The trade-off between economic performance and spatial equality

spatial equality

MSc Thesis Supervisor: S. Brakman Co-assessor: M. Gerritse

Keywords: Agglomeration, Spatial equality, Economic performance

Abstract

1) Introduction

The trade-off between economic performance and spatial equality is based on two different views. On the one hand, there is the argument that the distribution of opportunities for inhabitants of a country should be equally distributed, indicating spatial equality. On the other hand, there is the argument of maximizing resources to generate economic growth, indicating a higher economic performance. The trade-off theory is based on the assumption that a higher degree of agglomeration indicates a lower degree of spatial equality and that agglomeration has a positive effect on economic growth. The theory that this paper follows is based on the positive influences of being in an agglomeration on firm’s and worker’s productivity. Agglomeration is the gathering of different firms (from different sectors) and workers. The positive agglomeration influences on firms and workers translates into higher economic growth. Economic growth is an indication for economic performance. There are two main arguments that indicate a positive relationship between agglomeration and economic growth. The first argument states that in an agglomerated area spillovers (ideas, knowledge), thick labour market, tacit information, and linkages are relatively more present and have a positive influence on the performance of firms and workers. This will in turn increase their productivity. The second argument states that the limited land area in the agglomeration drives up the costs of being located in the agglomeration. The higher costs associated with agglomeration can only be paid by the more productive firms and workers. Hence, the second argument lead to higher productivity through the survival of the relatively more productive firms and workers and the leaving the market of the less productive firms and workers.

This paper will answer the research question by looking at the 15 EU-member that signed the Amsterdam Treaty in 1998. The Amsterdam Treaty states that countries will promote strengthening its economic and social cohesion, in particular periphery regions (European Union, 1997). This paper examines the time period 1995-2014. The 15 EU-member countries include countries with high and low economic growth as well countries with either a high degree of agglomeration or low degree of agglomeration.

Economic growth is measured by the growth in GDP per capita. In the robustness check GDP per capita is replaced by value added growth. For agglomeration there is not one measurement that stands out in the literature. This paper examines six different measurements for agglomeration instead of one. The different measurements are the share of the population living in the largest city, share of the population living in an agglomeration of one million or more inhabitants, the share of the population that lives in an urban area, the adjusted geographic index based on value added, and Zipf’s law as proxies. Further, this paper introduces a new measurement for agglomeration; the adjusted geographic index based on population.

This paper finds evidence for the trade-off between economic performance and spatial equality. Of the six different proxies for agglomeration only the share of the population living in an agglomeration of one million or more inhabitants has a positive effect on GDP per capita growth. This is thus the only proxy in line with the trade-off between economic performance and spatial equality. This result has been confirmed in the robustness test. The share of the population living in an agglomeration of one million or more inhabitants is the only agglomeration proxy that has a significant positive effect on value added growth in a country. The robustness measurement however also shows that agglomeration, when measured in the adjusted geographic index based on value added, has a significant negative effect on economic growth. This negative effect indicates that the opposite of the trade-off between economic performance and spatial equality is the case. This could have implications for policy, since policies that are designed to increase economic growth through agglomeration do not necessarily lead to higher economic growth.

in this paper. The fifth section focusses on the results of the analysis. This paper concludes with the findings and limitations in the sixth section.

2) Literature review

This section explains different theories and models for economic growth. After the general growth theory this section will focus on the contribution of agglomeration to economic growth, how this has changed throughout the years and how this could influence the policies of today and tomorrow. This section will end with the hypothesis and the research question that this paper tries to answer.

This section will start with a brief description of the Solow model -one of the most influential economic growth models- and his predecessor the Harrod-Domar model (Solow, 1956). The Solow model is based on the Harrod-Domar model (Solow, 1956). The Harrod-Domar model explains the GDP per capita growth of a country by using investments (that are equal to savings in this model), population growth, and the depreciation of capital. In the Harrod-Domar model investments have a positive effect on GDP per capita growth while population growth and depreciation of capital have a negative effect. The Harrod-Domar model is a neutral theory of economic growth because there is no difference in the model between relatively high and relatively low GDP per capita.

(Nelson & Pack, 1999; Foster, Haltiwanger & Krizan, 2005). The availability of human capital and government policies in combination with technological change -Solow model- is essential for economic growth (Barro, 1999). It is important to realize that human capital has to be employed otherwise it will not contribute to economic growth. Next to human capital the presence of physical capital influence the economic growth of a country (King & Rebelo, 1990).

At the end of the 19th century the first unbundling took place, the world transformed from the

so called “village economies” (production and consumption take place at the same location) towards a more open economy (Baldwin, 2013). The main reason for this change was the invention of the steam engine, that made it possible to separate production and consumption. This change resulted in economic growth, industrialization, and the clustering of consumption in cities (Baldwin, 2013).

During the 1980s the second unbundling started, new developments in information and communication technologies (ICT) boosted specialization. The first and second unbundling both had an effect on trade cost by lowering them. The change in trade cost had an effect on the location choice of firms, new theories were developed to describe the behaviour of firms after the change in trade costs. A leading theory for the explanation of firm localisation is the new economic geography (NEG) theory. The location choice of firms is based on two factors in the NEG theory: dispersion and agglomeration (Krugman, 1991; Baldwin, 2013).

firms. An important addition to the dispersion and agglomeration forces is the income level of workers. Workers will locate where the real wage is highest (Blanchard & Kiyotaki, 1987; Krugman, 1991; Brakman, Garretsen & van Marrewijk, 2009). When the agglomeration forces are strong workers and firms want to be located in the centre of the agglomeration. The movement of firms and workers to the agglomeration will increase the prices of housing and goods in the agglomeration and thus lower real wages. To compensate for the higher prices -lower real wages- the nominal wage will have to go up. The higher nominal wage can only be paid by the relatively more productive firms, the less productive firms will leave the market (Alonso, 1964; Melitz, 2002). The leaving of relatively less productive firms and staying of the relatively more productive firms will lead to an increase of the productivity of the agglomeration. The fact that only the more productive firms survive in an agglomeration increases the pressure for firms to perform.

