Free Mobility with the EU and Immigration of North American Brains to Switzerland: What Consequences?

2012, Vol. 148 (4) 497–530 a Professor, School of Public Policy, Simon Fraser University, Harbour Centre, 515 West Hastings St., Vancouver, V6B 5K3, Canada. I thank Kemi Alfolabi and Andriy Baransky for their invaluable assistance in compiling the dataset, seminar participants at the University of Neuchâtel and a referee, for their comments. I also thank the School of Economics at the University of New South Wales, Sydney, Australia, for its hospitality while writing this paper. 1 High-skill immigrants/expatriates/workers and brains are used interchangeably in this paper. Free Mobility with the EU and Immigration of North American Brains to Switzerland: What Consequences?


Introduction
In this paper, I investigate the factors that drive North American brains to Switzerland before and after the implementation of the free mobility agreement with the European Union in 2002. The analysis is about immigrants in occupation categories requiring high education levels who came from Canada and the United States between 1990 and 2009. The main question is: Has the free mobility agreement led to substitution away from North American brains 1 and did it change factors influencing non-European high-skill workers to come to Switzerland? The focus of the paper is the impact of the new policy on immigration flows from North America in terms of magnitude and incentives; some potential consequences for Swiss firms' competitiveness on the world market are then discussed.
After a decade of increasing free mobility with the European Union, there are mixed reactions in the Swiss society about the perceived impact of the policy especially on high-skill expatriates. Some argue that not only foreign university professors are taking jobs of Swiss citizens but the diversity of foreign academics' origins is extremely low. Muller (2010), p. 78, for example, states: "The Swiss People's Party […], a right-wing group notorious for its nationalist politics, accused the University of Zurich of having too many German nationals on its faculty." Others, the business sector in particular, hold opposite views. When 3 For a detailed history of immigration policy in Switzerland and its impact on the foreign resident population, see Gross (2006). 4 The word "permanent residents" covers foreigners residing in Switzerland for a period of one year or longer (OFS, 2009, p. 5). This definition is different from the one used by settlement countries such as Canada and the United States where only people with unlimited authorizations are considered permanent.
those two countries. The impact was not limited to a time-specific drop as the role of some factors influencing immigration changed significantly. Since 2002, North American highly skilled people are more influenced by a stimulating work environment at home and there has been a clear substitution away from them, in favour of Europeans and Swiss. In addition employers must now offer attractive financial conditions to North American high-skill candidates. So not only do Swiss employers have limited access to North American brains but they face stiffer competition to attract them. Finally, in addition to the adverse effects of the new policy, a negative trend in hiring of North American innovators due to labour market conditions might in the future diminish Swiss firms' competitiveness on world markets. The rest of the paper is organized in the following way: Section 2 presents a brief survey of Swiss immigration policy since the mid-1990s. Section 3 describes changes in the composition of immigration to Switzerland since the implementation of free mobility. In Section 4, a model of skilled immigration from Canada and the United States is estimated to evaluate the nature and magnitude of substitution between North American and European brains and economic implications are discussed. Section 5 concludes.

Swiss Immigration Policy
Swiss immigration policy experienced little change between the early 1970s and the late 1990s but the approval by popular vote of free mobility with the EU/EFTA in 2000 led to drastic changes to the law for citizens coming from third countries. 3 Swiss policy has always been demand-driven and thus requires foreigners obtain a one-year or longer job contract prior to applying for a "permanent" work/residence permit. 4 Another avenue, until 2002, was through conversion of temporary status (i.e., seasonal permits). Generally one-year permits (i.e., sojourn permit) were automatically renewed conditional on employment. After a certain number of years of residence with such permits, immigrants and their family could apply for an unlimited, unconstrained permit (i.e., establishment permit). Yearly annual quotas were set for the number of new sojourn permits, not including seasonal conversions. Until the mid-1990s, there was no defined priority for the hiring of foreign workers. In 1995, the government developed the new circle-policy with priority given to citizens of countries similar in culture to Switzerland. Also, anticipating on the signing of an agreement with the EU, in November 1998, the government introduced the dual recruiting system by advising employers to give preferences to workers from the European Union; they could however, still prospect worldwide for skilled workers.
