“Unraveling the existence of Unemployment
across Dutch Regions”
By Daan Alberdingk Thijm*
JEL-code: J21, J23, J60
Abstract
This paper tries to identify how certain variables differ in their effect on regional unemployment. It is found that the determinants differ substantially in their effect on regional unemployment. Moreover, the coefficients of the determinants estimated by a national model differ substantially from a regional model when neglecting their significance.
Hence, it is suggested with caution that the Netherlands should apply clustered regional unemployment policies instead of a national unemployment policy.
1. Introduction
Reducing the level of unemployment remains to be one of the key problems governments try to resolve. Since the early 70s, economists have been trying to explain how unemployment works on a regional level (Blanchard and Katz, 1992; Layard et al., 1991; Vega and Elhorst, 2012). It has been found that ’’the variation in unemployment rates between regions within countries is even larger than the variation between countries’’ (Vega and Elhorst, 2012). Such a finding suggests that countries should aim to fight unemployment on a regional base instead of implementing a common policy on national level (Zeilstra and Elhorst, 2012).
This paper deals with the explanation of unemployment across regions in the Netherlands. The geographical focus of the Netherlands was chosen as, unlike most countries, it is small in size and people can easily migrate from one region to another.
According to neoclassical economic theory (Todaro, 1969), demand and supply of labour determine the extent of regional migration as the unemployed tend to move to places where there is enough demand for labour. Therefore, one would expect that in a small country, like the Netherlands, regional unemployment differences remain relatively low. This is conventional to the matching theory.
The matching theory describes the formation of matches out of unmatched agents (Mortensen and Pissarides, 1994). In labour economics, it suggests the extent to which new matches are created out of job seekers and the number of vacancies. Intuitively, the matching rate can be higher for a small country like the Netherlands as people can easily shift between regions and migrate to regions where jobs exist. Despite the fact that the overall level of unemployment in the Netherlands is one of the all-time lowest within the OECD (Howell et al. 2007), certain regions remain to have problems in reducing their regional level of unemployment. This can be seen in figure 1. The histogram shows that during the 30 years of the four datasets presented (1981, 1991, 2001 and 2011) unemployment is consistently higher in certain provinces than in others.
For instance, the northern province of Groningen always faces a higher percentage of unemployed (5.5% < Unemployment < 9.0%) than the southern provinces of Zeeland (Unemployment < 5.0%) or Zuid-Holland (Unemployment < 6.0%). This makes it interesting to investigate the regional level of unemployment in the Netherlands and how it differs from the national level of unemployment.
Figure 1: Percentage of unemployed across Provinces from 1981-2011
Various macroeconomic studies use different explanatory variables and models to identify the causes of unemployment on a country-base level. Many studies mentioned migration, labour participation, commuting (to travel regularly over some distance to work), wage level, the level of employment and the industry mix as possible factors that influence unemployment.
This paper tries to identify the possible factors that determine region specific unemployment across forty regions in the Netherlands. In addition, the paper tries to explain to what extent these factors differ from those determining national unemployment. This is done by investigating annual data from 1995 to 2014 for every COROP region. The outcomes are compared and evaluated to seek a decisive conclusion whether the government should pursue national policies for fighting unemployment or should start adopting regional clustered policies.
The results indicate that certain explanatory variables suggested by empirical and theoretical literature have an effect on the determination of regional unemployment. Moreover, the paper contributes to the literature as it suggests that the effect of regional unemployment determinants differs across regions, but also from the national trend in the Netherlands. Hence, regional clustered policies seem to be better suited than national policies to fight unemployment in the Netherlands. But, caution should be given to this suggestion due to the low amount of observations which results in many insignificant results.
The paper is structured as follows. The second section is composed out of the theoretical and empirical framework. In the theoretical framework, the types of unemployment are explained and their theoretical causes. Moreover, it also identifies theoretical models for regional unemployment. The empirical literature identifies what has been found in previous papers as causes of regional unemployment. This provides the investigation with possible factors that could explain regional unemployment. These can be incorporated into the model.
Section three consists of the methodology and data. In this section, the sample and the model are explained. In addition, section three gives an overview of the data and how certain variables are measured. In section four, the results are presented, while section five provides limitations. Finally, section six concludes.
2.1 Theoretical Framework
2.1.1 Types of unemployment
In order to understand what might cause unemployment in theory, familiarity with the different concepts of unemployment is essential.
According to Keynes (1936), there exist three types of unemployment; cyclical, structural and frictional unemployment. In the 1960s Milton Friedman and Edmund Phelps came up with a fourth type of unemployment, namely the natural rate of unemployment. This section demonstrates all four types of unemployment, explains the difference between frictional unemployment and the natural rate of unemployment, and clarifies how the Beveridge Curve can serve as a medium through which unemployment can be illustrated.
Cyclical Unemployment, or demand-deficient unemployment, is explained by the business cycle. When a country’s economy experiences growth, labour supply increases (more jobs are created) and unemployment falls (keeping other factors constant) as job searchers can find a job more easily. Vice versa, in a recession, firms lay off workers and unemployment increases. This is known as cyclical unemployment (Keynes, 1936). Structural Unemployment is a more permanent level of unemployment that isn’t caused by an effect of the business cycle. This type of unemployment namely consists of a mismatch between job seeker and job vacancies. Technological advances and trade agreements intuitively increase this mismatch. Technological progress can make human labor unnecessary, while trade agreements cause factories to reallocate to other countries,
lowering the initial amount of job vacancies (Kennedy, 2000). Possible causes of structural unemployment are; differences in the level of education, density of population, technological progress, globalization, GDP and employment growth.
Frictional Unemployment indicates the segment of unemployed that arises from labor market turnovers. It basically involves people that are in-between jobs or actively seeking for a job. The jobs are available, but the worker and the job haven’t been matched yet. Hence, it is a short-term type of unemployment (Reder, 1969). In addition, frictional unemployment could find its origin in; differences in the level of education, the extent to which unemployed have migration possibilities to other regions, the height wage level (higher wages, make it more easily for job searchers to accept a job) and the extent of commuting (jobs that are less far away from their house are more easily accepted).
The Natural Rate of Unemployment is the final type of unemployment. Milton Friedman defined the natural rate of unemployment as;
‘‘The level that would be ground out by the Walrasian system of general equilibrium equations, provided they enclosed the actual structural characteristics of the labor and commodity markets, including market imperfections, stochastic variability in demands and supplies, the cost of gathering information about job vacancies and labor availabilities, the costs of mobility, and so on (Friedman, 1968).’’
It primarily says that the natural rate is the level of unemployment that occurs due to real economic forces. It is the rate of unemployment that a country always experiences. It is the level of unemployment without the disturbance of any temporary frictions.
Both frictional unemployment and the natural rate of unemployment are caused by changing economic forces. The difference between the natural rate of unemployment and frictional unemployment is that the natural rate of unemployment is considered to be long-term, whereas the level of frictional unemployment is short-term as people are in-between jobs and quickly find a new job.
Unemployment can be illustrated by the Beveridge curve. This curve depicts an inverse relationship between the unemployment rate and the vacancy rate (Beveridge, 1944) and is shown in figure 2. A movement along the curve indicates a change in cyclical unemployment; an increase in the vacancy rate, initiates a fall in the level of unemployment. Hence,
unemployment is directly influenced by changes in GDP and variables that influence GDP. In
terms of regional unemployment, it is called GRP. Changes in structural unemployment shift the entire Beveridge curve. Long-term structural unemployment shifts the curve outward from the origin (Beveridge, 1944).
2.1.2 Theoretical models for regional unemployment
This paragraph describes the models applied by previous papers to explain regional unemployment. Moreover, the assumptions of these models are shortly discussed. In addition, on every model a small conclusion is drawn whether it is sufficient in explaining regional unemployment.
The NAIRU model (Blackley, 1989) explains regional unemployment through the use of the wage-setting curve (Philips curve). When applying this model, one uses region-specific wage-setting curves to obtain region-specific non-accelerating inflation rates of unemployment (NAIRUs). In other words, the dependent variable is the unemployment rate while maintaining stable inflation. In order to estimate a region specific NAIRU an assumption is needed. This assumption is that there exists variation between regions in the wage level / wage determination. These regional differences in wage could be explained by regional differences in for instance political entities (Johnes and Hyclak, 1989). Such a way of modelling regional unemployment is questionable in the Netherlands as our regional economies are open and follow one national entity. Moreover, the Netherlands easily allows for migration and commuting and so variations across regions in the wage level can easily be overcome. Therefore such a way of modelling won’t be applied. However, understanding their line of reasoning is essential in forming an own model.