History has shown that economic growth -an increase in economic performance- is only possible if there is growth of cities. History has also shown that economic activities get more concentrated when the overall development of a country goes up, this happened to the now developed world during the first and second unbundling. With higher development the equality in welfare between rural-urban and within-urban increases (World Bank, 2009). The productivity growth in agglomeration does not only come from the survival of relatively more productive firms and leaving of less productive firms. The productivity growth in agglomeration is also driven by spillovers, linkages, thick labour market, and tacit information (Marshall, 1920; Krugman, 1991; Sturgeon, Biesebroeck & Gareffi, 2008; Redding, 2009; World Bank, 2009). Spillovers refer to the learning from each other, firms can learn from each other and this goes easier when multiple firms are close to each other. With linkages the presence of supplier and costumer of firms are described. A thick labour market is the presence of a large labour force with relatively high knowledge of tasks that are important for the firms in the agglomeration. Tacit information is information that is difficult to explain without to-face contact. Because tacit information becomes increasingly important and so does face-to-face interaction (Sturgeon, Biesebroeck & Gareffi, 2008). Thus next to the pressure to perform that increases productivity there are the spillovers, thick labour market, linkages, and tacit information.

economy of a country. This inidcates that agglomeration has a positive effect on the performance of firms and firm performance has a positive effect on economic performance. In extension this would mean that agglomeration has a positive effect on economic performance of a country and thus a higher degree of agglomeration stimulates economic growth of a country. The positive relationship between agglomeration and economic growth indicates that countries should focus on trying to achieve and stimulate agglomeration. On the other side of agglomerations are the peripheries in a country. The peripheries are areas with less economic activities, lower population, and less firms. The relatively smaller degree of economic activities in the peripheries compared to the agglomerations contributes to a lower spatial equality in a country. Spatial equality refers to distribution of resources and the quality of those resources over different regions in a country. In the case of perfect spatial equality it does not matter in which region a person lives because, for example, the available welfare is the same. The lower spatial equality in a country that arises from the difference between agglomerations and peripheries becomes larger when agglomerations in a country experience relatively higher growth than the peripheries. To achieve spatial equality countries should stimulate the peripheries in a country. This creates a trade-off between economic performance and spatial equality. The assumption in this case is that a higher degree of agglomeration indicates a lower degree of spatial equality.

(crowdedness and pollution) outgain the positive effect of agglomerations (Dijkstra, Garciazo & McCann, 2013).

The possibility of a trade-off between economic performance and spatial equality can have political consequences. This paper explained multiple theories that suggest that there is a trade-off between economic performance and spatial equality. The main reason why this potential trade-off is important for politicians is because both economic performance and spatial equality are desirable objectives politicians try to achieve. Spatial equality is not as prominently present in the news like political debates about economic performance. Still most countries have different policies in place that address the spatial equality. Research that explains the trade-off between economic performance and spatial equality and what the magnitude of this trade-off is can have substantial implications for future policies. The importance of spatial equality is, for example, highlighted in the Amsterdam Treaty. Article 158 of the Amsterdam Treaty is the first article under the section economic and social cohesion and states the following:

“In order to promote its overall harmonious development, the community shall develop and pursue its actions leading to the strengthening of its economic and social cohesion. In particular, the community shall aim at the reducing disparities between the levels of development of the various regions and the backwardness of the least favoured regions or islands, including rural areas.” (European Union, 1997, p112).

This paper will focus on the 15 countries that signed the Amsterdam Treaty (Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Sweden, and the United Kingdom). The 15 selected countries are different from each other when it comes to size, population density, agglomeration, GDP per capita, and GDP per capita growth that it makes interesting group of countries to test. Compared to each other, these 15 countries show a different paces and states of economic growth and development. As mentioned before this uneven development of spatial areas lead to a decrease in spatial equality and spatial equality has been a struggle for politicians. The two baseline arguments were politicians have to deal with are based on two different visions. On the one hand there is the social equality argument that focusses on the distribution of opportunities for inhabitants. On the other hand, the economic performance argument focusses on the maximization.

The literature described in this paper points to a trade-off between economic performance and spatial equality. Countries that are more agglomerated experience more rapid economic growth, this economic growth is a result of a lower degree of spatial equality. This statement leads to the key hypothesis of this paper; countries experience more rapid economic growth if

the country is more agglomerated.

The hypothesis will be used to answer the research question of this paper; is there a trade-off

between economic performance and spatial equality? If these EU-15 countries experience a

faster economic growth when the country is more agglomerated, then this finding supports the theory of a trade-off between economic performance and spatial equality.

3) Empirical model

This section will look at the empirical model that this paper uses to test the hypothesis explained in the literature review. The different variables that are used in the empirical model are explained and what the predicted effect of those variables are on GDP per capita growth. This section will conclude with the description of multiple tests that increase the strength of the empirical model.

the hypothesis and uses different proxies for agglomeration. The empirical model is based on the paper written by Brülhart and Sbergami (2009) and the paper written by Gardiner, Martin, and Taylor (2011). The advantage of the model used in those papers is that the model is specially designed to test the relationship between agglomeration and economic growth. The model is not overly complicated and focusses on the essence of the relationship between

agglomeration and economic performance. The model used by Gardiner, Martin, and Taylor

(2011) is based on the model by Brülhart and Sbergami (2009), this means that there are multiple similarities between the two models. The models use the same basic model but use different proxies for agglomeration. Brülhart and Sbergami (2009) use the share of the population living in an agglomeration with 750.000 or more inhabitants in the year 2000, the share of the population that lives in the largest city, and the share of the population in an urban area. Gardiner, Martin, and Taylor (2011) use geographic concentration index and the adjusted concentration index as proxy for agglomeration. The empirical model that this paper uses to test if countries experience a more rapid growth if the country is more agglomerated is the following one:

∆"#$%_&' = )_*+ )_,-""_&' + )_./0"#$%_&'1,+ )₂3%%_&'+ )₄/0$%%_&'+ )₅6789_&'+ )_:;#_&'1.+ <_&+ =_'+ >_&'

The GDP per capita growth is expressed in ∆"#$%_&' where i stands for the specific country in year t. The GDP per capita growth is the difference between GDP per capita in year t compared to the GDP per capita in year t-1 expressed in percentage. The GDP per capita growth is a common measurement that captures the change in economic performance well. At the same time, it controls for differences between countries with regards to the population size and the population growth. This paper looks at 15 different countries with large differences when it comes to the population size, for example, between Germany and Luxembourg.

differences are explained after the explanation of the control variables. An overview of the variables and the expected effect can be found in appendix 1. The expectation for the agglomeration variable, based on the literature review, is that agglomeration has a positive effect on GDP per capita growth. This means that the expectation is that there is a positive relationship between agglomeration and GDP per capita growth shown by a positive outcome for the agglomeration variable in the model.

