• No results found

The effects of immigration on the income of native born workers: Evidence from Sweden

N/A
N/A
Protected

Academic year: 2021

Share "The effects of immigration on the income of native born workers: Evidence from Sweden"

Copied!
43
0
0

Loading.... (view fulltext now)

Full text

(1)

School of Business, Society and Engineering Bachelor Thesis in Economics

Mälardalens Högskola Spring 2021

The effects of immigration on the

income of native-born workers:

Evidence from Sweden

Supervisor: Johan Lindén

Examiner: Christos Papahristodoulou Date: June 3

Author:

(2)

Abstract

Course: NAA305 Bachelor Thesis in Economics 15 ECTS

University: Mälardalen University, School of Business, Society and Engineering, Västerås Title: The effects of immigration on the income of native-born workers:

Evidence from Sweden

Author: Arnold Schanzer-Larsen Supervisor: Johan Lindén

Problem: Sweden has experienced a lot of immigration, and the phenomenon has received a

great deal of attention in the public and political debate. There is, among other things, fear that immigration could be harmful for the labor market outcome of the receiving country. Researchers from a variety of countries have tried to address this issue by estimating the effect of immigration on the native wage of the receiving country. The results have varied strongly and no universal conclusion can be drawn. For what can be said about Sweden, there is no paper (of our knowledge) that has done any similar estimates. For that reason, it is of great importance that there is some research which could bring empirical evidence and shed light on the debate.

Purpose of the Research: The aim of the thesis is to quantitatively measure immigrations

effect on the wage of native workers in Sweden.

Methodology: Conducting a panel study, observation of the average native income from 290

municipalities over 2011-2019 was collected. The effect was estimated using OLS regression technique and a fixed effect model.

Conclusion: From a 10% increase in the share of foreign-born within a municipality, led on

average to a 2.89% increase in the native average income in that municipality.

Keywords: Immigration, Income, Wage, Unemployment rate, Panel study, Fixed effect

(3)

Acknowledgement

I would like to begin by expressing my gratitude towards my supervisor Johan Lindén. Who, despite an ongoing pandemic, were able to ensure that I received the most valuable feedback. Without his tips and insights, I would not have been able to complete this thesis.

(4)

Table of Contents

1. Introduction ... 1 1.1. Background ... 1 1.2. Problem ... 4 1.3. Aim ... 4 1.4. Limitations... 4 2. Theory ... 4

2.1. The Production Function & Labor Market Equilibrium ... 4

2.2. Immigration ... 6

2.3. The Solow Growth Model ... 8

2.4. Human capital ... 10

2.5. Unemployment ... 11

3. Methodology ... 12

3.1. Regression analysis ... 12

3.2. Panel study ... 13

3.3. The Fixed Effect Model ... 13

3.4. Specifying the model ... 14

3.5. Hypothesis of each variables effect ... 15

3.6. Data collection ... 16

3.7. Model evaluation ... 17

4. Results ... 20

4.1. Descriptive Statistics ... 20

4.2. Evaluation of the model ... 21

4.3. Final Results ... 24

5. Discussion ... 26

6. Summary and conclusion ... 29

7. References ... 30

(5)

1

1. Introduction

The rise of immigration in Europe and fear of potential negative outcomes in the economic and social landscape for the receiving country, has led the phenomenon to become one of the most important discussions among the public (Brücker & Jahn, 2008). Sweden, for example, experienced in 2016 their largest ever influx of 163 005 immigrants, which was

approximately 1.63% of the entire population (SCB, 2021). In their two years later

parliamentary election Novus (2018) among with SVT (2018) reported that immigration had been the second most important question for the voters.

One issue that caught researchers and policy maker’s interest has been the consequences on the labor market opportunities and wages of native workers. The theoretical understanding has for long been well understood, but the empirical evidence is not as clear. Starting off in the US Grossman (1982) released a seminal paper in which he estimated a small negative effect from immigration on the wages of native workers. From that point, a large scale of

researchers has tried to quantify and estimate the effects with different approaches and techniques.

1.1. Background

This section reviews the most well-known papers that have tried to estimate the effects of immigration on the wages of the receiving country. It will mostly cover the papers source country, approach, estimating technique and results.

United States

The impact of immigration on wages in the receiving country has been widely studied, with the majority of studies coming from the US. In the 80’s, Grossman (1982) estimates a

production function to compute the elasticity of substitution between the stock of immigrants and the native workforce. The study concludes that a 10% increase in the number of

employed immigrants would reduce native wages by 1%.

A couple of years later Butcher and Card (1991) tried to estimate the effect of immigration on the different percentiles within the wage distribution. In their estimate they controlled for population growth rate, the share of immigrants initially living in the city and the initial wages in the city. Their study could not prove that immigration had any effects on the different percentiles.

(6)

2

Later, some researchers started to become more interested in how wages within different occupations and/or skill groups were affected by immigration. For example, Altonji and Card (1991) used an instrumental variable (IV) technique with data from the 1970 and 1980’s US census to estimate the wage effects on less skilled natives. Their conclusion was that a 1% increase in the foreign share of the population in a city reduces the wages of unskilled native by 1.2%. Also, Card (2001), used the same technique to study the effect of immigrant inflows on the wages of six different occupations during 1985-1990. His conclusion was that a 10% increase in the immigrant inflow reduced wages of low-skilled service native workers in metropolitan cities by 1-3%. Once again, but this time smaller, Card is able to point to a negative effect from immigration. With a similar study, Camorata (1997) compared wages of natives in occupations with different proportions of immigrants and found that a 1% increase in immigration reduced wages of low-skilled natives by 0.8%. Another study on the issue, Borjas (2003), examined immigration over 1980-2000. Borjas defined skill groups in the form of education acquirement and years of work experience. He also used an IV approach and could conclude that the average native wage, due to immigration, was reduced by about 3% over the period. However, for the unskilled natives, the impact was the highest with a

reduction of 9%. Similar findings were done by Orrenius and Zavodny (2007). They used data on native’s wages within different occupations over the period 1994-2000 and estimated the effect from immigration inflow. Their findings reveal that only the least skilled occupations such as manual labor were negatively affected.

Also using the skill-cell approach Ottaviano and Peri (2012) found quite contradictory results to previous studies from the US. They found that during 1990-2006, immigration had a small positive effect on native workers. Mostly affected were the least skilled natives that

experienced an increase of 1.7%. Meanwhile, the average native worker only experienced a 0, 6% increase in the wage.

United Kingdom

Dustmann et al. (2005) was one of the first papers from the UK to estimate the impact of immigration on the aggregated wages. The study used a variety of estimation techniques such as OLS, IV and generalized method of moments (GMM) to analyze data on wages in different skill groups over the period 1983-2000. According to their study, wages seemed to be slightly positively correlated with immigration. From another perspective, Dustmannet et al. (2013), estimated the impact of immigration on the different percentiles in the wage distribution during 1997-2005. With the use of OLS and IV estimates, they found negative correlations

(7)

3

between immigration and low income earners and positive for the high income earners. On average, the effect seemed to be positive.

Nickell and Saleheen (2009) investigated with OLS and generalized least squared (GLS), how the immigrant-to-native ratio in an occupation was correlated to the average wage. Mostly small negative effects were estimated. Except for the unskilled service sector, where a 10% rise in the share of immigrants was associated with a 5% lower paycheck.

