The impact of immigration on the unemployment of natives in the Netherlands

Faculty of Economics and Business

Bachelor Thesis Version: 2

The Impact of immigration on the unemployment of natives in The Netherlands

Miloud Ourahou, 10881603

BSc Economics and Business

Supervisor: Simon ter Meulen

Abstract

This bachelor thesis is about the impact of immigration on unemployment. There still no consensus among academics about the right approach to estimate this impact. There has been also little research about impact of immigration on unemployment in the Netherlands and none about the recent influx of Syrian immigrants. The differences in immigration influx between provinces in the Netherlands has been the reason to use a spatial approach in this paper using a difference-in-difference estimation model to estimate the impact of immigration on unemployment. Using regional data from the Netherlands between 2010 and 2016, I investigated whether the shock of immigrants in the Netherlands in 2013 has led to more unemployment. I find that provinces who experienced a growth of immigration of more than 0,1 percentage point also experienced an increase of 0,156 percentage point in

Table of content

1. Introduction

The last couple of years the debate about immigrants has been an pressing topic. The ‘Syrian refugee crisis’ where many refugees from Syria fled the brutal war and were

searching for a safe haven, has been the most important topic. In western Europe, particular Germany and the Netherlands it has been one the major social debate topics. The National institute for Social Research (SCP) researched the sentiment among Dutch civilians in 2015 and 2016. 64 percent in 2015 and 46 percent in 2016 of respondents answered that

immigration is the biggest problem in the Netherlands. This dissatisfaction has led to (or maybe enforced by) politicians that expressed their social and economic concerns of immigration. The focus of this study is to investigate the effect of immigration on

unemployment which is relevant in the social discussion. It could prove statements made or sentiments felt by Dutch citizens.

This research is also academic relevant since there is still no consensus among academics about the effect of immigration on the labor market. There has been a lot of research and the results vary widely. There is especially no consensus about the method of estimation. The basic canonical model of labor is not consistent with empirical findings. Many researchers have tried therefore to enhance the canonical model.

The aim for this study is to investigate what impact the immigration shock of 2013 had on the unemployment rates in Netherlands. This will be done by using regional data and compare the differences using a difference-in-difference estimation method. I investigate whether provinces that received more than 0,1 percent point increase in immigration after the shock of 2013 will experience larger increase in unemployment than provinces with a lower increase. I find a slightly positive coefficient of 0,156. Provinces that had an increase of immigration of more than 0,1 percentage point experienced 0,156 percentage point more unemployment compared to provinces that did not receive that much immigration. However this coefficient is not significantly different from zero.

I will first discuss the relevant literature and show the different estimation models and why they differ in results. Next I will discuss the data and its limitations. I will then explain my

methodology and the results obtained. Finally the conclusion be discussed in which I will comment on my findings.

2. Literature review

The effect of immigration on employment and labor market in general is a widely studied area. The classic canonical textbook model predicts that inflow of immigrants will lower the wages of the natives since supply of labor will exceed the demand for labor. This may seem logic and intuitive but there is ample empirical evidence that support this model. Research on this topic have showed a mixed and often confusing results. Friedman and Hunt (1995) conclude that the effect of immigration on the labor market fluctuates around zero or is at the least very small. Since then new approaches have introduced in an attempt to enhance the classic canonical model.

Altonji and Card (1991) used a spatial correlation model to estimate the effect of immigration on low-skilled labors across cities in the U.S. They measured the effect of immigration in a city on the wages of the natives in that particular city. They used a cross-section analysis and a difference-in-difference analysis. The latter is less likely to be biased by omitted city specific factors that affect immigration density and native employment. Using data between 1970-1980 in the U.S. they estimate that one percentage increase point reduces the wage of low skilled labors on average with 1.21 percent. This result might be country specific since wage rigidity and its effect on unemployment vary heavily across country Grubb (1983). Many more studies followed up on this approach and the results vary widely even between positive and negative effects. Most recently Foged and Peri (2016) studied the data of Denmark between 1991 and 2008 ( 2016) and the found a suprising 1.8 percent increase in wage on average as a result of 1 percentage point increase in refugees.