Figure 2: the Beveridge curve
The Blanchard and Katz model (1992) incorporates more regional aspects that NAIRU. According to this model, regions generate various bundles of goods, which are sold on the national market. In addition, it allows firms and labor to be portable beyond regions. The theoretical model applied by Blanchard and Katz to explain regional unemployment is shown below:
𝑤𝑤𝑖𝑖𝑖𝑖 = −𝑑𝑑(𝑛𝑛𝑖𝑖𝑖𝑖∗) + 𝑧𝑧𝑖𝑖𝑖𝑖; (1)
𝑐𝑐𝑤𝑤𝑖𝑖𝑖𝑖 = −𝑢𝑢𝑖𝑖𝑖𝑖; (2)
𝑛𝑛𝑖𝑖,𝑖𝑖+1∗ − 𝑛𝑛𝑖𝑖𝑖𝑖∗ = 𝑏𝑏𝑤𝑤𝑖𝑖𝑖𝑖+ 𝑥𝑥𝑠𝑠𝑖𝑖+ 𝜀𝜀𝑖𝑖,𝑖𝑖+1𝑠𝑠 ; (3)
𝑧𝑧𝑖𝑖,𝑖𝑖+1− 𝑧𝑧𝑖𝑖𝑖𝑖 = −𝑎𝑎𝑤𝑤𝑖𝑖𝑖𝑖+ 𝑥𝑥𝑑𝑑𝑖𝑖+ 𝜀𝜀𝑖𝑖,𝑖𝑖+1𝑑𝑑 ; (4)
In this model all variables are measured in logarithms and calculated are measured against their national equivalent.
Hence, The variable 𝑛𝑛𝑖𝑖𝑖𝑖∗ stands for the relative employment in region i at time t
divided by the relative level of employment in the whole country. In addition, 𝑧𝑧𝑖𝑖𝑖𝑖 is the
long-run effect of labour demand. Moreover, 𝑢𝑢𝑖𝑖𝑖𝑖 is the unemployment rate in the same region at
time t. The first equation is known as the short-run labour demand relation. The second equation identifies a negative relation between unemployment and wages; higher unemployment leads to lower wages, following economic theory. The third equality represents the labour supply, in which 𝑥𝑥𝑠𝑠𝑖𝑖 is a positive shock in labour
supply. The equation shows that labor mobility is dependent on relative wages and relative unemployment. The final identity explains the change in the long-run effect of labour demand in a region. According to the equation, the wage rate has a negative effect on labour demand within a region. In addition, 𝑥𝑥𝑑𝑑𝑖𝑖 represents a positive shock in relative labor
demand. This leads to a positive relative effect on employment and higher than average wage.
The main target of the Blanchard and Katz model was to explain regional unemployment through the migration of firms and people. Due to the focus on migration of firms and people, multiple other factors, possibly affecting regional unemployment, weren’t incorporated into the model. In forming a sufficient model explaining regional
unemployment, more factors possibly explaining regional unemployment should be incorporated. In addition, Blanchard & Katz mainly identify the indirect effects of their variables on regional unemployment through 𝑛𝑛𝑖𝑖𝑖𝑖∗ and 𝑧𝑧𝑖𝑖𝑖𝑖. This gives an indecisive conclusion
regarding the effect of the variables besides 𝑛𝑛𝑖𝑖𝑖𝑖∗ and 𝑧𝑧𝑖𝑖𝑖𝑖. Hence, it can be stated that the
Blanchard and Katz model is insufficient for identifying regional unemployment as it lacks the direct effects of a large amount of possible explanatory variables.
The accounting identity model is an unemployment models that can be used for all
geographical labour markets; local, urban and regional (Elhorst, 2000). The identity is shown below:
𝑈𝑈𝑙𝑙= 𝑊𝑊𝑝𝑝∗ 𝐿𝐿 + 𝐼𝐼𝐼𝐼𝑛𝑛 − 𝐸𝐸 (5)
𝛥𝛥𝑊𝑊𝑝𝑝 = 𝐺𝐺 + 𝐼𝐼𝐼𝐼𝑛𝑛 (6)
In equation (5) the variable 𝑈𝑈𝑙𝑙 represents the level of unemployment, 𝑊𝑊𝑝𝑝is the working age
population, 𝐿𝐿 is the participation rate of the labour force, 𝐼𝐼𝐼𝐼𝑛𝑛 is the rate of net inward
commuting and E I the level of people who are employed. In equation (6), 𝐺𝐺 indicates the balance of new entrants and departures into the working age population in the specific
region and 𝐼𝐼𝐼𝐼𝑛𝑛 is the net inward migration. Moreover, 𝑊𝑊𝑝𝑝∗ 𝐿𝐿 measures labour supply,
thereby taking into account changes in migration and commuting, whereas the last variable incorporates labour demand into the model (Elhorst, 2000). This model only incorporates a few influential factors determining unemployment. Therefore, one should extend this model and incorporate multiple causes that explain regional changes in unemployment.
As is justified in this section, a different type of modelling is needed to explain regional unemployment. This is further explained in the methodology and data section. 2.2 Empirical Literature
This paragraph identifies previous empirical work on the factors determining regional unemployment. It gives an overview which variables cause regional unemployment.
The effect of the Labour Participation Rate is shown to be different than theoretically expected; Partridge and Rickman (1995) found that the male participation rate has a negative significant effect on the unemployment rate. This isn’t in line with the expectancy of the accounting identity, an increase in the labour participation rate is expected to
increase unemployment. But, the claim of Partridge and Rickman (1995) has the support of others (Sieger, 1983; Fleisher and Rhodes, 1976; Van der Veen and Evers, 1983). Also, many studies reported the same result for the female participation rate; a negative significant effect on unemployment (Malizia and Ke, 1993; Hofler and Murphy, 1989; Holzer, 1993). An increase in female labour participation would decrease the level of unemployment. This could be due to reverse causality as a lower level of unemployment could encourage an increase in the labor participation rate. This is known as the “encouraged worker” effect.
According to economic theory, Employment growth has a negative effect on unemployment. Most of the time, empirical research shows the effect of employment growth on unemployment to be negative and significant (Hyclak, 1996; Summers, 1986; Fleisher and Rhodes, 1976). However, one should bear in mind that the effect of employment growth doesn’t need to be negative following economic theory. For instance, if unemployment benefits are large, people are easily tempted to remain unemployed instead of taking on a job. This is one of the possible reasons why higher employment growth doesn’t necessarily cause the unemployment rate to decrease. Furthermore, Employment growth captures the effect of the Gross Regional Product on regional unemployment. This is due to the direct relation between GRP and Employment growth.
Following Layard et al. (1991) and, Vedder and Gallaway (1996), a region could experience an increase in unemployment when the Natural Population Growth Rate is larger than its employment growth rate. Over time, less people obtain a job as there are fewer jobs available for more people. The variables that explain the natural population growth rate are the birth rate and death rate in the region. Studies that investigated the relation between the birth/death rate and the unemployment rate, found that a region with a relatively young population faces more problems in finding work (Johnson and Kneebone, 1991; Elhorst, 1995).
Following the accounting identity, net inward commuting causes regional unemployment to rise. Due to the increasing level of welfare, people are capable to leave urban areas (Simpson, 1987). This change in welfare has realized a new trend; high-skilled worker live most of the time outside the city and travel to work on a daily basis. Low-skilled workers stay in the urbanized area (Hanson and Pratt, 1988). According to empirical research, this leads to higher unemployment rates close to cities due to a higher inflow of commuters (Burridge and Gordon, 1981; Van der Veen and Evers, 1983).
According to economic theory, unemployment is considered to have a negative effect on the Wage Level. Following the wage curve, a lower level of job searching unemployed results in higher wages. The contrary holds when the level of job searching unemployed is considered to be high. The relation between the Wage Level and the rate of Unemployment is subject to reverse causality. Multiple studies have done research on the relationship between the Wage Level and the rate of unemployment and found similar evidence as economic theory. Their results show that the relation needs to be considered negative and significant (Blackaby and Manning, 1992; Gripaios and Wiseman, 1996; Murphy, 1985). In some cases the relation was found to be positive, but the results remained insignificant (Molho, 1995). Blanchflower and Oswald (1994) extended the research on the wage curve by incorporating the bargaining position of the employee. According to economic theory, a higher level of unemployment raises the level of wages. However, Blanchflower and Oswald (1994) found this relation to be negative; a higher level of unemployment lowers the level of wages. They considered this to be the effect of an employees bargaining position; as unemployment rises, many job-searching unemployed are willing to take on a job against a relative lower level of pay. Hence, the exact effect of the wage curve on unemployment remains an empirical question. This empirical question may be answered through this paper as the wage curve might be more important on a regional level, while less important on a national level due to a lower level of vacancies available (the employees have a worse bargaining position).
