residence, r the region of residence, s the sector^[21] and t the time. We estimate the equation above using as outcome the probability to cross-border commute, which takes value one if the individual commutes across regions towards a foreign country (cross-border) and value zero otherwise. Post-Schengen is a dummy variable equal to zero for the years before Switzerland joined the Schengen area (2005–2008) and equal to one for the years after (2009–2015). Treated is the dummy variable which identifies the treated group, and takes value one for those individuals who reside in a treated region, and value zero for those individuals living in a control region. The matrix X _i,r,t includes a set of individual pre-treatment characteristics which may affect the individual probability to cross-border commute. The control variables include age, gender, marital status, education as well as work characteristics such as employment status, sector,^[22] occupation, type of contract, and firm size. Finally, ρ _r captures region fixed effects, σ _s sector fixed effects, μ _t year fixed effects, β _ct country times year fixed effects, γ _st sector times year fixed effects and ϵ _i,r,s,t is the individual iid error term.

In line with the recent literature (Beerli et al. 2021; Dustmann, Schönberg, and Stuhler 2016) and due to the direct exposition of regional labour markets to the treatment, we cluster the standard errors at region level (Abadie et al. 2017). Given that our key regressor is a cluster-specific binary treatment dummy and there are few treated groups, we adjust for the small number of clusters (Cameron and Miller 2015).

6 Results

In our baseline DiD model (Column 1 of Table 2) we compare the average outcome in the post-Schengen period with the entire pre-reform period (2005–2008), without any additional controls. We find a positive, and statistically significant effect of Schengen on cross-border commuting.^[23] Specifically, the probability to commute across the border with Switzerland is estimated to be 0.48 percentage points higher among treated versus control individuals. This Schengen effect is also confirmed when we control for the language differences^[24] and for the quality of the road infrastructure (Column 2 and 3 of Table 2).

Over the period under consideration, there are two main confounding effects which might have had an impact on the probability to cross-border commute: the 2009 Great Recession (Davies 2011) and the exchange rate appreciation (Bello 2020). Data show that after 2009 labour demand in Switzerland was stronger than in most European countries and as a result, it is possible that residents in the treated group started commuting more to Switzerland because of relatively better employment opportunities. On the other hand, the Swiss Franc appreciated relative to the Euro by almost 30 % between 2008 and 2011, then remained constant until 2014 and appreciated again in 2015. For cross-border commuters, this implied a very large increase in real wages over a period of two to three years, providing an incentive to work in Switzerland.

While the Schengen agreement made commuting to and from Switzerland easier and hence more attractive in both directions across the Swiss border, the crisis and the appreciation of the exchange rate created more incentives to commute towards Switzerland. By including in the treated group labour force individuals living in Swiss regions sharing the border with France, Italy and Germany, we take into account country-specific business cycle fluctuations.^[25]

To further exclude the confounded effect of the 2009 economic crisis, which varies at the origin-destination pair level, we add to our baseline specification several controls for bilateral labour demand conditions at both origin and commuting destination. We include variables such as the unemployment rate, the youth unemployment rate and the employment share by sector, constructed per each region as the ratio between the value in the origin region and the average value among all potential destination regions abroad.^[26] The unemployment rate is chosen as financial crises have a much stronger impact on unemployment (Reinhart and Rogoff 2009; Scott et al. 2008) compared to other economic recessions (Column 4 of Table 2). As the sensitivity to business cycle is found to be twice as high for young workers below the age of 24 than for older workers (Brian and Patrick 2010; Jimeno and Rodríguez-Palenzuela 2003), we separately consider youth unemployment (Column 5 of Table 2).^[27] In Column 6 of Table 2 we include the employment share by sector to account for the highly asymmetric nature of the crisis at sector level.^[28] Ultimately, the 2009 crisis was triggered by the burst of a housing bubble which led to the collapse of the US house prices, with strong international repercussions. In Europe, the crisis hit countries at different times, to a different extent and for shorter or longer periods, leading to very different housing market adjustments and government responses (Whitehead, Scanlon, and Lunde 2014). Within our setup, an increase in the housing price in e.g. France relative to Switzerland may have lead French residents to migrate to Switzerland, instead of commuting. We control for regional house prices, by including a variable constructed as the ratio between the house price in the origin region and the average value among all potential destination regions abroad (Column 7 of Table 2). The coefficient capturing the Schengen effect is positive, statistically significant and of similar magnitude across all specifications.

Finally, to address the issue of the exchange rate appreciation (Bello 2020), we construct a measure of the “real” exchange rate as the ratio between the exchange rate between the region of residence and the region of work and the price level of the region of residence (Column 8 of Table 2). Even in this specification, the coefficient of interest is positive and statistically significant.