Following approval by popular vote in 2000, the implementation of the free mobility agreement with the EU/EFTA countries started on June 1 st , 2002 and a new immigration legislation was in place by December 2005. 5 The most relevant component of the new legislation for this study is the specification of search priorities for employers who want to hire foreign workers: Third countries' citizens can be considered only if they are skilled and no worker is available on the domestic or EU/EFTA labor markets. Hence, priority must be given to Swiss citizens, foreigners living in Switzerland with permanent status and residents in EU/EFTA countries. Employers must prove they have searched extensively before submitting a request for hiring from third countries. Furthermore, only skilled candidates (i.e., managers, specialists and workers with tertiary education) can be considered with a few exceptions, i.e., in case of intra-firm transfers or exchanges, acute shortages of labor in economically key occupations, or firm's creation leading to local employment growth. The conditions may also be relaxed to gain access to new markets, expand exports or for non economic activities such as arts, culture, religion and international organisations. Once their hiring has been approved, third-country citizens receive a sojourn permit (livret B) valid for one year, renewable. They are eligible for an establishment permit generally after ten years, in some cases after five.
From 2002, there was a five year transitory period with quotas before implementation of complete free mobility. Initially, the agreement concerned only 15 EU member countries and the EFTA countries. On June 1 st , 2004 it was extended to the ten new EU members. 6 By June 1 st , 2007, the transition period was completed for the initial 15 countries plus Cyprus and Malta, and complete free mobility applied to their citizens. Since May 1 st 2011, it applies to the remaining 7 In Figure 1, skilled is defined according to the International Standard Classification of Occupation (ISCO) from the International Labour Office (ILO) and includes legislators, senior officials and managers (ISCO-1), professionals (ISCO-2) and technicians and associate professionals, (ISCO-3; ILO, 1990). eight members who entered in 2004. The latest step in the process has been the start of the transitory period for Bulgaria and Romania on June, 1 st , 2009. By 2014, all EU/EFTA countries are expected to benefit from full free mobility with Switzerland. It must be noted that free mobility does not exempt foreign citizens from applying for permits. Their sojourn permit (livret B) is valid for five years after which period they can apply for an establishment permit. To obtain a sojourn permit they must be employed, show proof of independent activity or if inactive, have financial resources and health insurance coverage. The major change that is likely to have affected brain flows from third countries is priorities imposed on employers for hiring. Non-European skilled individuals are still eligible for immigration to Switzerland but only after potential candidates in Switzerland and the EU countries have been considered. Such regulation is expected to lead to substitution among skilled immigrants from various geographical areas. The remaining of this paper provides a descriptive and econometric analysis to identify whether the new legislation has actually led to substitution away from North American brains in favor of European and Swiss ones after 2002.

Brain Immigration to Switzerland
The choice of Canada and the United States (US) to represent third countries is guided by the fact that the two countries have historically provided the most educated immigrants to Switzerland. In 2000, 49.7% of Canadian and 57.6% of American expatriates in Switzerland had tertiary education. Only immigrants from the United Kingdom and Sweden had a comparable levels of education with 50.1% and 52.2% respectively (Docquier and Marfouk, 2005). Also, their education standards are relatively similar to those of Europe and Switzerland which is likely to have made it easier for employers to follow the new legislation and substitute away from them in favour of Europeans and Swiss skilled workers.
The legislation targets new immigrant workers and Figure 1 shows the evolution of the inflows of workers (total and skilled) from Canada and the United States from 1990 to 2010. 7 Not surprisingly at the beginning of the 21 st century, there is a sharp drop in the total annual number of North Americans coming to Switzerland. In 2003, after nineteen months of implementation of  8 The sharp falls in the early 2000s followed a long period of steady increases especially in North American immigration. From 1994, a year before the introduction of the circle policy, until 2001 the total annual inflow of immigrants from Canada and the United States rose by 110.7% and 31.8% respectively (Table 1, column 5). During that period, immigration from EU15 countries grew only by 8.2%. However, when only skilled workers are considered in Table 2, column 1, the growth rates for European countries are much larger than for total immigration while they are close for North America. Hence, before free mobility at the end of the 20 th century, the skill intensity of European migrants to Switzerland was growing quite fast while that of North American, already the highest, was constant.