The Educational Level of the Population in the region has a negative impact on unemployment. A more educated population leads to an increased probability of a match according to the matching function and as a result the rate of unemployment falls. With some concluding the relation to be significant (Simon, 1988; Holzer, 1993; Burridge and Gordon, 1981), others found it to be insignificant ( Blackley, 1989; Murphy, 1985; Hofler and Murphy, 1989).
Net-Migration tends to impact unemployment. According to economic theory, net-migration directly impacts unemployment through the labour supply effect and the productivity effect. The labour supply effect states that as more migrants enter the country, they take on jobs previously held by national inhabitants because they are considered to be cheaper. This would increase national unemployment and in some cases regional unemployment (as certain regions are more accessible to immigrants). The productivity
effect involves high-skilled immigrants. This effect exists when these high-skilled immigrants can make a substantial contribution to economic growth (Borjas, 1994) and hence impact unemployment. These migrants can raise productivity of the national human capital by transferring know-how (Borjas, 1994). Therefore, the effect of the productivity effect on unemployment remains ambiguous as some migrants take national jobs (previously held by national inhabitants), increasing unemployment, while other migrants increase human capital through the transfer of knowledge, decreasing unemployment as people become more suitable for a various type of jobs. Bear in mind that the above is immigration while this paper also incorporates migration between regions. Chalmers and Greenwood (1985) consider the effect of Net-Migration to be an empirical question, as both regional labour supply and demand would increase. Empirical studies on the matter found various results. Bilger et al. (1991) concluded the relation to be negative. But, the results were obtained by an OLS estimates. This yields biased and inconsistent estimates. Chalmers and Greenwood (1985) found net migration to have a positive effect and Van der Veen and Evers (1983) considered it to be negative when using OLS, and significantly positive when applying FML. In the Netherlands, specific industries are sometimes located in limited amount of regions. As a result, one could expect that the inhabitants of a region have a specific skill and have problems adapting when the extent of certain industries decrease or increase in their region. One should take the Industry Mix into consideration when seeking for the factors that determine unemployment as it can influence the level of unemployment. Following Martin (1997), it is expected that the composition of the industry mix is capable of explaining differences in labor demand and regional unemployment across regions.
All the variables mentioned above are investigated whether they fit into the model that evaluates to what extent the variables have an impact on regional unemployment and on national unemployment. This is done as previous empirical literature found these variables to have a certain effect on the regional level of unemployment. Therefore none of the variables explained in the empirical literature section can be left out as it might cause an omitted variable bias.
3. Methodology
This section is divided in three parts. First, it starts with illustrating the sample that is used to explain the differences in the factors explaining regional unemployment and
national unemployment. Second, a summary of the data is supplied. Third, the variables are investigated by checking for possible high levels of correlation. Fourth, the model is provided, demonstrating how the effect on regional unemployment is examined. This section incorporates which estimation method is preferred: pooled OLS, fixed effects or random effects. Finally, the descriptive statistics are supplied.
3.1 Sample
In this study, all forty COROP regions in the Netherland are used. The regions are shown in figure 3. These COROP regions are compared across each other and compared with the national level effects that the variables might have on unemployment. This is done in order to differentiate between the effect of regional determinants and national determinants. The time period considered is 1995-2014. This is done on an annual base. This time period is selected as it takes into account all economic situations the Netherlands has been in: an improving economy and a financial downturn. The economic condition of the country could possibly have influenced the effect of certain factors on unemployment. 3.2 Data
As explanatory variables, the model incorporates factors that had a significant impact on regional unemployment in previous empirical and theoretical literature. These variables were explained in section 2.2. Moreover the variables used are briefly discussed below in order to understand how they were measured.
The Industry Mix is based on the added value of the following industries within every
region; agriculture, health sector, minerals, industry, construction, financial services,
construction, traffic and communication, and trade and hospitality. The industry mix is illustrated through entropy and is calculated by the following measure;
𝑆𝑆𝑖𝑖𝑖𝑖 =𝑙𝑙𝑙𝑙𝑖𝑖𝑖𝑖
𝑖𝑖 (7)
𝐸𝐸𝑛𝑛𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝑖𝑖 = − ∑𝑖𝑖=1𝑖𝑖 𝑆𝑆𝑖𝑖𝑖𝑖 ∗ log (𝑆𝑆𝑖𝑖𝑖𝑖) (8)
into a region specific entropy value that can be used for equations (9) and (10) to examine the effect the industry mix has in a specific region.
Figure 3: COROP regions in the Netherlands
The Education variable is split in three types of education levels; high (representing hbo, a Dutch education level, and university), medium (representing the education level where people attended the last three years of their secondary havo/vwo education or obtained a mbo 2, 3 or 4 diploma, this is also a Dutch education level) and low (representing those who finished their vmbo, or attended the first three years of their secondary education, havo/vwo). This is done in order to investigate the effect different educational
groups have on a regional unemployment. Intuitively, it could be expected that a certain changing educational group has a larger impact on unemployment than others. For instance, when the region is characterized by low-skilled labour, a lower level of education is demanded as more high skilled unemployed don’t necessarily take on these jobs.
Due to a limited amount of data on Net-inward Commuting, Spatial Mobility is used as an explanatory variable. The Spatial Mobility variable signifies the average distance between working place and housing residence. It is expected that traveling over distance becomes easier as time progresses. This results in unemployment to fall as people can easier take on jobs over longer distances.
Emp reflects the yearly employment growth in the COROP. Additionally, migr1 represents the regional migration surplus divided by the population in the COROP multiplied by a thousand. Wage resembles the real average wage of an employee divided by a thousand within the COROP. 𝐼𝐼𝑛𝑛𝑑𝑑𝑢𝑢𝐼𝐼𝐸𝐸𝐸𝐸𝐸𝐸 represents entropy measured by equation (8). Higheduc signifies the amount of workers that have hbo or university as their educational level within the COROP divided by the population in the COROP. Lpr is the amount of those participating in the labour market divided by the population in the COROP. Finally, pop signifies the population growth rate.
3.3 Variables
The empirical and theoretical literature describes many variables that impact the level of unemployment. To obtain a sufficient amount of explanatory variables that are significant, empirical and theoretical literature is followed. Moreover, if two explanatory variables are highly correlated, the variable that has the largest effect on regional unemployment, following economic theory, is added while the other one is dropped out of the regression. This leaves the most important variables that determine regional unemployment for investigating.
The correlation matrix is shown in Appendix A. Appendix A indicates that there exists high correlation between multiple variables. In order to avoid multicollinearity some variables are dropped. 𝐸𝐸𝐸𝐸𝐸𝐸 and 𝑚𝑚𝑚𝑚𝑚𝑚𝐸𝐸 are highly correlated. This can be due to the natural increase in the population (due to birth), which might explain part of the surplus within every COROP. As a result, 𝐸𝐸𝐸𝐸𝐸𝐸 is dropped. The expectation is that 𝑚𝑚𝑚𝑚𝑚𝑚𝐸𝐸 is more important in determining regional unemployment due to the matching theory as explained in the
introduction. Consequently, 𝑙𝑙𝐸𝐸𝐸𝐸 has a high correlation with 𝑤𝑤𝑎𝑎𝑚𝑚𝑤𝑤 and ℎ𝑚𝑚𝑚𝑚ℎ𝑤𝑤𝑑𝑑𝑢𝑢𝑐𝑐. This could be due to reverse causality. As a result, 𝑙𝑙𝐸𝐸𝐸𝐸 is dropped, as the expectancy is that the other two explanatory variables are more important in their effect on regional unemployment.