6.2 Robustness Checks

Figure 5 reports the event study on the effect of the implementation of the Schengen agreement on the probability of cross-border commuting. Each dot represents the coefficient of the interaction between the treatment variable and each year in the period 2005–2015.

Figure 5:

Event study for the implementation of the Schengen agreement on the probability of cross-border commuting. Each dot represents the coefficient of the interaction between the treatment variable and the corresponding year in the period 2005–2015, when controlling for year, region, sector, year dummies and the sector-year and country-year interactions. Per each coefficient, the 90 % confidence intervals are reported. Year 2009 has been set as the reference year.

The absence of a significant effect before 2009, which is set as the reference year, is particularly important for two reasons. First, it confirms that our strategy identifies the impact of Schengen net of the effect of the 2007 extension of the freedom of movement to non-border regions of Switzerland and to all EU/EFTA citizens (Section A.1 in the Appendix A).^[31] In addition, it serves as a placebo test, to rule out the presence of heterogeneous trends between treatment and control regions before 2009. After 2009 the event is positive, although statistically significant and persistently higher only from 2011.

In order to test for the robustness of our results and in particular to further control for the potential confounding effect of the crisis, we estimate the baseline model by adding additional controls for bilateral labour demand conditions at both the origin and commuting destinations (Table 4). Differently from the controls used in Table 2, we include the youth unemployment rate interacted with the corresponding age dummy (Column 1), the unemployment rate (Column 2) and the employment shares by sector (Column 3), all constant at the 2005 value and interacted with year dummies. The advantage of these regressors, which capture the time-specific impact of labour demand shocks, is to be pre-determined with respect to the Schengen policy, as in principle they could have been as well affected by labour demand conditions.

Table 4:

Robustness checks of the effect of Schengen on cross-border commuting – labour force.

	(1)	(2)	(3)	(4)
Treated Post-Schengen	0.0049 *	0.0050 *	0.0070 **	0.0048 *
	(0.0027)	(0.0027)	(0.0032)	(0.0027)
Observations	3,099,277	3,099,277	3,099,277	2,850,085
Adjusted R2	0.040	0.040	0.040	0.039
(Youth unem rate in 2005*age 16–24)*year dummies	YES	NO	NO	NO
Unem rate in 2005*year dummies	NO	YES	NO	NO
Employment share in 2005*year dummies	NO	NO	YES	NO
No Spain	NO	NO	NO	YES

1. The dependent variable is the probability of cross-border commuting. Standard errors are clustered at region level and adjusted for the small number of clusters. In Column 1 we use the baseline model and also control for origin-destination youth unemployment rate in 2005 interacted with time dummies, in Column 2 for origin-destination unemployment rate in 2005 interacted with time dummies, in Column 3 for origin-destination sectoral employment in 2005 interacted with time dummies. ^* p < 0.1; ^** p < 0.05; ^*** p < 0.01. The baseline cross-border commuting rate is 0.9 %. The bold values refer to the diff-in-diff coefficients of interest, i.e., the coefficients of the interaction of the treated dummy and the treatment dummy across different specifications.

In Column 4, we exclude from the control group the regions which share the border with Spain, as this is the country which according to the statistics has been hit most severely by the crisis (Appendix A.3). The large drop in construction employment in Spain after 2008 could have deteriorated, e.g. the employment prospects of French workers living close to the French-Spanish border and intending to commute to Spain for work relatively more than of French or German workers intending to commute to Switzerland.

In all specifications, the Schengen effect is positive, significant and of similar magnitude, confirming the robustness of our estimates.

7 Potential Schengen effects

So far we have provided evidence of a positive and mostly significant effect of the implementation of Schengen on cross-border commuting, across different samples. In what follows we are going to explore the channels which led to the increased flow of cross-border commuters. Having pooled cross-section data, we are not able to follow individuals over time, but we can use descriptive statistics to get some insights.

7.1 Destination of Inter-Regional Commuters

The strong effect found in the sample of inter-regional commuters could be ascribable to the increased number of individuals who commute across regions (and were not commuting before) or it could be due to workers who were already inter-regional commuters internally before the implementation of Schengen, but started commuting across the border after the policy.^[32] To disentangle the two effects, first we look at the evolution of the number of inter-regional commuters and cross-border commuters in treated regions over time. Figure 6 shows that both numbers have increased after 2009, with the number of cross-border commuters increasing more. This means that among cross-border commuters after 2009 there could be both individuals who were not commuting (across regions) before the implementation of Schengen and individuals who were already commuting across regions, but internally. To further explore the redirection of inter-regional commuters channel, we look at the destination countries of inter-regional commuters before and after the implementation of Schengen. If the redirection hypothesis was the major driver, we would observe a smaller share of workers commuting across regions internally and a larger share commuting across the border towards Swiss regions. We observe in Table 5 that the redirection effect plays a major role. In most regions, such as FR43, FR71, and ITH1, the share of cross-border commuters to Switzerland increased significantly, at the expenses of the share of commuters internally, which largely decreased. In other regions, such as FR42, the share of inter-regional commuters towards Switzerland increased mostly at the expenses of commuting towards other countries. For other regions, such as DE10, ITC2 and ITC4 the changes have been minimal. Overall, we find a clear-cut evidence of a substitution between commuting from contiguous national (and foreign) regions to commuting to Switzerland for inter-regional commuters living in treated regions as a consequence of the implementation of Schengen.