Germany

De New and Zimmermann (1994) examined the effects of immigration on white- and blue-collar native’s wages. With estimations from a random-effects panel model, their study implied that 1% increase in the share of immigrants correlated to a reduction of 4.1% for blue-collar workers hourly wage. In contrast, the same increase implied a 3.5% increase for white-collar workers. In the same year, Pischke and Velling (1994), with data from the same period (1985-89), but with an IV approach, found no significant effects from immigration on native’s wages. Later, Winter-Ebmer and Zimmermann (1999) with the same technique, but data from 1986–94, also found no significant negative effects.

With the use of a wage curve approach and econometric techniques such as two-stage least squares and GMM Brücker and Jahn (2008) managed to estimate small effects from immigration on wages. Their results suggest that a 1% increase in the labor force from immigration decreased the average wage by less than 0.1%. Of late, D’Amuri et al. (2010) presented, from their expectations, contrary findings. Based on a labor market equilibrium model and data over the period 1992-2001, immigration seemed to have a positive effect on wages of the less educated and negative on the highly educated.

Spain

More recently, a few studies have been developed in Spain. Two similar studies Carrasco et al. (2008) and González and Ortega (2011) found no significant negative impact of

immigration on the wages of native workers. Both studies applied OLS and IV estimates, but from two different time periods.

The most recent study carried out by María Gutiérrez-Portilla et al., (2020) takes on a slightly different approach as they made use of a spatial fixed effects model. With panel data from Spanish regions over the year 2004-15 they estimate that the relative stock of foreigners in a

(8)

4

province seems to have a small negative impact on the natives wage within that province and those nearby

1.2. Problem

With the large body of literature, it is no exaggeration to state that the results are somewhat ambiguous. Whether the differences in results are a consequence of a specific time, country or estimation technique could all be up for debate. But what is most concerning is that Sweden with its political and public interest in immigration has, to our knowledge, no study

quantifying the effects of immigration on the wages of native workers. A study of that sort would bring some light on the issue and hopefully clear some obscurities among the public.

1.3. Aim

The aim of the thesis is to quantitatively measure immigrations effect on the wage of native workers in Sweden. By doing so, the paper hopes to strengthen the Swedish public’s

knowledge about the labor market outcome of immigration.

1.4. Limitations

The study has been wildly limited in the availability of data. Therefore the study will only cover the period 2011-2019. For the same reason, the study had to be based on aggregated data instead of individual. The estimations are based on panel data with observations over demographic entities. Under such circumstances, it could be useful to control for spatial autocorrelation but those techniques are beyond the scope of our knowledge.

2. Theory

In the theoretical section, a framework for the empirical model will be made. Starting off with a general derivation of how wages are set in accordance with the production function and the equilibrium model. Thereafter follows theories that explain variation in the inputs of the production function such as human capital, unemployment and immigration.

2.1. The Production Function & Labor Market Equilibrium

The majority of studies that tries to explain the impact of immigration on wages, starts off with a somewhat simple production function. Most commonly is the assumption of a constant elasticity of substitution (CES) production function, see for example Borjas (2003),

Ottaviano and Peri (2006), Card and Lemieux (2001), Brücker and Jahn (2008). This

assumption is also used in student textbooks, in which they implement the CES Cob-Douglas production function (Borjas, 2019; Gottfires, 2013; Mankiw, 2019). They then derive the

(9)

5

demand for labor combined with labor supply in order to determine the wage. Consequently, this study will take the same approach and base its theoretical reasoning on the papers and textbooks mentioned above.

The Cob-Douglas production function could consist of many inputs, but for the sake of simplicity let’s assume only three: capital, labor and technology. The aggregated production function would then be

𝑌 = 𝑓(𝐿, 𝐾, 𝐴) = 𝐴𝐾𝛼𝐿1−𝛼 (1)

where Y denotes output, L is labor, K is capital, A stands for technology and α is a constant between 0 and 1. The essential concept related to the production function is the marginal product. Particularly of interest is the marginal product of labor which tells the change in output resulting from employing an additional worker, everything else being constant. Typically it is assumed that the marginal product of labor follows the law of diminishing returns and therefore declines as more labor is added. Coming back to this later, let’s further make two more assumptions. The first is that the market for goods is somewhat characterized by perfect competition. Even if the assumption is not that realistic, it is commonly used when making theoretical derivations (Grossman, J. B., 1982, p597). Perhaps because the

consequence of disturbances in the market has the same direction of change as the more realistic assumption of monopolistic competition, but differs in magnitude. It is therefore not so important whether a monopolistic or competitive approach is used. The second assumption is that the firms are somewhat trying to maximize their profits. The profits are then given by

𝜋 = 𝑝𝑌 − 𝑤𝐿 − 𝑟𝐾 (2)

where p is the price of output, w is the real wage and r is the rental rate of capital. The fact that perfect competition is assumed, a typical firm cannot influence the price and in order to maximize profits, it has to hire and rent an optimal amount of labor and capital.

However, a distinction needs to be made in relation to the time aspect. In the short run, firms are less able to adjust their capital, hence fixed capital is assumed. The firm’s optimal choice of inputs in the short-run is therefore the amount of labor. To determine the profit maximizing amount of labor firms will need to hire to the point where no more profits can be made. This is achieved when the value of the marginal product of labor equals the marginal cost of labor. For a small firm that is unable to affect the wage, the condition is given by

(10)

6

𝑉𝑀𝑃𝐿 = 𝑃 ∗ 𝑀𝑃𝐿 = 𝑤 (3) that is the short-run labor demand curve.

On an aggregated level, in a competitive labor market, the wage is assumed to be determined according to the equilibrium model as can be seen in Figure 1. Where labor demand has been

derived above, is in equilibrium with labor supply.

Now, what exactly determines the shape of the labor supply curve is debatable as the literature is somewhat uncertain (Murayama, 2010).But according to Borjas (2019), either a perfectly

inelastic or more likely slight upward sloping is a realistic assumption. Because the higher the wage, the more hours people prefer to work.

Anything that alters the wage determining condition explained above will change the wage. Therefore the remaining part of the theory section will focus on three essential concepts that could alter the parameters in the equilibrium model.

2.2. Immigration

The first question that needs to be addressed is, whether immigrants and natives are

substitutes or complements. If they were to be perfect substitutes, they would compete for the same type of jobs and shift the supply curve to the right. In the short-run, when capital is held fixed, this would result in a decrease i006E wages for the natives. If assumed to complement, immigrants would actually raise the value of marginal product of native workers and therefore shift their demand curve out. The effects are illustrated in Figure 2 and Figure 3. An

implication from Keynesian theory on wage rigidity is that this is a gradual procedure and wages will not change immediately.

Figure 1. Equilibrium in a competetive labor market. Source: Coursehero.

(11)

7 Figure 2. An increase in the supply of labor.

Source: (Borjas, 2019)

Borjas (2019) makes a clear distinction of the long term and short term effects when

immigrants and natives are assumed to be perfect substitutes. This is where the assumption of a CES production function, profit maximization and competitive labor market, mentioned in the previous section, becomes crucial. It requires that the return of capital and the wage is given by their respective value of marginal product. Assuming that the price of output is 1 their respective conditions is

𝑟 = 𝛼𝐴𝐾𝛼−1𝐿 (4)

𝑤 = (𝛼 − 1)𝐴(𝐾

𝐿)𝛼 (5) .When immigration increases the supply of labor, the

labor market experience more competition which drives down the wages and more labor is employed. However, the return to capital will increase. This will stimulate an increase in the capital stock and thus increase the wage. If the capital ratio is constant in the long run, so is the wage. This phenomenon is illustrated in Figure 4. As can be seen, the wage initially is brought down due to the supply shock of foreign labor and then brought back as the demand for labor increases with the influence of increased capital investment.