3.4 Model
To examine the effect of the variables discussed above on regional unemployment, the following models are estimated:
𝑈𝑈𝑖𝑖𝑖𝑖 = 𝛾𝛾𝑖𝑖+ 𝜕𝜕𝑖𝑖 + 𝛽𝛽1∗ 𝑈𝑈𝑛𝑛𝑤𝑤𝑚𝑚 𝑑𝑑𝑤𝑤𝐸𝐸𝑤𝑤𝐸𝐸𝑚𝑚𝑚𝑚𝑛𝑛𝑎𝑎𝑛𝑛𝐸𝐸𝐼𝐼𝑖𝑖,𝑖𝑖+ 𝜖𝜖𝑖𝑖𝑖𝑖 (9)
𝑈𝑈 𝑖𝑖𝑖𝑖 = ∑40𝑘𝑘=1𝛽𝛽𝑘𝑘∗ 𝐼𝐼𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑘𝑘+ ∑𝑗𝑗=140 𝛽𝛽𝑗𝑗 ∗ 𝐼𝐼𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑗𝑗∗ 𝑈𝑈𝑛𝑛𝑤𝑤𝑚𝑚 𝑑𝑑𝑤𝑤𝐸𝐸𝑤𝑤𝐸𝐸𝑚𝑚𝑚𝑚𝑛𝑛𝑎𝑎𝑛𝑛𝐸𝐸𝐼𝐼𝑖𝑖,𝑖𝑖+ 𝜕𝜕𝑖𝑖+ 𝜖𝜖𝑖𝑖𝑖𝑖 (10)
𝑈𝑈𝑖𝑖𝑖𝑖 is the regional unemployment rate for COROP i at time t. Moreover,
𝑈𝑈𝑛𝑛𝑤𝑤𝑚𝑚 𝑑𝑑𝑤𝑤𝐸𝐸𝑤𝑤𝐸𝐸𝑚𝑚𝑚𝑚𝑛𝑛𝑎𝑎𝑛𝑛𝐸𝐸𝐼𝐼𝑖𝑖,𝑖𝑖 is a vector of all explanatory variables of COROP i at time t which in
this case are; 𝑚𝑚𝑚𝑚𝑚𝑚𝐸𝐸, 𝑛𝑛𝑚𝑚𝑐𝑐, 𝑤𝑤𝑎𝑎𝑚𝑚𝑤𝑤, 𝑚𝑚𝑛𝑛𝑑𝑑𝑢𝑢𝐼𝐼𝐸𝐸𝐸𝐸𝐸𝐸, ℎ𝑚𝑚𝑚𝑚ℎ𝑤𝑤𝑑𝑑𝑢𝑢𝑐𝑐 and 𝑤𝑤𝑚𝑚𝐸𝐸. In equation (9), 𝛽𝛽1 can be
interpreted as the marginal impact of the unemployment determinants of interest of COROP i at time t on the unemployment rate of region i at time t. Equation (9) is the region-fixed panel model.
In specification (10), the model includes interaction dummies for every COROP with
the main explanatory variable of interest. 𝛽𝛽𝑗𝑗 captures the region specific impact of
unemployment determinants thereby measuring the regional specific trends in
unemployment. 𝛽𝛽𝑗𝑗, estimated in equation (10), is compared to 𝛽𝛽1, estimated in equation (9),
to analyze whether regional specific trends in unemployment differ from the estimated
national trends with respect to the unemployment determinants. 𝜀𝜀𝑖𝑖𝑖𝑖 represents the error
term and the indices denote the specific region of interest, i and the year, t. 𝛾𝛾𝑖𝑖 represents
the fixed effects thereby controlling for time-invariant unobservables. 𝜕𝜕𝑖𝑖 is the time fixed effects which controls for temporal variation in the dependent variable.
estimator, is rejected. In that case, fixed effects estimation is preferred. If the coefficients estimated by the efficient random effects estimator aren’t the same as the ones estimated by the consistent fixed effects estimator, then the null hypothesis isn’t rejected and a random effects estimator is preferred.
The results generate prob>chi2 = 0.000. Therefore one can reject the null hypothesis at a 1% significance level and fixed effects estimation is preferred. Consequently, in determining whether a fixed effects estimator is preferred to a Pooled-OLS estimator one looks at value of prob > F when running a fixed effects estimation. This is an F-test of the joint significance of its intercepts. If the null hypothesis is not rejected, then the fixed effects intercepts are zero. If it is rejected, a fixed effects method is preferred over Pooled OLS estimation. The estimates indicate prob > F = 0.000. Hence, a fixed effects estimator is preferred to Pooled OLS estimation and is applied in determining the region specific effect per COROP in model specification (9). In model specification (10) the fixed effects are incorporated by including ∑40𝑘𝑘=1𝛽𝛽𝑘𝑘∗ 𝐼𝐼𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑘𝑘.
3.5 Descriptive Statistics
The descriptive statistics for all COROP are shown in the table below. Variable Obs Mean Std. Dev. Min Max
unem 800 5.52 1.88 1.54 14.1 emp 800 1.04 4.92 -10.9 117.7 Nic 800 14.6 3.79 2.44 30.0 Wage 800 25.8 4.09 17.4 40.4 migr1 800 1.04 4.17 -13.9 34.1 industry 800 2.42 0.07 2.22 4.4 higheduc 800 0.12 0.04 0.036 0.25
Table 1: Descriptive Statistics
All variables are measured in levels except for 𝑤𝑤𝑚𝑚𝐸𝐸. Thereby I follow the literature that suggests incorporating employment growth in determining its effect on unemployment. 𝐸𝐸𝑚𝑚𝐸𝐸 exhibits an odd value for its maximum, namely 117.7%. This could be due a possible new establishment of large firms in 2014 in COROP 25.
Furthermore, by applying unit root tests for all variables one can determine whether the variables are stationary or non-stationary. If a unit root is present, the variable is considered to be non-stationary. One can conclude from the table 2 that all variables should be measured in terms of levels, except for wage, as the null hypothesis that a unit root is present, and hence being non-stationary, can be rejected at a 1% significance level. The presence of a unit root in 𝑤𝑤𝑎𝑎𝑚𝑚𝑤𝑤 might resemble the fact that there exists a trend in 𝑤𝑤𝑎𝑎𝑚𝑚𝑤𝑤. By incorporating time-fixed effects the presence of a unit root is of no further issue for the remainder of this paper.
Emp Nic Wage Migr Industry higheduc Residual
Unit root
present No No Yes No No No No
Table 2: Presence of a Unit Root 4. Results
Section four is divided into three parts. First, an estimation of equation (9) is estimated (section 4.1) and analyzed. Second, equation (10) is estimated (section 4.2) and analyzed. The results are illustrated by means of a map of the Netherlands In which a green region represents the explanatory variable of interest to have a positive significant effect on regional unemployment and a red region to have a negative significant effect on regional unemployment. A grey COROP region indicates the coefficient to be insignificant. Finally, section 4.3, compares the estimates of section 4.1 and 4.2 and analyzes whether the coefficients of the determinants on a regional level differ substantially from the national level.
4.1 Evaluation of entire Panel
The results of equation (9) are presented in table 2 below. The output indicates the national trend of the explanatory variables that determines regional unemployment.
Moreover, the output illustrates the results for random effects, fixed effects and OLS estimation. In order to make the estimations in table 2 clear, stars are applied to indicate their significance. The following is used: one star indicates the variable and its coefficient to be significant at a 10% level, two stars indicates it to be significant at a 5% level and three stars resembles the variable to be significant at a 1% level. As was indicated by the
2011.YEAR -0.264 -1.993*** -5.494*** (0.361) (0.544) (0.849) 2012.YEAR 0.665* -1.058* -4.633*** (0.370) (0.563) (0.871) 2013.YEAR 2.344*** 0.796 -2.853*** (0.391) (0.605) (0.924) 2014.YEAR 0.336 -1.709*** -5.795*** (0.400) (0.634) (0.992) Constant -2.583 4.240 -7.186 (1.987) (3.555) (5.338) Observations 800 800 800 R-squared 0.600 0.717 Number of corop 40 40
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Table 3: Determinants of Regional Unemployment (National Trend)
The coefficient of 𝑤𝑤𝑚𝑚𝐸𝐸 isn’t in line with previous research, which showed the relationship to be negative (Hyclak, 1996; Summers, 1986). The coefficient is found to be positive and insignificant. Even though the coefficient is insignificant an explanation for the positive coefficient is supplied.
As mentioned in the empirical literature section, the effect of employment growth doesn’t need to be negative following economic theory. In section 2.2, unemployment benefits were given as a possible explanation for the relation to be non-negative. In the Netherlands there exist high unemployment benefits. In this country, these benefits are considered to be higher than in other countries. Hence, due to the strong institutional framework of the Netherlands the positive sign of the coefficient of 𝑤𝑤𝑚𝑚𝐸𝐸 could originate as people like to remain unemployed instead of applying for a job.