Generally, the government's recommendation to give priority to European workers at the end of 1998 (i.e., dual recruiting) does not appear to have had  much of an impact on employers' hiring strategies. However, when the recommendation became law in 2002, the flows from many countries and North America in particular were reduced drastically and remained low throughout the decade. So, it is likely that substitution occurred in favour of European countries since the drop in North American immigration was permanent. Consistent with the new legislation, the shift in the origin of foreign workers has been accompanied by a shift in the skill composition: By 2010 the shares of high-skill immigrants in total were 86.4% for Canadians and 87.3% for Americans up from an average of 68.3% and 78.1% respectively between 1994 and 2001 (Table 2, column 7). Thus, North American immigration to Switzerland which had been historically high in skill content became even more skill intensive after the introduction of free mobility with the EU. Since employers could hire low-skill workers only from EU countries, it is not surprising that these countries have experienced a fall in skill intensity since 2001 after the 1990s' rise. France and Germany show the largest falls: -12.5 and -25.2 percentage points respectively (Table 2, columns 5 and 7). 9 In short, North Americans coming to Switzerland are now more educated but their number has dropped drastically compared to Europeans. Next the econometric analysis focuses on identifying the impact of the change in legislation on standard factors driving immigrants and possible implications for the future of the Swiss economy are discussed.

What Drives North American Brains to Switzerland?
For this analysis, skilled immigrants are classified in eleven occupation categories based on ILO-ISCO. They cover managers (F), professionals (A, C, D, E, H, I, J, L) and technicians (G, B; see Appendix I, Table A.I.1. for details). Table 3 shows the levels of occupation flows from Canada and the United States to Switzerland during nine years before (1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001) and nine years after (2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010) the start of the implementation of the free mobility agreement. The total flows from the two countries fell by about the same magnitude between the two periods (-31.3% from Canada; -31.8% from the US) and there are similarities as well as differences between occupation categories.
For both countries the maximum positive change occurred for managers (+89.4% from Canada; +73.9% from the US); a result which is not surprising since managers' migration is likely to result from intra-firm transfers that are exempted from the hiring priority rule set by the new legislation. Also, American companies have long found Switzerland attractive for establishing regional or world headquarters. 10 The only other occupation categories with positive growth are Canadian engineers (E; +42.9%) and American architects and related specialists (A; +14.3%). A major difference between the two countries is the magnitudes of the drops across occupation categories. Three categories from Canada experienced more than fifty percent decline (commercial and financial technicians, -60.6%, teachers, -62.6%, and scientists, -51.3%) while five from the US did so (social scientists and humanities related professionals, -75.7%, health and science technicians, -73.0%, teachers, -69.9%, scientists, -68.1%, and commercial and financial technicians, -62.6%). Finally, the flow of academics declined but relatively modestly, i.e., -21.0% from Canada and -7.0% from the United States.
The consequence of these changes has been a re-distribution of the categories' shares between the two periods ( Figure 2). Managers now dominate the skilled immigrant flows from North America (59.3% for the US in the 21 st century; 32.1% for Canada). But the flows from Canada remain more diversified than those from the US. No other occupation category represents more than 10% of skilled immigration from the United States, while from Canada health and science technicians (G) represent almost the same share as managers (30.8% vs. 32.1%). For both countries architects and related specialists (A) engineers (E), IT professionals (I) and academics (L) saw their share increase slightly. In short, the drop in immigration was not uniform. a From March to June 2008, a much larger proportion of immigrants did not declare an occupation (between 14% and 31% instead of about 2%) and thus the 2008-flow is slightly underestimated. In 2008, the share of workers not declaring an occupation was 12.7% against 11.7% in the preceding year. So, the overall effect on total worker immigration is negligible and it is likely to be even smaller on skill-and country-specific flows. Sources: see Appendix I for category definitions; ODM (various years).
11 Modeling the flows of skilled workers from North America to Switzerland in the context of brain drain would be difficult to justify the main reason being that the Swiss labor market, while attractive, is very small. Between 1990 and 2000, the rate of skilled emigration (defined as the number of emigrants per 100,000 skilled member of the labor force in the source country) of Canadians to the US was more than ten times larger than that of Canadians to Switzerland (2,867 versus 227); the rate for the Swiss going to the United States was 445 while the rate of the Americans coming to Switzerland was 35 (Gross, 2011).