According to Borjas (2019), the assumption of perfect substitution is unrealistic. It would require the skill composition of natives and immigrants to be the same. Data from Statistics of Sweden (2021) tells that in the year 2011, 20% of the foreign-born population in Sweden at

Figure 3. Immigration increases VMP for natives. Source: (Borjas, 2019)

Figure 4. The long-run adjustment of a supply shock. Source: (Borjas, 2019)

(12)

8

ages of 25-64 had not attended high school. An additional 7% had no information about their education. In comparison, the natives that had not attended high school were only 10%. Also, Brücker and Jahn (2008) mention that it is common for high skilled foreign-born workers to work in fields that require no formal education. Their reasoning is that due to barriers of entry, such as language or verifying your education, it is easier and less costly to settle with a low-skilled work. Given what is said, it is arguable to assume that an inflow of immigrants would somewhat disrupt the skill composition in the labor market. Hence, foreign-born workers compete to a greater extent than natives for less skilled work occupations. Now Borjas (2019) explains that there most certainly is a complementary effect between high and unskilled workers, which is described in more detail under the human capital section. A combination of both the substitution and complementary effect should therefore be expected as immigration increases. Depending on how large the differences are among the foreign-born and natives should determine which effect that dominates and hence the final outcome for the native workers.

Borjas (2013) modelling of the long run effects is perhaps a bit ambiguous and suits best as an explanation for a shock. In the case for Sweden, immigration is continuous and has been the major reason for population growth (SCB, 2021). For this reason, the inclusion of The Solow Growth Model could give some insight into how population growth affects the income.

2.3. The Solow Growth Model

To explain Solow’s theory on population growth, a brief summary of Mankiw’s (2016) derivation of the model will be given. The model builds on the CES production function

𝑧𝑌 = 𝐹(𝑧𝐾, 𝑧𝐿) (6) , where z is any positive number. Setting z = 1/L will give

𝑌 𝐿 = 𝐹 (

𝐾

𝐿, 1) (7)

.The equation says that output per worker is a function of capital per worker and a constant 1. Labeling quantities per worker with lowercase letters gives y = K/L and k=K/L, which gives the production function

𝑦 = 𝐹(𝑘, 1) = 𝑓(𝑘) (8)

.The demand for goods and services is assumed to come from consumption and investment. Hence,

(13)

9 𝑦 = 𝑐 + 𝑖

.The model also assumes that each year people save a fraction of s and consume (1-s) of their income. The consumption model could then be modelled

𝑐 = (1 − 𝑠)𝑦

, with a fixed and given savings rate that goes from 0 to 1. To see what this consumption function implies for investment, substitute (1- s)y for c in the demand for goods and services

𝑦 = (1 − 𝑠)𝑦 + 𝑖 (9) and by rearranging terms one gets

𝑖 = 𝑠𝑦 = 𝑠𝑓(𝑘) (10)

The intuition of what has been said so far is that for any given capital stock k, the production function determines how much output the economy produces, and the saving rate s determines the allocation of that output between consumption and investment. Whatever changes the capital stock will also change the income.

To explain changes in the capital stock one could use the following equation ∆𝑘 = 𝑠𝑓(𝑘) − (𝛿 + 𝑛)𝑘 (11)

, where δ is the deprecation rate, and n is the population growth rate. The equation says that the capital increases with new

investments but decreases with deprecation of the current capital stock and the bigger the population the less capital per worker. This is illustrated in Figure 4.

The long run implications are that the economy will reach a steady state level and that steady state will be lower if the economy experience population growth.

(14)

10

2.4. Human capital

It has already been mentioned, foreign-born or not, workers differ in skills. To observe how immigration affects native wages one must therefore take into consideration, factors within the native population, that cause variation in the wage.

One such thing is human capital. Becker (1964, p11) defined human capital as "activities that influence future monetary and psychic income by increasing resources in people" and argued for schooling and on-the-job training to be its main form. Originated from the human capital theory comes “The Wage-Schooling Locus” and “Age-Earning Profiles”. These concepts provide a theoretical foundation for how a workers wage is correlated to years of schooling and age. The insight from these concepts is that the more years of education a worker possesses, the more will the worker earn, and the older the worker is the more experience from on-the-job training he gains and thus a higher wage (Borjas, 2019).

To relate these statements to the production function, one could assume two types of labor, unskilled and skilled. The production function could then look like

𝑌 = 𝑓(𝐿, 𝐾, 𝐴) = 𝐴𝐾𝛼𝐿

𝑠𝑏𝐿𝑢1−𝑎−𝑏 (12) where 𝐿𝑠 is skilled labor and 𝐿𝑢 unskilled labor.

The marginal product of skilled and unskilled labor is 𝑀𝑃𝐿𝑠 = (1 − 𝛼)𝐴𝐾𝛼( 1 𝐿𝑠𝛼∗ 𝐿𝑢1−𝛼−𝑏) (13) 𝑀𝑃𝐿𝑢 = (1 − 𝛼 − 𝑏)𝐴𝐾𝛼(𝐿𝑠1−𝛼 ∗ 1 𝐿𝑢𝛼+𝑏) (14) and if 𝐿𝑠 = 𝐿𝑢, then 𝑉𝑀𝑃𝐿𝑠 > 𝑉𝑀𝑃𝐿𝑢

. These equations say that the value of the marginal product for a skilled labor is higher compared to an unskilled labor. They also suggest a complementary effect between unskilled and skilled labor, because increasing the amount of unskilled labor will increase the marginal product for high skilled labor. This was the kind of effect that was presented for immigrants and natives.

(15)

11

2.5. Unemployment

The supply of labor does not only change due to immigration, another cause could be the fact that the market is in disequilibrium and experience unemployment. It is commonly known that most labor markets in the world experience some level of unemployment. It could be frictional, structural, seasonal or cyclical. So the outcome of unemployment depends mostly on what caused it. For example, theories on, the bargaining of wages, minimum wages and efficiency wages all comply with the marginal productivity theory. That is, when the market is in disequilibrium due to increased wages, employment is reduced. See equation (5) for example. But these factors rarely change from year to year. Instead, it’s the cyclical unemployment whose effect is shown yearly.

Gotffries (2013) explains the following reasoning. When unemployment is high, a lot of workers are looking for a job, making it more difficult to find a job. If unemployment is low, employed workers who search for a new job will more easily find a job. Hence, increasing the job turnover cost for firms. To keep workers from not applying for a new job, the firm will have to pay higher wages. This could be explained with a setting equation. The wage-setting equation is ∆𝑊𝑡𝑑 𝑊𝑡−1= ∆𝑊𝑡 𝑊𝑡−1− 𝑏(𝑢𝑡− 𝑢 𝑛) (15) .Where ∆𝑊𝑡 𝑑

𝑊𝑡−1 is the firms desired wage increase,

∆𝑊𝑡

𝑊𝑡−1 is the average wage increase, b is a

constant and (𝑢𝑡− 𝑢𝑛) is the difference between this year’s unemployment and the natural rate of unemployment. If unemployment is higher than the natural rate, firms will want to raise wages less than the average wage increase.

The theoretical implication is that an increasing unemployment rate has a negative impact on the market wage, but at the same time, an increase in the market wage has a positive impact on the unemployment rate. Therefore, the correlation is indeterminate. However, the empirical evidence seems to indicate a negative relationship from an increase in the unemployment rate (María Gutiérrez-Portilla et al, 2020).