The Solow model (Solow, 1956) predicts that the level of GDP per capita (/0"#$%_&'1,) has

an influence on the economic growth of a country. In the Solow model the prediction is that a higher GDP per capita is responsible for a lower GDP per capita growth because of diminishing returns on capital. This would mean that there is a negative relationship between GDP per capita growth and GDP per capita. GDP per capita is expressed in a logarithm to control for large difference between countries and between years. This paper assumes that the effect of GDP per capita on GDP per capita growth is not immediately visible but takes time, this is the reason GDP per capita is expressed in t-1.

Human capital (3%%) and physical capital ($%%) are expected to have an influence on GDP per capita growth. In both human capital and physical capital, the expected relationship with GDP per capita is positive. This means that the expectation is that when human capital or physical capital increases, GDP per capita growth also increases. For human capital is it important that inhabitants of a country are not unemployed. The moment that inhabitants (providers of human capital) become unemployed, human capital no longer contributes to economic growth. To control for the the effect of unemployment on economic growth the percentage of the labour force without work is included in the model. The unemployment is

expressed in 6789_&' and expected to have a negative influence on economic growth.

This paper will test for time fixed effects and country fixed effects. The expectation is that both time specific effects and country specific effects influence GDP per capita growth. There could be unobservable time and country fixed effects that the model does not account for. To control

for those unobservable effects on GDP per capita growth the model uses time fixed effects (=_')

and country fixed effects (<&). The error term in the model is expressed by >&'.

As explained before this paper uses multiple proxies for agglomeration in the model. The different proxies have their own advantages and disadvantages as a measurement for agglomeration. The share of inhabitants living in the largest city is a proxy that gives a quick indication of the agglomeration of a country. This paper uses the share of the population to account for different population sizes in the 15 countries. This makes it possible to compare the different countries and compare over time. The disadvantage of the share of inhabitants living in the largest city is that it only takes one city in to consideration. This means that the second city is not accounted for while this city is most likely also an agglomerated area. When looking at Germany -where the largest city is Berlin- this method does not include cities like Munich, Hamburg, or Cologne while those cities contributed to the agglomeration of Germany. Another disadvantage of this variable is that city borders are mode on paper; this means that part of the agglomeration can be outside the city.

The share of inhabitants in an agglomeration of one million or more inhabitants is like the largest city variable a quick indication of the agglomeration of a country. A second similarity with the largest city variable is the measurement in share of the population so that also for the agglomeration of one million or more inhabitants variable is controlled for differences in population size between countries and over time. Compared to the largest city share does the agglomeration with one million or more inhabitants variable take multiple agglomerations in consideration. The disadvantage of this variable is the cut-off of one million inhabitants. This implies that agglomerations with fewer than one million inhabitants are not taking into account. The agglomeration with fewer than one million inhabitants can still have a significant impact when it comes to the agglomeration of a country.

largest city and population share in agglomeration of one million or more inhabitants. The urban share has the tendency to overestimate agglomeration while the other two variables have the tendency to underestimate the agglomeration of a country. If a country has relatively a high existence of small cities who have little effect on agglomeration, in this situation the urban share variable overestimates agglomeration.

The adjusted geographic concentration (AGC) index based on value added is based on the Herfindahl index, Ellison and Gleaser (EG) index, and geographic concentration (GC) index (Spiezia, 2002). The ability to compare between countries makes the AGC index the superior option compared to the other three indexes.

"% = |@_&

A

&B,

− D_E|

The number of NUTS 2 regions is expressed by N. The production share of each NUTS 2 region is calculated using the value added generated in that specific NUTS 2 region and is expressed by @_&. The area share of the NUTS 2 region is expressed by D_&. To indicate the absolute value of @_&− D_& the formula is between vertical bars. The GC index itself would be a decent measurement of agglomeration but is still sensitive to overestimating or underestimating agglomeration, this is especially the case when there are large differences between countries when it comes to the size of NUTS 2 regions. The AGC index improves the GC index by including a maximal GC index. This maximal GC index is the value of the GC index if all production, in this paper value added, would take place in the smallest NUTS 2 region. The maximal GC index is calculated using the following formula (Spiezia, 2002):

"%FGH _{= D}

& &IF&J

+ 1 − DF&J = 1 + 1 − 2DF&J = 2(1 − DNEO)

The relative smallest NUTS 2 area is expressed by D_F&J. The GC index and the maximal GC

are used to calculate the AGC index. The formula for the AGC index is the following one (Spiezia, 2002):

-"% = "%/"%FGH

The Herfindahl index, EG index, GC index, and AGC index are based on production share. The production share is a measurement that focusses on firms and their production. In the case of this paper it is a measurement that focusses on value added within a NUTS 2 region. Production and value added can be generated, for example, by robots, this could lead to a relative high production share and value added share in a NUTS 2 region. In this case the AGC would indicate that there is a relative high degree of agglomeration. However, this would be a distorted picture if all production and value added is generated by robots and there is little agglomeration present.

(2)

(3)

To account for the disadvantage of the AGC index based on value added this paper makes an adjustment to the GC index and in extension to the AGC index. The new GC index will be based on population share instead of production share. The GC index based on population is calculated with the following formula, this formula is an adjustment to formula (2):

"%RSR _{= |T}

& A

&B,

− D_E|

As in formula (2) the number of NUTS 2 regions is expressed by N, the area share is expressed by D_&, and the vertical lines indicate the absolute value. The difference with formula (2) is that the production share is replaced by population share. The population share of NUTS 2 region

is expressed by T&. The GC index based on population is used to calculate the AGC based on

population. The maximal GC index calculated with formula (3) does not change for the AGC based on population. The AGC based on population is calculated as:

-"%RSR _{= "%}RSR_/"%FGH

The AGC index based on population focusses on in which NUTS 2 region inhabitants of a country life. This measurement controls for the disadvantage of the AGC based on production share because it looks at the location of people. The AGC index based on population makes the assumption that inhabitants life and work in the same NUTS 2 region. The AGC based on production share and the AGC based on population share have a common disadvantage. Both the production AGC and the population AGC is based on NUTS 2 level data while an agglomeration could be in different NUTS 2 regions. The Randstad in the Netherlands is an example of an agglomeration that is located in multiple NUTS 2 regions. The Randstad is an agglomeration that exist out of four different cities (Amsterdam, The Hague, Rotterdam, and Utrecht) and is located in three different NUTS 2 regions (Noord-Holland, Utrecht, and Zuid-Holland).