Background Model and Empirical Implementation
Aside from the literature on brain drain which tends to focus on the impact of skilled migration on source countries' welfare (see for example, Ivlevs and de Melo, 2008;Commander, Kangasniemi, and Winters, 2004;Beine, Docquier, and Rapoport, 2001), few studies focus on international migration of the highly skilled. 11 There is however a growing literature on the location decision by high-skill migrants to which this paper is a contribution because understanding what drives skilled migrants to move among rich countries is fundamental to designing policies to attract them. The basic theoretical framework to analyze such flows is the standard push/pull model that originates in internal migration,  (4) 12 See for example, Helliwell (1997), Karemera, Oguledo, and Davis (2000), Lewer and Van den Berg (2008) and Gross (2011). 13 All the variables, how they are measured and data sources, are described in detail in Appendix II.
where workers maximize returns to labor (Sjaastad, 1962;Harris and Todaro, 1970;Borjas, 1987). Given, the questions investigated in this study, the analytical framework must be able to accommodate individual choices, i.e. supplyside, as well as policy issues, i.e., demand-side and the push/pull model has been recently developed in that direction. For example, Clark, Hatton, and Williamson (2007) combine Borjas (1987)'s model of migration driven by human capital return with various immigration policies implemented by the US. A commonly used empirical specification for migration flows consistent with the individual's utility maximizing present value decision and flexible enough to accommodate policy factors is the gravity model, 12 the general equation of which is, with Fl jk the flow between country j and k which is a function of weighted mass (wM) in each country and distance between the two countries (d). The weighted mass in the case of skilled migration is educated population corrected by the standard of living (Isard, 1998); distance can be physical, cultural, political or professional, depending on the case under investigation. A log linear specification of (1) is thus: The dependent variable (lFlRate CH jit ) is the log of immigration rate of skilled individuals in occupation i (i = 1 to 11), from country j (j = Canada, US) to Switzerland (CH), during year t (t = 1990 to 2009). The model is estimated over 440 observations. The fixed effect (c ji ) accounts for time-invariant, source country-specific and occupation-specific factors.
Immigration rate is the annual inflow from Canada and the United States of skilled people as a share of the Swiss university-trained labor force in thousands during the previous year. 13 Destination rather than source labor force is used for standardization to ensure consistency across panels and the previous year 14 Also, the Swiss skilled labour force may be endogenous to the immigration flows if changes in policy have an impact on incentives to study for Swiss residents. The results of the Grangercausality test show that in this sample, the dependent variable does not Granger cause LFUni CH t (F-statistics = 0.12, p-value = 0.89) and thus, endogeneity is unlikely.
value is used to avoid simultaneity problems. 14 The average immigration rate for Canada is 0.033 and for the US 0.090. The variations across occupation groups are quite large with a minimum of zero for both countries and a maximum of 0.219 for Canada, and 0.425 for the US (see Table 4). Not surprising both maximum values are reached in the manager categories, in 2001 for Canada and in 2007 for the United States. The explanatory variables representing individual maximisation are the log of real occupation-specific earnings and high education unemployment rates in the source countries (lEarn jit ; URUni ji ) and in Switzerland (lEarn CH it ; URUni CH t ). Relatively higher earnings and lower unemployment among tertiary-educated people in Switzerland are expected to increase the flow of highly skilled foreigners as relative financial rewards and probabilities to find a job improve. The Swiss unemployment rate is expected to capture the labor market driven immigration policy (demand-side) rather than the perception by applicants of the job market (supply-side) since new immigrants must have a job contract.
Tertiary-educated labor forces in the source countries and in Switzerland (lLFUni jt ; lLFUni CH t ) are expected to have a positive sign in the traditional gravity model as bigger masses generate larger flows. In the context of brains, larger pools can be seen as proxies for professional networks. So, a larger skilled labor force at destination (i.e. Switzerland) can be seen as attractive for North Americans and can increase the flow of immigrants. However, larger networks at home for the same reason may decrease the incentive to migrate. In addition, the policy since 2002 that requires employers to choose among local candidates first would also decrease the flow of immigrants when the Swiss skilled labor force increases as employers have more choice domestically. Hence, because of the opposite effect from size (mass effect) and quality (network effect) of labor forces, the signs are indeterminate. Nevertheless, after 2002, the negative impact of the Swiss labor force through priority in hiring is likely to have increased.