Now there is also a possibility that immigration causes unemployment, especially for the native workers. It has been argued that immigration could potentially decrease the wage and if so, then some native workers might not consider it worthwhile to work at that wage and instead start looking for a new job. This is graphically illustrated in Figure 2. This positive

(16)

12

relationship between unemployment and immigration is exactly what María Gutiérrez-Portilla et al. (2020) found in their regression estimation. According to Bauer and Zimmermann (1999) however, the evidence is somewhat mixed. They mean that factors like labor market institutions and country specific characteristics could influence the outcome.

3. Methodology

The purpose of this section is to clarify the empirical approach that has been taken and how it helps to answer the research question. It will include an explanation of the choice of data, statistical method and how the model has been evaluated. The statistical program that has been used is SPSS.

3.1. Regression analysis

To be able to analyze and quantitatively estimate a theoretical relationship, one should use a regression analysis (Studenmund, 2017). The implications of the literature review and the theory suggest that the wages of natives are affected by a variety of reasons. For example, both unemployment and immigration could affect the wage. Then it was said that immigration could affect unemployment. So, to isolate the effect from immigration, a multivariate

regression model was of interest. Because a multivariate regression model allows for variation in one variable as the others are kept constant (Studenmund, 2017).

A wildly common technique in regression analysis when estimating a model is the Ordinary Least Squares (OLS). The main goal of using OLS is to minimize the sum of the squared residual. The technique is said to be, if all requirements are fulfilled, the best linear unbiased estimator (BLUE) (Studenmund, 2017). To obtain the best possible results, it is sensible to use and comply with the demands of OLS. The author will therefore describe, in the following part of the chapter, how this has been done.

First, recommendations in applied regression analysis given by Studenmund (2017) were used as a guideline, but with a few exceptions. For example, before even deciding upon a model, a review of the available data had to be done. The six steps he recommends are:

1. Review the literature and develop a theoretical model.

2. Specify the model: Select the independent variables and the functional form. 3. Hypothesize the expected signs of the coefficients.

4. Collect the data. Inspect and clean the data. 5. Estimate and evaluate the results.

(17)

13 6. Document the results.

The literature and theory were already reviewed; hence a selection of data and models could be assembled.

3.2. Panel study

The fact that there were problems in finding a sufficient amount of data to be analyzed led to the decision of conducting a panel study. A study that uses panel data combines time-series and cross-sectional observation. When the same variable is measured both cross-sectional and over time a lot of observations are collected. The best way to estimate panel data equations according to Studenmund (2017) is by a fixed effect model or a random effect model, but recommends beginning researchers to use the fixed effects model instead because of the difficulties in applying the random effect model. With regards to his recommendations, a fixed effect model was estimated.

The entities in this study were the 290 Swedish municipalities, which were observed over the years 2011-2019.

3.3. The Fixed Effect Model

Before going into detail about the choice of variables, some words need to be said about the properties of the fixed effect model. A typical fixed effect model could look like following:

𝑌𝑖𝑡 = 𝛼0+ 𝛽1𝑋𝑖𝑡+ 𝛼𝑖𝐸𝐹𝑖 + 𝜌𝑡𝑇𝐹𝑡+ 𝜖𝑖𝑡

Where 𝐸𝐹𝑖 is a dummy variable and represents the entity fixed effects and 𝑇𝐹𝑡 is a dummy variable to control for time fixed effects, 𝛼𝑖and 𝜌𝑡 are constants.

Now Studenmund (2017, pp. 494-495) argues that the major benefit of using such a model is that it absorbs the effects of the omitted variables, even the ones that are not possible to measure. For example, let’s say that there is a different workplace culture in the north of Sweden compared to Stockholm. The work culture of Stockholm could be to work very hard and productive, so in general, everybody earns more. An entity fixed effect would then absorb the effects that the cultural differences have on the dependent variable. An example of a time fixed effect could be the Covid-19 pandemic, which has a considerable impact on the

(18)

14

3.4. Specifying the model

The model was specified with respect to theory and the literature review. Most of the ideas came directly from a similar study by María Gutiérrez-Portilla et al., (2020) as it complied well with the available data as well as the theoretical reasoning presented in this study. The estimated model was

ln 𝐼𝑛𝑐𝑜𝑚𝑒𝑖𝑡 = 𝛼0+ 𝛽1𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡𝑖𝑡+ 𝛽2𝐼𝑚𝑚𝑖𝑔𝑟𝑎𝑡𝑖𝑜𝑛𝑖𝑡+ 𝛽3𝐴𝑔𝑒𝑖𝑡

+ 𝛽4𝐻𝑖𝑔ℎ𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛𝑖𝑡+ 𝛽5(𝐼𝑚𝑚𝑖𝑔𝑟𝑎𝑡𝑖𝑜𝑛 ∗ 𝐻𝑖𝑔ℎ𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛)𝑖𝑡+ 𝜌𝑡𝑌𝑒𝑎𝑟𝑡 + 𝛼𝑖𝑀𝑢𝑛𝑖𝑐𝑖𝑝𝑖𝑎𝑙𝑖𝑡𝑦𝑖+ 𝛼290𝐹𝑒𝑚𝑎𝑙𝑒 + 𝜀𝑖𝑡

, where:

Income – is the average income among the working age natives, adjusted to 2011 prices and measured in thousands of Swedish crowns. The natural logarithm has been used in order for the coefficients to be interpreted as percentage changes.

i – is a subscript for which municipality the regression hold for and ranges from 1-290 t – is a subscript for which year the regression hold for and ranges from 2011-2019 Unemployment – is the share of the workforce that is unemployed, measured in percent. Immigration – is the share of the working age population that is foreign-born, measured in percent.

Age – is the average age among the working age native population.

HighEducation – is the share of the working age population that has a high school education of at least three years or more, measured in percent.

Year– has a dummy variable for every year except for 2011

Municipality – has a dummy variable for every municipality except for Stockholm Female – is a dummy variable that takes the value of 1 if the regression is for a women

When two variables have a combined effect, an interaction term such as a multiple of the two variables could be used (Studenmund, 2017). To absorb a potential complement effect

between high skilled native workers and immigrants an interaction term, in the form of a multiple, was added between Immigration and HighEducation.

By working age, the study refers to those in the age of 16-64.

For convenience, the study defines immigrants similar to its source of data. That is someone who was born abroad (SCB,2021). This is also a common definition among the literature; see for example Nickell and Saleheen (2009) and Borjas (2003).

(19)

15

3.5. Hypothesis of each variables effect

The table shows an overview of the expected signs of the coefficients and is followed by an explanation for each hypothesis.

Table 1. Hypothesis

Explanatory variable Expected sign

Unemployment - Immigration - Age + HighEducation + Interaction term + Female - 3.5.1. Unemployment

It’s been said that firms desire to increase the wage in order to reduce turnover cost is at the most when the unemployment is low, because the low unemployment rate makes it easier for on the job searchers to find a new higher paying job. This would imply a negative

relationship.

3.5.2. Immigration

The insights have been that immigrants and natives could somewhat be viewed as substitutes. When immigration increases there is more competition for the available jobs. More

competition should lead to a decrease in the market wage and thus lowering the average wage of the natives.

The Solow Growth Model predicts that a somewhat constant population’s growth will depress income. Statistics from SCB (2021) tells that immigration is the major reason for Sweden’s population growth so the variable was also thought to capture this negative impact.