There might be many reasons why studies do find different outcomes when using the spatial correlation approach. First a region or a city is not a closed economy so migration exist between regions and is more likely if wage rates are higher in a different region (Borjas

2003). If all wages decrease, workers can only choose between employment and

unemployment (Dustmann et al. 2016). Firms are also mobile and might want to move into a region that has lower wages due to the inflow of immigrants. This might even result in a positive correlation between an increase in immigration share and job opportunities (Carrasco et al. 2008).

In a very influential paper of Borjas (2003) introduces a new approach. In this research Borjas (2003) investigates the effect of immigration based on their characteristics on natives with same characteristics. Based on their education level and experience natives and immigrants are put in the same so called ‘skill-cell’. Since this is a short run model Borjas (2003) assumes that there is no mobility between skill-cells because attaining a higher education level will cost several years. Using national data the migration bias between regions does no longer matter. The research shows that an increase in supply of 10 percent in a certain skill category would result in a 3,2 percent decrease in wage on average in that particular skill group.

Many studies have followed up on this national-skill cell approach and these studies start showing more similar results. Most studies show a negative effect of immigration on wages of natives.

Ottaviano and Peri (2008) made some alterations in the national skill-cell model. They relax the assumption made by Borjas (2003) and many other studies that followed, that capital is fixed in the short run. Ottavaino and Peri (2008) find it more realistic to let capital increase ten percent every year as a response of firms to the increase in labor stock.

In the study of Borjas (2003) immigrants and natives are divided in to education-experience groups and are treated as perfect substitutes. Ottaviano and Peri (2007) challenge this assumption. Immigrants might tend to choose a different set of occupations because they are a selected group or because they have some culture specific skills. There is also strong empirical evidence that immigrants downgrade upon arrival (Dustmann et al. 2016). This means that they accept lower paying jobs than the natives in the same education and experience group.

A third approach on the estimation of immigration of wages introduced by Card (2001) is the mixture approach. This approach exploits variation in immigration inflow on both skill-cell and region. It is a mixture of the previous two mentioned approaches using a triple difference estimator where differences are taken over time, across region and across education groups. Dustmann et al. (2016) analyzed some studies that used this approach and found that most studies show a negative effect of immigration on the wages of natives, but these effect are less strong compared to the national skill-cell approach. Borjas (2006) investigated this approach and found a decrease of 0,06 percent by percentage point increase in immigration compared to a decrease of 0,57 percent in his national skill-cell approach study in 2003.

All these approaches give different results and a consensus on which approach is most suited to estimate the impact of immigration on the labor market has not been reached among academics. Dustmann et al. (2016) argue that the contradictory results come from the fact that the specifications used in the models differ and that the results cannot be compared with each other.

Furthermore, two major assumptions are tacitly made. First it is assumed that the labor supply elasticity is homogenous across all groups. Second immigrants are placed in the same education -experience group as natives if the immigrants report the same education and age group. As mentioned before there is strong evidence that immigrants downgrade upon arrival. To solve this bias it is better to look at the total effect of immigration so the downgrading is excluded since the immigrants are not put in to groups. Dustmann et al.(2016) proposes to use overall instead of group specific immigration shock on wages and employments of various native groups. It is assumed that the allocation of immigrants to different groups is independent to shocks in wages and employment. Downgrading leads to a bias in the national skill-cell approach and the mixture approach. It overestimate the immigration shock in high-skilled group and under estimates the immigration shock in the low skilled group.

There has not been conducted many studies about the effect of immigration on unemployment rates in Netherlands. The last study was conducted by Galloway and

Jozefowicz (2008) who investigated the volatility of Dutch employment rates in relation with an increase in foreign labors in the Dutch labor market and found a positive statistically significant effect.