Figure 6:

Inter-regional commuters and cross-border commuters in treated regions. CB-comm is the number of cross-border commuters in treated regions; Inter-comm is the number of inter-regional commuters in treated regions. Ratio is the ration between the number of cross-border commuters and the number of inter-regional commuters in treated regions.

Table 5:

Destination of inter-regional commuters living in treated regions.

	Internal		Switzerland		Elsewhere
	Pre-2008	Post-2008	Pre-2008	Post-2008	Pre-2008	Post-2008
DE10	0.70	0.69	0.28	0.29	0.01	0.02
FR42	0.22	0.20	0.42	0.52	0.36	0.28
FR43	0.79	0.64	0.21	0.36	0.00	0.00
FR71	0.56	0.51	0.44	0.49	0.00	0.00
ITC1	0.86	0.80	0.12	0.19	0.01	0.01
ITC2	0.92	0.94	0.04	0.03	0.03	0.04
ITC4	0.73	0.74	0.27	0.26	0.00	0.00
ITH1	0.72	0.62	0.10	0.17	0.18	0.21

1. Share of inter-regional commuters living in treated regions and commuting to an internal region, or to Switzerland or to another foreign country on the total number of inter-regional commuters before and after 2008. The bold values refer to the diff-in-diff coefficients of interest, i.e., the coefficients of the interaction of the treated dummy and the treatment dummy across different specifications.

8 Local Labour Market Effects

The implementation of Schengen might have had effects beyond the cross-border commuting phenomenon. Two are the effects we are interested in exploring: the improved attractiveness of living close to the Swiss border after the implementation of Schengen and the effects of Schengen on local labour markets, i.e. labour markets in treated regions. In order to address the first issue, we estimate the following regression at regional level:

where α _r is region fixed effects, γ _t is year fixed effects, β _ct is country-year fixed effects and the outcomes of interest include the total population, the working age population, and the labour force. We find no significant effect on all three variables (Table 9), pointing to no evidence of any increased level of attractiveness of treated regions after 2009.

To dig further and explore the presence of local labour market effects, we perform additional individual level regressions. Swiss residents could have gained incentives to shop across the border, affecting labour demand and employment opportunities in local industries in the bordering regions of neighboring countries. This effect could be strengthened by the reduced local labour supply, as people take up jobs across the border. If this hypothesis is confirmed, we should see a reduction in the probability to be unemployed and perhaps a decrease in the probability to be inactive in treated regions, compared to control regions. We therefore estimate Equation (1) using the probability to be unemployed as the outcome of interest (Table 10).^[33]

We find that when we use the baseline specification, the probability to be unemployed is higher among individuals living in treated regions after the implementation of Schengen (Column 1 of Table 10). However, when we include control variables to take into account the effect of the crisis, such as the youth unemployment rate (interacted with the corresponding age category dummy), the unemployment rate and the employment by sector, the coefficient of interest is not statistically different from zero (Columns 2, 3 and 4 of Table 10). Then, we use the working age population sample to estimate Equation (1) using the probability to be inactive as the outcome of interest (Table 11). We find that across all specifications, even when controlling for the crisis, the coefficient of interest is not statistically significant. We interpret these results as evidence that the local labour market conditions in treated regions (compared with control regions) did not change significantly after the implementation of Schengen, as both the probabilities of individuals to be unemployed or inactive are unchanged.

9 Conclusions

Our results point to an increase in the individual probability to commute cross-border of approximately 0.5 percentage points among individuals in the labour force due to the implementation of the Schengen agreement. These results are robust to controls for the potential impact of the economic crisis and the exchange rate fluctuations.

We find that the effect is up to 4 percentage points when we restrict our sample to inter-regional commuters. When looking into specific channels, we find that the increased cross-border commuting flow is likely to come both from an increased number of inter-regional commuters after the implementation of Schengen and to a redirection of existing inter-regional commuters from national and other foreign destination regions to Swiss ones. We find little evidence in support of any interplay between migration and cross-border commuting either at national or international level. These findings provide evidence that border controls, even if only perceived, represent a serious obstacle to cross-border commuting. In fact, the increase in the flow of cross-border commuters happened even though the changes at the frontier have been minimal. As such, our estimates are likely to be an under-estimation of the true effect of the implementation of Schengen on cross-border commuting.