The move to free mobility with the EU is tested in several different ways. First, a simple dummy variable (Policy 2002) with value 0 until 2002, 0.5 in 2002 and 1 afterwards is used to capture a fixed impact of the introduction of the new policy. The dummy variable is also interacted with the explanatory variables to evaluate potential changes in elasticities. Second, the impact of free mobility is measured through the sudden direct access to a much larger pool of skilled workers including not just residents of Switzerland but also residents of the European Union. Two variables are considered: The supply provided by the border countries (LFBord t ) and the supply provided by EU countries eligible to free mobility with Switzerland (LFEU t ). 15 The use of the supply from border countries only (i.e., Austria, France, Germany and Italy) is based on geographical closeness and on the sharing of languages with Switzerland, two characteristics likely to favour their 16 Statistically, it is not possible to enter the European and the Swiss labor force measures separately because of high correlation (about 0.980; see Table A.II.1). 17 Part of the reason for using a parsimonious model lies in the fact that the two North American economies are very similar in terms of the distribution of residents' characteristics relevant to migration decision such as age or human capital intensity.
citizens within the new priority set-up for immigrants. Also, immigrant workers from these four countries represented more than 90% of the European annual inflow and 60% of total annual inflow. The measure is the weighted sum of university trained labor supplies from the four countries with the weights equal to the share of each country in the total length of the border with Switzerland. The supply from eligible EU countries is the weighted sum of university trained labor supply with the weights equal to the inverse of the distance between the capital of each country and Bern, the capital of Switzerland. Each weighted labor force is introduced with the constraint of equality of coefficients with the Swiss labor force after 2002 which is consistent with the fact that employers must consider them as perfect substitutes. 16 Finally a dummy is also tested for the introduction of the dual recruiting system (Policy 1998) which recommended giving priority to Europeans. The model is first estimated in its basic form then various policy specifications are used to test the robustness of the results. Because the time dimension of the sample is relatively long (20 years), before starting the estimations I ran unit-root tests. Table 5 shows that the hypothesis of individual unit roots can be rejected at 5%. The case of a common unit-root process cannot be rejected but such a restriction is unlikely to be valid since hiring in different occupation categories has no reason to follow the same dynamics. Hence, the estimations are run in levels and spurious correlation is unlikely. The model is estimated by OLS even though migration rate is bounded at zero. In such context, zero-values represent information about the types of North American brains in demand in Switzerland and thus, cannot be eliminated. Since they do not represent censored data, estimations using Tobit are not considered appropriate. To avoid biased results and the possibility of negative estimated values, the estimations are done in log linear form such that the dependent variable is ln[( flow + 1) / LFUni ch t−1 ], even though the number of zeroes is small (14 observations).
Some robustness tests and various experiments with the policy dummy for the introduction of free mobility are presented in Table 6. Note that the results of an AR(2) serial correlation test require the use of serial correlation robust standard errors. One of the concerns with the rather parsimonious basic specification is potential omitted variables. 17 For example, the empirical literature on migration flows shows that geographical distance (proxy for cost of moving) and cultural determinants such as common language or former colonial ties are important determinants of the flows (see Clark, Hatton, and Williamson, 2007;Mayda, 2010). In column 1, the fixed-effect test shows a strong contribution of source-country non time-varying factors to the size of the individual occupation-category flow rates. These factors may include the geographical distance and the fact that French is a common national language in Canada and Switzerland. Another factor may be international transferability of skills; employers for some occupations may have a preference for applicants of a given source country because of perceived better compatibility of training with demand.
However, the fixed effect captures only mean effects and if time-varying source-country effects are not adequately taken into account, endogeneity translates into correlation between the fixed effect and the error terms (Wooldridge, 2002, Chapter 10). So, in column 2, I introduce a time variable for each crosssection to control for possible source-country specific aggregate time-varying factors. They are not significant which is consistent with the fact that the Canadian and American economies are closely linked through high levels of bilateral trade flows. Also, this likely generates multi-collinearity as the value of the coefficients of the source-country labor force variable shifts substantially (even though it is still not significant) because it is trended.