The last proxy for agglomeration this paper uses is based on Zipf’s law (Gabiax, 1999). Zipf’s law assumes that there is a relationship between the city size (measured in population) and the city’s rank in size. The relationship between the city size and city rank is explained in the following formula (Brakman, Garretsen & van Marrewijk, 2009):

(5)

VWX 9_Y = log ] − ^ log (;_Y)

The city size by its population is expressed with 9_Y, where 9 stands for the population and _

for the city. ;_Y stands for the rank of city _ and ] is a constant. Since this paper uses the slope

of Zipf’s law as proxy for agglomeration, the variable ^ is especially important. For Zipf’s law to hold ^ has to be equal to one, if the slope is smaller than one the distribution of cities is more even. When the slope is larger than one the distribution of cities is less even. This means that a higher slope indicates a higher degree agglomeration in a country. This paper makes a distinction from the Zipf’s law described here, instead of cities this paper looks at the distribution of NUTS 3 population. The disadvantage of the city distribution is the data availability. The disadvantage of the NUTS 3 distribution is that NUTS 3 regions can be defined as parts of large cities or as regions where agglomeration is present. This could change the results. The decision for the NUTS 3 distribution instead of the city distribution is based on data availability.

This paper uses multiple proxies for agglomeration to test if countries experience a more rapid economic growth if the country is more agglomerated. The different proxies that are used to explain agglomeration have their own advantages and disadvantages. For some proxies the advantages outweigh the disadvantages by a larger margin than for some other proxies. The proxy for agglomeration that in this paper has the biggest advantages, is the proxy “share of the population living in an agglomeration with one million or more inhabitants”. The biggest advantage of this agglomeration proxy compared to the other agglomeration proxies is that this proxy is not based on borders but can change in area size. The other agglomeration proxies, especially the proxies based on NUTS regions, are bounded by borders that could divide agglomerations.

strength of the model and results is for autocorrelation. The test, their outcome, and the resulting action is described in the section analysis with the exception of the first test.

4) Data

In this section the data that is used for the empirical model, described in the previous section, and the limitations of the data are explained.

4.1) Data

The data used in this paper can be divided in three different levels: the country level, NUTS 2 level, and NUTS 3 level. The data is collected for 15 different countries, this include the NUTS 2 and NUTS 3 level, for the time period 1995-2014. Unfortunately, not all data was available for all 15 countries or for the (complete) time period. The time period of 1995-2014 is on the one hand based on the availability of data, on the other hand on the fact that agglomeration takes time and a longer time period would be able to catch the change in agglomeration more extensive. The data based on NUTS 2 and NUTS 3 is collected from 2000 until 2014, the reason is the data availability. The data is collected from three different sources: Eurostat, World Bank, and Penn World Table.

country. The population data is collected from the World Bank (2016) and describes the number of inhabitants of a specific country in a specific year.

In the empirical model section, it is explained that this paper used multiple proxies for agglomeration: the share of inhabitants living in the largest city, the share of inhabitants living in an agglomeration of one million or more inhabitants, the share of the population that lives in an urban area, the AGC index based on value added, the AGC index based on population, and the Zipf’s law slope.

The share of inhabitants living in the largest city, the share of inhabitants living in an agglomeration of one million or more inhabitants, and the share of the population that lives in an urban area are collected from the World Bank (2016). The three variable are expressed as a percentage of the total population of a country in a specific year. The share of inhabitants living in an agglomeration of one million or more inhabitants is based on agglomerations that met this criterion in the year 2000.

The data for the AGC index based on value added and AGC index based on population is collected on the NUTS 2 level. The NUTS 2 level data is collected from Eurostat (2016). For the calculation of the AGC index based on value added, the following data is used: value added per NUTS 2 region and the area size of the NUTS 2 regions. For the AGC index based on population the following data is used: population per NUTS 2 region and the area size of the NUTS 2 region.. Both the AGC index based on value added and population share take a value between 0 (no agglomeration) and 1 (complete agglomeration).

The data for the Zipf’s law slope is collect from Eurostat (2016) at the NUTS 3 level. The population level of the largest cities in a country are not available for most of the years in the time period this paper uses. The NUTS 3 regions per country are ranked based on population in that specific year. The population and population ranking of the NUTS 3 regions are used to calculate the slope described in the empirical model as Zipf’s law. A steeper Zipf’s law slope indicates a higher degree of agglomeration in a country.

4.2) Descriptive statistics

observations while the AGC index based on population has 188 observations. This paper looks at 15 countries for a time period of 20 years, this means that only the the share of the population that lives in an urban area and unemployment has data available for the complete times period for all 15 countries (300 observations).

Table 1: Descriptive statistics

Variables Observations Mean Std. Dev. Min Max

GDP per capita 284 29,003 10,599 12,000 70,400

GDP per capita growth 280 1.53 3.01 -9.00 22.60

Largest city 280 22.71 12.17 5.6 50.73

Agglomeration of one million 280 20.81 7.87 6.25 40.04

Urban share 300 76.79 10.53 51.11 97.82

AGC index (production share) 192 0.39 0.09 0.27 0.57

AGC index (population share) 188 0.32 0.09 0.21 0.53

Zipf’s law 192 0.79 0.16 0.47 1.14

Human capital per capita 255 2.89 0.21 2.31 3.32

Physical capital per capita 255 75,255 23,956 26,045 170,417

Unemployment 300 8.22 4.35 1.8 27.2

R&D 255 1.87 0.81 0.43 3.91

The change in GDP per capita is missing observations for the first year (1995) because the GDP per capita for 1994 was not available for all countries in Eurostat (2016). Both GDP per capita as GDP per capita growth is missing the last year of the time period (2014). The data for the share of the population in the largest city and the share of the population in an agglomeration of one million or more inhabitants is not available for Luxembourg. The variables based on NUTS 2 data is only available for the period 2000-2014 and is not available for Greece and Luxembourg. The availability of the NUTS 2 data lowers the amount of observations for the AGC index based on value added and the AGC index based on population. Human capital per capita and physical capital per capita are not available for the last three years of the time period (2012, 2013, and 2014); this lowers the amount of observations to 255. The NUTS 3 data was only available for the time period 2000-2014. Luxembourg has no NUTS 3 data available and multiple countries miss data on the NUTS 3 level. The unavailability of NUTS 3 data is visible in the 192 observations for the slope based on Zipf’s law.

subtracting this outcome from the mean to find the lower boundary and adding the outcome to the mean to find the upper boundary (Osborne, 2016). If the observation value is outside of those boundaries it is an outlier. There are outliers in GDP per capita, GDP per capita growth, physical capital per capita, and unemployment variables. To control for the effect of the outliers the GDP per capita variable and the physical capital per capita variable are expressed in their logarithm. For the outliers for GDP per capita growth and unemployment are no extra actions needed, the detected outliers show a realistic value.