3. Data

I have collected data from the Central Bureau of Statistics (CBS) from 2010 till 2016 from the Netherlands. The data consist of immigration and unemployment rate of each province. The immigration is expressed as percentage of the total population within in each province. The unemployment rate is the percentage of the labor force that seeks but does not have a paying job for at least 12 hours a month (CBS). The CBS includes everybody between the age of 15 and 75 that earns wages by having a job or those who do not have job but are currently searching for a job, to the labor force.

The definition of immigrants might be different from definitions used by other statistical bureaus and might also be country specific. According to the CBS and thus this paper, not all the people that enter The Netherlands are considered immigrants. Immigrants in this case are all the people that enter the Netherlands while they are not registered and seek to live there for a longer period of time and are granted the right ( or obligation) by the government to register themselves at the municipality or residence.

The Syrian refugees as referred to in this paper as well as refugees in general can seek asylum and are put in emergency shelters upon arrival. They can apply for an asylum upon arrival. In awaiting of approval of their asylum they will be put in Asylum Seeker Centers (Asiel Zoeker Centra). No later than 6 months the asylum seekers have to be registered in the municipality of residence regardless of having received an approval of asylum. From this moment on they are considered immigrants and accounted for in immigration number of the CBS and in this paper.

What will be explained will be important for the unemployment rate of immigrants. The same definition criteria are used for immigrants and natives. However their practical situation is different. While the refugees wait for approval of their asylum up until they

receive their working permit they are not been considered part of the labor force by the CBS. Immigrants whose motive is economical to migrate and have a working permit, like Polish workers for example, will be considered part of the labor force. This is an important aspect of this paper since the influx of Syrian immigrants does not lead to correlation by default on unemployment rates.

After been granted asylum the immigrant does not necessarily enter the labor force. Some immigrants keep receiving government allowance or have other income. Therefore to estimate the labor force participation of immigrants we look at the most important source of income. If the income is wage, profit or from unemployment benefits received from the government, than those immigrants are considered part of the labor force. Note that there is a difference between allowance and unemployment benefits from the government. The latter is granted unemployed persons according to before mentioned definitions and the allowance is for every other reason the government decides to aid the individual.

In the graph below you can see the labor force participation rate of immigrants in the Netherlands in years since their arrival. In their first year of immigration on average 36 percent of immigrants between 15 and 65 years old actively search for or having a paying job. After approximately 7 years of residence in the Netherlands the participation rate is about 65 percent on average where it stagnates. The average labor force participation rate in the Netherlands is 69% (CBS)

Graph 3.1

There are also some limitation of the data set used. Undocumented immigrants are not considered but they have an impact on unemployment rates if they take undeclared jobs.

Also other characteristics of immigrants are absent in the data set. Characteristics like education level would give more insights about the employment chances of immigrants. It would be also interesting to know where immigrants migrate after arrival. It could be that a low unemployment rate in a certain province lures immigrants to that area. This could lead to endogeneity. Firms that seek low skilled labor could also locate themselves where there are unemployed low skilled workers.

4. Methodology

To estimate the effect of immigration on the unemployment rates of an certain region I will perform a difference-in-difference regression. This method is used to run an experiment like research but then with observatory data, therefore it is a quasi-experiment. I investigate whether the unemployment rate changes if a province (treatment groups) experience a higher influx of immigration compared to other provinces (control groups). We are looking for

the average difference over time between the control groups and the treatment groups, because they might be different to begin with.

The estimation model will look like this:

𝑌"#$ = 𝛼#+ 𝛽$+ 𝛿𝐷+$+ 𝜀"#$

𝑌_"#$ is the dependent variable which in this case is the unemployment rate at data point i given region r, and year t. 𝛼_# and 𝛽_$ stand for the intercept with region and time respectively. 𝐷+$ is a dummy variable that indicates whether a treatment is received or not and 𝛿 is the

effect on the treatment received. 𝜀_"#$ is the error term. 𝛿 is the variable of interest. First we have to define the treatment dummy 𝐷_+$. With observatory data we cannot give a treatment to a group like an in an experiment. We have to look for changes over time. In our case with immigration, we are looking at intensity or variation differences in immigration over time. For this example we are comparing the province Noord-Holland (n) with Zuid-Holland (z). They are both quite similar. They are most populated provinces of the Netherlands and both very urbanized. There are some differences though. The capital Amsterdam is located in Noord-Holland and that attracts the most tourist. Zuid-Holland has the second largest city of the Netherlands, Rotterdam which has the biggest port in Europe. The assumption made here is that these factors are constant over time and should not affect our results.