In column 1 free mobility has a significant adverse effect on immigration flows from North America and results in column 3, confirm there is no significant difference in the average fall in inflow of brains from Canada and the United States in 2002-2003; also the recommendation by the government to give priority to Swiss and EU citizens in 1998 does not have a significant impact (column 4).    Hence, only legislated priorities have affected the inflow of North American brains. Finally, the 2002-policy, in addition to the impact effect, is modelled by augmenting the Swiss labor force with the two measures of the EU labor force alternatively (columns 5 and 6). In both cases, the variable is significant but the coefficients are smaller and weaker than for the Swiss labour force alone. Overall these results suggest that the introduction of free mobility has led to changes beyond a simple impact effect and the basic model is likely to be under-specified. In Table 7 all elasticities are allowed to vary after 2002 and several hypotheses are tested. 18 In column 1, the negative impact of Swiss unemployment embodying the condition of having a job contract does not vary after 2002; however, the gravity dynamics (i.e., Swiss and North American labour forces) does change.
In columns 2 to 4 alternative specifications for the new policy are tested and the results are robust. Before free mobility the network effect dominated even though it is statistically weak. An increase in skilled labor force in North America and in Switzerland (i.e., expansion of networks) attracted brains. After the introduction of free mobility with EU/EFTA, there is substitution with Swiss labor force and the European highly skilled labor force however it is measured. In column 2, Swiss and EU labor force are allowed to be perfect substitutes after 2002. The magnitude of the post-2002 change in substitution is given by the coefficient on the Swiss-EU labor force (-1.36 significant at 1%). The weak attraction from Swiss skilled labor force before 2002 for North American brains is completely offset 19 and growth in EU labor force lowers their inflow. Those are clear indications that employers have been following the priority rules in hiring. All the other results of the model pre-and post-2002 are stable with respect to column 1 except the coefficient on Swiss earning which is now positive, significant; so, North American brains after 2002 respond to financial incentives in their decision to come to Switzerland. Limiting the labor force to border countries generates similar results (column 3). This is not surprising since the border countries include three of the largest European economies and the simple correlation between the two measures is extremely high (0.998). Finally, rather than considering the whole European labor force as the potential pool of applicants for jobs in Switzerland, only unemployed Europeans are taken into account (column 4) implying that only the unemployed would be interested in working in Switzerland. Even though this is an extremely restrictive assumption, there is a significant yet smaller elasticity (0.997 versus 1.36 for the whole labor force) probably because the unemployed are less diversified and may not fit the demand characteristics as well.
Because North American brains seem to care about network effects, I also test if the skilled labour force can be a proxy for quality of work environment (i.e. network). In column 5, the source-country skilled labour force is weighted with innovations [(1 / patent) × LFUni]. The results clearly show that when innovation intensity increases at home (when patent applications by residents increase), immigration from North America decreases but when the labor force increases, immigration increases. So there is a quality and a quantity effect and the magnitude did not change with free mobility but it is strongly significant. It must be noted that the traditional factors of the push/pull model have very weak impact on North American skilled flows to Switzerland except for Swiss income and unemployment rate which characterizes the demand-driven policy. These results are, however, consistent with the findings of a paper comparing semi-skilled and skilled immigrant workers from 49 countries to Switzerland between 2000 and 2004. When many more source countries are considered, very similar results are found for skilled migrants (see Gross and Schmitt, 2006). Interestingly, a study of low-skill and high-skill immigration to France from 63 countries between 1983 and 2000 also finds that only financial attractiveness of the destination country is relevant to skilled immigrants (see Gross and Schmitt, 2012). The strong influence of financial incentives is also identified by Docquier, Lohest, and Marfouk (2006) for skilled immigrants to OECD countries. So high-skill migrants clearly do not respond to the same factors as low-skill migrants; their emphasis is on financial return to human capital and work environment.