The coefficient of 𝑤𝑤𝑎𝑎𝑚𝑚𝑤𝑤 indicates the effect on regional unemployment to be positive and significant at a 1% significance level. As the wage level rises, the level of regional unemployment increases. A one unit (implies 1000 euro) increase in 𝑤𝑤𝑎𝑎𝑚𝑚𝑤𝑤 is associated with a 0.420 percentage point increase in 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚. As already explained in section 2.2, different results, namely negative, were found by previous literature (Blackaby and Manning, 1992; Gripaos and Wiseman, 1996; Murphy, 1985).
experience an increase in the level of unemployment. Moreover, the income effect might dominate the substitution effect. As wage increases, people are triggered to start searching for jobs and take on jobs as their marginal utility of working has increased relatively to their marginal utility of leisure. As a result, the level of regional unemployment falls.
𝑛𝑛𝑚𝑚𝑐𝑐 is found to have a negative effect on the rate of unemployment. The negative coefficient is significant at a 5% significance level. This means that as the extent of spatial mobility increases, regional unemployment decreases. A one unit increase in 𝑛𝑛𝑚𝑚𝑐𝑐 is associated with a 0.052 percentage point decrease in 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚. This effect might be explained by the fact that people are more capable of taking on a job in other parts of their region. This leads to a better matching of vacancies and those unemployed. On a COROP specific level, the expectancy is that less urbanized regions (e.g. COROP 2 and 5) might experience a positive effect on regional unemployment as the distance from home to work is longer compared to more urbanized region (e.g. COROP 21 and 22). This is further investigated in section 4.2.
𝑚𝑚𝑚𝑚𝑚𝑚𝐸𝐸1 is found to have a negative effect on the rate of regional unemployment. Moreover, the coefficient is significant at a 1% significance level. This negative effect on the rate of regional unemployment means that as the rate of net-migration increases, the rate of regional unemployment tends to fall. The estimated marginal impact of 𝑚𝑚𝑚𝑚𝑚𝑚𝐸𝐸1 on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 implies that a one percent increase in 𝑚𝑚𝑚𝑚𝑚𝑚𝐸𝐸1 is associated with a 0.049% drop in the regional unemployment rate.
Empirical and theoretical literature on the effect of net-migration remained indecisive due to possible labour supply and productivity effects. Moreover, the estimation results are in line with previous literature like van der Veen and Evers (1985) and Bilger et al. (1991). The results in this paper contribute to the empirical literature as the negative effect of net-migration on the rate of regional unemployment is found by applying a fixed effects estimator while Bilger et al. (1991) and van der Veen and Evers (1985) both obtained their results from an OLS estimator, which yields biased and inconsistent estimates. Moreover, Pissarides et al. (1990) found that low-unemployment regions mainly experience positive net-migration, while the opposite holds for high-unemployment regions.
Following the estimates of 𝑚𝑚𝑛𝑛𝑑𝑑𝑢𝑢𝐼𝐼𝐸𝐸𝐸𝐸𝐸𝐸, one can conclude that as entropy increases, regional unemployment increases. Moreover, the coefficient is significant at a 10% significance level. An interesting feature of the effect of the industry mix on regional
unemployment is that various regions have specific types of industries located in their region. As a result, inhabitants have specific skills and experience problems in finding a new job when entropy increases. The direct effect of the industry mix can be further explained by looking at a COROP level as it indicates whether certain regions experience more entropy than others. This further explains the possible effects of an increase or decrease in the industry practices on regional unemployment.
As shown in the output of the fixed effects estimator, ℎ𝑚𝑚𝑚𝑚ℎ𝑤𝑤𝑑𝑑𝑢𝑢𝑐𝑐 illustrates a negative effect on regional unemployment. In addition, the coefficient is found to be significant at an 1% significance level. This finding is in line with theoretical and empirical literature; a more educated population leads to an increased probability of a match, following the matching function, and as a result regional unemployment falls (Simon, 1988; Holzer, 1993; Burridge and Gordon, 1981).
The year dummies which were incorporated by applying a time-fixed effects estimation are found the be significant at a 1% significance level except for 1996, which is insignificant. Moreover, 1996 is close to zero indicating that it has no real effect on the intercept when the dummy variable takes on the value of 1. Moreover, the rest of the coefficients of the year dummies are negative and significant, indicating that they lower the intercept when the dummy takes on the value of 1.
The estimation of equation (9) generated what most economic and empirical literature on the determinants of regional unemployment expected. The analysis of section 4.1 gives an incentive to investigate regional unemployment on a smaller level. Hence, in section 4.2 the specific 𝛽𝛽𝑗𝑗’s of the explanatory variables for every COROP are investigated to
examine the specific causes of the level of regional unemployment per COROP. 4.2 Regional differences in the determinants of unemployment
In this section maps of the Netherlands are used in order to illustrate how the explanatory variables affect regional unemployment within the COROP. This section illustrates the results of equation (10) instead of equation (9), which was examined in table 3. Moreover, every explanatory variable is taken apart to look for trends in regional unemployment, while the regression is on the entire sample. The results for every COROP are shown in table 4 and figure 4 to 9.
COROP emp nic wage migr1 Industry higheduc 1 .0766 -.3512 -.1318 -.0596 4.341 22.33 2 .1772*** -.1446 .1543 -.0933** -17.22*** -13.71 3 -.1040 -.1298 -.2298 -.1179* -23.53 -17.56 4 .0331 .3318 -.8298 -.0857 54.38 28.68 5 -.1080* .1761 -.1800 .0915 -28.21* -9.271 6 .0940 .1865 -.5258 -.1110 -3.650 10.08 7 -.0474 -.0060 -.2551 -.1510 26.63 -6.463 8 .0440 -.7024** -.1708 -.0458 41.38*** -3.441 9 -.0156 .8586*** -.8109 .1422 24.97 0.589 10 -.0367 -.1674 -.0175 .1066 -27.39 4.587 11 .0786 .1277 -.5116 -.1776*** 21.95 22.81 12 -.0181 -.0050 -.3324 -.1424 13.03 7.289 13 -.0049 -.0986 -.2892 .0141 16.03 5.459 14 -.0246 -.2303 -.2372 .0878 -8.294 27.59 15 .0144 -.0202 -.1219 -.0907 -30.59 -25.98 16 .0441 -.1233 -.1081 -.1056 9.137 -8.801 17 -.0561 -.3222 -.0439 .2390 -18.52 -0.511 18 -.0588 .0813 -.1666 .0094 -15.26 -35.30 19 .0426 .2176 -.0806 .0240 2.746 -17.58 20 .0304 -.8693 -.3107 -.0275 41.48 -20.01 21 .0607 .4524 -.0853 -.0783 -12.66 -6.362 22 .0032 -.9190 .1768 -.0436 -17.68 -8.824 23 .0197 -.9038 -.0702 .0268 -11.90 -17.58 24 -.0811 -.8347 .1216 -.0565 15.66 -42.59** 25 -.0006 -.8027 -.1240 .1259 24.51 49.33** 26 -.0735 -.0349 .0406 .0191 -17.99 -5.982 27 -.0486 .0743 -.1533 -.0632 28.22 -20.14 28 .0123 -.1361 -.1839 -.0803 -17.43 9.581 29 .0306 .3261 -.3145 -.0234 -21.70 27.93 30 -.1032 -.3260 -.0114 .1485 47.41 -27.93 31 .0722 -.2465* -.0473 .1112 24.49** -9.067 32 .1091 .0290 -.2909 .2091 17.36 -8.392 33 -.0478 -.3863 -.3353 .0130 22.16 28.30 34 .0033 -.1911 .0710 .0197 -5.141 -18.77 35 -.0099 -.4239 .1628 -.0679 -11.95 -31.83 36 -.0713 -.0001 -.1274 .0170 3.633 -23.11 37 -.0730 .0206 -.3179 .0088 -4.001 4.509 38 .1120 .1052 -.6622 .0175 42.97*** -2.043 39 -.1111 -.0982 .0499 -.2756* -31.22 6.808 40 .0299 1.711*** -.4591 -.0102 22.86 23.16 Nr. sign 2 4 0 4 5 2
Table 4: the specific 𝜷𝜷𝒋𝒋 of every explanatory variable per COROP
Nr. Sign represents the number of significant coefficients for every explanatory variable. Moreover, the same interpretation of the starts is applied as in section 4.1
In the left map of figure 4 to 9, a green region signifies the explanatory variable of interest to have a positive significant effect on unemployment within the COROP, while a red region illustrates the determinant of interest to have a negative significant effect on unemployment within the COROP. A grey region signifies the specific 𝛽𝛽𝑗𝑗’s of the explanatory
variables to be insignificant. The fact that almost all specific 𝛽𝛽𝑗𝑗’s of the explanatory variables
in equation (10) are found to be insignificant might be due to the small amount of observations, namely twenty per COROP. More attention to this is paid in the limitations of this paper and in section 4.3. As a consequence, the right map is incorporated which illustrates the effect of all specific 𝛽𝛽𝑗𝑗’s when neglecting their significance.