We are investigating the effect on unemployment after the treatment is given. We will use 2013 as our shock. We are going to use the difference in immigration between Noord-Holland and Zuid-Noord-Holland from before the treatment e.g. the shock of Syrian immigrants to the Netherlands and after and use it as an estimator for the difference in employment after the treatment.

Graph 4.1

Between 2010 and 2013 the immigration rate moves the same in Noord-Holland and in Zuid-Holland. On the graph you clearly see that the immigration increases more in the Noord-Holland after the shock in 2013 than in Zuid-Noord-Holland. The difference before and after the shock in the difference between the two provinces is the treatment.

Graph 4.2

The difference between Noord-Holland and Zuid-Holland in employment rate also changes after 2013. Using a difference-in-difference estimation we hope to infer a causal relationship between the change in immigration and the change in unemployment.

0,00% 2,00% 4,00% 6,00% 8,00% 10,00% 2010 2011 2012 2013 2014 2015 2016

Unemploymen rate

yearly immigration

Now follows the formal explanation of the difference-in-difference estimation. Time can take two values. Before the shock (b) or after the shock (a). Before the shock is mean of the data between 2010 and 2013 and after is the data form 2014 to 2016.

(𝑌/₀₁− 𝑌/₀₃) − (𝑌/₅₁− 𝑌/₅₃)

=[(𝛼0+ 𝛽1+ 𝛿𝐷01+ 𝜀01) − (𝛼0+ 𝛽3+ 𝛿𝐷03+ 𝜀03)] − [(𝛼5+ 𝛽1+ 𝛿𝐷51+ 𝜀51) −

(𝛼5+ 𝛽3 + 𝛿𝐷53+ 𝜀53)]

= 𝛿(𝐷₀₁− 𝐷₀₃) + 𝛿(𝐷₅₁− 𝐷₅₃) + 𝜀₀₁ + 𝜀₀₃+ 𝜀₅₁+ 𝜀₅₃

Using ordinary least square and thus assuming that error term are zero. We get:

𝐸 (𝑌/01− 𝑌/03) − (𝑌/51− 𝑌/53) −= 𝛿(𝐷01− 𝐷03) + 𝛿(𝐷53− 𝐷51 )

The treatment dummy is received in Noord-Holland after 2013 so 𝐷53 = 1 and

𝐷₅₁, 𝐷₀₁, 𝐷₀₃ are equal to zero. That leads to:

𝛿: = (𝑌/₀₁− 𝑌/₀₃) − (𝑌/₅₁− 𝑌/₅₃)

Synthetic control groups

A more suitable way to investigate the shock effect of immigration on the

employment levels of provinces in the Netherlands, is by use of a synthetic control

groups. By combining different groups and assigning them different weights, we can

simulate the treatment group before the shock. After the shock the weights of the

control groups stay the same. We could then see what would happen if the treatment

group would not have received the treatment. The difference found would then be

the causal effect of the treatment.

Instead of using just Zuid-Holland as a control group for Noor-Holland we

could use other provinces as well and assign them different weight so that the total

weight of all the control provinces simulates Noord-Holland precisely before the

shock moment. And then we look with the same weights what the synthetic control

group would have behaved like after the shock moment. Due to a lack of time I could

not perform this method, but it would be an interesting method to study further.