The next two experiments (columns 6 and 7) consider whether the labor forces from the three regions (Europe, North America and Switzerland) were treated similarly when the legislation allowed employers to do so. In column 6, the European labor force is assumed to be considered perfect substitute for the Swiss labor force before and after 2002 because of the Government's recommendation about priorities in hiring in 1998. Before 2002, the effect is not significant and after 2002, the result hardly changes and there is substitution against North Americans in favour of Europeans and Swiss. So, European skilled workers were not considered equivalent to Swiss workers before 2002. In column 7, North American and European workers are allowed to be considered substitutes before 2002 and European and Swiss, after 2002. Again, only the policy change after 2002 is significant. It is interesting that in these two experiments, after 2002, Swiss unemployment shows a much stronger impact. So, these two experiments suggest that prior 2002, skilled workers from the three regions were considered distinct      in hiring: Employers chose the most suitable high-skill candidates regardless of origin and only job availability was a constraint. After 2002, Swiss and EU skilled workers had to be given priority over North Americans and their incentives changed. North Americans now expect a financial reward in terms of higher earnings and they pay much more attention to the stimulating work environment at home. In conclusion all the experiments are consistent with a significant substitution away from North American brains in favour of European and Swiss ones once hiring priorities became legislated.

Future Economic Implications.
One questions arising naturally from these results is: Does the evolution in the distribution of geographical origins of skilled immigrants to Switzerland have a potential impact on the country's future economic performance? While this question is not directly addressed through estimation, it is shown that occupations which matter for innovation have been adversely affected and this may have consequences for the future of the Swiss economy.
During the past decade evidence that a diaspora facilitates various economic links between their home and host countries has been growing. Expatriates have been shown to contribute to the expansion of trade between countries (Rauch, 2001). This aspect appears to have been taken into consideration by Swiss legislators through the provision that priorities in hiring can be relaxed for economic reasons such as access to new markets or expansion of exports (Confédération Suisse, 2011a). There is also growing evidence that ethnic networks contribute to foreign direct investment (see for example Buch, Kleinert and Toubal, 2006;Javorcik et al., 2011) and recently a new argument has arisen: Ethnic networks can increase firms' competitiveness on world markets (Foley and Kerr, 2011). Skilled expatriates can provide an information advantage (directly or indirectly) to firms that employ them which lowers entry barriers on foreign markets. Foley and Kerr present evidence supporting the following three points: (1) Ethnic innovators have information advantage about the market and product in their home country and they can increase competitiveness of their firm abroad through better targeting; (2) firms with ethnic innovators are more likely to collaborate with inventors in their home countries associated with that ethnicity and gain information; (3) those gains in insight about foreign markets decrease the need for local partners in joint ventures which often generates additional costs. Foley and Kerr conclude that such evidence has clear implications for immigration policy. Consequently, if Swiss firms are constrained by law to hire skill immigrants mostly from the same region (i.e., the European Union) they might be 20 Foley and Kerr (2011) use engineers and scientists. Academics are added because in Switzerland they are major contributors to innovations especially at the EPFL and ETHZ. 21 The sample being much smaller (120 observations) the efficiency of the results is expected to be weaker so coefficients with a p-value around 6% (absolute t-value around 1.9) can still be considered to be significant in their impact.
penalized in their attempts to penetrate markets in other regions of the world. This could be a particularly important adverse effect of the immigration law given the rapid development of Asian economies and the fierce competitiveness among foreign firms to serve those markets. So, the next step is the estimation of the model for occupation categories of innovators (i.e., Engineers, E, Scientists, J, and Academics, L). 20 Innovators have experienced smaller than average declines after 2002 but mixed evolutions. In Table 3, the average annual inflow of Canadian engineers increased by 42.9% while that from the US fell only by 5.4%. Flows of academics decreased but their shares in total inflow grew: From 1.8% to 2% for Canada and from 2% to 2.8% for the United States (see Figure 2). Scientists from both countries however experienced sharp declines.
The results of the estimations for the innovators are somewhat different from those for all skilled immigrants which is consistent with the fact that other professionals, such as executives for example, are likely to respond differently since they are not necessarily subjected to the new constraints. However, the new immigration legislation did have some adverse effects on innovators. 21 In Table 8, while there is no significant fixed impact of the legislation (column 1) there have been shifts in elasticities and the focus of discussion is column 4 with the more favorable Schwarz criteria.
There are three important points about the results. First, the impact of the state of the Swiss labor market on the probability to get a job is much larger after 2002 (i.e., the elasticity of the Swiss unemployment rate increases in absolute value from 0.170 to 0.944). So the hiring of North American innovators has become much more sensitive to the state of the labor market while it was slightly above average before 2002. The mean tertiary-educated unemployment rate while still low has risen over the sample from 2. 27% between 1990 and 2001 to 2.49% between 2002 and 2009; this may have caused a lasting decline in North American brain inflows. Second, substitution with Swiss brains is lowered ceteris paribus after 2002 as the impact of the Swiss labor force changed from 6.78 to 5.69. So, the new legislation did not benefit local innovators even though European applicants from border countries may be complement to North American (i.e. positive coefficient after 2002). One possible explanation is the high level of specialisation in those occupations and the fact that such individuals from different regions with different training and experience are unlikely to be perfect substitutes. Finally, the influence of push factors changes after 2002 and the incentive to leave North America is higher when the market deteriorates but the financial incentive is somewhat weaker. Overall these results suggest that the new legislation has been beneficial neither to North Americans nor to Swiss innovators.

Conclusion
In 2002, Switzerland started to implement the free mobility agreement with the EU/EFTA member countries. Simultaneously, it established priorities in nationalities for hiring that restricted skilled immigration from the rest of the world. As a result, Swiss employers saw the pool of potential applicants increase 40-fold 22 but at the same time, they no longer had the freedom to hire the best of all potential candidates since they had to search sequentially across geographical areas.
Clearly that policy affected Canadian and American brains (and probably also brains from Australia and New-Zealand) and their chance to obtain a job in Switzerland. Globally, the new policy, as expected, led to substitution in favour of Swiss and Europeans. But it has also had some undesirable side effects. North American high-skill migrants pay more attention to the stimulating innovative environment at home and earnings in Switzerland have become a significant determinant in their decision, increasing competitive pressure on Swiss employers. The new policy may also penalise Swiss firms' competitiveness on world markets in the future. Since 2002, hiring of North American engineers, scientists and academics is much more dependent on the state of the Swiss labor market which exhibits higher average unemployment. So, increased volatility over the business cycle and a possible negative trend in unemployment may have adverse consequences for flows of non-European high-skill people in addition to the more restrictive policy. Growing evidence on the role of ethnic innovators in stimulating links between their home and host countries hints at the fact that less geographical heterogeneity among skilled immigrants may be costly to Swiss firms as they want to enter world markets. This effect seems particularly relevant for Switzerland as between 2004 and 2006 the country had the highest proportion of patent applications with co-inventors located in the European Union among OECD countries (OECD, 2009, Table 4.1). Thus, by raising barriers on hiring from North America and other regions of the world such as Asia, the new legislation may have decreased the possibility of similar collaboration with innovators from those regions and thereby decrease firms' competitiveness on those markets.
In conclusion, the problem is not too many Europeans as nationalists argue but rather it might be too much homogeneity in regions of origin of skilled immigrants. This is a first evaluation of the impact of free mobility with the EU on brain diversity; clearly the question should be investigated further especially at a more disaggregated level and in conjunction with international research cooperation and foreign direct investment.

Appendix I
The 11 categories of occupations for skilled immigrant workers are based on the American classification (USDJINS, 2002). ISCO codes up to 4 digits were used to determine the components of the occupation categories.

Policy 2002
Dummy with value 0.5 for 2002, 1 from 2003 to 2010 and 0 otherwise; measures the introduction of free mobility with EU and EFTA.

Policy 1998
Dummy with value 1 from 1999 to 2010 and 0 otherwise; measures the circle policy with recommendation to give priority to EU workers (i.e., dual recruiting) after November 1998.

UnempEU t
Weighted unemployment for tertiary educated people in 9 of the EU15 countries computed from E, LF and UR (see LFBord t , LFEU t ). URUni CH t , URUni jt Unemployment rate in destination and source country j, at time t. In Canada, unemployment rates both sexes, university degree, 15 years and over (Statistics Canada, 2011, Table v2627998). In the United States, unemployment rate for tertiary level of education computed from labor force and employment level data 25 years and over (US BLS, 2011b). In Switzerland, unemployment rate for tertiary level of education computed from labor force and unemployment level data (OFS, 2011).    (2011a, 2011b, 1999), ILO (1990), HRDC (2006).