This paper has shortcomings and limitations when it comes to the data that is used. The data is collected from different sources; this could lead to differences. The different sources did not have all the data needed for this research individually. The biggest data limitation this paper faces is that not all data is available for the complete time period or for all 15 countries. As mentioned before the NUTS 2 data is not available for Greece and Luxembourg. For the NUTS 3 data is not available for Luxembourg, the reason why the NUTS 2 and NUTS 3 data are not available for Luxembourg is because Luxembourg does not have NUTS 2 and NUTS 3 regions. Throughout the data set multiple observations are missing without a specific reason. This paper leaves the missing values blank.

5) Results

This section begins with the tests describe in the empirical model to increase the strength of the empirical model and the results. The empirical model is used to test the hypothesis and the results of this test are analysed.

5.1) Preliminary tests

effect regression support the conclusion of heteroscedasticity. To control for heteroscedasticity this paper uses robust standard errors to minimize the effect of heteroscedasticity. The fifth test is the Hausman tests which test if the fixed effect model or the random effect model is the appropriate model. The Hausman test shows that the fixed effect model is the most appropriate model for this analysis. This means that there are country fixed effects that have an influence on GDP per capita growth. The six test, test if there are besides the country fixed effects also time fixed effects. The results show that there are time fixed effects, this means that different years have an influence on GDP per capita growth. To control for the country fixed effects and time fixed effect the empirical model will include both in the regressions. The seven and last test is for autocorrelation, to test for autocorrelation this paper uses the Wooldridge test for autocorrelation in panel data. The Woolridge test shows that there is autocorrelation. To control for autocorrelation this paper uses the clustered robust standard errors.

5.2) Main results

The empirical model described in combination with the adjustments to increase the strength of the results are presented in table 2. The results in table 2 are with the time fixed effect (not visible), the complete results including the time fixed effect are presented in appendix 2. This paper tries to answer the question if there is a trade-off between economic performance and and spatial equality? This is tested by six different regressions with six different proxies for agglomeration, the six results are presented in table 2. The six different proxies are visible at the top of the table.

In table 2 in column 3 the regression is shown with the share of the population living in an agglomeration of one million or more inhabitants, it shows that if a larger share of the population lives in an agglomeration of one million or more inhabitants GDP per capita growth increases. This is visible by the positive and significant, at five percent level, number of 1.540. The 1.540 indicates that an increase of one percent point in share of population in an agglomeration of one million or more inhabitants the GDP per capita growth increases by 1.540 percent points. Measuring agglomeration in one of the other five proxies does not generate significant results. This means that even though four out of the five are positive there is no significant evidence for a relationship between the five different proxies for agglomeration and GDP per capita growth.

Table 2: Main results; the effect of agglomeration on GDP per capita growth

Largest city Agglomeration Urban share AGC production AGC population Zipf’s law

Variables (GDP per capita

growth) (GDP per capita growth) (GDP per capita growth) (GDP per capita growth) (GDP per capita growth) (GDP per capita growth) Agglomeration 0.343 1.540** 0.0767 -37.187 49.147 20.115 (0.368) (0.654) (0.107) (27.949) (88.651) (23.984)

GDP per capita (log, 1 lag) -13.794* -15.089** -12.084* -12.849 -12.334 -18.751

(7.121) (5.389) (6.210) (10.068) (9.745) (10.949)

Human capital 1.287 3.900 0.746 2.548 0.610 -2.157

(2.979) (2.896) (3.523) (2.484) (3.038) (5.574)

Physical capital (log) -3.635 0.334 -8.433** -9.786*** -9.178** -6.144

(6.390) (5.556) (3.114) (2.579) (3.985) (3.648)

Unemployment -0.214 -0.273 -0.228 -0.139 -0.115 -0.297

(0.148) (0.155) (0.156) (0.106) (0.110) (0.194)

R&D (2 lags) 1.100 0.235 1.504 1.267* 1.078 2.091*

(0.779) (0.511) (0.887) (0.696) (0.628) (1.133)

Time fixed effects Yes Yes Yes Yes Yes Yes

Constant 170.717** 110.693* 210.172*** 248.912** 212.046** 250.095**

(59.130) (61.674) (40.617) (89.805) (90.389) (102.984)

Observations 187 187 195 149 146 144

R-squared 0.809 0.822 0.812 0.856 0.854 0.804

Number of Countries 14 14 15 13 13 14

GDP per capita is negative for all six regressions and three out of those six are significant. If agglomeration is measured as share of the population living in the largest city, share of the population living in an agglomeration of one million or more inhabitants, or the share of the population that lives in an urban area the relationship between GDP per capita and change in GDP per capita is significant. GDP per capita is expressed in a logarithm, this means that when GDP per capita goes up by one percent, in the regression for the share of the population living in an urban area, the GDP per capita growth goes down by approximately 0.121 percent points. In the regression for share of the population in the largest city and share of the population in an agglomeration of one million or more inhabitants the effect of one percent increase in GDP per capita leads approximately to a 0.138 and 0.151 percent point decrease in GDP per capita growth, respectively. GDP per capita is expressed in a lag in the six regressions, this means that the GDP per capita effect on GDP per capita growth is one year later, for example, if GDP per capita goes up by one percent in 2007, in share of the population in an urban area, the effect of this increase will be a GDP per capita growth decrease of approximately 0.121 percent points in 2008.

Human capital is insignificant in all six regressions; this means that human capital has no significant influence on GDP per capita growth according to the empirical model used in this paper. Physical capital is significant negative for the urban share, AGC index based on value added, and AGC index based on population regressions. The negative relationship between physical capital and GDP per capita growth could be explained by the same theory, the Solow model, as the negative relationship between GDP per capita growth and GDP per capita. In the AGC index based on population regression an increase of physical capital by one percent leads approximately to a 0.092 percent point decrease in GDP per capita growth.

increase in R&D expenditures in the year 2002 of one percent point leads to an increase of 2.091 percent points in GDP per capita growth in the year 2004.

5.3) Robustness

To increase the strength of the results, this paper uses a different proxy for economic performance. The proxy for the robustness empirical model is value added growth instead of GDP per capita growth. Value added is a measurement that looks at the value that a country adds (Johnson, 2014). The model used for the robustness of this paper is the following one:

∆"#_$% = '₍+ '_*#++_$%+ '_,-.+/01_$%2*+ '₃411_$%+ '₅-.011_$%+ '6789:$%+ ';</$%2,+ =$ + >%+ ?$%

The value added is expressed in ∆"#_$%. The other variables are the same variables as described

for model (1).

For the robustness measurement this paper uses the value added growth. The value added growth data was not available per capita and for multiple countries the data is missing, hence the value added growth includes 222 observations.