The regression specific impact of 𝑤𝑤𝑚𝑚𝐸𝐸 on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 is shown in figure 4.
Figure 4: The effect of 𝒆𝒆𝒆𝒆𝒆𝒆 on 𝒖𝒖𝒖𝒖𝒆𝒆𝒆𝒆 within every COROP from 1995-2014.
As shown in the left map of figure 4, the relationship between 𝑤𝑤𝑚𝑚𝐸𝐸 and 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 is found to be insignificant in most COROP regions. Only two COROP specific 𝛽𝛽𝑗𝑗’s for 𝑤𝑤𝑚𝑚𝐸𝐸 are
force is aging and becoming smaller in the Northern market. So when 𝑤𝑤𝑚𝑚𝐸𝐸 increases, 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 also increases due to aging and a labour force that is becoming smaller. The reasoning behind this is that more jobs become available, while the labor force decreases, causing higher regional unemployment. Furthermore, COROP 5 indicates 𝑤𝑤𝑚𝑚𝐸𝐸 to have a negative effect on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚. This follows the theoretical and empirical literature. Additionally, the right map indicates the North and West of the Netherlands to be mainly characterized by a positive effect of 𝑤𝑤𝑚𝑚𝐸𝐸 on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 while for the South and East this relation is negative effect. No further interpretation can be provided due to insignificant results.
The regression specific impact of 𝑚𝑚𝑚𝑚𝑚𝑚𝐸𝐸1 on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 is illustrated in figure 5. As indicated by the left map of figure 5, 𝑚𝑚𝑚𝑚𝑚𝑚𝐸𝐸1 tends to have a negative effect on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 in the COROP where the coefficient is found to be significant. The finding suggest the same negative relation as in most empirical literature (Pissarides et al., 1990; Bilger et al., 1991). As shown by the UWV (Vreeburg et al., 2015) these COROP are characterized by low levels of demand for labor. Additionally, the amount of vacancies is expected to decrease in these COROP according to UWV’s research. As a result, people might be reluctant in moving to these places as they expect that obtaining a jobs is harder within these COROP due to the low level of demand for labor. This might imply reverse causality as the demand of labor effect regional unemployment and vice versa. However, these regions are also characterized by low labour supply (UWV) and therefore unemployment might decrease when people move to these regions. Following the same reasoning, the right figure is especially characterized by a clustering of a negative effect on regional unemployment in the Northern regions. The extent and effect of net-migration might be determined by the labor demand and labor supply within the COROP. This follows Todaro (1969).
The regression specific impact of 𝑤𝑤𝑎𝑎𝑚𝑚𝑤𝑤 on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 is exhibited in Figure 6. As can be seen from the left illustration in figure 6, 𝑤𝑤𝑎𝑎𝑚𝑚𝑤𝑤 has an insignificant effect on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 in all COROP. Therefore a possible explanation for its effect in certain COROP cannot be supplied when taking into account its significance. In addition, the right map indicates the effect of 𝑤𝑤𝑎𝑎𝑚𝑚𝑤𝑤 to be negative in most cases while seven regions are characterized by a positive effect. The positive and negative effect on regional unemployment can be explained by means of the trade-off between the substitution and income effect. More attention is paid to this trade-off in section 4.3.
Figure 5: The effect of 𝒆𝒆𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎 on 𝒖𝒖𝒖𝒖𝒆𝒆𝒆𝒆 within every COROP from 1995-2014
Figure 6: The effect of 𝒘𝒘𝒘𝒘𝒎𝒎𝒆𝒆 on 𝒖𝒖𝒖𝒖𝒆𝒆𝒆𝒆 within every COROP from 1995-2014 The regression specific impact of 𝑛𝑛𝑚𝑚𝑐𝑐 on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 is shown in Figure 7.
Figure 7: The effect of 𝒖𝒖𝒎𝒎𝒏𝒏 on 𝒖𝒖𝒖𝒖𝒆𝒆𝒆𝒆 within every COROP from 1995-2014
As illustrated by the left map in figure 7, 𝑛𝑛𝑚𝑚𝑐𝑐 has two positive and two negative significant effects on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 across the COROP. The positive significant effect in COROP 9 and 40 is explained by the fact that the inhabitant are less mobile than the average of the Netherlands. The right map illustrates the positive effect to be clustered to some Southern and Northern regions. These regions are considered to be less mobile and have a high level of labor supply compared to the Dutch average. As a result, an increase in mobility doesn’t have the same effect on their regional level of unemployment as more urbanized regions. This results spatial mobility to have a positive effect on regional unemployment in certain COROP.
The negative significant effect in COROP 8 and 31 can be explained by people becoming more mobile and as a result they can easier take on jobs in other COROP as the distance between work and home is less of an issue. When interpreting the right map, the negative effect of spatial mobility on unemployment might be illustrated by the fact that these regions are strongly urbanized and therefore the average work-home distance is relatively low which makes it easy to commute to cities for work. This leads unemployment to decrease in these regions as spatial mobility increases. However, further research is needed to conclude on this statement.
The regression specific impact of 𝑚𝑚𝑛𝑛𝑑𝑑𝑢𝑢𝐼𝐼𝐸𝐸𝐸𝐸𝐸𝐸 on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 is shown in Figure 8.
Figure 8: The effect of 𝒎𝒎𝒖𝒖𝒊𝒊𝒖𝒖𝒊𝒊𝒊𝒊𝒎𝒎𝒊𝒊 has on 𝒖𝒖𝒖𝒖𝒆𝒆𝒆𝒆 within every COROP from 1995-2014 The left map indicates that most COROP experience 𝑚𝑚𝑛𝑛𝑑𝑑𝑢𝑢𝐼𝐼𝐸𝐸𝐸𝐸𝐸𝐸 to have an insignificant effect on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚. In COROP 2 and COROP 5, the results provide a significant negative effect; as 𝑚𝑚𝑛𝑛𝑑𝑑𝑢𝑢𝐼𝐼𝐸𝐸𝐸𝐸𝐸𝐸 increases, 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 tends to fall. This resembles the fact that as entropy increases regional unemployment tends to fall. This is in line with Mason (2011), who investigated regional unemployment disparities and the effect of industrial diversity. The reasoning behind his result is that regional diversity is associated with a low level of regional instability (Mason, 2011). The right map indicates what Mason suggest; the Western regions, which are associated with more regional diversity and stability, also experience entropy to have a negative effect on regional unemployment. Caution should be paid to this interpretation as the results in the right map are insignificant.
these COROP. This can explain the positive effect of the industry mix on regional unemployment. The right map indicates the same pattern and shows that the positive effect is often clustered. This is as expected as regions close to one another often have the same core industry.
The regression specific impact of ℎ𝑚𝑚𝑚𝑚ℎ𝑤𝑤𝑑𝑑𝑢𝑢𝑐𝑐 on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 is shown in Figure 9.
Figure 9: The effect of 𝒉𝒉𝒎𝒎𝒎𝒎𝒉𝒉𝒆𝒆𝒊𝒊𝒖𝒖𝒏𝒏 on 𝒖𝒖𝒖𝒖𝒆𝒆𝒆𝒆 within every COROP from 1995-2014 Figure 9 indicates that in COROP 25 an increase in ℎ𝑚𝑚𝑚𝑚ℎ𝑤𝑤𝑑𝑑𝑢𝑢𝑐𝑐 also increases 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚. This concurs with previous theoretical and empirical literature as one would expect that as the amount of highly educated in the COROP increases, regional unemployment falls. This finding is consistent with a trend that experiences great attention in the Netherlands. A big point of discussion in the Dutch government is the fact that it becomes harder for highly educated to obtain a job once graduated. The positions for high-educated are scarce and the competition is fierce. The supply of high-educated for high-skilled jobs is larger than the demand for these positions in the Netherlands. As a result, some highly educated people are taking on jobs below their marginal utility in order to obtain an income (Ponds et al., 2015). The highly-educated start competing for jobs with medium-educated people, who become unemployed as the competition is unfair. These medium-educated people then try to take
on jobs at a lower level of the labour market1. Hence, an increase in the amount of
highly-educated people can lead to a lower overall level of regional unemployment in the Netherlands.