3.5.3. Age

From the human capital theory, it was said that the amount of experience increases workers productivity and hence increases the value of the marginal product. A positive relationship should therefore be expected.

3.5.4. High Education

(20)

16

native population possesses a higher education i.e. more years of schooling, the average wage should increase.

3.5.5. Interaction term

The intentions behind the interaction term is to capture the possibility that an increase in immigration has a positive and more considerable effect in those municipalities were the share of highly educated is large or wise verse. This idea is based on the assumption that low skilled labor and high skilled labor are complements. It was argued that immigrants, to a larger extend than natives, compete for low skilled jobs. Therefore, it is possible that the outcome from immigration actually depends on the share of highly educated in the municipality. The interaction term is, therefore, a multiple of the share of foreign-born and the share of highly educated. The effects from an increase in either variable will be determined by the level of the other variable. Another idea is that the greater the skill difference there is between natives and foreigners, the less of a substitute they will be. If one then assumes that the skill difference is greatest in municipalities were the share of highly educated is large, there should then be less negative impact from immigration in those municipalities.

3.6. Data collection

In this section, a brief description of the origin of the data is given. In order to save time and money, all of the data in this study was collected from secondary sources. The data was in some cases computed into means or shares.

3.6.1. The dependent variable

Due to the fact that data on wages were limited, an approximation had to be done. First of all, the mean of aggregated earned income was used instead of wages. Aggregated earned income consists of income from employment and income from business activities. Employment income includes both salary income as well as income from pensions, sickness benefits and other taxable benefits (SCB, 2021).

The second dilemma is that there were no data for the ages 16-19. It was a bit unfortunate because almost all of the variables in the regression contain data for the ages 16-64, but it was not considered to be a major problem.

Another concern was that income would not be as affected by immigration as for example wages, because the theories that have been presented are based on wages and not income.

(21)

17

However, it is reasonable to assume that it has the same impact. After all, if the wages are reduced due to competition from immigration and at the same time more people apply for unemployment benefits and sickness benefits. It is arguable that it would lead to an average decrease of aggregate income, exactly as the impact on wages.

The major benefit of using aggregate income instead of wages was the access to data. It allowed for much more observations, as data on wages only allowed for regional averages. Another benefit was the ability to filter income between native and foreign-born population. This allowed the study to better achieve the aim of the thesis.

The observations of aggregated income were separated for men and women and also adjusted for inflation with the use of the consumer price index from SCB (2021).

3.6.2. The independent variables Unemployment

There are two common sources for unemployment data in Sweden. Either from the labor force survey by SCB (2021) or the enrolled unemployed cover by Arbetsförmedlingen (2021). The major differences are that SCB collects their data from surveys and Arbetsförmedlingen bases their statistics on the persons registered as unemployed in the authority's databases. Another difference is that SCBs broader definition of unemployed results in a higher unemployment rate for youths (Arbetsförmedlingen, 2021). Now that the data for income was missing for the age of 16-19, it was considered better to use Arbetsförmedlingens as it puts less weight on the unemployment rate of the youth. Another reason to use Arbetsförmedlingens data was the ability to filter data into the age range 16-64 instead of SCB’s 15-74.

Immigration & Age

Population statistics was collected from SCB (2021). The data has been filtered by origin, gender and the age group 16-64.

High Education

The level of education among the population was also collected from SCB (2021). The data was filtered by gender and the age group 16-64. Unfortunately, there was no option to filter the education level by origin at a municipality level. As natives are in majority in most municipalities, it was considered to cause no major impact on the average level of education.

3.7. Model evaluation

(22)

18

classical assumptions need to be fulfilled (Studenmund, 2017). This section will therefore present which methods that’s been used in order to evaluate to what extent the classical assumptions are met.

3.7.1. Hypothesis testing

To test the validity of all the expected coefficients, a t-test was run. SPSS automatically conducts a two-tailed test with significance levels of both 5% and 1%.

3.7.2. The classical assumptions

I. According to Studenmund (2017), a way to ensure that the model is correctly specified is to investigate the underlying theory. The theoretical insight was that some linearity is to be expected, but no clear evidence as for perfect linearity. Theories on wages say that wages are rather sticky and take time to adapt to the changes in the market. Therefore, an attempt to lag the explanatory variables was made.

Another recommendation is to check the relationship with a scatterplot, but should be

interpreted with caution due to omitted variable bias (Studenmund, 2017). A model with poor fit such as low R2, insignificant F-score or coefficients could also be an indication of non-linearity and incorrect functional form (Studenmund, 2017). All of these procedures were done.

II. The mean of the error term was computed using excel.

III. To examine collinearity between the explanatory variables, a correlation matrix was used.

IV. Serial correlation is common in time-series data and is tested using a Durbin-Watson or LM test (Studenmund, 2017). However, in the case of panel data, it is a bit more difficult to

I. The regression model is linear, correctly specified, and has an additive

error term.

II. The error term has a zero population mean.

III. All explanatory variables are uncorrelated with the error term.

IV. Observations of the error term are uncorrelated with each other (no serial correlation).

V. The error term has a constant variance (no heteroscedasticity)

VI. No explanatory variable is a perfect linear function of any other explanatory variable(s) (no multicollinearity).

VII. The error term is normally distributed (this assumption is optional but unusually is invoked)

(23)

19

test. It requires the tests to be modified at an advanced level (Born & Breitung, 2016;

Bhargava et al.,1982; Baltagiet al., 2007). As the statistical software SPSS, did not allow for these types of modification a normal Durbin-Watson test was run as the data was sorted by year and not municipality.

V. There are normally 3 ways to test for homoscedasticity; examine the residuals against predicted values, run a Breusch–Pagan test or a White-test (Studenmund, 2017). All of the three procedures were done.

VI. From the theories, one should expect some degree of multicollinearity, for example, between unemployment and immigration. To detect multicollinearity, Studenmund (2017) recommends computing Pearson-Correlation coefficients and conduction a VIF-test and that was what was done.

VII. To check for a normally distributed error term Studenmund (2017) recommends using a

Histogram and check for a bell-shaped form. The graphical technique by Chambers et al., (1983) of a normal probability plot for assessing whether or not a data set is approximately normally distributed was also used.

(24)

20

4. Results

This section begins with a presentation of the descriptive statistics from the collected data. Thereafter a

presentation of the model evolution. The evaluation will primarily focus on the estimated coefficients, the degree of fulfilling the classical assumptions and the remedies that have been taken to correct for any violations or statistical errors. Finally, an end result is given in which the estimated effects are interpreted.

4.1. Descriptive Statistics

Table 2 contains measures of central tendencies and variabilities of the collected data as well the number of observations. The logarithm of income has been replaced with only native income as it is more interpretive, also the interaction term and fixed effects are left out for the same purpose.

Table 2 . Descriptive Statistics

Variables Min Max Mean SD

Average native income (tkr) 207,40 853,34 305,13 57,95

Average age 36,63 44,99 41,11 1,27

Share of highly educated (%) 33,22 81,56 55,09 8,90

Unemployment rate (%) 2,14 17,83 7,72 2,73

Share of foreign-born (%) 3,93 53,93 16,33 7,45

N=5220

The table shows that most of the variables differ quite a bit across the municipalities or time.

To give a general picture of the data, this section also includes a presentation of timeline diagrams displaying averages of the dependent and independent variables computed from all the municipalities. The values in the diagrams are separated for men and women except for unemployment as it has been measured aggregately.

In Figure 7, for both genders, a steady positive trend in income throughout all time periods is observed. Figure 8 reveals that women of working age in the dataset is typically older than men and that there has been a slight increase in the average age over the years, except in 2016. Figure 9 tells that a large share of the population tends to acquire higher education and that the share has increased over the years. A larger fraction of women acquires higher education compared to men. Figure 10 shows a 1,5 percentage point decrease in the

unemployment rate over the nine years. The last and perhaps most important Figure 11 prove that the immigrant share has generally increased at a rapid rate in most municipalities. The immigrant share is greater for women than men, but the dispersion between them is strongly

(25)

21

reduced as the years go by. From what can be seen of the descriptive statistics, there is no indication of errors in the data.

4.2. Evaluation of the model

Specification, coefficient estimates and linearity (I)

The Fixed Effect Model was run with a set of different specification, which all can be seen in Table 3; however, the fixed effects have been moved to the Appendix. All of the specified regressions included the dummy variables, but varied with different sets of non-binary variables. An attempt to lag the independent variables was made, but it resulted in a worse fit for Regression 6 and a decrease of the estimated coefficients and this is not what theory would have suggested. As a consequence, the variables remained unlagged. Disregarding the theoretical implication of a time lag, the 6th regression with the most theoretical support proved to be the one with the best fit and significant coefficients that were in line with their expected signs.

Table 3. Estimations of the Fixed Effect Model.

Dependent variable: Lg(NativeIncome)

Variables Regression 1 Regression 2 Regression 3 Regression 4 Regression 5 Regression 6 Share of foreign-born 0.00081* (0.00032) 0.0012** (0.00036) 0.0018** (0.00034) 0.0021** (0.00035) -0.0022** (0.00061) -0.0024** (0.00061) Unemployment - -0.0016** (0.00059) -0.00051 (0.00056) 0.000802 (0.00057) 0.0011* (0.00057) -0.0049* (0.00186) Share of highly educated - - 0.0057** (0.00028) 0.0057** (0.00028) 0.0051** (0.00029) 0.005** (0.00029) Average Age - - - 0.0065** (0.0015) 0.0081** (0.0015) 0.0083** (0.0015) Interaction term - - - - 0.000092** (0.000011) 0.000096** (0.000011) Unemployment² - - - 0.0002* (0.0000905) Female -0.234** (0.0009) -0.235** (0.0009) -0.290** (0.0028) -0.292** (0.0029) -0.301** (0.0031) -0.301** (0.0031) Constant 5.950** (0.011) 5.951** (0.011) 5.570** (0.021) 5.315** (0.062) 5.245** (0.062) 5.253** (0.062) R² 0.971 0.972 0.974 0.974 0.974 0.974 Adjusted R² 0.970 0.970 0.972 0.972 0.973 0.973 F-score 560.132** 559.066** 605.588** 605.886** 613.049** 611.490** Standard errors within parenthesis

**Significance level at 1%, two-tailed *Significance level at 5%, two-tailed

A squared unemployment term was added in Regression 6 as the scatterplot matrix, found in the Appendix, showed signs of a non-linear relationship. The term seemed to improve the overall fit and was significant at a 5% level. With the inclusion of more explanatory variables,

(26)

22

the equation is assumed to be as linear as possible and cleared of problems caused by omitted variables. Except for the squared term, all explanatory variables were significant at 1% level and therefore all null hypotheses can be rejected. The number of explanatory variables in Regression 6, including all dummy variables, is 305. The degrees of freedom is therefore 5220 – 305 -1 = 4914. Such a high number should free potential problems with hypothesis testing that could be caused by low degrees of freedom

The F-value’s is above 560 for all regressions. Comparing that to the critical value of 1, found in Studenmund (2017) Table B-3, leads to a rejection of the null hypothesis. The equations are statistically significant at the 1-percent level.

The overall fit can be read from the table as well. The regression models are able to explain a large share of the variation of the dependent variable around its mean, with 5 and 6 having the best fit. They are able to fit the data to 97.4%. This is an indication of a very good fit.

Computing the mean of the error term (II)

By summarizing all the predicted residuals in excel, a mean extremely close to 0 was computed.

No severe correlation between the explanatory variables and the error term (III) Table 4. Bivariate Correlation Matrix

Correlation variable: Unstandardized Residual

Unstandardized Residual Average Age Share of highly educated Unemployment Unemployment² Share of foreign-born Pearson Correlation 1 0 0 0 0 0

The bivariate correlation matrix indicates no correlation between the explanatory variables and the residuals.

(27)

23

When running a Durbin-Watson test to check for potential serial correlation of the error term, SPSS computed a Durbin Watson statistic of 1,217 .

Table 5. Durbin-Watson Test

Durbin-Watson Test

Regression Statistics

6 1.217

Using the sample size and the number of explanatory variables, the upper and lower critical value for a 5-percent one-sided test found in Studenmund (2017) table B-4 is dL:1,53 and dH:1,83. As the d-value falls within the rejection region, d<dL, the null hypothesis of no positive serial correlation is rejected. It is, therefore, possible that the classical assumption IV is violated. As mentioned in the methodology chapter, this might not be the best way to test for serial correlation in a fixed effect model based on panel data.

Constant variance of the error term, i.e. no heteroscedasticity (V)

To test for homoscedasticity the standardized residuals were plotted against the standardized predicted value. The resulting Scatterplot can be found in the Appendix. From the Scatterplot, it is possible to detect heteroscedasticity as the variance of the residuals increases for larger predicted values of the dependent variable.

For further analysis, a Breusch-Pagan Test was run, with the results summarized in Table 6. Table 6. Breush-Pagan Test for Heterskedasticity

Breush-Pagan Test for Heterskedasticity

Chi-square Significance

232.385 0.000

The null hypothesis of homoscedasticity was rejected at less than 1% significance level, as can be seen from the table.

A White-test was also conducted, however, SPSS had a problem launching the test and resulted in a software crash. No time was spent in trying to solve the issue as the evidence of heteroscedasticity is clear. To deal with heteroscedasticity, robust standard errors were used. Multicollinearity (VI)

The Scatterplot Matrix, Correlation Matrix and VIF-Test can all be found in the Appendix. A glance at the Scatterplot Matrix seems to indicate a negative correlation between Average Age

(28)

24

and Share of Foreign-born, but Average Age is measured as the average native age and makes therefore no sense. A potential positive correlation between Share of Foreign-born and

Unemployment is quite visible. After all, the matrix is quite difficult to interpret as the plots could be affected by variation in other variables.

Analyzing the Correlation Matrix instead, two conclusions can be made. If the correlations from the multiple explanatory variables are disregarded, there is no problem with severe multicollinearity as none of the Pearson Correlation Coefficients exceeds 0.7. But most variables are correlated at a strength between 0.15 < p < 0.6, which all are significant at a 5% level.

Similar logic is applied to the test. Without the inclusion of composed variables, the VIF-score for all variables were lower than 5 and according to the rule of thumb; there should be no problem with multicollinearity. If all variables are regarded, all non-binary variables except for age show VIF-score well above 10. Either way, no remedies were taken to correct for multicollinearity as it did not lead to sufficiently low t-scores.

Testing for normality in the distribution of the error term (IIV)

Examining the Histogram in the Appendix, a bell-shaped and symmetrical pattern can clearly be seen and thus strengthen the possibility of a normally distributed error term. It is also possible to observe an almost fully linear line in the P-P Plot from the Appendix and combined with the Histogram; draw the conclusion that the error term probably is normally distributed. This finding supports the use of the t-test and F-statistic as they are not truly applicable unless the error term is normally distributed (Studenmund, 2017, p117).

4.3. Final Results

After adjusting for heteroscedasticity with robust standard errors, the regressions suffer from less biasness when estimating the standard errors. A comparison with and without robust standard errors for regression 4-6 is presented in Table 7, where 4.1-6.1 are with robust standard errors. Once again the fixed effects are left out. After the table follows an interpretation of the coefficients from regression 6.1 as it’s the most complete regression equation.

(29)

25 Table 7. Estimations with and without robust standard errors.

Dependent variable: LN(NativeIncome)

Variables Regression 4 Regression 5 Regression 6 Regression 4.1 Regression 5.1 Regression 6.1 Share of foreign-born 0.0021** (0.00035) -0.0022** (0.00061) -0.0024** (0.00061) 0.0021** (0.00047) -0.0022** (0.00067) -0.0024** (0.00066) Unemployment 0.000802 (0.00057) -0.0011* (0.00057) -0.0049** (0.00186) -'0.000802 (0.00065) -0.0011 (0.00065) -0.0049* (0.00224) Share of highly educated 0.0057** (0.00028) 0.0051** (0.00029) 0.005** (0.00029) 0.0057** (0.00034) 0.0051** (0.00035) 0.005** (0.00029) Average Age 0.0065** (0.0015) 0.0081** (0.0015) 0.0083** (0.0015) 0.0065** (0.0017) 0.0081** (0.0017) 0.0083** (0.0016) Interaction term - 0.000092** (0.000011) 0.000096** (0.000011) - 0.000092** (0.000011) 0.000096** (0.000011) Unemployment² - - 0.0002* (0.0000905) - - 0.0002 (0.00011) Female -0.292** (0.0029) -0.301** (0.0031) -0.301** (0.0031) -0.292** (0.00352) -0.301** (0.0035) -0.301** (0.00353) Constant 5.315** (0.062) 5.245** (0.062) 5.253** (0.062) 5.315** (0.071) 5.245** (0.069) 5.253** (0.069) 0.974 0.974 0.974 0.974 0.974 0.974 Adjusted R² 0.972 0.973 0.973 0.972 0.973 0.973 F-score 605.886** 613.049** 611.490** 605.886** 613.049** 611.490**

Standard errors within parenthesis ** Significance level at 1%, two-tailed

* Signficance level at 5%, two-tailed

The parameter estimates with robust standard errors are all, except for squared unemployment, significant at less than 5%.

As for the interpretation, starting with Average Age, a 1-year increase in the average age among the working age natives in the municipality correlates to a 0.83% average income increase for the natives in that municipality.

A 1 percentage point increase in the unemployment rate in a municipality will, on average, decrease the average native income in that municipality by 0.45%.

The interpretation of an increase in either the share of foreign-born or the share of highly educated depends on the value of the other. Firstly, it consists of the main effect from the coefficient that is in front of the increased variable. Secondly, it depends on the effects from the interaction term, which is dependent on the variable that is held constant. A common way is to assume the mean value of the constant variable (Williams, 2015). The interpretation will then be for the average municipality. To illustrate, let’s assume, Y = B1X1 + B2X2+

(30)

26

computations for a 10 percentage unit increase in the share of highly educated, therefore, becomes 𝛽4+ 𝛽5(𝑆ℎ𝑎𝑟𝑒 𝑜𝑓 𝑓𝑜𝑟𝑒𝑖𝑔𝑛𝑏𝑜𝑟𝑛) = 10(100 ∗ 0.005 + 100 ∗ 000096 ∗ 16.33) ≈ 6.57%.

Therefore, the total effect from a 10 percentage point increase in the share of highly educated among the working age in a municipality, will on average, lead to a 6.57% increase in the average income for the natives in that municipality.

A similar computation can be done for a 10 percentage unit increase in the share of foreign-born and is hence 𝛽2+ 𝛽5(𝑆ℎ𝑎𝑟𝑒 𝑜𝑓 ℎ𝑖𝑔ℎ𝑙𝑦 𝑒𝑑𝑢𝑐𝑎𝑡𝑒𝑑) = 10(100 ∗ −0.0024 + 100 ∗ 0.000096 ∗ 55.09 ≈ 2. .89%. The total effect from 10 percentage point increase in the share of foreign-born in a municipality, will on average, lead to a 2.89% increase in the average income for the natives in that municipality.

As for the gender dummy variable, the interpretation is; the average native wage for women in the municipality is approximately 30.1% lower than for men.

5. Discussion

The perhaps most interesting result of this study is the positive net effect on native income from immigration. It is fair to say that the magnitude of the effect could be reasonable, considering that Dustman (2013) estimated a 1-3% increase in average wage from a 10% increase in the share of foreign-born within the population. In his earlier paper Dustman (2005) found a slightly higher increase of between 8.02-9.09%, but which he stated was an overestimation. However, compared to Ottaviano and Peri (2012) who estimated a 0.6% increase in the average wage of natives from a 10% increase in the share of foreign-born, it’s quite a big difference. Perhaps one should not compare the findings too much as they are from different times, countries and are based on different approaches, but at least it gives a sense of reliability. What is interesting though is the cause of these variations.

From a theoretical standpoint, it is arguably the way immigration affects the skill-composition in the country and how well the economy adjusts to these changes. It has been mentioned that native swedes possess, to larger extent than foreigners, a higher degree of education. It has also been said that even highly educated foreigners, due to barriers of entry, does not always find a job that matches their competence. In that sense, they compete less for the same type of jobs and are for that reason less of a substitute to natives. If workers with different

(31)

skill-27

compositions are complements, to a certain degree, then immigration could perhaps also have a positive outcome. This is probably what’s been captured in the estimated regression. Firstly, the coefficient in front of the immigration variable is negative. The explanation could be that, without considering the cross product between the shares of highly educated among the municipality and the share of foreign-born, immigration causes more competition of jobs and hence a substitution effect is captured. It could also capture the negative effects of constant population growth as the Solow Growth Model implies. Sweden has been experiencing a lot of population growth due to immigration, so it is reasonable to assume this. But then comes the interaction term that is positive. It increases with the share of highly educated in the municipality. So a municipality that has a lot of highly educate will experience a higher positive effect from immigration compared to a municipality with a lower share. This makes complete sense, since the complementary effect should be larger where the skill-composition differs the most and the substitution effect should be smaller and smaller, all of this is

potentially captured by the interaction term. However, the equation does not capture the differences between natives and immigrants skills. Instead it has just been assumed that this is the case, due to the aggregate statistical evidence. This aspect would be a good inclusion for any other researcher in Sweden that conducts a similar research.

A potential criticism is the fact that the lagged estimations were not presented as the final results even though theory strongly suggests, that wages are rigid. The argument for not including it was that the fit was worse and that the effect became smaller which shouldn’t be the case. However, if one were to consider the total effects, that is both from the interaction term and the partial effect of immigration, they are actually larger by a few decimals. There is also another critic that could be directed toward the regression and it is the potential of having a positive spurious regression. For instance, even though there is a correlation between immigration and native income, the causal effect could be somewhat different. Borjas (2013) argues that immigrants probably want to settle in high-wage cities with robust labor markets and it would therefore be difficult to detect any negative impact from increased competition as theory suggest (when they are substitutes). Borjas (2013) also mentions that a common way to deal with this problem is to use a lagged measure of immigrants living in the city. It was pointed out that the lagged estimates did not significantly change the estimates, and for that reason a positive spurious regression is less likely. One could also speculate about the immigrants settling pattern in Sweden. A lot of immigrants in Sweden are actually

(32)

28

settle in cities which has capacity to receive them rather than having the option to choose. Refugees who are victims of war have potentially left a lot of their assets and commodities behind and settling in high-wage cities with expensive housing prices could be difficult. Even if these are all speculations, it is fair to say that these suggestions would make a positive spurious regression less likely.

Before drawing any conclusions about the results, it is worth to stressing that there could be potential flaws in the execution of the econometrical method. For example, the test for serial correlation in a fixed effect model should differ from that of a simple regression. If serial correlation is present, as suggested, it would lead to overestimated standard errors and thus an unreliable hypothesis testing. On the bright side, robust standard errors have been used and should weaken the impact of the biasness caused by positive serial correlation.

However, the strength of this study is that the results are significant and has strong support by the presented theory. It also manages to explain a lot of the variation in native income as it has such a high value of fit. The estimates are also based on a lot of data, which should make the estimated equation somewhat closer to the true equation.

A lot of attention has been drawn to the limitation of available data. This limited the study to the extent that not much could be said about how different groups in the society are affected by immigration. Instead, the primary focus was on the aggregate population. The majority of previous studies take on a slightly different approach when they estimate the effect of

immigration. They are more interested in how different percentiles or different skill groups are affected, which gives a detailed picture of immigrations impact on wages. In the future, there might be more data available in Sweden and a strong recommendation for any

researcher that is interested in deepening our understanding of immigrations impact on labor market outcome should make use of such data.

To focus on the aim of the thesis, there will not be any discussion on the effects of the other explanatory variables.

(33)

29

6. Summary & Conclusion

This paper has studied the effect of immigration on the average income of the natives in Sweden. To achieve this aim, the paper took the shape of a panel study and estimated the effects with a fixed effect model. The model was built on aggregate data from 290 municipalities over 2011-2019. The key variables in the estimated equation were average native income, share of foreign-born population, unemployment, average age and share of highly educated among the population. The estimates were then evaluated and robust standard errors had to be used to control for homoscedasticity. As a final finding; the estimated effect from a 10% increase in the share of foreign-born in a municipality led to an increase of 2.89% in the average income of the natives in that municipality.

(34)

30

7. References

Altonji, J. G., & Card, C. (1991). The effects of immigration on the labour market outcomes of less-skilled natives. In J. M. Abowd, &R. B. Freeman (Eds.), Immigration, trade, and the labour market (pp. 208–234). Chicago: University of Chicago Press.

Arbetsförmedlingen. (2021). Två viktiga skillnader. Därför är SCB:s och Arbetsförmedlingens siffror så olika. Retrieved 11 May from:

https://arbetsformedlingen.se/om-oss/press/nyheter/nyhetsarkiv/2020-07-24-darfor-ar-scbs-och-arbetsformedlingens-siffror-sa-olika

Arbetsförmedlingen. (2021). Inskrivna arbetslösa 2008 - 2020. Tidigare statistik. Retrieved 10 April from: https://arbetsformedlingen.se/statistik/sok-statistik/tidigare-statistik

Baltagi, B. H., Seuck, H. S., Jung, B. C., & Koh, W. (2007). Testing for serial correlation, spatial autocorrelation and random effects using panel data. Journal of Econometrics, 140(1), 5

Bauer, T. K., & Zimmermann, K. F. (1999). Report no. 3: Assessment of possible migration pressure and its labour market impact following EU enlargement to central and eastern Europe. St. Louis: Federal Reserve Bank of St Louis.

Becker G: Human Capital. 2nd edition. Columbia University Press, New York; 1964. Bhargava, A., Franzini, L., & Narendranathan, W. (1982). Serial Correlation and the Fixed Effects Model. The Review of Economic Studies, 49(4), 533-549.

Borjas, G. J. (2003). The labor demand curve is downward sloping: Reexamining the impact of immigration on the labor market. Quarterly Journal of Economics, Cxviii(4), 1335-1374. Borjas, G.J. (2019). Labor Economics . McGraw-Hill Education.

Born, B., & Breitung, J. (2016). Testing for serial correlation in fixed-effects panel data models. Econometric Reviews, 35(7), 1290-1316.

Brücker, H., & Jahn, E. J. (2008). Migration and the wage curve: A structural approach to measure the wage and employment effects of migration. St. Louis: Federal Reserve Bank of St Louis

(35)

31

Butcher, K., & Card, D. (1991). Immigration and wages: Evidence from the 1980s. American Economic Review, 81(2), 292–296.

Camarota, S. A. (1997). The effect of immigrants on the earnings of low-skilled native workers: Evidence from the June 1991 current population survey. Social Science Quarterly, 78(2), 417–431.

Card, D. (2001). Immigrant inflows, native outflows, and the local labour market impacts of higher immigration. Journal of Labour Economics, 19(1), 22–64.

Card, D., and T. Lemieux (2001): “Can Falling Supply Explain the Rising Return to College for Younger Men? A Cohort Based Analysis," Quarterly Journal of Economics, CXVI, 705-746.

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, 627–648.

Chambers, J., et al. (1983). Graphical Methods for Data Analysis. Wadsworth.

D’Amuri, F., Ottaviano, G. I. P., & Peri, G. (2010). The labor market impact of immigration in western Germany in the 1990s. European Economic Review, 54(4), 550–570.

De New, J. P., & Zimmermann, K. F. (1994). Native wage impacts of foreign labour: A random effects panel analysis. Journal of Population Economics, 7, 177–192.

Dustmann, C., Fabbri, F., & Preston, I. (2005). The impact of immigration on the British labour market. Economic Journal, 115, F324–F341.

Dustmann, C., Frattini, T., & Preston, I. (2013). The effect of immigration along the distribution of wages. Review of Economic Studies, 80(1), 145–173.

González, L., & Ortega, F. (2011). How do very open economies adjust to large immigration flows? Evidence from Spanish regions. Labour Economics, 18, 57–70.

Gottfries, N. (2013). Macroeconomics . Palgrave Macmillan.

Gutiérrez-Portilla, M., Villaverde, J., Maza, A., & Hierro, M. (2020). A spatial approach to the impact of immigration on wages: Evidence from spain. Regional Studies, 54(4), 505-514.

Figure

Figure 3. Immigration increases  VMP for natives. Source: (Borjas,  2019)
Figure 5. The Impact of Population Growth. Source: (Mankiw, 2016)
Table 1. Hypothesis
Figure 6. The classical assumptions. Source: (Studenmund, 2017)
+3

References

Related documents

Däremot är denna studie endast begränsat till direkta effekter av reformen, det vill säga vi tittar exempelvis inte närmare på andra indirekta effekter för de individer som

The literature suggests that immigrants boost Sweden’s performance in international trade but that Sweden may lose out on some of the positive effects of immigration on

where r i,t − r f ,t is the excess return of the each firm’s stock return over the risk-free inter- est rate, ( r m,t − r f ,t ) is the excess return of the market portfolio, SMB i,t

This thesis argues that, in the present situation, the relative size of immigration from both other EU countries and from countries outside the Union might impact the public

There could also be other possibility of a lower income despite the human capital characteristics are similar between natives and immigrants; the possibility of

Using a Swedish nationwide policy experiment implemented in 2015, he finds a positive and significant impact of immigration on the increased vote share of the

We estimate the elasticity of substitution between workers across experience classes and educational attainment groups; we use these outcomes to calculate the factor

- is it any impact of asylum seekers, as separated and recently growing type of immigrants, on trade in Sweden. This paper is organized by following way. First Background is