5. Results

All the provinces experience a continuous inflow of immigrants. It is not so obvious to select control and treatment groups. The yearly influx of immigrants as a percentage of the

population ranges between 0,41% and 2,52%. The yearly influx of immigrants rose in all the provinces. Some more than others. We are investigating whether an increase of more than 0,1 percentage point of the growth of immigration has an significant impact on

unemployment rates. This will be our treatment. For this model is does not matter with how much more the immigration rose after the shock of 2013. This way outliers do not harm the model. For instance the Immigration and Naturalization Service (IND) is located in

Groningen at the border with Drenthe. Those provinces had biggest increases in immigration but a more than average fraction of those immigrants were refugees. They most likely will leave the provinces later on migrate to other provinces.

The table below shows the immigration influx difference before and after the shock. We see that Friesland, Overijsel and Zeeland experience an increase in the percentage points of immirgants of 0,051%, 0,076% and 0,056 respectively. These three provinces will be the control groups. The other nine provinces receive the treatment of having an increase of immigration of more than 0,1 percentage point.

Table 5.1 Average immigration rate in percentage points.

Province GR FR DR OV FL GE UT NH ZH ZE NB LI Before 2013 1.081 0.512 0.438 0.617 1.068 0.600 0.780 1.337 1.257 0.905 0.810 1.034

After 2013 1.870 0.563 1.247 0.693 1.647 0.784 0.921 1.691 1.403 0.960 1.135 1.271

Difference 0.790 0.051 0.809 0.076 0.579 0.183 0.141 0.354 0.146 0.056 0.325 0.237

When using Ordinary Least Square (OLS) estimationwe get the following regression results. The R Square is 0,712 which means that 71,2 percent of the variance of the dependent variable is explained by the model.

Table 5.2 Regression coefficients

Dependent variable

The dependent variable is the unemployment rate of each province before 2013 and after 2013 including natives and immigrants.

Constant

The constant is the average unemployment rate if all the other variables are equal to zero. That means that the time dummy variable is equal to zero meaning the period before 2013 and treatment dummy variable is zero meaning that the increase in yearly immigration is lower than 0,10 percent. This is the average unemployment rate of Friesland, Overijssel and Zeeland which 4,7 percent.

Time Dummy

The time variable indicates whether we look at the unemployment rate before or after the shock in 2013. Before the shock the average of the unemployment rates of each region is taken between 2010 and 2013. After the shock it is the period between 2014 and 2016. This coefficient indicates the difference in unemployment rate on average of all the provinces between before and after the shock of 2013. The coefficient is 1,8% meaning that on average the unemployment rate was 1,8 percent higher after 2013 than before 2013. This captures all time trend effects. This coefficient is significantly different form zero with a P-value of 0,004.

Treatment dummy

The treatment dummy indicates whether a province experienced an increase of 0,10 percent or larger of growth in immigration. The coefficient gives the average difference in

unemployment rates between provinces that experiences the growth increase of more than 0,10 percent versus the provinces that receive less. The coefficient is 0,389. That means that the unemployment is 0,389 percentage point higher in the treatment groups than in the control groups. However this coefficient is not significantly different from zero.

Interaction variable

The coefficient of interest is the interaction between the time and treatment. The differences are taken over time and across regions hence the name difference-in-difference. This coefficient estimates whether unemployment changes between the treatment groups and control groups over time. The coefficient is 0,156. The interpretation of this coefficient is that on average provinces with an increase of yearly immigration of more than 0,10 percentage point experience 0,156 percentage point more unemployment after the shock of 2013 than the control provinces. This result is in line with the hypothesis however the coefficient is not significantly different from zero.

6.Conclusion

The objective of this study was to investigate the impact of immigration on the

unemployment rates in the Netherlands after the Syrian refugee crisis in 2013. The impact of immigration on the labor market has been widely studied especially in the United States but there is yet to be found consensus about the right method of approach to estimate the impact of immigration on unemployment.