In addition, the right map illustrates this trend to exist in the Northern and Eastern COROP. These COROP are signified by a working force with a lower than average educational level than a substantial part of the rest of the COROP. Therefore, the reasoning above can suit as an explanation for the positive effect of ℎ𝑚𝑚𝑚𝑚ℎ𝑤𝑤𝑑𝑑𝑢𝑢𝑐𝑐 on regional unemployment. However, it needs more research to be a decisive answer. In addition, COROP 24 follows the empirical and theoretical literature as a higher level of ℎ𝑚𝑚𝑚𝑚ℎ𝑤𝑤𝑑𝑑𝑢𝑢𝑐𝑐 decreases 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚.
It is hard to draw a resolute conclusion on the effect of the explanatory variables on regional unemployment due to the limited amount of significant results. However, one trend seems to flow throughout the significant 𝛽𝛽𝑗𝑗’s: characteristics of the COROP play a
considerable role in determining regional unemployment. Hence, for further research it is suggested to include certain demographic trends of the COROP that vary over time and include a larger time-span. This can provide more significant results.
4.3 Regional Unemployment vs. National level of regional Unemployment
As already highlighted at the start of section 4, this section compares the estimates of section 4.1 and 4.2 and shows whether the determinants of specification (9) differ from specification (10).
By adding and subtracting the standard deviation of 𝛽𝛽1 of the explanatory variables,
two values are supplied. These serve as boundaries. If the specific 𝛽𝛽𝑗𝑗’s fall within the
boundaries of 𝛽𝛽1, then the COROP is colored blue and 𝛽𝛽1 = 𝛽𝛽𝑗𝑗. A red region signifies
𝛽𝛽1 > 𝛽𝛽𝑗𝑗. A green region illustrates 𝛽𝛽1 < 𝛽𝛽𝑗𝑗. This indicates how the regional trend differs
from the national trend that determines regional unemployment.
For explanatory convenience, statistical significance of coefficients is not discussed individually due to the large amount of estimated coefficients. Overall, coefficients indicate weak statistical significance. However, the results do not imply that the explanatory variables exhibit no relation with the dependent variable due to a low amount of
1
file:///X:/My%20Downloads/kamerbrief-reactie-op-artikel-uit-de-volkskrant-baan-en-salaris-van-laagbetaalde-steeds-onzekerder.pdf
observations per COROP (twenty observations). Consequently, equation (9) and equation (10) are compared based on their coefficients. Their significance isn’t taken into account.
Moreover, the maps are shown for all explanatory variables, however one should bear in mind that these differ in their significance. Therefore, it is first noted what the
significance level is of the 𝛽𝛽1 ′𝐼𝐼 in equation (9) before analyzing the illustration. In addition,
the boundaries and the significance of all explanatory variables are shown in table 5. These boundaries are used to indicate whether the coefficients of the COROP differ substantially from the national level.
Variable Significance Lower boundary Upper boundary
migr1 1% -0.060 -0.039 wage 1% 0.34 0.50 higheduc 1% -13.24 -6.56 nic 5% -0.077 -0.028 industry 10% 1.35 5.02 emp - 0.0032 0.017
Table 5: Variable Significance and Boundaries
In table 6 the coefficients are shown of all 𝛽𝛽𝑗𝑗 and whether they fall within the
boundaries supplied in table 5. A green coefficient illustrates 𝛽𝛽1 < 𝛽𝛽𝑗𝑗. A red coefficient
shows 𝛽𝛽1 > 𝛽𝛽𝑗𝑗 and if no colour is attached to the coefficient, 𝛽𝛽1 = 𝛽𝛽𝑗𝑗.
COROP emp nic wage migr1 Industry higheduc
8 .0440 -.7024** -.1708 -.0458 41.38*** -3.441 9 -.0156 .8586*** -.8109 .1422 24.97 0.589 10 -.0367 -.1674 -.0175 .1066 -27.39 4.587 11 .0786 .1277 -.5116 -.1776*** 21.95 22.81 12 -.0181 -.0050 -.3324 -.1424 13.03 7.289 13 -.0049 -.0986 -.2892 .0141 16.03 5.459 14 -.0246 -.2303 -.2372 .0878 -8.294 27.59 15 .0144 -.0202 -.1219 -.0907 -30.59 -25.98 16 .0441 -.1233 -.1081 -.1056 9.137 -8.801 17 -.0561 -.3222 -.0439 .2390 -18.52 -0.511 18 -.0588 .0813 -.1666 .0094 -15.26 -35.30 19 .0426 .2176 -.0806 .0240 2.746 -17.58 20 .0304 -.8693 -.3107 -.0275 41.48 -20.01 21 .0607 .4524 -.0853 -.0783 -12.66 -6.362 22 .0032 -.9190 .1768 -.0436 -17.68 -8.824 23 .0197 -.9038 -.0702 .0268 -11.90 -17.58 24 -.0811 -.8347 .1216 -.0565 15.66 -42.59** 25 -.0006 -.8027 -.1240 .1259 24.51 49.33** 26 -.0735 -.0349 .0406 .0191 -17.99 -5.982 27 -.0486 .0743 -.1533 -.0632 28.22 -20.14 28 .0123 -.1361 -.1839 -.0803 -17.43 9.581 29 .0306 .3261 -.3145 -.0234 -21.70 27.93 30 -.1032 -.3260 -.0114 .1485 47.41 -27.93 31 .0722 -.2465* -.0473 .1112 24.49** -9.067 32 .1091 .0290 -.2909 .2091 17.36 -8.392 33 -.0478 -.3863 -.3353 .0130 22.16 28.30 34 .0033 -.1911 .0710 .0197 -5.141 -18.77 35 -.0099 -.4239 .1628 -.0679 -11.95 -31.83 36 -.0713 -.0001 -.1274 .0170 3.633 -23.11 37 -.0730 .0206 -.3179 .0088 -4.001 4.509 38 .1120 .1052 -.6622 .0175 42.97*** -2.043 39 -.1111 -.0982 .0499 -.2756* -31.22 6.808 40 .0299 1.711*** -.4591 -.0102 22.86 23.16 Nr. 𝜷𝜷𝒎𝒎 = 𝜷𝜷𝒎𝒎 4 1 0 4 2 5
Table 6: the specific 𝜷𝜷𝒋𝒋 of every determinant per COROP compared to the national trend
Nr. 𝜷𝜷𝒋𝒋= 𝜷𝜷𝒎𝒎 represents the number of coefficients falling in the boundaries of the coefficient of specification (10)
The results on how the effect of 𝑤𝑤𝑚𝑚𝐸𝐸 on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 differs between a COROP specification and a national specification is illustrated in figure 10. In specification (9) 𝑤𝑤𝑚𝑚𝐸𝐸 is found to be insignificant on a national level.
In figure 10 four COROP exhibit 𝛽𝛽𝑗𝑗 to be equal to 𝛽𝛽1. The other COROP don’t fall
within the boundaries supplied in table 5. The fact that certain COROP don’t follow empirical literature, and see 𝑤𝑤𝑚𝑚𝐸𝐸 to have a positive effect on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 is explained in section 4.2;
Figure 10: the comparison of 𝜷𝜷𝒋𝒋𝒘𝒘𝒎𝒎𝒊𝒊𝒉𝒉 𝜷𝜷𝒎𝒎 for 𝒆𝒆𝒆𝒆𝒆𝒆, 𝒎𝒎𝒖𝒖𝒊𝒊𝒖𝒖𝒊𝒊𝒊𝒊𝒎𝒎𝒊𝒊 𝒘𝒘𝒖𝒖𝒊𝒊 𝒖𝒖𝒎𝒎𝒏𝒏
Figure 11: the comparison of 𝜷𝜷𝒋𝒋𝒘𝒘𝒎𝒎𝒊𝒊𝒉𝒉 𝜷𝜷𝒎𝒎 for 𝒘𝒘𝒘𝒘𝒎𝒎𝒆𝒆, 𝒆𝒆𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎 𝒘𝒘𝒖𝒖𝒊𝒊 𝒉𝒉𝒎𝒎𝒎𝒎𝒉𝒉𝒆𝒆𝒊𝒊𝒖𝒖𝒏𝒏
possible explanations are an aging population and a labor force that is becoming smaller in
these COROP. However, in both specification (9) and specification (10) the coefficients of 𝛽𝛽𝑗𝑗
and 𝛽𝛽1are found to be insignificant. Therefore, the question remains whether 𝑤𝑤𝑚𝑚𝐸𝐸 has an effect on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚.
The results on how the effect of 𝑚𝑚𝑛𝑛𝑑𝑑𝑢𝑢𝐼𝐼𝐸𝐸𝐸𝐸𝐸𝐸 on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 differs between a COROP specification and a national specification is illustrated in figure 10. In specification (9) 𝑚𝑚𝑛𝑛𝑑𝑑𝑢𝑢𝐼𝐼𝐸𝐸𝐸𝐸𝐸𝐸 is found the be weakly significant, namely at a 10% significance level.
Figure 10 illustrates that 𝛽𝛽𝑗𝑗 is substantially different from 𝛽𝛽1 as only three COROP
indicate 𝛽𝛽𝑗𝑗 = 𝛽𝛽1. This suggest that in overcoming the problem of regional unemployment,
the Dutch government should investigate every COROP specific and seek how possible changes in the industry mix can be applied to increase industry diversity. In addition, section 4.2 provides a possible explanation for the positive effect of 𝑚𝑚𝑛𝑛𝑑𝑑𝑢𝑢𝐼𝐼𝐸𝐸𝐸𝐸𝐸𝐸 on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚; these regions might be characterized by very specific industries and demand for labor. This increases regional unemployment as the unemployed job seekers are unable to provide the means demand by firms.
Without taking into account the significance, figure 10 illustrates that a common national policy on unemployment is subservient to a clustered regional policy as regions differ a lot in their effect of the industry mix on unemployment.
Figure 10 also illustrates on how the effect of 𝑛𝑛𝑚𝑚𝑐𝑐 deviates between specification (9) and specification (10). In equation (9), the coefficient of 𝑛𝑛𝑚𝑚𝑐𝑐 is found the be significant at a 5% significance level. The results presented in figure 10 illustrate a clustering of a negative effect in the center and far North of the Netherlands. It also indicates that for only one
COROP 𝛽𝛽𝑗𝑗 = 𝛽𝛽1. The reason that most COROP exhibit a more negative relation on a regional
level than the national trend is that a substantial amount of the inhabitant commute to big cities for their work. As a result, they are subject to more labor demand which reduces the level of unemployment within these regions. These regions are considered to be strongly urbanized. COROP that are less mobile can’t take advantage of this and have to deal with labor demand close to their home. If labor demand is small, then an increase in spatial mobility might not have a direct effect on unemployment. As a result these regions can be characterized by a more positive effect on regional unemployment.
The results on the differences between specification (9) and specification (10) for 𝑤𝑤𝑎𝑎𝑚𝑚𝑤𝑤, 𝑚𝑚𝑚𝑚𝑚𝑚𝐸𝐸1 𝑎𝑎𝑛𝑛𝑑𝑑 ℎ𝑚𝑚𝑚𝑚ℎ𝑤𝑤𝑑𝑑𝑢𝑢𝑐𝑐 is illustrated in figure 11. The figure indicates that the effect of
𝑤𝑤𝑎𝑎𝑚𝑚𝑤𝑤 is always smaller in a COROP specification compared to a national specification. This is
an odd finding as the specification (9) showed 𝛽𝛽1= 0.42 and significant at a 1% significance
level. The steps leading to these results are consulted but no misspecification is found. This is left for further research as these results might be due to all 𝛽𝛽𝑗𝑗 being insignificant for 𝑤𝑤𝑎𝑎𝑚𝑚𝑤𝑤 in specification (10). Consequently, the following interpretation is provided for the possible negative effect; all COROP in the Netherlands experience the substitution effect to dominate the income effect. As wage increases, people are triggered to start searching for jobs and take on jobs as their marginal utility of working has increased relatively to their marginal utility of leisure. As a result, the level of unemployment falls more in these COROP compared to the national indicated trend. Following this line of reasoning, the bargaining position of the unemployed may play a considerable role in the Netherlands. As there are less vacancies available, the job-searching unemployed is more willing to take on a job even though its socially inefficient.
𝑚𝑚𝑚𝑚𝑚𝑚𝐸𝐸1 is significant at a 1% significance level in specification (9). The map illustrating the comparison of specification (9) and specification (10) on 𝑚𝑚𝑚𝑚𝑚𝑚𝐸𝐸1 indicates central and Southern Netherlands to experience a more positive effect on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚. In addition, the Northern and Eastern COROP represent a more negative effect on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 than specification (9). The reasoning behind this is explained in section 4.2: the positive and negative effects of 𝑚𝑚𝑚𝑚𝑚𝑚𝐸𝐸1 on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 are probably due to the trade-off between labour demand and labour
supply. The fact that only four COROP signify 𝛽𝛽𝑗𝑗 = 𝛽𝛽1 calls for regional clustered policies in
terminating unemployment instead of one common national policy.
ℎ𝑚𝑚𝑚𝑚ℎ𝑤𝑤𝑑𝑑𝑢𝑢𝑐𝑐 is significant at a 1% significance level in specification (9). Figure 11 shows
five COROP that experience 𝛽𝛽𝑗𝑗 = 𝛽𝛽1. Additionally, ℎ𝑚𝑚𝑚𝑚ℎ𝑤𝑤𝑑𝑑𝑢𝑢𝑐𝑐 is found to have a more
negative and positive effect on regional unemployment in specification (10) than specification (9). As mentioned in section 4.2, the negative effect follows empirical and theoretical literature whereas the positive effect could possibly be explained by the increased competition in high-educated positions which results in high-educated people to take on jobs below their marginal utility, leading to a chain reaction that causes total unemployment in the COROP to increase. An interesting observation is that the negative effect is clustered to specific regions. It might be that these regions are characterized by a high demand of labor for highly-educated. However, this doesn’t explain the more negative effect in the North and hence further research is needed on this matter.
5. Limitations of research
According to Vega and Elhorst (2012) regional disparities in unemployment should be modelled in a spatial framework. Therefore, the omission of spatial autocorrelation probably provides an inadequate measure as regional observations are not assumed to be independent. Regional independence is assumed in this paper.
Furthermore, there might exist a possible relationship between 𝑤𝑤𝑚𝑚𝐸𝐸 and 𝑛𝑛𝑚𝑚𝑐𝑐1. The reasoning behind this is that the average distance from work to home affect employment growth. This may lead workers to search employment outside the region they live in which therefore effects the region specific employment. Moreover, as highlighted by Brown (1972), 𝑤𝑤𝑚𝑚𝐸𝐸 and 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 provide reverse causality, implying endogeneity problems. 2SLS or GMM estimation might overcome this problem. The spatial framework to GMM and 2SLS lies outside the scope of this paper and will be left for further research.
In addition, the findings of specification (10 in this paper do not imply any causal relation between regional unemployment and the explanatory variables as the coefficients predominantly provide insignificant results. However, as section 4.3 indicated, by analyzing the boundaries of the national trend, one can infer that specification (9) and specification (10) differ substantially from an economic perspective.
A final limitation of this paper could be the fact that certain demographic trends aren’t incorporated in the study. An example that could effect 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚 is the aging of the population which would leave omitted variable bias. As explained in section 4.2, aging could have a positive effect on 𝑤𝑤𝑚𝑚𝐸𝐸 as it might be expected that regions having a more aged labor force experience a larger inflow of new workers. This is in line with SNN. Therefore it could
explain the positive effect of 𝑤𝑤𝑚𝑚𝐸𝐸 on 𝑢𝑢𝑛𝑛𝑤𝑤𝑚𝑚. 6. Conclusion
This paper has tried to identify the factors determining region specific unemployment in the Netherlands from 1995 to 2014. Moreover, the paper tried to explain the effect of these factors across forty COROP regions. Finally, it is investigated whether the regional trend differs from the national trend. This indicates if the Dutch government should pursue national unemployment policies to overcome unemployment or start fighting unemployment on a regional clustered level.