The results presented in table 2 are supported by the robustness test results, the robustness results are presented in table 3. The robustness results are presented without the time fixed effects, the complete robustness results are presented in appendix 3. The robustness results show the effect of the different variables and different measurements on value added growth instead of GDP per capita growth.

Table 3: Robustness; the effect of agglomeration on value added growth

Largest city Agglomeration Urban share AGC production AGC population Zipf’s law

Variables (Value added

growth) (Value added growth) (Value added growth) (Value added growth) (Value added growth) (Value added growth) Agglomeration 0.709 2.237* 0.146 -62.169** 44.915 25.133 (0.753) (1.048) (0.141) (27.764) (74.743) (22.955)

GDP per capita (log, 1 lag) -21.516 -22.892** -17.953 -14.173 -11.907 -20.368

(13.686) (9.706) (11.541) (11.361) (10.630) (11.985)

Human capital -2.686 0.517 -5.064 0.145 -1.529 -5.253

(5.926) (5.263) (6.856) (3.021) (3.815) (6.777)

Physical capital (log) 5.022 9.081 -4.807 -6.292** -5.762 -2.057

(12.686) (9.009) (4.067) (2.730) (3.708) (4.133)

Unemployment -0.378* -0.452** -0.390** -0.249** -0.228* -0.425**

(0.190) (0.187) (0.181) (0.111) (0.112) (0.191)

R&D (2lags) 0.669 -0.468 0.957 0.193 0.300 1.241

(1.158) (0.723) (1.100) (0.668) (0.582) (1.179)

Time fixed effects Yes Yes Yes Yes Yes Yes

Constant 157.918 90.016 242.861** 243.282** 179.916* 228.557*

(121.320) (112.063) (93.264) (99.248) (95.486) (112.231)

Observations 160 160 168 146 143 141

R-squared 0.702 0.722 0.701 0.751 0.761 0.721

Number of Countries 14 14 15 13 13 14

The share of the population living in an agglomeration of one million or more inhabitants is, just as for GDP per capita growth, the only agglomeration proxy that has a significant positive effect on value added growth. An increase of one percent point of the share of population living in an agglomeration leads to a 2.237 percent point increase in value added growth. It stands out that, in opposite of the GDP per capita results, there is a significant negative relationship between the AGC based on production and value added growth. This means that when the AGC index based on value added goes up by 0.1 value added growth decreases by 6.217 percent points. The AGC index has a value between 0 and 1, this means that changes in this index are small. The robustness results indicate that when production takes place in a few (small) areas the value added growth of a country decreases. The decrease in value added growth by a more concentration of production is the opposite of the theory explained in this paper. This indicates that there is no trade-off between economic performance and spatial equality but that spatial equality has a positive effect on economic performance. This negative relationship could be explained by negative effects of a higher degree of agglomeration. Pollution, oversupply of workers, and higher living costs are a few examples of negative effects that Dijkstra, Garciazo, & McCann (2013) find. When the negative effects of agglomeration are stronger than the positive effects the relationship between agglomeration and economic performance could become negative.

to a value added growth decrease of 0.425 percent points. The effect of R&D is insignificant when it comes to the relationship between R&D and value added growth, while R&D was significant for two out of the six regressions in the main results.

The main results from table 2 and the robustness results from table 3 do not show strong evidence for a positive relationship between economic performance and agglomeration. The main results and robustness results include twelve regressions and out of those twelve two show a positive relationship and one a negative relationship between economic performance and spatial equality. The agglomeration proxy “share of population living in an agglomeration of one million or more inhabitants” indicates that an increase in degree of agglomeration positively influence economic growth. In the empirical model this proxy is described as the most realistic proxy for agglomeration, this means that the results of this regression weigh relatively heavier than the other results.

The 15 countries that are being researched in this paper are countries with relatively high GDP per capita. High GDP per capita is an indication for high quality infrastructure and high quality communication infrastructure. The presence of quality infrastructure could increase the spreading of positive agglomeration effects to periphery area. Brülhart and Sbergami (2009) find that higher GDP per capita reduces the effect of agglomeration on economic performance. This could be a reason why the results do not fully support the hypothesis of a trade-off between economic performance and spatial equality. A second reason could be that not all agglomeration proxies capture agglomeration. The second reason is supported by the significance of the strongest agglomeration proxy; the share of the population living in an agglomeration of one million or more inhabitants.

6) Conclusion

economic growth. To test this trade-off this paper answers the following research question; is there a trade-off between economic performance and spatial equality?

This paper explained two main arguments that argue that agglomeration has a positive effect on economic performance. The first one argues that spillovers, thick labour market, linkages, and tacit information are relatively more present in an agglomeration and that those factors have a positive influence on economic performance. The second one argues that prices in an agglomeration are higher and that only the more productive firms and workers can pay the higher costs associated with an agglomeration.

This paper looked at the economic performance of the 15 countries that were members of the European union in 1998 when the Amsterdam Treaty was signed. The economic performance of those 15 countries is analysed over the time period 1995-2014. To measure agglomeration within a country this paper used six different proxies for agglomeration. This paper uses share of the population living in the largest city, share of the population living in an agglomeration of one million or more inhabitants, the share of the population that lives in an urban area, the adjusted geographic index based on value added, adjusted geographic index based on population, and Zipf’s law as proxies for agglomeration. The adjusted geographic index based on population is a new measurement for agglomeration that this paper introduces. Next to the six different proxies for agglomeration this paper used two proxies for economic growth. GDP per capita is used as economic growth proxy in the main regression while value added growth is used as a robustness test.

Policies that target economic performance through agglomeration have a negative effect on spatial equality. Policies that target spatial equality have a negative effect on economic performance.

This paper also finds evidence that contradict the trade-off between economic performance and spatial equality. The agglomeration measurement AGC index based on value added has a negative effect on the robustness measurement value added growth. This indicates that a higher degree of agglomeration leads to a decrease in economic growth. This result indicates that there is no trade-off between economic performance and spatial equality but that economic performance and spatial equality go hand in hand.

This paper has different limitations. The data needed for this research was not available for all countries or all years of the time period. The countries analysed in this paper are different from each other when it comes to economic performance and spatial equality, but the 15 countries are all members of the European Union. Being a member of the European Union creates a cohesion with each other, this means that the 15 countries show similarities in many aspects. The biggest limitations that this paper faces and tried to account for is the measurement of the degree of agglomeration in a country. Agglomeration is difficult to measure and for that reason it is different to test the relationship between economic growth and the degree of agglomeration in a country.

For future research it would be interesting to see if the results of this paper change when looking at different regions in the world. This paper focusses on European countries while the results could be different for countries located in other regions of the world. Countries outside Europe show different patterns than European countries, this could have effect on the trade-off between economic performance and spatial equality. Countries in Asia show rapid economic growth, while countries in Africa still struggle to growth economically. That are just a few factors that could have an impact on the trade-off between economic performance and spatial equality.

References

Alonso, W. (1964) Location and Land Use. Harvard University Press

Baldwin, R. (2013) Factory free Europe? A two unbundlings perspective on Europe’s 20th century manufacturing miracle and 21st century manufacturing malaise. Graduate institute,

Geneve and Oxford University

Barca, F., McCann, P., Rodríguez-Pose, A. (2012), The case for regional development intervention: place-based versus place-neutral approaches. Journal of Regional science, 52 (1): 134-152

Barca, F. (2009) An Agenda for A Reformed Cohesion Policy: A Place-Based Approach to Meeting European Union Challenges and expectations. Independent report: prepared at the

request of the European Commissioner for regional policy, Danuta Hübner, European Commission, Brussels.

Barro, R. (1999) Determinants of Economic Growth: Implications of the Global Evidence for Chile. Cuadernos de Economia, 107: 443-478

Blanchard, O., Kiyotaki, N. (1987) Monopolistic competition and the effects of aggregate demand. The American Economic Review, 77 (4): 647-666

Brakman, S., Garretsen, H., van Marrewijk, C. (2009) The new introduction to geographical economics. United Kingdom, Cambridge University Press.

Brülhart, M., Sbergami, F. (2009) Agglomeration and growth: cross-country evidence.

Journal of Urban Economics, 65: 48-63

CAF. (2010) Desarrollo Local: Hacia un Neuvo Protagonismo de las Ciudades y Regiones. Caracas: corporación Andina de Fomento

Dall’erba, S., Hewings, G. (2003) European regional development policies: the trade off between efficiency-equity revisited. Regional Economics Applications Laboratory, Discussion paper 03-T-2

Dijkstra, L., Garcilazo, E., McCann, P. (2013) The Economic Performance of European Cities and City Regions: Myths and Realities. European Planning Studies, 21 (3): 334-354 Dupont, V. (2007) Do geographical agglomeration, growth and equity conflict. Papers in

Regional Science, 86 (2): 193-213

Eurostat, European Commission. Data. Accessed April 11, 2016 http://ec.europa.eu/eurostat/data/database

Feenstra, R., Inklaar, R., Timmer, T. (2015) The Next Generation of the Penn World Table.

American Economic Review, 105 (10): 3150-3182

www.ggdc.net/pwt

Foster, L., Haltiwanger, J., Krizan, C. (2006) Market selection, reallocation and restructuring in the US retail trade sector in the 1990s. Review of Economics and Statistics, 88(4): 748-758 Fujita, M., Thisse, J. (2003) Does geographical agglomeration foster economic growth? And who gains and loses from it? The Japanese Economic Review, 54 (2): 121-145

Gabiax, X. (1999) Zipf’s law for cities: an explanation. The Quartely Journal of Economics 114 (3): 739-767

Gardiner, B., Martin, R., Tyler, P. (2011) Does spatial agglomeration increase national growth? Some evidence form Europe. Journal of Economic Geography, 11: 979-1006

Johnson, R. (2014) Five facts about value-added exports and implications for macroeconomic and trade research. Journal of Economic Perspectives, 28 (2): 119-142

King, R., Rebelo, S. (1990) Pulbic policy and economic growth: developing neoclassical implications. Journal of Political Economy 98 (5): 5126-5150

Krugman, P. (1991) Increasing returns and economic geography. Journal of Political

Economy, 99 (3): 483-499

Marshall, A. (1920) Principles of Economics; An Introductory Volume. Macmillan and Co.

London, U.K.

Martin, R. (2008) National growth versus spatial equality? A cautionary note on the new ‘trade-off’ thinking in regional policy discourse. Regional Science Policy & Pratice, 1 (1): 4-13

McCann, P., Rodríguez-Pose, A. (2011) Why and When Development Policy Should Be Place-Based. OECD Regional Outlook 2011: building Resilient Regions for Stronger

Economies, OECD Publishing

Melitz, M. (2002) The impact of trade on intra-industry reallocations and aggregate industry productivity. NBER working paper 8881

Nelson, R., Pack, H. (1999) The Asian Miracle and Modern Growth Theory. The Economic Journal, 109 (457): 416-436 OECD. (2006) Competitive cities in the global economy. Paris: Organisation for Economic Cooperation and Development Osborne, J. (2004) The power of Outliers (and why researchers should always check for them). Research Gate https://www.researchgate.net/publication/242073851

Redding, F. (2009) Economic Geography: a Review of the Theoretical and Empirical Literature. CEP discussion paper no 904

Solow, R. (1956) A Contribution to The Theory of Economic Growth. The Quarterly Journal

of Economics, 70 (1): 65-94

Spiezia, V. (2002) Geographic concentration of production and unemployment in OECD countries. Cities and Regions. OECD, Paris

Sturgeon, T., Biesebroeck van, J., Gereffi, G. (2008) Value chains, networks and clusters: reframing the global automotive industry. Journal of Economic Geography, 8: 297-321 World Bank. (2003) World Development Report: 2003: transforming Institutions, Growth and Quality of Life. Washington DC: World Bank

World Bank. (2009) world development report 2009 Reshaping Economic Geography. The

International Bank for Reconstruction and Development / The World Bank

Appendix

Appendix 1: overview of different variables

Variables variable name in regression Expected sign Source

GDP per capita growth ∆"#$%_&' Eurostat

GDP per capita ()"#$%&'*+ Negative Eurostat

Share of the population living in the largest city ,""_&' Positive World Bank

Share of the population living in an agglomeration of one million or more inhabitants

,""_&' Positive World Bank

Urban share ,""_&' Positive

World Bank

AGC index based on value added ,""&' Positive Eurostat

AGC index based on population ,""_&' Positive Eurostat

Zipf's law ,""_&' Positive Eurostat

Human capital per capita -%%_&' Positive

Penn World Table

Physical capital per capita ()$%%_&' Positive Penn World Table

Unemployment ./01&' Negative World Bank

Appendix 2: Main results; the effect of agglomeration on GDP per capita growth (incl. years)

Largest city Agglomeration Urban share AGC production AGC population Zipf’s law

Variables (GDP per capita

growth) (GDP per capita growth) (GDP per capita growth) (GDP per capita growth) (GDP per capita growth) (GDP per capita growth) Agglomeration 0.343 1.540** 0.0767 -37.187 49.147 20.115 (0.368) (0.654) (0.107) (27.949) (88.651) (23.984)

GDP per capita (log, 1 lag) -13.794* -15.089** -12.084* -12.849 -12.334 -18.751

(7.121) (5.389) (6.210) (10.068) (9.745) (10.949)

Human capital 1.287 3.900 0.746 2.548 0.610 -2.157

(2.979) (2.896) (3.523) (2.484) (3.038) (5.574)

Physical capital (log) -3.635 0.334 -8.433** -9.786*** -9.178** -6.144

(6.390) (5.556) (3.114) (2.579) (3.985) (3.648) Unemployment -0.214 -0.273 -0.228 -0.139 -0.115 -0.297 (0.148) (0.155) (0.156) (0.106) (0.110) (0.194) R&D (2 lags) 1.100 0.235 1.504 1.267* 1.078 2.091* (0.779) (0.511) (0.887) (0.696) (0.628) (1.133) 1999 0.285 0.0343 0.351 (0.502) (0.588) (0.447) 2000 1.340*** 0.915 1.421*** (0.399) (0.539) (0.410) 2001 -0.916* -1.485* -0.827* -2.182*** -2.107*** -1.833*** (0.492) (0.699) (0.400) (0.368) (0.342) (0.369) 2002 -1.126* -1.820** -0.938 -2.327*** -2.258*** -1.545** (0.567) (0.775) (0.663) (0.466) (0.524) (0.675) 2003 -0.880 -1.693* -0.700 -2.449*** -2.245** -1.430 (0.688) (0.920) (0.870) (0.666) (0.754) (1.054) 2004 0.468 -0.531 0.727 -0.746 -0.574 0.0160 (0.847) (1.162) (0.843) (0.728) (0.896) (1.058) 2005 0.505 -0.643 0.882 -0.570 -0.484 0.203 (0.991) (1.330) (0.982) (0.958) (1.119) (1.332) 2006 1.893* 0.533 2.257** 0.679 0.844 1.728 (0.986) (1.391) (1.008) (0.994) (1.202) (1.525) 2007 1.924 0.413 2.447* 1.013 1.220 1.999 (1.259) (1.684) (1.156) (1.351) (1.514) (1.749) 2008 -0.813 -2.440 -0.528 -1.583 -1.443 -0.599 (1.396) (1.810) (1.189) (1.530) (1.737) (1.940) 2009 -4.886*** -6.556*** -4.593*** -5.835*** -5.810*** -4.501* (1.450) (1.637) (1.383) (1.624) (1.854) (2.297) 2010 0.675 -1.098 1.183 0.377 0.362 0.789 (1.482) (1.933) (1.363) (1.242) (1.407) (1.694) 2011 0.267 -1.559 0.696 0.0366 0.0501 0.380 (1.543) (2.075) (1.428) (1.380) (1.632) (1.821) Constant 170.717** 110.693* 210.172*** 248.912** 212.046** 250.095** (59.130) (61.674) (40.617) (89.805) (90.389) (102.984) Observations 187 187 195 149 146 144 R-squared 0.809 0.822 0.812 0.856 0.854 0.804 Number of Countries 14 14 15 13 13 14

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Appendix 3: Robustness; the effect of agglomeration on value added growth (incl. years)

Largest city Agglomeration Urban share AGC production AGC population Zipf’s law

Variables (Value added

growth) (Value added growth) (Value added growth) (Value added growth) (Value added growth) (Value added growth) Agglomeration 0.709 2.237* 0.146 -62.169** 44.915 25.133 (0.753) (1.048) (0.141) (27.764) (74.743) (22.955)

GDP per capita (log, 1 lag) -21.516 -22.892** -17.953 -14.173 -11.907 -20.368

(13.686) (9.706) (11.541) (11.361) (10.630) (11.985)

Human capital -2.686 0.517 -5.064 0.145 -1.529 -5.253

(5.926) (5.263) (6.856) (3.021) (3.815) (6.777)

Physical capital (log) 5.022 9.081 -4.807 -6.292** -5.762 -2.057

(12.686) (9.009) (4.067) (2.730) (3.708) (4.133) Unemployment -0.378* -0.452** -0.390** -0.249** -0.228* -0.425** (0.190) (0.187) (0.181) (0.111) (0.112) (0.191) R&D (2lags) 0.669 -0.468 0.957 0.193 0.300 1.241 (1.158) (0.723) (1.100) (0.668) (0.582) (1.179) 1999 1.373*** 1.383*** 1.474*** (0.266) (0.227) (0.235) 2000 1.489 1.488* 1.887** (0.869) (0.748) (0.681) 2001 0.0236 -0.158 0.518 -1.366** -1.709*** -1.090* (1.165) (1.116) (0.780) (0.622) (0.423) (0.515) 2002 -0.451 -0.790 0.0978 -1.902** -2.053*** -0.965 (1.059) (1.037) (0.695) (0.638) (0.633) (0.877) 2003 -0.236 -0.730 0.531 -2.129** -2.157** -0.888 (1.202) (1.208) (0.889) (0.774) (0.798) (1.293) 2004 0.749 0.0169 1.679 -0.770 -0.407 0.563 (1.591) (1.592) (0.983) (0.786) (0.922) (1.187) 2005 0.879 -0.0484 1.860 -0.420 -0.442 0.618 (1.846) (1.885) (1.183) (1.009) (1.126) (1.504) 2006 2.671 1.468 3.893** 1.131 1.119 2.411 (1.940) (2.056) (1.352) (1.133) (1.237) (1.762) 2007 2.603 1.203 4.122** 1.326 1.101 2.377 (2.348) (2.478) (1.671) (1.410) (1.574) (1.955) 2008 0.498 -1.056 1.664 -0.672 -1.052 0.334 (2.572) (2.682) (1.830) (1.716) (1.869) (2.265) 2009 -3.187 -4.805* -1.976 -4.480** -5.012** -3.193 (2.597) (2.638) (2.058) (2.017) (2.134) (2.776) 2010 1.286 -0.451 3.200* 1.111 0.615 1.437 (2.943) (2.937) (1.782) (1.388) (1.481) (2.005) 2011 0.805 -1.018 2.408 0.603 0.157 0.909 (3.086) (3.134) (1.864) (1.512) (1.667) (2.142) Constant 157.918 90.016 242.861** 243.282** 179.916* 228.557* (121.320) (112.063) (93.264) (99.248) (95.486) (112.231) Observations 160 160 168 146 143 141 R-squared 0.702 0.722 0.701 0.751 0.761 0.721 Number of Countries 14 14 15 13 13 14

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1