1
International labour mobility
What is the impact of the European Union enlargement on the Dutch labour
market?
By Marleen Swart Student ID: 10184376Bachelor thesis: Economics, specialization: immigration Faculty of economics and business
30-12-2014
Supervisor
Mr. L. Ziegler MSc
Faculty of Economics and Business Department of Human Capital
2
Contents
1. Introduction
3
2. Stylized model of the labour market impact of immigration
5
3. European Union’s enlargement
7
4. Economic theory and methodology
11
5. Data descriptive
18
6. Empirical model and results
21
7. Conclusion
25
8. Limitations and suggestions for future research
26
9. Reference list
27
10. Appendices
29
10.1 Appendix 1 Graphs
29
10.2 Appendix 2 Residuals
31
3
1. Introduction
In august 2013 the Dutch minister of social affairs Assher and the English Publist Goodhaart announced code Orange for the present European Labour market (Herderschee, 2013). We are almost at the point where the last straw is breaking the camel’s back (Herderschee, 2013). New members of the European Union, especially Eastern-European countries, will, according to Assher negatively affect the Dutch labour market. Unscrupulous employers take advantage of the relative low wages the immigrants are willing to work for, which nourishes the abhorrence of natives against foreigners (Herderschee, 2013).
Since 2004, there has been an increasing pressure of immigrants on the European labour markets. The European Union has admitted 12 new countries that enlarged the EU labour market significantly. The expansion contained Cyprus, Malta, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Slovakia, Slovenia, Bulgaria and Romania; most of them are located in Eastern Europe. These countries are characterized by a purchasing power parity of roughly 40% of the founding and wealthy members (Brücker, 2007). This parity can be seen as a drive for these new EU member-countries to move to the wealthy countries such as the Netherlands, and as a result affect its labour market in a negative manner.
In January 2014 the transitional arrangement in the Netherlands will change for Romania and Bulgaria, which will lower the barriers for these immigrants to migrate to the Netherlands. As of 2007 this was already the case for the other ten countries stated above (Rijksoverheid, 2013). Renowned economists argue that the winners of migration are the sending countries because their total income will increase by opening up to trade, leaving the host countries damaged (Krugman, Obstfeld & Melitz, 2012). Therefor free movement of labour must be restricted according to Assher, especially due to the negative welfare effects on the less qualified population (Herderschee, 2013). On the other hand, Angrist and Kugler (2003) state that by reducing labour market flexibility the negative effects on equilibrium employment will increase. The new transitional arrangement of 2014 and the discrepancies within the academic literature demand new insights on this matter. The goal of this thesis is therefore to answer the question ‘What is the impact of the European Union enlargement on the Dutch labour market?’
The outcome of this thesis will provide new understanding on the matter of labour immigration and the negative and/or positive effects this might have. It will contribute to the
4 existing academic literature and the social debate concerning the influence of labour migration. Due to the fact that the Dutch market, specifically, has yet to be researched this thesis focuses on this matter. To answer the main question, it is first of all important to look at economic models, which describe the causes and effects of international labour mobility. In order to fully comprehend the effects of the Eastern expansion on the Netherlands, the effect will be examined by looking at the change in immigration numbers to the Netherlands combined with country specific effects. Thirdly the economic theory and methodology will be described including omitted variable bias, simultaneous causality and the functional form. Section five contains the data descriptive followed by an empirical model and results. The final section discusses the conclusion of this research and the limitations and shortcomings of the findings. It also provides the reader with suggestions for future study.
5
2. Stylized model of the labour market impact of immigration
The effects of international labor mobility can be analysed by the specific factor model that is developed by Paul Samuelson and Ronald Jones (Krugman, 2012). This model explains the economic forces that drive workers to move across borders. To focus on international migration, two assumptions are made: the two separate countries produce only one good, with labour which is mobile between the countries, whereas land is immobile. Because there is one good present, there is no trade between the countries. The countries differ in the marginal productivity of labor because of wage differences. In figure 1 the Marginal productivity of labor (MPL) is defined as price times wage; when labor increases the marginal productivity declines due to the law of diminishing marginal returns (Krugman, 2012). The area under the MPL curve equals the total output; the total wage is determined by labor times real wage, the red rectangle is the income for landowners; the total of rents and wages is the total income of a country.
Figure 2 additionally includes foreign employment; in this way the effects of international labor mobility is shown. In autarky, where countries are self-sufficient and the international labor market does not exist, the wages in an Eastern European country are lower (point C in figure 2) than the wages in the Netherlands (point B in figure 2). When opening up for international labor mobility the workers employed in Eastern Europe want to move to the Netherlands because they can earn more income due to the higher wages. This movement
6 of workers will increase the supply of labor in the Netherlands and therefor change the equilibrium and reduce the real wage. The opposite is happening in Eastern Europe where the supply decreases which in turn increases the real wage. This means that in the short run when some production factors are immobile and the demand is not perfectly elastic the labor market cannot clear, but in the long run when all factors are mobile and they can all adjust the market will clear in point A of figure 2. Thus, employees in the Netherlands move from point B to point C and the Eastern European country moves from point C to point B. This results in a higher MPL for Eastern European natives while the MPL for Dutch native workers decrease. The red triangle that arises in the graph is the increase in world production, because the gain for the Netherlands in production is bigger than the loss of production in Eastern Europe.
Figure 2 causes and effects of international labour mobility
source: international economics
The ‘winners’ of this model are the workers who stayed in Eastern Europe, because they receive a higher wage, move from point C to point A, and landowners are worse off because of the decrease in labor supply leading to an increase in the wages. In the Netherlands the labor force is worse off because they receive a lower real wage and landowners are better off by having a larger labor supply. The main conclusion is that with international labor mobility the general cake is bigger but there are people who get a smaller piece, this conclusion will not change in a more complex model.
7
3. European Union’s enlargement
In 2004 and 2007 the European Union admitted 12 new members, being: Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Slovak Republic, Slovenia, Cyprus, Malta, Bulgaria and Romania (Europese Unie, 2012). This changed the European economic environment for the wealthy countries in the European Union with relatively homogenous income levels before the enlargement. The incumbent states had a gross national income per capita which lies 21 per cent above the GNI of the new members, which is significantly higherthan in any previous accession episode (Brücker, 2007). Due to the income gap there is a substantial increase in migration from “new” members to the founding members.
Important data for labour immigration to the Netherlands
1 May 2004 Admitting “A10” * countries Free movement of people with
transitional restrictions and a labor market test
1 May 2006 Easing the transition regime Phased exemption of the labor market
test by sector
1 January 2007 Accession of Romania and
Bulgaria
Free movement of people with transitional restrictions and a labor market test
1 May 2007 End transitional arrangements
“A10” countries
Cancellation of the transitional restrictions for employees “A10” countries
1 January 2014 End transitional arrangements
Romania and Bulgaria
Cancellation of the transitional restrictions for Romanian and Bulgarian Employees
* Cyprus, the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Malta, Poland, Slovakia and Slovenia.
When looking at the numbers in figure 3, it becomes clear that the biggest amount of migrants that move to the Netherlands are European residents. In 2002, 43.3% of the migrants who moved to the Netherlands were European. This rate has grown with more than 20% in ten years to 63.6% and will continue to grow because of a changing immigration policy and the continuing enlargement of the European union. Both will favour the
8 movement of labor forces between the European countries (Cörvers et al, 2009). Furthermore, the overall migrants costs to move to the Netherlands were lowered due to a lot of vacancies in the Netherlands and because of the technological developments what makes it easier to keep in touch with their families (Cörvers et al,. 2009). Additionally, the short term abolishment of restrictions for the new EU countries with relatively low wages, high unemployment and already settled migrant networks made the Netherlands popular for immigrants (Cörvers et al., 2009)
Figure 3 Total immigration to the Netherlands
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Total all countrie s 121,25 0 104,514 94,019 92,297 101,150 116,819 143,516 146,378 154,432 162962 158,374 Total Europe 52,505 49,333 49,445 49,680 55,995 68,492 84,464 84,442 91,094 99,951 100,727 Total America 22,853 19,700 16,832 15,929 17,235 18,104 20,542 20,371 21,275 20,911 19,814 Total Asia 21,751 18,446 15,186 15,446 16,815 18,557 24,127 24,467 24,720 26,277 24,999 Total Africa 21,378 14,910 10,712 9,444 8,819 8,994 11,492 14,478 13,213 11,877 9013 Total Oceania 2,078 1,654 1,565 1,552 2,091 2,513 2,659 2,300 2,346 2,302 2,193 Europe rate 43.30 47.20 52.59 53.83 55.36 58.31 57.70 57.69 58.99 61.33 63.60 source: CBS
Figure 4 shows how absolute immigration in the Netherlands changes when countries become a member of the European Union. For example Poland became member in 2004 and immigration grew with factor 2.2 from 2003 to 2004. In 2007 Europe admitted Romania (among others) that initiated growth from 718 immigrants in 2006 to 2,347 immigrants in 2007, which is a growth factor of 3.3. Together with Romania, Bulgaria became a member of the European Union and grew from 2006 to 2007 with a factor of 10.7 (from 451 to 4,837). In comparison to countries like Germany that became a member in 1957 and the United Kingdom that became member in 1973, it is a significant expansion; hence it shows the effect of admitting new members in the European Union.
When examine the overall figuresfrom 2002 to 2012 the absolute immigration grew with different factors. For the “founding” members; United Kingdom and Germany, the factor grew in 10 year with respectively 0.09 and 0.69. That Germany experienced a bigger growth factor can be attributed to the fact that they are nearby residents (Rijksoverheid, 2013). In comparison to the founding members, the new members: Poland, Romania and Bulgaria grew respectively with factor 7.07, 2.92, and 10.03. One can conclude that the new
9 members utilize the free movement of labor policy of the European Union, and that the Netherlands is perceived as an attractive country to migrate to.
Figure 4 Immigration from different European countries
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Germany 7,959 7,921 8,671 9,134 10,424 10,981 12,929 12,818 13,914 13,851 13,462 UK 6,805 5,871 5,405 4,903 5,550 6,368 7,536 7,376 7,068 7,455 7,446 Poland 2,275 2,106 5,073 6,672 8,214 10,126 13,890 13,027 14,782 18,937 18,348 Romania 627 703 711 559 718 2,347 2,390 2,155 2,594 2,721 2,459 Bulgaria 440 480 418 416 451 4,837 5,148 4,272 4,121 5,213 4,852 source: CBS
The Eastern European Union enlargement results in income gains for the immigrants; their income increases by more than 100 per cent (Brücker, 2007). The losers are the Dutch native workers, due to an extended labor market which results in less jobs available and lower wages (Angrist and Kugler 2003). These feared negative effects of immigration might result in new restrictions implemented by the host country aimed at reducing the number of Eastern European immigrants’ (Brücker, 2007). Angrist and Kugler (2003) state that although employment protection and entry barriers may reduce job loss in the short run, this reduced flexibility may amplify negative consequences for natives by having stricter rules for them in the long run. Conjointly, immigrants create jobs in the Netherlands by purchasing goods and services, regardless of their participation in the labor market (Islam, 2007). This leads to benefits for the Dutch workers as demand for labour increases, which in turn creates jobs related to higher wages. It is important to take into account that the impact of migration in the Netherlands will be small on the short term; while on the long run the effects are more clearly visible. These effects are more compelling due to the fact that transitional periods of free movement expire for Romania and Bulagaria. Additionally, adjustments to new economic opportunities by migrants can take some time (Brücker, 2007).
Important in the immigration discussion is the difference between highly skilled labour migrants and lower skilled labour migrants. The highly skilled labour migrants contribute to the innovation and growth potential of the Dutch Economy (Cörvers et al, 2009). They increase production, productivity, give an impulse to entrepreneurship.
10 Furthermore, the likelihood that the migrants need alimony from the Dutch government is weak (Cörvers et al, 2009). Additionally, Bevelander (2010) states that higher educated immigrants have fewer problems finding a job in a new country, because the adjustment is easier and the need to acquire country-specific skills is bigger. For higher educated people the informal competences like culture-specific proficiency, language skills, understanding of different patterns of expected behaviour in team work and in relations with public authorities are easier to pick up and most of them speak English (Bevelander, 2010). In comparison, the lower skilled immigrants are more often substitutes due to the fact that less education and adjustment is necessary. This means that the lower skilled Dutch labor force is more affected because the immigrants replace the natives more easily (Dustman, Tomasso & Glitz, 2008). Furthermore there is a bigger chance they need alimony from the Dutch government. Also, lower educated immigrants find it harder to adjust to the Dutch customs (i.e. culture) and contribute less to the growth potential of the Dutch economy(Cörvers, 2008). All together, it makes this group less attractive for the Dutch Government. Subsequently, the Dutch government tries to encourage complementary highly skilled immigrants while discouraging the lower skilled immigrants. This can be noticed by for example giving higher skilled immigrants tax benefits while making family immigration more restrictive (Cörvers, 2008).
11
4. Economic theory and methodology
According to the theoretical framework of international labour mobility an area analysis has been done by the Ordinary Least Square regression. As the estimators of the regression minimize the sum of squared residuals, the regression coefficients will be as close as possible to the observed data so that the estimators will be as precise as possible (Stock & Watson 2012). To attain the effects of immigration on unemployment, the dependent variable will be unemployment, whereas immigration will be the independent variable. An area analysis has been performed because migration populations are concentrated in particular geographic areas. By looking at the diversity of the provinces, the effects on regional labour markets of migration can better be estimated. If regions with more immigrants have higher unemployment rates it would confirm the hypothesis that immigrants have a negative effect on the unemployment rate (Okkerse, 2008). A possible problem for the area analysis concerns endogenity. When migrants have to choose where to settle in the Netherlands, an important motive is the unemployment rate in a province or city (Okkerse, 2008). As a result, the causality can run in both directions; being from immigration to unemployment but also from unemployment to immigration. The second problem which could arise regarding the area analysis is that while e.g. migrants can live in Groningen, they could work in Friesland and do groceries in Drenthe. This will result in a labour market effect which is diffused over three provinces. The effect for Groningen is really small and therefor seen as almost zero while in reality the effect on the Dutch economy is bigger. This difference can account for between 40% and 60% of the difference in measured unemployment impact of immigration between the national and local labour market (Okkerse, 2008) .
The general linear regression model is 𝑌! = 𝛽! + 𝛽!𝑋! + 𝑢!, where the subscript 𝑖 runs across observations, 𝑖 = 1, … , 𝑛; . 𝑌! is the dependent variable, 𝑋! the independent variable, 𝛽! is the intercept hence the value of the regression line when 𝑋 = 0. Most of the time this is not of any importance unless economic theory proves it is. 𝛽! Gives the slope of the regression line. Together 𝛽!+ 𝛽!𝑋! gives the relationship that holds between Y and X on average of the population. 𝑢!is the error term/residual and incorporates all of the factors that determine the value of de dependent variable Y (Stock & Watson, 2012). To get an appropriate estimator for the unknown coefficients there are four assumptions, which need to hold. The first assumption is that the conditional distribution of 𝑢! given 𝑋!!, 𝑋!!, … , 𝑋!" has a mean of zero. It implies that there is no omitted variable bias. From a theoretical perception
12 this would be reasonable to assume, in practice it is shortsighted to assume that there is no omitted variable at all. The second assumption is that 𝑋!!, 𝑋!!, … , 𝑋!", 𝑌! , 𝑖 = 1, … , 𝑛, are independently and identically distributed variables. This assumption holds whenever the sample is randomly selected. The data used in this research comes from CBS, which uses at least 30.000 observations that are randomly assigned and uses an imprecise margin, so we can treat the sample as randomly selected. The third assumption is that large outliers are unlikely. This means that expected values of immigration will not have extreme values; this is reasonable. On the other hand it is a difficult decision to make. The final assumption is that there is no perfect multicollinearity (Stock & Watson, 2012). In plain English one regressor should not be a perfect linear function of another regressor. Multicollinearity is avoided by leaving out two dummies: one for the province and one for periods.
To start with the equation and test the effect of immigration on unemployment, data from CBS data centre are used. The data used runs for the period from 2002-2012 and cover the 12 provinces in the Netherlands. Together this results in 132 observations used for the Area analysis: 𝑈𝑁!" = 𝛽!+ 𝛽!∗ 𝐼𝑀𝑀!"+ 𝑢!", where 𝑈𝑁!" denotes unemployment, with key explanatory variable immigration (𝐼𝑀𝑀!") and 𝑢!" the last component represents the error term that incorporates all other factors responsible for the difference in unemployment. The error term 𝑢!" is homoskedastich when the variance of the conditional distribution of
𝑢!" given 𝐼𝑀𝑀!" is constant for the 12 provinces and 11 years used in this research and in
particular does not depend on 𝑋!. There is no theory that gives reason to believe that this holds in practice, so it is provident to assume that the errors might be heteroskedastic and therefor use the robust standard error option in STATA, which is also appropriate when the error is homoscedastic. The hypothesis to be tested is if the estimate of parameter 𝐼𝑀𝑀!" is significantly different from zero and exhibits the expected positive sign, which means that when immigration increases, unemployment rates increase as well. To state mathematically, the hypothesis tested is 𝐻!: 𝛽!"" = 0, 𝐻!: 𝛽!"" > 0, reject 𝐻! if p-value is < 0.1.
𝑈𝑁!" is the dependent variable and therefore the variable to be explained in the
regression. In this research the dependent variable is the unemployment rate per province in the Netherlands from 2002 until 2012. The unemployment rate is the unemployed labor force as a percentage of the labor force. Unemployment is defined as people (15-65 year) who are without work for less than 12 hours a week and are actively seeking for a paid job for more than 12 hours a week; they are directly available. The lower age limit is determined by the assumption that the Dutch youth is able to start working and enter the labour market when
13 they are 15 years old. The upper age limit of 65 is chosen due to the start of retirement when they reach this specific age. The unemployment rate is a suitable variable to measure labour market performance and is easy to understand, it can be argued that is should be relative to unemployment rates in the Eastern European countries. But Dutch immigration is characterized by a higher inflow of immigrants in comparison to demand, so it does not seem to be necessary (Okkerse, 2008). Changes in unemployment rates in Eastern European countries do not seem to influence the move to the Netherlands.
For this research the key independent variable is 𝐼𝑀𝑀!", this means that by knowing the value of immigration it does not provide information about the other variables. Immigration is defined as: foreigners who establish in the Netherlands and are enlisted at the “Gemeentelijke Basis Administratie”. It is registered as soon as they are expected to stay longer than four months. Although some foreigners are not registered (but are expected to stay within The Netherlands for over four months), the CBS takes these people into account by implementing a data correction. The immigration rate is the immigration per province as a percentage of the total immigration.
A problem that could arise is focusing solely on the immigration rate, potentially ignoring important deteminants in the empirical analysis. Important in an OLS regression is that causal effects are only binding for the population studied when the regression is internally valid (Stock & Watson, 2012). When studying causal effects there are two aspects that need to be taken into account. One: the OLS estimators should be unbiased, which indicates that the estimator of the error term should equal the real error term. And secondly, the OLS estimator should be consistent, which implies that when the sample size is large, the variations between the sample error term and real error term are very small (Stock & Watson, 2012). It is essential to get causal effects that are binding. Therefore, it is of great importance that the internal threats of validity are taken into account. There are five reasons why a coefficient might be biased, these are: omitted variable bias, misspecification of the functional form of the regression function, measurement error and errors-in-variables bias, missing data and sample selection, simultaneous causality (Stock & Watson, 2012). There is no reason to believe that data is missing or that the variables are measured imprecisely because the data used and the resources are simple and trustworthy.
The most common threat to internal validity is the omitted variable bias. Omitted variable bias occur when the dependent variable is correlated with one or more of the included regressors (Stock & Watson, 2012). When this appears, the first least square
14 assumption does not hold, hence the regressor will be biased. To solve the omitted variable bias in this research, two determinants will be added to control for time and area effects. Because immigration and unemployment will differ from time to time and from province to province, the linear regression will extend with two fixed effects. The control variables added are the estimators’ 𝐸𝐷𝑈!" and 𝑃𝑂𝑃!". These variables are integrated to prevent that the regressor 𝐼𝑀𝑀!" captures effects that are actually due to differences in education or population. The vector 𝑃𝑂𝑃!" denotes the highly skilled population and 𝐸𝐷𝑈!" controls for the labor force rate, all in the 𝑝th province in the 𝑦th year.
The abbreviation 𝐸𝐷𝑈!" stands for education, which is defined as the percentage of high skilled labor force per province. Where high skilled workers are people who achieved University or an HBO degree. Borjas (2001) states that immigrants tend to move to regions where the potential return to their level of education is especially high. In the Netherlands most of the highly skilled labor force live in and around the Randstad (CBS, 2013). They determine the level of unemployment, while they also create jobs for the lower skilled labor force. In figure 5, the size of the circle displays the amount of immigration per province in comparison to other provinces. In figure 6 the circles show the rate of highly skilled people per province in comparison to other provinces. Together one can state that popular destinations for immigration are the same as destinations with a large amount of highly skilled people. The effect of immigration on unemployment becomes hard to interpret as it may be picking up the correlation between unemployment and the percentage for highly skilled labor force and therefor it is important to include it in the model.
Figu re 5 P erc en tag e Im mig rat ion p er reg ion 2 01
15 With 𝑃𝑂𝑃!", population ratio is meant. This is the labor force as a percentage of the total population per region. Dutch inhabitants move relatively easily for there work, as distances in the Netherlands are short and the adjustments uncomplicated (Cörvers et al, 2009) . An important driver to move inside a country is the demand for labor. This makes the population ratio correlated with factor for unemployment. To get a clear estimation for immigration it is important to correct the influence of population on unemployment. The regression will expand to 𝑈𝑁!" = 𝛽!+ 𝛽!∗ 𝐼𝑀𝑀!" + 𝐸𝐷𝑈!"+ 𝑃𝑂𝑃!"+ 𝑢!"
When adding the education and population variables, the regression doesn’t take into account that the dependent variable changes over time and per province in this research paper; panel data can monitor and control omitted variables that differ from one province to the other and can also control for variations through time. With dummy variables for time and province unobserved effects can be taken into account. This is necessary because, for example, Noord-Holland has a completely different working environment with all the big market players covered, whereas Friesland still has a lot of farmers (CBS, 2013). Furthermore, the absolute number of a working force differs substantially; noticeable in figure 7. The time is also rapidly changing, As an example, the crisis of 2008 changed the economic landscape drastically.
Figure 7 Workingforce per province
source: Centraal Bureau Statistiek
Because all variables are observed for each province and each year, there is a balanced panel and fixed effects can cover the unobserved effects (Stock & Watson, 2012).
16 The regression will extend to a fixed effects regression: 𝑈𝑁!" = 𝛽!+ 𝛽! ∗ 𝐼𝑀𝑀!"+
𝐸𝐷𝑈!"+ 𝑃𝑂𝑃!" + 𝛿!+ 𝜂!+ 𝑢!".. The key insight this analysis provides is that if the
unobserved variable does not change over time, any changes in the dependent data must be due to influences of other factors. The binary variables 𝛿! & 𝜂! makes it necessary to exclude one province and one year due to the dummy trap. When this is not excluded, there will be perfect multicolinearity. The excluded province is Drenthe and the excluded year is 2010, because they both have the lowest significance rate.
Another threat to internal validity which is especially important for an area analysis is simultaneous causality. In this paper the effect of immigration on unemployment is being tested. However, migrants choose their destination also because it is likely to find a job. So they tend to move to destinations where the economic circumstances are favourable and low unemployment prevails. Some studies argue that there is a significant relationship found for unemployment causing immigration, but that there is no evidence that immigration causes unemployment (Marr & Siklos, 1994). When there is simultaneous causality, an OLS regression picks up both effects, so the OLS estimator is biased and inconsistent. The correlation between the two variables will measure a net effect and not a causal relationship. To have an effective OLS the explanatory variables need to be exogenous. With the instrumental variable regression 𝐼𝑀𝑀_𝑙𝑎𝑔!", which stands for immigration time lag, this problem can be solved. Marr and Siklos (1994) discovered that past immigration increased current unemployment, while future immigration rates were statistically unaffected by the contemporaneous unemployment. Also Borjas argues that immigrants have strong incentives to accumulate country-specific skills just after migration. A consequence is low levels of employment for immigrants upon their arrival, mostly because of informal competence like cultural-specific proficiency, language skills, understanding expected behaviour etc.
(Bevelander, 2010.) Okkerse (2008) points out that one year in most of the cases is not sufficient to get accustomed to a new country, therefor 𝐼𝑀𝑀_𝑙𝑎𝑔2!" is included as instrument variable. The last variable that will be added in the regression is 𝐸𝐷𝑈!"! , which is the abbreviation for Education * Education. Since the effect of highly skilled people will probably weaken over time, as for a certain point there are too many highly skilled people that there affect on the unemployment rate won’t be worth mentioning (Cörvers et al., 2009).
To finish the methodology, the research takes into account the law of large number, technically written 𝑌 → 𝜇!. The law of large numbers states that under general conditions the sample average will be near the real mean with a very high probability when the research is
17 done by a large number of random variables (Stock & Watson, 2012). The reason the law of large numbers holds is because the larger the sample, the more reliable the sample average will be. The observations in this research are 132, so it can be useful to do the research in some years again and thereby increase the observations and its reliability.
18
5. Data descriptive
Figure 8: Summary data used for OLS
Variable Observations Mean Std. Dev. Min Max Immigration 132 8.333 8.1187 1.258 28.761 Immigration lag 132 8.333 8.0127 1.348 28.761 Immigration lag2 120 8.333 7.994 1.379 28.761 Education 132 29.291 5.812 16.827 44.073 Education2 132 891.510 364.788 283.145 1942.469 Population 132 45.896 1.881 42.251 49.860
source: Centraal Bureau voor Statistiek
Figure 8 summarizes the main features of the variables obtained by data from CBS. In addition, figure 9 present the mean of the unemployment rate and the immigration rate for the Netherlands in general and for the separate provinces. As can be seen in figure 9 the unemployment rate is more stable and fluctuates less than the immigration rate does between the provinces. The immigration rate fluctuates significantly between the provinces, which can be attributed to the five main reasons for immigrants regarding where to settle.
Figure 9: Mean immigration and unemployment rate per area
Netherlands Groningen Friesland Drenthe Overrijssel Flevoland Gelderland Unemployment 5.517 7.400 5.882 6.191 5.500 6.536 4.809
Immigration 8.333 3.592 2.439 1.598 4.813 2.482 7.836
Utrecht Noord-‐
Holland Zuid-‐Holland Zeeland Noord-‐Brabant Limburg Unemployment 4.555 5.100 5.509 4.009 4.800 5.918
Immigration 6.256 22.793 27.276 1.985 12.421 6.508
source: Centraal Bureau voor Statistiek
Figure 10 presents the five reasons to migrate to the Netherlands: the labor market, family, friends and social networks, organizations which cope with migrants and the public domain like social security (Corvers et al., 2009). Noord-Holland and Zuid-Holland are the provinces in the Netherlands that cover most of those aspects and are therefore popular for migrants. Unemployment has less deviation between the different provinces due to the fact that on one hand regional government tries to keep this rate as low as possible and on the other hand the people are more flexible moving in a country in comparison to people who move between countries (Corvers et al., 2009).
19
Figure 10: Migration motives of non-‐Dutch immigrants in the Netherlands
source: Centraal Bureau voor Statistiek
Cörvers et al. (2009) shows that there is a strong correlation between unemployment and immigration. For example, when the labor market grows because of more vacancies than immigration will follow after approximately one year. This cyclically sensitivity especially counts for migration from European countries (Cörvers et al., 2009). In the graphs for the different provinces (appendix 1) it can be seen that in every province there is almost the same time trend for the unemployment. From 2002 to 2004 there is an increase in unemployment, subsequently there is a decrease until approximately 2008, in 2008 the start of the credit crunch and the crisis can be seen again by an increase in the unemployment. This time trend is an inference of the economic environment, and is supported by the GDP growth rate that can be seen in figure 11. From 2002 to 2004 the GDP growth fluctuates around zero, from 2004 there is a growth, which fluctuates around 0.2 and 2, while from 2008 you can clearly see the impact of the crisis.
20
Figure 11: The Dutch GDP growth rate
source: Trading economics
In contrast to immigration there is no real time trend that can be seen comparing the different provinces. This seems odd due to the expansion of the European Union and the reduction of the restrictive admission policy in 2007, which should encourage immigrants. But it makes sense when looking at the other factor which influences labour migration: vacancies. Especially immigrants from Eastern European countries are sensitive for
vancancies and job opportunities. This put together there is no real time trend (Cörvers et al, 2009). The provinces with the biggest immigration rates are Noord-Holland and Zuid-Holland due to the described factors above. Together with Utrecht and Noord-Brabant they house the six biggest cities of the Netherlands, which are: Amsterdam, Rotterdam, Den Haag, Utrecht, Eindhoven and Tilburg (CBS, 2013). Those provinces have in common that their immigration rate lies above the unemployment rate, while the other provinces have a higher unemployment rate than immigration. Brücker (2007) attributes this to the fact that big cities are attractive for migrants. In former times those cities had the most vacancies, which have been attracting a lot of immigrants. Nowadays new immigrants are attracted by among others the existing social networks and follow immigration.
21
6. Empirical results and model
The results for 𝑈𝑁!" = 𝛽!+ 𝛽!∗ 𝐼𝑀𝑀!"+ 𝐸𝐷𝑈!"+ 𝑃𝑂𝑃!" + 𝛿!+ 𝜂!+ 𝑢!" are summarized in figure 12. Each column represents a different regression, and each row reports a coefficient estimate and the standard deviation is between the brackets. At the bottom of figure 11 the F statistic is summarized, that test joint hypothesis of more than one of the regression coefficients is significant (Stock & Watson, 2012). Additionally the R2 is summarized, which is the fraction of the sample variance of the dependent variable that is explained by the regressors, important to beer in mind is that whenever you add more variables to an equation the R2 will increase without making necessarily it more precise
(Stock & Watson, 2012). Finally the residuals are tested on normality with the Shapiro-Wilk test, if the significance of the Shapiro-Wilk test is greater than 0.05 the residuals are normal. In appendix 2 the histograms for the normality of the residuals for the different regressions are summarized in order to find if the value would indicate that heteroskedasticity. In this example the chi-square of regression 6 indicated heteroskadasticity. Hence all regressions are performed with the usage of robust standard errors. The hypothesis to be tested with these results is 𝐻!: 𝛽!"" = 0, 𝐻!: 𝛽!"" > 0 the hypothesis will be rejected if the p-value is less than 10%. In the figure the bold outcomes are significant with 10%, bold and one star represents significance with 5% and two stars symbolizes a significance level of 1%.
Figure 12: Regression analysis of the effect of immigration on unemployment rate
Dependent variable: unemployment rate
Regressor (1) (2) (3) (4) (5) (6) Immigration -‐0.025 (0.013) -‐0.020 (0.015) -‐0.101 (0.094) Immigration 1 year lag (0.068) 0.004 Immigration 2 year lag (0.061) 0.072 (0.070) 0.061 Education 0.034 (0.034) (0.462) 0.025 (0.043) 0.036 (0.043) 0.042 (0.111) -‐0.015 Population -‐0.170 (0.118) 0.072 (0.090) 0.083 (0.091) 0.087 (0.098) 0.085 (0.100) Education2 (0.002) 0.001 Groningen 1.050** (0.348) 1.296** (0.327) 1.406** (0.344) 1.378** (0.349) Friesland -‐0.426 (0.380) (0.279) -‐0.007 (0.265) 0.294 (0.184) 0.246
22 Drenthe -‐0.123 (0.499) 0.419 (0.373) 0.776* (0.354) 0.513* (0.276) Overrijssel -‐0.648** (0.175) -‐0.487** (0.006) (0.200) -‐0.374 -‐0.706* (0.371) Flevoland -‐0.075 (0.404) (0.375) 0.297 (0.425) 0.591 (0.448) 0.554 Gelderland -‐1.138** (0.335) -‐1.328** (0.324) -‐1.447** (0.314) -‐1.414** (0.332) Utrecht -‐2.021* (0.801) -‐2.197** (0.795) -‐2.280* (0.792) -‐2.388** (0.851) Noord-‐Holland 0.271 (1.920) -‐1.591 (1.482) -‐2.749* (1.303) -‐2.601 (1.381) Zuid-‐Holland 1.407 (2.114) (1.526) -‐0.838 (1.379) -‐2.225 (1.571) -‐1.975 Zeeland -‐2.230** (0.514) -‐1.715** (0.409) -‐1.427** (0.383) -‐1.509** (0.408) Noord-‐Brabant -‐0.765 (0.758) (0.629) -‐1.442 -‐1.863** (0.571) -‐1.570** (0.303) 2002 -‐0.744 (0.411) -‐0.638 (0.403) 2003 0.341 (0.349) 0.429 (0.341) 0.473 (0.345) 0.460 (0.350) 2004 1.473** (0.273) 1.536** (0.271) 1.566** (0.263) 1.569** (0.266) 2005 1.523** (0.239) 1.575** (0.233) 1.601** (0.229) 1.607** (0.231) 2006 0.384 (0.207) (0.206) 0.433* (0.206) 0.456* (0.178) 0.462* 2007 -‐0.696** (0.171) -‐0.660** (0.174) -‐0.640** (0.176) -‐0.635** (0.178) 2008 -‐1.466** (0.168) -‐1.448** (0.171) -‐1.437** (0.181) -‐1.432** (0.182) 2009 -‐0.501** (0.155) -‐0.490** (0.156) -‐0.483** (0.157) -‐0.479** (0.159) 2011 -‐0.982 (0.144) (0.146) -‐0.089 (0.139) -‐0.085 (0.138) -‐0.084 2012 0.840** (0.221) 0.840** (0.227) 0.841** (0.218) 0.839** (0.220) Constant 5.727 (0.174) 12.466 (4.653) (4.745) 2.601 (4.659) 1.112 (4.848) 0.327 (4.970) 1.290
F-‐statistics & p-‐values testing immigration
F-‐statistic 3.78 2.14 60.57** 61.80** 51.32** 50.34**
𝑹𝟐 0.024 0.052 0.919 0.918 0.915 0.915
Shapiro-‐Wilk. 0.127 0.084 0.014 0.032 0.030 0.004
Observations 132 132 132 132 120 120
source: Centraal Bureau voor Statistiek
When looking at the results in column (1) until column (3) the outcomes show an optimistic effect from immigration. The coefficient presents a negative relation between immigration and the unemployment rate, implying that when immigration increases the
23 unemployment rate drops. This can be contributed to two effects: most of the immigrants do not have a job upon arrival and need to acclimatize first as discussed above. Immigrants need to look for job opportunities, apply for jobs and get accepted. Secondly, they make use of the Dutch economy, by purchasing goods and services. They provide extra demand and therefore provide jobs for the native workforce. However, the coefficients of equation (2) and (3) are not significantly different from zero, hence once cannot make a decision about the hypothesis to be tested. For column (1) Immigration is significant, but the R2 is considerably low, indicating that the fraction of the sample variance of the dependent variable explained by immigration is very little. Which makes sense as the empirical analysis ignores some important determinants of unemployment. As explained in economic theory and methodology, the percentage of high skilled labour force and the population rate can probably affect the unemployment rate. Additionally unemployment can differ widely from province to province but also from time to time, therefore the R squared jumps from 0.052 to 0.919 when fixed effects for year and province are included ( Column (3)). This means that the fraction explained by the regressors increased from 5.2% tot 91.9%. Evidently, the province fixed effects and year fixed effects account for a large amount of the variation in the data.
For the purpose of an additional check and to solve the internal threat of reverse causality the Immigration coefficient gets replaced by Immigration lag. The estimations (4), (5) and (6) now show the necessity of immigrants to adapt for at least one year before they are completely settled. When compensating for the time lag in immigration the dependent variable changes the negative relationship with unemployment into a positive one. An important factor, as mentioned before, is that it takes some time to get accustomed to the economic environment. This adjustment is harder for lower educated people as it is harder to learn for example a new language. In some cases they cannot speak proper English yet (Heyma et al, 2008). On the contrary, highly educated immigrants adapt more quickly, due to, among others, they already speak English and are used to international environments by participating at an university (Heyma et al., 2008). To generalize Cörvers (2009) states that lower educated people need two years to fully acclimatize, whereas higher educated people need one year top. The results in figure 12 supports previously mentioned while immigration with one year lag is positive but really small and not significantly differs from zero. As a result of the higher educated immigrants who are adapted in the economy where only some of the lower educated immigrants have been. Replacing the 𝐼𝑀𝑀_𝑙𝑎𝑔!" with 𝐼𝑀𝑀_𝑙𝑎𝑔2!" is
24 encouraging as more coefficients are significant and the dependent variable immigration lag 2 becomes more significant, the P-value changes from 0.958 to 0.237. Important is to keep in mind that the estimates are not significant. When looking at 𝐼𝑀𝑀_𝑙𝑎𝑔2!" the positive relation becomes bigger and more significant, which support the fear of Assher, that immigration will increase the unemployment rate. Including 𝐸𝐷𝑈! as a control variable has little effect and is not significant, It can be concluded that 𝐼𝑀𝑀_𝑙𝑎𝑔2!" is not really affected by inclusion, meaning that the coefficient education is more linear than theory would predict.
Broadly speaking, there is a positive relationship between unemployment, education and population nevertheless the outcomes are not significant hence it is likely that the data is a true null. In none of these specifications have the dynamics of lower and higher skilled labour immigrants been explored. It was not possible to find statistically reliable and well determined estimates of dynamic specifications and have therefore refrained from commenting on differences between the different educated immigrants. It is good to note that literature suggest that there is a difference in replacing and adding up immigrants for lower and higher skilled labor force,
When deriving a conclusion based on the outcomes provided in the table. When testing 𝐻!: 𝛽!"" = 0, 𝐻!: 𝛽!"" > 0 with a significance level of 10% for the p-value. The regressor 𝐼𝑀𝑀_𝑙𝑎𝑔2!" in column (5) has a value of 0.703 with a p-value of 0.25. Therefor the null hypothesis is not rejected, and the probability that immigration had no influence on unemployment is likely. The outcome estimates that immigration probably does not displace the current native workforce neither create jobs. When looking at column (1) the announcement of the Dutch minister of social affairs Assher “immigration has a negative impact on the unemployment of the Dutch labour market” is not true when they arrive in the Netherlands and nothing else is taken into account. However when looking at the effects when immigrants stay longer than one year in the Netherlands the negative relationship between immigration and unemployment changes into a postive one. In other words immigration leads to more employment in the Netherlands the first year they arrive, whereas they lead presumably to unemployment after intergration or have no real effect on the economy. An important side note is the Shapiro-Wilk test of the residuals. The outcome is 0.005, which rejects the hypothesis that the residuals are normally distributed and therefor it
is necessary to carefully handle the outcomes provided.
25
7. Conclusion
Despite the concern occasionally expressed by policy makers and the popular press about the impact of immigration on the unemployment rate, few have studied the problem empirically. This thesis provides a first examination of the impact of immigration on the Dutch labour market focusing on unemployment. The main finding of the specific factor model is that with international labour mobility the general cake is bigger but there are people who get a smaller piece. The research about the Eastern European enlargement proved that the Netherlands is perceived as an attractive country to migrate to.
The salient conclusions from the area analysis are as follow. First of all when only looking for the effect of immigration it appears to have a significant negative relationship with unemployment amplifying when immigration increases, the unemployment rate drops. However this empirical analysis ignores some important determinants of unemployment and hence has a low R2. When fixed effects got included this problem is solved. However immigration is not significant anymore. An important factor is that it takes time for immigrants to get accustomed to a new economic environment. This adjustment differs between lower and higher skilled migrants but takes between one and two years. When compensating for this time lag in immigration the dependent variable changes from a negative relationship with unemployment into a positive one. Which would support the fear of Assher, that immigration will increase the unemployment rate for Dutch natives and thereby has a negative effect for the Dutch economy. Nonetheless because of a p-value higher than 10% it is presumably that immigrants have no real effect on the economy.
To conclude and to give an answer on the question “What is the impact of the European Union enlargement on the Dutch labour market?”. The European enlargement does change the economic landscape of the Netherlands. It significantly increases the labour force and the demand for goods and services. The area analysis does not give a significant outcome to prove the effect of immigration on unemployment. But if there is any evidence of negative effects on the unemployment rate in any group, then this is for those with lower education levels. However this negative effect is offset by the contribution to the Dutch innovation and growth potential from highly skilled immigrants. To finalize it is desirable to further examine the effects of immigration, find solutions for the negative effects on the lower skilled labour force and attract higher skilled immigrants.
26
8. Limitations and suggestions for future research
This thesis comes with various shortcomings. First of all, the sample issue using a sample of the CBS restricts the research because of the accurate numbers available from 2002 till 2012. This is a small time frame and effects to the enlargement can take up more time because of the limited flexibility of migration. Therefor measurement can be entirely different when looking twenty years from now. Also, between the provinces there are large differences, so it is best to divide them in smaller parts like for example in municipalities. The migration pattern in the capital is completely different to that of a small village. So effects of immigration can be entirely different in the future and more specific regions. Additionally measurement issues may arise with the choice of the dependent variable and how it is measured. Using one variable for immigration instead of dividing it between low skilled and high skilled immigration or distinguish between seasonal/short term immigration and long term immigration will lead to an overall effect which does not hold for specific groups.
Future research should focus on exploring more detailed effects of immigration. This paper only shows the overall effect of immigration on the unemployment rate but does not distinguish between age, sex or job experience. Besides it would be valuable to add social aspects so the conclusion could be used for creating new immigration policies also a valuable addition could be to investigate the impact of the credit crunch on immigration.
27
9. Reference list
Angrist, J. D., & Kugler, A. D. (2003). Protective or counter productive? Labour market institutions and the effect of immigration on EU natives, The Economic Journal, 488, 302-331.
Bonin, H. (2005). Wage Employment effects of immigration to Germany: Evidence
from a skill group approach (IZA bonn Discussion Paper No. 1875). Retrieved from: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=870272
Borjas, G.J. (2001). Does immigration grease the wheels of the labor market? Brooking papers on Economic Activity, 1, 69-133.
Brücker, H. (2007). Labor mobility after the European union’s Eastern enlargement: who wins, who loses? The German Marshall Fund of the United States.
Bevelander, P. (2010). The employment integration of immigrants in Sweden. Journal of Ethnic and Migration Studies, 25, 445-468
Centraal Bureau Statistiek. (2013). Beroepsbevolking, Kerncijfers provincie. Retrieved from: http://statline.cbs.nl/statweb/
Cörvers, F., Muysken, J., Neubourg, C., Schliwen, A. (2009). Migratie naar en vanuit Nederland. Cashier 2009-3, 1-180
Dustmann, C., Fabbri, F. & Preston, I. (2005). The impact of immigration on the Britisch labour market, The Economic Journal 507, 324-341.
Europese Unie. (2012). Statistics. Retrieved from:
http://europa.eu/publications/statistics/index_nl.htm
Herderschee, G. (2013, august 17). Asscher: bescherm Nederlandse werknemers tegen invasie Oost-Europa. De Volkskrant.
Heyma, A., Berkhout, E., Werf van der, S. & Hof, B. (2008). De economische impact van arbeidsmigratie uit de MOE-landen, Bulgarije en Roemenie (SEO-rapport nr 2008-70). Retrieved from: http://www.seo.nl/uploads/media/200870_
De_economische _impact_van_ arbeidsmigratie_uit_de_MOE landen_ _Bulgarije_en_Roemenie.pdf
Islam, A. (2007). Immigration unemployment relationship: the evidence from Canada, Australian Economic Papers 46, 52-66.
Krugman, P. R., Obstfeld, M. & Melitz, M. J. (2012). International economics: theory & policy. Edinburgh Gate, England: Pearson.
Mar, W. L. & Siklos, P. L. (1994). The link between immigration and unemployment in Canada. Journal of Policy Modeling 16(1), 1-25.
28
Okkerse, L. (2008). How to measure labour market effect of immigration, Journal of Economic Surveys 22, 1-30.
Rijksoverheid, (2013). Terwerkstellingsvergunning voor buitenlandse werknemers. Retrieved from: http://www.rijksoverheid.nl/onderwerpen/buitenlandse-werknemers/tewerkstellingsvergunning-voor-buitenlandse-werknemers
Stock, J. H., Watson, M. M. (2012). Introduction to econometrics. Edinburgh Gate, England: Pearson.
29
10. Appendices
10.1 Appendix 1 Graphs
30
31