In this study I have adopted a difference-in-difference estimation method to estimate the impact of immigration on the unemployment rate in the Netherlands. More specifically I investigated what happened to the unemployment rates of provinces with an higher increase in immigration after the shock of 2013 compared to provinces with lower immigration growth. Using a OLS regression I found a coefficient of 0,156. This coefficient indicates that a

province with more than 0,1 percentage point increase in immigration experienced an 0,156 percent point average increase in unemployment. However this coefficient is not significantly different form zero.

This model has its limitations. It would be more suitable to use synthetic control group instead of using other provinces as control. There was also a lack of other control variables. It would be more informative to control for education and age of the immigrants.

Another factor which is not been controlled for is the fraction of the population with immigrant background. There is a strong correlation between immigration inflow and fraction of the population with immigrant background. (see Appendix)

There are also some endogeneity concerns. The mobility of workers between

provinces is not controlled for. Workers could migrate to provinces with lower unemployment rates. Or firms could relocate to provinces with high unemployment if the wages are lower.

These limitations could be further investigated and could lead to improvements to future studies.

.

7. References

Altonji, J. G., & Card, D. (1991). The effects of immigration on the labor market outcomes of less-skilled natives. In Immigration, trade, and the labor market (pp. 201-234). University of Chicago Press. Amnesity International: Ongedocumenteerden (illegalen) en uitgeprocedeerden. (2018, june 20). Retrieved from:

https://www.amnesty.nl/encyclopedie/ongedocumenteerden-illegalen-en-uitgeprocedeerden

Ben-Gad, M. (2004). The economic effects of immigration—a dynamic analysis. Journal of Economic

Dynamics and Control, 28(9), 1825-1845.

Borjas, G. J. (2003). The labor demand curve is downward sloping: Reexamining the impact of immigration on the labor market. The quarterly journal of economics, 118(4), 1335-1374.

Borjas, G. J., Grogger, J., & Hanson, G. H. (2008). Imperfect substitution between immigrants and

natives: a reappraisal (No. w13887). National Bureau of Economic Research

Carrasco, R., Jimeno, J. F., & Ortega, A. C. (2008). The effect of immigration on the labor market performance of native-born workers: some evidence for Spain. Journal of Population Economics,

21(3), 627-648

Card, D., & Peri, G. (2016). Immigration economics by George J. Borjas: A review essay. Journal of

Economic Literature, 54(4), 1333-49

Centraal bureau voor de Statistiek (CBS): immigration and unemployment data 2010-2016. Retrieved on 25 mei 2018 from: http://statline.cbs.nl/Statweb/publication/?DM=SLNL&PA=70072ned&D1=0-88,158,294-295&D2=0,4,8,20,91,312,768&D3=21-23&HDR=T&STB=G1,G2&VW=T

Chassamboulli, A., & Palivos, T. (2013). The impact of immigration on the employment and wages of native workers. Journal of Macroeconomics, 38, 19-34.

Dustmann, C., Schönberg, U., & Stuhler, J. (2016). The impact of immigration: Why do studies reach such different results?. Journal of Economic Perspectives, 30(4), 31-56.

Foged, M., & Peri, G. (2016). Immigrants' effect on native workers: New analysis on longitudinal data.

American Economic Journal: Applied Economics, 8(2), 1-34

Galloway, R. M., & Jozefowicz, J. J. (2008). The effects of immigration on regional unemployment rates in the Netherlands. International Advances in Economic Research, 14(3), 291-302

Grubb, D., Jackman, R., & Layard, R. (1983). Wage rigidity and unemployment in OECD countries.

European Economic Review, 21(1-2), 11-39.

Longhi, S., Nijkamp, P., & Poot, J. (2010). Joint impacts of immigration on wages and employment: review and meta-analysis. Journal of Geographical Systems, 12(4), 355-387.

Ortega, J., & Verdugo, G. (2011). Immigration and the occupational choice of natives: A factor proportions approach.

Ottaviano, G. I., & Peri, G. (2008). Immigration and national wages: Clarifying the theory and the

empirics (No. w14188). National Bureau of Economic Research

Sociaal Cultureel Planbureau (2016): Publication on civilian perspecives. Source: