Location, Urbanization and Economic Performance

(1)

Location, Urbanization and Economic

Performance

Master thesis International Economics & Business Name: David van Ommen

Student Number: 2231093 Supervisor: Prof. dr. S. Brakman Co-assessor: dr. M.J. Gerritse Date: 14-06-2016

Student email: d.w.van.ommen@student.rug.nl

(2)

2 Contents Abstract ... 3 Introduction ... 4 Research ... 4 Literature review ... 5 Localization economies ... 5 Urbanization economies ... 6

Knowledge intensive industries: NEG and Florida ... 7

Policy ... 9

Empirics ... 10

Research question and studied relationship ... 11

Hypotheses ... 13 Method ... 14 Data collection ... 14 Dependent variable ... 15 Independent variables ... 16 Control variables... 18 Model ... 19 Results ... 22 Conclusions ... 24 Policy implications ... 25

Shortcomings and further research ... 26

Further research ... 26

References ... 27

Appendix ... 32

Figure 1: Urbanization degree in The Netherlands ... 32

Table 1: Knowledge intensive-countries with KEI-Rank, from World Bank ... 33

Table 2: Urbanized countries in EU & percentage of urban population, from World Bank 33 Table 3: Explanation variables and abbreviations ... 33

Table 4: Descriptive statistics ... 34

Table 5: Regression hypothesis 1 ... 35

(3)

3

Abstract

In this paper agglomeration economies (urbanization and localization) are looked at on two scale levels. The economic performance of subsidiaries in knowledge intensive industries is tested to see the effect of agglomeration economies. For this the urbanization degree and knowledge intensity of the surroundings of subsidiaries both on the country level and city level are used. Both interpretations of agglomeration economies seem to have a positive effect on the economic performance of the subsidiaries studied. Localization economies are shown to be extra relevant for older and larger subsidiaries, more so than urbanization economies.

Key words: Subsidiary performance, Urbanization economies, Localization economies

(4)

4

Introduction

There are many determinants of economic performance of companies. Differences between companies in this regard can be contributed to intuitive aspects like size and industry. When comparing similar companies the need to look at other aspects come into play. One of these aspects, location, is the center of this research. Locational aspects that influence economic performance are described to be, for example, differences on an institutional level (Jackson & Deeg, 2008; Nunn & Trefler ,2014) or differences in international agreements and trade integrations. (Dreher et al., 2014; Baier et al., 2015).

In this research the relation companies have with their surroundings is the focus. In the literature the different aspects of these surroundings are thought to have influence on

economic performance of companies. Marshall (1890) described the positive effects economic activities in similar industries have on an individual company. Jacobs (1969) build on this by adding that any form of economic activity and their institutional advantages have positive effects. The clustering of economic activity creates a sharing of ideas and knowledge which is essential for innovation according to Porter (2000). And according to Florida (2005), an ultimate manifestation of this clustering is an urban area with its creative environment and tolerant nature. There is no consensus in the academic literature on the localization versus urbanization debate and also no straight answer on this matter from empiric studies (Duranton & Puga, 2004). This research doesn‘t aim to give a definite answer to this question, but rather aims to shine a light on the strength of both effects on different scale levels and on different companies.

Research

Thus, in this paper we are interested in the relation between agglomeration and economic performance. As the literature describes, agglomeration generate positive external effects for companies, with different ideas about the most beneficial surroundings. Panel data from 2004-2015 on foreign subsidiaries in the EU is used in this research to test the different

(5)

5

Literature review Localization economies

Companies have a close connection to their surroundings. In the academic literature this connection has been studied intensively. When the surroundings are more agglomerated and urbanized, companies and workers are more productive (Puga, 2010)

Marshall (1890) was among the first to describe the benefits from agglomeration. He primarily discusses benefits for companies that share the same industry. In his view the benefits came from the sharing of certain resources. The closeness of companies in the same industry makes it possible to share information and also resources such as labor. In essence, this entails the lowering of transaction costs as these costs can be shared. Transport costs are one of the main drivers of location choice according to geographical economics, driving the formation of cities and agglomerations (Brakman et al., 2009). In shorts, external economies of scale in production are available when companies come together. The sharing of transport, infrastructure and other resources makes this possible. Marshall mentions that clusters create external economies of scale in production through the togetherness of firms. Also, because of the close interconnections between the firms, there will be knowledge-spillovers which create clustering of ideas. This can lead to more innovations in an area. The idea that many similar firms benefit from being located in a close vicinity of each other is called localization economies (Marshall, 1890; Henderson, 2006).

(6)

6 institutional climate of a country can also strongly influence the performance and

competiveness of a cluster and its companies. When thinking about these institutions

especially the structure of the government and legal system come to mind. However, among institutions cultural aspects and traditions should also be counted. (Visser & Atzema, 2008). Thus, a shared way of doing business can be extremely important. The ´´rules´´ affect the decision making and the freedom of these decisions for a company in a geographical area. Furthermore, laws and the legal system influence these decisions more directly.

When addressing localization economies and the surrounding debate on this topic, it is sheer impossible to overlook the works of Porter (2000) in this regard. Porter mentions that the performance of companies and countries is anything but evenly and widely spread. One of the reasons he gives for this are the benefits of geographical clusters. These benefits of clustering make the clustering of companies in the same geographical area useful. These clusters are essentially a relatively uneven spread of firms in the same industry. Their existence makes it clear that it is worthwhile for a firm in a certain industry to locate in a specific area. Porter describes four pillars of benefits of geographical clustering in his ‗‘Competitive Diamond‘‘. The main idea is that the dynamic created in a limited geographical area creates competition on a small scale. This will lead to the four corners of the diamond: ―Firm Rivalry‖, ―Demand Conditions‖, ―Supporting Industries‖ and ―Input Conditions‖. Porter‘s main explanation can be formulated as follows: The close proximity of similar businesses will lead to a more competitive environment in which no company can take it easy but is constantly pushed towards innovation (Porter, 2000). In these ways, a location close to a company in the same industry is beneficial.

Urbanization economies

(7)

7 Because relevant knowledge spill-overs can also exist between different industries

urbanization economies are beneficial. The spill-overs of knowledge between companies in different industries strengthen innovation if this knowledge is complementary. In short, the increasing returns come from different companies and their different resources. Jacobs argues that diversification rather than specialization creates stronger externalities that improve productivity. Here the emphasis also is on innovation, as Jacobs argues that the competition for ideas and human capital between different industries acts as a driver for innovation. Puga (2010) describes evidence for urbanization economies and its manifestations in the following way: The fact that there are cities and that there is therefore a clustering of

economic activity in urban areas is evidence that urbanization economies exist. There is more clustering of economic activities then you would expect if there weren‘t any agglomeration economies. Also there is substantial evidence for agglomeration economies in the fact that both rents and wages are higher in cities. The fact that the higher wages can overcome the higher rent prices shows that it is more beneficial to live and work in urban areas.

Puga further describes what these urbanization economies would entail. The urbanization economies share some traits with the benefits for clusters described earlier. But more commonly urbanization economies are associated with benefits due to the availability of certain facilities in cities. These can be shared between companies in a geographically limited area. The same hold for the sharing of a labor pool and of suppliers. The availability of these aspects in an urban area and their reduced costs because of the sharing aspects is the main reason producing in urban areas is beneficial

Duranton and Puga (2004) suggest that the main benefits from urbanization economies come from sharing, matching and learning mechanisms present in agglomerated areas. They attribute to the growth of cities and to the strong performance of companies located there, through the heterogeneity of both companies and workers. The authors state that although empirical studies do not solve the debate between urbanization and localization economies, cities have a wide variety of firms in size, productivity and industry. This indicates the presence of urbanization economies.

Knowledge intensive industries: NEG and Florida

(8)

8 monopolistic competition model as described by Dixit and Stiglitz‘ (1977). An important variable in the model is the costs of trade. These costs mean that trading over distance comes with extra costs, dependent on the size of the trade. An assumption crucial in these models is that firms can select a location, because certain locations are more beneficial. Because of increasing returns, companies have an incentive to locate to certain locations. This is related to the fact that the locations of firms are related to the location of demand. The NEG assumes the location of demand to be endogenous (Krugman,1998). So, choice of location is vital and used especially by more innovative companies stimulated by tacit knowledge and trade costs. This gives extra weight to the importance of the agglomeration economies for knowledge intensive industries.

Florida (2005) sees an important role for knowledge intensive and creative sectors in the economic stimulation of urban regions. He sees creativity as the main driving force for the economic and urban development. This creative sector is growing both in size and in importance. According to Florida this ‗‘creative class‘‘ encompasses people engaging in technological and scientific activities as well as art and cultural activities. The attraction of creative capital in this regard is seen as the key to growth in productivity, the bettering of living conditions and sustainable development (Florida ,2005).

Florida sees a clustering of the creative class. The idea is that this clustering will attract a clustering of companies as well. Then the earlier mentioned benefits in clustering will come to fruition. The most important argument is that the jobs follow the people in this idea: The creative class (the people) moves to a tolerant and diverse location and the companies follow. In this idea these companies are knowledge intensive and follow the human capital relevant to their respective sector.

(9)

9

Policy

A strong link between agglomeration and productivity as described by Porter (2000), leads him to suggest policy regarding this. This idea to generate ‗‘clusters‘‘ in certain areas to improve the competiveness of areas is called ‗‘cluster policy‘‘. Some authors argue for these policies. Institutions and government spending and subsidies can create clusters, for example by investing in physical and digital infrastructure. This makes it beneficial to locate a

company to a certain area. Because these clusters are constructed artificially its long term existence and benefits to the whole country can be questioned (Kline & Moretti, 2014). This is a relevant problem in assessing the casualization of the benefits of clusters

However, in the academic literature there is a lot of criticism on the cluster policies (Brakman & van Marrewijk, 2013). For example, Duranton (2011) has strong criticism on some of the assumptions made by authors in the cluster literature and subsequently on their advice on cluster policies by governments. Duranton describes clusters as ‗‘a complex second-order issue that wrongly receive first-order attention‘‘. According to him clusters are associated with market failure. Therefore there will be local economic development policies needed to deal with the inefficient outcomes of clustering. However, as the benefits of the local composition of economic activity are not clear policies should not focus on clustering. Duranton sees clustering as an intermediate outcome and not as a recipe for success or guaranteed driver of prosperity. The idea that clusters should be the focus of local economic development is therefore disregarded as they can be difficult to implement correctly, while the benefits are unclear.

Furthermore, the ‗‘evolutionary economic geography‘‘ describes that successful clusters develop organically and evolutionary. A policy cannot successfully imitate this process. (Coe, 2010; Boschma & Frenken, 2006; MacKinnon et al., 2009). This theoretical idea explains the spreading of economic activity by looking at the historical perspective of

(10)

10 The main idea from this relevant to this research is that the model from Arthur(1987)

incorporated in the evolutionary economic geography theory, describes that the agglomeration effects only come into play when a agglomeration reaches critical mass. The formation and the growth of these agglomerations depend on coincidence related to the spin-offs of the existing companies. Only when the agglomeration effects develop in an area, companies locate in this area. The development of successful agglomeration do not necessarily form when companies locate in an area, but only if these companies become successful and more successful companies spin off. Therefore, a policy based on clustering does not necessarily have the desired effect.

Empirics

Most empirics studies show agglomeration economies to be present ( Ciccone and Hall ,1996; Rosenthal and Strange ,2001; and Ciccone (2002). However, consensus on the contribution of urbanization or localization economies the empiric literature does not provide (Duranton & Puga 2003). Carlino and Kerr (2015) provide an extensive overview on empiric literature showing the relation between agglomeration and innovation, which studies show to be positive. Combes & Gobillon (2014) conclude in their overview on empirical studies on agglomeration economies that ‗‘Most of the literature identifies the overall impact of local determinants of agglomeration economies but not the role of specific mechanisms that generate agglomeration effects‘‘ ( Combes & Gillion, 2014). Furthermore, Head & Mayer (2015) provide an overview of the empirics regarding the NEG, with the conclusion that the literature has generated mixed results regarding this idea.

Empirical research such as Ciccone (2002) and Brulhart& Mathys (2007) find positive relationships between agglomeration and labor productivity, while Rosenthal and Strange (2001) find a positive relation between innovation and agglomeration.

(11)

11

Research question and studied relationship

The premise that both types of agglomeration effects have a positive effect on economic performance is tested on different scale levels. On the country level the countries are differentiated between based on knowledge intensity and urbanization. Then the focus will zoom in to the community or city-level. Both localization and urbanization economies are tested on these scale levels. On the country level localization economies are looked at by testing the relationship between knowledge intensive countries and knowledge intensive industries. Then urbanization economies are tested on the country level by looking at the urbanization degree of countries and the performance of companies. On a smaller scale urbanization economies are tested by looking at companies in urban communities and their performance. Localization economies are tested by looking at companies in urban

communities in the knowledge intensive countries.

Thus, this paper will also address the scale of these agglomeration effects. The data provides the opportunity to look at the different scale levels. Regarding these scale levels, the

expectation is to see a stronger influence of agglomeration economies when looking at a smaller scale.

For the testing of the localization economies, knowledge-intensive industries are addressed. The extra relevance for these sectors was already mentioned by looking at the NEG and Florida‘s creative class. When talking about geographical area´s and clustering, tacit

knowledge sharing jumps out as one of the major benefits for clusters. However, knowledge sharing will not be equally important overall all industries. It should be intuitive that

industries with the propensity to place more value on knowledge benefit more from tacit knowledge sharing and therefore from clustering. Because industries differ in knowledge-intensity, the determination of an industry for research on clusters is vital (Gupta &

Govindarajan, 2000). The focus will lie on knowledge intensive industries, as their knowledge use is more specific and therefore these industries are more likely to depend on tacit

(12)

12 will see a higher economic performance when located in a knowledge-intensive cluster.

(Egeln et al. 2004).

In short, the agglomeration economies are more relevant for knowledge intensive companies. Not in the least because the level of competition is higher in these sectors. In general, highly specialized, high- tech companies and companies in related service sectors show more competition. The concentration of these types of industries in a geographically area will therefore create a more competitive environment. These will lead to more innovation and therefore Athreye & Keeble (2002) argue that the innovation of firms will be stronger in locations with a concentration of these industries. The strengthening of the competitiveness in a region is helped by the technological infrastructure as well as the innovative environment. Areas where knowledge intensive companies are concentrated need institutions and

infrastructure to stimulate the innovative environment. A company‘s performance and capabilities depend on the level of these institutions in the direct geographical surroundings (Frost, 2001). These innovations are stimulated by spillovers in the area between the knowledge intensive firms. (Cantwell & Lammarino, 2000).

On the other hand, the strength of competition on a small scale in a particular industry does not necessarily have a positive influence on all companies involved. The performance of a newly formed company could does not benefit from strong competition. Kwok & Yin (2013) argue that this is especially the case in knowledge intensive industries because of the high startup costs in these sectors. The costs of the equipment, technology and human capital are higher and can therefore limit growth in the starting phase of a company and the performance of a company. Therefore, new companies in these sectors have problems to overcome, but should eventually benefit more from being located in an urban area or city because of the positive effects of clustering.

For this research, a selection of all companies in knowledge intensive sectors would be too broad. The size of the sample would be too large and there would be difficulty in comparing different types of companies. Therefore we will limit the sample by size by looking at foreign subsidiaries. This is done to be able to compare the companies internationally (Frost,2001). With subsidiaries we can assume that a conscious location choice has been made, as the parent company has now emotional or historic bond with a specific area and can be assumed to make the decision based on economic motives. This is an important factor for the

(13)

13 Another reason subsidiaries are chosen is their relation with their respective geographical surrounding area, a relation this research is interested in. As opposed to other types of companies , such as for example multinational companies, the location of a subsidiaries is restricted to one specific country or area. Therefore they are more rooted in a geographical area and can be assumed to give more relevant information about the influence this area and its characteristics has on the company. Furthermore, subsidiaries are assumed to be more homogenous in size, structure and other relevant aspects and differ less in the nature of their relationship with their surroundings as opposed to multinationals (de Jong et al, 2011). Thus, data on subsidiaries provide better means for comparing companies based on geographical and locational aspects because of their similarities in aspects relevant for this research.

Hypotheses

Following this our hypotheses are constructed. The first hypothesis concerns itself with the localization economies on the country level:

1: Being located in an urbanized country has a positive effect on the economic performance of foreign subsidiaries in knowledge intensive sectors.

There is an expectation that companies in a knowledge intensive sector that are located in urban or agglomerated areas outperform other companies in these sectors. The level of

urbanization in a country according to the World Bank is used as a first proxy of the extent to which the company is embedded in an urban area. A subsidiary in an urbanized country should therefore have an advantage over other subsidiaries.

2: Being located in a knowledge intensive country has a positive effect on the economic performance of foreign subsidiaries in knowledge intensive sectors.

(14)

14

3: Being located in an urbanized area has a positive effect on the economic performance of foreign subsidiaries in knowledge intensive sectors.

This hypothesis focuses on a smaller scale of urbanization, as is perhaps more useful when studying this phenomenon. The relation studied here is similar to hypothesis 1, but the focus lies on the urbanization degree of the ‗‘city‘‘ of location instead of the country. This poses a greater challenge for the gathering of data, as the city of every country needs to be linked to their respective urbanization degree. However, this information will be more specific and more robust in testing the relationship. The locations of the subsidiaries and the urbanization degree of these locations will be tested to see their influence on the economic performance of the companies in the data sample.

4: Being located in an urbanized area in a knowledge intensive country has a positive effect on the economic performance of foreign subsidiaries in knowledge intensive sectors.

This hypothesis tries to look at localization economies on the smaller scale as it studies companies in urban areas in knowledge intensive sectors, but only in country where these knowledge intensive sectors are abundant. Urban areas in knowledge intensive countries are assumed to have agglomerated companies in knowledge intensive industries. Thus,

localization economies are captured when studying companies operating in these sectors in these areas. Essentially data used in hypothesis 2 and 3 are paired, as specific data on the knowledge intensity of urban areas is not available.

Method Data collection

(15)

15 This is data that can all be found in the Orbis database, containing information about

companies worldwide. In this database, we make a selection of relevant subsidiaries and the needed information about these companies. Differentiation on knowledge intensive industry is made using NACE. Rev codes. Companies defined by these codes to operate in knowledge intensive industries or high-tech industries are selected in the database.

The data from Orbis contains information on the respective locations of the subsidiaries, on different levels of scale. The selection of subsidiary companies in this database is limited to data for companies in the EU. This is because the sample needs certain limits, but mostly because of the comparability of the data on companies in the EU. Data from Eurostat will be paired with data from Orbis and the integration of the two databases is helped by conformity of the data on the country and city level. Eurostat has data available on the urbanization degree and knowledge intensity of countries and cities. This means we use data on subsidiaries from the 27 countries in the EU. Croatia, as the newly appointed 28th EU member, is excluded because of the lack of relevant data in Eurostat.

Furthermore, as time specific data will be used, the subsidiaries need to have at least data available on one extra year other then 2015. The database from Orbis contains information over a 10-year span. Therefore the database for this research contains data on the operating revenue of foreign subsidiaries from 2004-2015. The idea behind the use of time specific data is that is allows for a control of variables that influence the economic performance of

companies that have changed over time. New international guidelines, agreements and other similar changes can be controlled for when using time-specific data. Hill et al. (2012) suggest using time-specific data when there are individual differences between companies that are difficult to capture in a econometric model, as these differences can be accounted for by incorporating the data from previous years of these companies.

Dependent variable

(16)

16 are the most important factor for this study. Indicators that relate to accounting tools are less relevant because they illustrate the performance of managers and the strategic behavior of the company (Selling and Stickney, 1989). Indicators that capture the linkages with the

surroundings use in empiric analyses are mostly sales-based (Tomohara and Yokota, 2013). Therefore for this research the operating revenue is chosen, as it is sales-based and available in the Orbis database. This variable contains the ‗‘net sales income form the main business operations‘‘. This variable is chosen because it is a relative good proxy for the total sales from a company that can be compared between different subsidiaries. Furthermore, the benefits and advantages a subsidiaries receives from its location are reflected in the total sales and this variable thus contains the locational advantages a subsidiary has. Given the use of this

variable, subsidiaries are only selected when there is data on their operating revenue available. The natural logarithm of this variable is taken to function as the dependent variable in the regression analysis. This is done to make the operating revenue of the subsidiaries follow and normal distribution and not to have much outliers. In this way, the results will be more reliable.

Independent variables

This selection of data needs to be paired with data about the respective locations and their characteristics. These will serve as the main explanatory variables. Data provided by Eurostat (2016) and the World Bank (2012) ;( 2016) is used for this. This data contains information about the respective locations of the subsidiaries, on different levels of scale. For hypothesis 1 data on urbanization on the country level is used. The explanatory variable used as a proxy for this will be the urbanization degree of countries according to the World Bank (2012). This data from the World Bank is used and paired to the Orbis data on the country level. In the appendix, table 2 shows the 5 most urbanized countries in the EU according to the World Bank. A dummy variable is created to incorporate subsidiaries located in these urbanized countries. This will serve as the main independent variable to test hypothesis 1.

(17)

17 intensive countries according to the World Bank. A dummy variable is created for

subsidiaries located in a country that is knowledge intensive. This will serve as the main independent variable to test hypothesis 2.

For hypothesis 3, data on urbanization on a smaller scale is needed. On the city level Eurostat provides data on every community in the EU, as the smallest scale on which information is available. It contains information about the urbanization degree of each of these communities. This degree is divided in three categories, with the most urbanized communities being in category 1. Communities in this category one will be sees as located in an urban environment in this research. Orbis has information about the city or village a company is located in. The data form these sources needs to be paired. This poses a challenge as there are significant differences in the categorization between both data sources. Data from Eurostat is on the municipality level for some countries and on city level for the others. Orbis provides data on the city level and not on the municipality level. Therefore municipality level data and city level data from the two data sources has to be paired manually. Having done this, a dataset is created with data on each subsidiary and the degree of urbanization of their direct

surroundings. As an example on this degree of urbanization, the appendix contains figure 1. This figure is a map depicting the urbanization degree in The Netherlands according to the categories from Eurostat, constructed with ArcMap. It describes the scale on which

differentiation is made in this research (N.B. for some of the countries in the EU information was available and used on an even smaller scale, but it was never larger than an local

equivalent of the communities seen on this map). A dummy variable based on the city level is created. As mentioned, the database contains information on the degree of urbanization of each community in the EU. The dummy variable will contain the subsidiaries located in communities which fall in category 1 of the degree of urbanization. This variable will serve as the main explanatory variable for hypothesis 3.

For hypothesis four data on the knowledge intensity of the community level is needed. As this kind of data is not available in Eurostat, a proxy for this is constructed with data that is

available. This is done by pairing data used for hypothesis 2 and 3. On the community level subsidiaries are differentiated between. Then they are differentiated between on the

(18)

18 created that contains subsidiaries in these areas. This dummy will be the main explanatory variable for hypothesis 4.

Control variables

In addition, more information about each of the subsidiaries is needed as they will tell more about the subsidiary and can be uses as control variables in order to have the model explain more of the differences in economic performance. The age of a subsidiary is assumed to be a determinant of the economic performance of this subsidiary, as the years that the subsidiary has been active is an indicator of the success of this subsidiary and of the knowledge they have gathered over the years in the market(Sapienza et al, 2003).. This is even more so for companies in knowledge intensive industries because of the presumed flatter learning curve of new companies these industries and their reliance on the surrounding companies,

institutions and resources (Sorenson & Stuart, 2000). The age of the subsidiary is also expected to increase the studied relationship and therefore will their individual influence on the relevant relationship will be studied. The age of the subsidiary is simply 2015 minus the year of incorporation and is concluded in the data set as a control variable.

The size of the subsidiary will also serve as a control variable. The larger the subsidiary the larger the economic performance, as larger firms have stronger resources and capabilities (Noorderhaven & Harzing, 2009). The size of the subsidiary is based on the distinction used in the Orbis database. There are four categories given, namely: very large, large, medium and small. As a control variable a dummy variable for a large subsidiary is created. This will contain subsidiary in the very large and large category.

Another variable that is supposed to have effect on the economic performance is the earlier economic performance for a subsidiary. In this research, the econometrics of the time specific data is based on Hill et al. (2012). A panel data set is created in order to incorporate the time specific effect. This way we have a dataset that includes different observations for every year of available operating revenue per subsidiary. Then, dummy variables based on time are created to differentiate between the years. This way it is possible to look at each individual year and revenue in that year and measure its effect on the revenue in 2015. In this manner the time specific data are analyzed and the issue of panel data is tackled.

(19)

19 Eurostat. However, Eurostat differentiates between these sectors. Based on this, a dummy based on subsidiary in ‗‘high-tech‘‘ sectors and a dummy for ‗‘high knowledge intensive‘‘ sectors is created. Furthermore dummies are created for the ten most knowledge intensive sectors are created, based on NACE codes available in Eurostat.

Lastly, several variables that incorporate cross effects are created. As mentioned, larger subsidiaries have stronger resources and more resilient capabilities. It could be interpreted that subsidiaries that are larger a less dependent on their direct surroundings in terms of knowledge spill-overs and sharing and overall less influenced by both positive and negative externalities(Noorderhaven & Harzing, 2009). Therefore the cross effect between country of location and size is added as control variable.

A similar argument can be made for older subsidiary. They have gathered knowledge, capabilities and linkages over the years and are therefore more successful(Sapienza et al, 2003), and can also be thought of to be less influenced by both positive and negative externalities. Thus, the cross effect between country of location and age is also used as a control variable. the names of the variables used in this research and what they incorporate can be found in table 3 in the appendix and the descriptive statistics can be found in table 4 of the appendix.

Model

Now that the relevant variables are available, the model to test the hypothesis is created. The dependent variable ‗‘operating revenue‘‘ will be explained in a regression analysis using the independent variables mentioned. For this the following baseline model will be used:

LnOperatingRevenue 𝑖,𝑡= 𝛽0+ 𝛽1locationdummy𝑖+ 𝛽2Timedummies+ 𝛽3Size𝑖 + 𝛽4Age 𝑖+

𝛽5Industrydummies𝑖 + 𝛽6Hightechdummy𝑖

+𝛽7Knowledgeindustrydummy𝑖+𝛽8Countrydummies𝑖 +𝛽9Size*country + 𝛽10Age*country𝑖 (1)

With time dummies meaning the time dummies for every year in the panel dataset, and industry dummies for the different industries in the sample.

Where 𝑖 = id of the subsidiary.

(20)

20 explanatory variables for their relevant hypothesis. Following the reasoning of Brullhart & Mathys (2007), a panel dataset is used in this research. They argue that a relationship between density and productivity warrant a panel data set to overcome a causality problem. The

inclusion of time specific data controls for the locational characteristics of an geographical area that could benefit economic performance or agglomeration. A way to overcome this is to add time dummies to a pooled OLS (Hill et al.,2011) to control for earlier performance of the subsidiaries.

For hypothesis 1, the main explanatory variable is the dummy variable for being located in an urbanized country. Subsidiaries in the 5 most urbanized countries are captured. The country dummies used in the regression analysis for this hypothesis will therefore be for these urbanized countries, in order to be able to see differences between these countries. One of these countries cannot be included in the regression because of the dummy trap (Hill et al.,2011). Because of this the dummy for Luxembourg is excluded. Furthermore the industry dummies are added with the extra specification for operating in high-tech or knowledge intensive sectors. Lastly the mentioned cross effects controls are added. For hypothesis 1 this gives us the following specification:

LnOperatingRevenue 𝑖,𝑡= 𝛽0+ 𝛽1UrbanizationCountrydummy𝑖+ 𝛽2Timedummies+ 𝛽3Size𝑖 + 𝛽4Age 𝑖+ 𝛽5Industrydummies𝑖 + 𝛽6Hightechdummy𝑖

+𝛽7Knowledgeindustrydummy𝑖+𝛽8MLT 𝑖+𝛽8NLD 𝑖+𝛽8FRA 𝑖+

𝛽8DEN𝑖+𝛽9Size*country𝑖+ 𝛽10Age*country𝑖 (2)

For hypothesis 2, a different locational dummy based on country is used. Here the interest lies in countries that have an abundance of knowledge intensive industries and as such are

(21)

21

LnOperatingRevenue 𝑖,𝑡= 𝛽0+ 𝛽1KnowledgeIntensiveCountrydummy𝑖+ 𝛽2Timedummies+

𝛽3Size𝑖 + 𝛽4Age 𝑖+ 𝛽5Industrydummies𝑖 + 𝛽6Hightechdummy𝑖

+𝛽7Knowledgeindustrydummy𝑖+𝛽8FIN𝑖+𝛽8SWE 𝑖+𝛽8NLD 𝑖+ 𝛽8DEN𝑖+𝛽9Size*country𝑖+

𝛽10Age*country𝑖 (3)

For hypothesis 3 the locational explanatory variable is constructed on a lower scale. The relevant variable is the urbanization degree of the direct surroundings of the subsidiary. The dummy variable for being located in a highly urbanized area is the main explanatory variable in this regard. Furthermore, as there is no specific differentiation made on country level, dummies for all 27 (N.B.26 are included because of the dummy trap) countries in the sample are used. This gives the following specification for hypothesis 3:

LnOperatingRevenue 𝑖,𝑡= 𝛽0+ 𝛽1KnowledgeIntensiveCountrydummy𝑖+ 𝛽2Timedummies+

𝛽3Size𝑖 + 𝛽4Age 𝑖+ 𝛽5Industrydummies𝑖 + 𝛽6Hightechdummy𝑖

𝛽10Age*country𝑖 (4)

With time dummies meaning the time dummies for every year in the panel dataset, industry dummies for the different industries in the sample, and country dummies for the different countries in the sample.

For hypothesis 4 the explanatory variable based on location incorporates subsidiaries in urban areas in knowledge intensive countries. Therefore a dummy variable for these subsidiaries is used. In this regression the country dummies of the knowledge intensive countries will be used, similar as for hypothesis 2. This gives the following specification:

LnOperatingRevenue 𝑖,𝑡= 𝛽0+ 𝛽1UrbanAreaInKnowledgeIntensiveCountrydummy𝑖+

𝛽2Timedummies+ 𝛽3Size𝑖 + 𝛽4Age 𝑖+ 𝛽5Industrydummies𝑖 + 𝛽6Hightechdummy𝑖

(22)

22 With time dummies meaning the time dummies for every year in the panel dataset, and

industry dummies for the different industries in the sample. Where 𝑖 = id of the subsidiary.

Per hypothesis the regressions tables will include five different ‗‘models‘‘, showcasing the different effects. The first model will include the different industry dummies, who will be divided in high-tech and knowledge intensive in the next two models, to show the differences between sectors. Then the fourth and fifth model will include the different cross effects respectively.

A first look at the data shows the potential for certain outliers. Looking at the data,

subsidiaries with an exceptional high age can be found. These types of observation can have a disruptive impact on the model that has incorporated age as one of its explanatory variables. Furthermore, these outliers seem counter intuitive as subsidiaries with an age of over 100 years seem unlikely and not relevant to this research. To account for this, subsidiaries with an age of over 100 years are disregarded. A first look at a regression still shows some values to have a disproportional effect on the results. These observations have high residuals in this regression. To account for this a robust regression will be used in this research, downgrading the effect of observations with a large residual (Berk, 1990). This will also help with issues of heteroskedasticity that the data shows signs of (White test : p (0,000). A regression with robust standard errors should be used for this kind of data (White, 1980).

Before the actual regressions can be done, the dataset is studied to see potentials problems or violations of the model (Hill et al., 2012). First the collinearity and correlation of the data is looked at. A matrix of correlation can identify multicollinearity. The matrix for our dataset shows that no values have a perfect correlation, as the matrix has no values close to 1 (Hair et al, 1995). Because of the size of this matrix, it is not included in this appendix. To assure that the multicollinearity does not exist in this sample, the variance inflation factors (VIF) are checked to see if they are greater than 10 (Hill et al., 2012). These values do not pass this border value and therefore multicollinearity is assumed to not be an issue in this dataset.

Results

(23)

23 is positive at the 1% level. Therefore hypothesis 1 seems be supported by the data. The

control variables all show to have a significant effect as well. The effect of the variable for urban countries is similar throughout the five different models tested. The second and third models makes distinction between the knowledge intensive industries the subsidiaries operate in. they show that high-tech subsidiaries outperform knowledge intensive subsidiaries in urban countries. The fourth and fifth model include the cross effects between size and country and age and country. These effects also are shown to be significant. Although these effects are small, they raise the r-squared and adjusted r-squared value relatively strong. The r-squared value is 0.47 for the first three models, and 0.52 when the cross effects are added. These values differ only slightly throughout the regression tables of the different hypotheses. The regression analysis for the second hypothesis can be found in table 6 of the appendix. This table is fairly similar to the table for the first hypothesis as there are similarities in the setup of these hypotheses. This analysis shows that the dummy variable for a location in an ‗‘knowledge intensive‘‘ country is positive at the 1% level. Thus, the second hypothesis seems to be supported by the data as well. When comparing these two hypotheses and regression analyses, overall the models show strong similarities with the models for hypothesis one. This is as expected as only the country dummy variable is different.

However, the effect of the variable for knowledge intensive countries shows to be less strong then the variable for urban countries in table 5, with coefficients of 0.51 vs 0.10. On the other hand, the effect of the variable for knowledge intensive countries shows to be stronger when the cross effect is added, indicating that large and older subsidiaries see relatively more benefits from being located in a knowledge intensive economy as opposed to larger and older subsidiaries in urbanized countries.

The regression analysis concerning hypothesis 3 can be found in table 7 of the appendix. This regression focuses on the smaller scale of the urban environment. At this scale level, the models show significant results at the 1% level for being located in an urban community. This indicates that across the EU, subsidiaries in urban environments outperform other subsidiaries. The strength of this relation is smaller than the relation on country level regarding urbanization (coefficient: 0.14). This could be due to other factors in the 5

(24)

24 The regression analysis for the fourth hypothesis, found in table 8 in the appendix, shows a positive effect from being located in an urban community in a knowledge intensive country. This effect is only a little stronger than the effect from hypothesis 3, indicating the strength of urbanization on the small scale across the EU regardless of country. The effect is stronger when the cross effects are added, on par with the results from hypothesis two. Thus, for larger and older subsidiaries an increase in strength of the relation can be seen when these

subsidiaries are located in urban communities in knowledge intensive countries. When comparing hypotheses 2 and 4 in table 8 and table 6, the results show a small increase in strength for subsidiaries in urbanized communities in knowledge intensive countries as opposed to all subsidiaries in these countries.

Overall the studied effects are not very strong compared to other determinants of economic performance, when looking at the coefficients. This is also shown by the fact that the R-squared value changes only slightly when differentiating between the different locational variables. The R-squared value also shows that still a part of the differences in economic performance are not explained by the variables in the model.

Conclusions

Following these results it can be concluded that location matters for the economic performance of international subsidiaries in a knowledge intensive sector. This result is intuitive and was expected. The main questions in this research concerns the way location matters. In this regard this research has brought forward some characteristics of locations that should positively influence the economic performance of subsidiaries. Following this, this paper concludes that a location that is knowledge intensive, urbanized or both benefits the performance of the local subsidiaries. Both localization and urbanization economies seem to be present on both of the studied scale levels.

(25)

25 effect seems to come from the knowledge intensity from the surroundings and is downplayed by the urbanization of the surroundings

As both knowledge-intensive areas and urban areas show to have positive influences on the subsidiaries, it seems that agglomeration effects (Marshall, 1890; Henderson, 2006) are in effect in for these subsidiaries in knowledge intensive sectors. These agglomeration effects come from clustering effects of these sectors and subsequent innovations (Porter, 2000). The age and size of subsidiaries is a strong determinant of economic performance, as expected (Sapienza et al, 2003).

As knowledge intensive industries have higher start-up costs, their performance lacks behind other sectors (Kwon & Yin, 2013). This idea is backed by the stronger performance of older and larger subsidiaries and even stronger performance of older and larger subsidiaries in knowledge intensive countries. In this research a distinction between high-tech companies and high-knowledge intensive companies has been made. In all regression models the high-tech companies are able to overcome this disadvantage due to agglomeration effects, while high-knowledge intensive companies are unable to do so.

Because also urbanized areas show to have a positive effect on economic performance, the ideas about cities by, among others, Jacobs (1969) and Florida (2005) seem to hold true in some extent. For urbanization economies, the focus in this research was the urbanization degree of localities on a small scale. As subsidiaries in these areas show better results, the positive effects or urban areas are clear. These results hold over the whole of the EU, overcoming differences between countries.

Policy implications

What also shows in the results of this research is the minimal strength of the locational effects. Although the proposed effects are all found to be positive, other characteristics of companies are found to explain more of the differences in economic performance. Age, size and industry are examples of these explanatory variables included in the model in this research. Already mentioned is the conclusion that ideas from cluster literature,

(26)

second-26 order issue that wrongly receive first-order attention‘‘. This statement seems accurate given the results of this research.

Shortcomings and further research

In this paper the urbanization degree of a country is used as a proxy for the level of

urbanization a subsidiary is located in. Countries across the EU differ in other regards then just the urbanization degree and knowledge intensity. It is of course no surprise that the countries in the EU are vastly different. Especially new member-states lack behind in

development of infrastructure and institutional quality. This research controls for differences in countries in order to deal with this and uses the knowledge intensity of the countries where urban areas are located as a control. A suspicion is that other factors in these urbanized

countries play a role in determining economic performance and therefore the relation between urbanization and economic performances is overstated. If all countries in the EU were more or less the same, zooming in on a smaller scale level would have had stronger effects on the economic performance. When looking at the different scale levels used in this research it shows that on the country level the results are stronger.

Furthermore, ideally data on the knowledge intensity of every urban area would be used in this research. This data would shine a stronger light on differences between localization and urbanization economies. Because of the lack of availability of this kind of data, only

information on the knowledge intensity of countries is used to differentiate between urban areas.

Further research

One of the main findings of this research is that the data sample shows that localization economies matter even more when subsidiaries are older and are larger, while for

(27)

27

References

Arthur, W. (1987) Urban systems and historical path dependence, in Ausubel, J. & Herman, R. Cities and their vital systems: 85-97. Washington : National Academy Press.

Athreye, S., Keeble, D. (2002).Sources of increasing returns and regional innovation in the UK Regional Studies, 36, 345–357

Audretsch, D.B., Keilbach, M.C Lehmann, E.E.(2006). Entrepreneurship and economic

growth. New York: Oxford University Press.

Baier, S. L., Bergstrand, J. H. and Feng, M. (2014). Economic Integration Agreementsand the Margins of International Trade. Journal of International Economics, 93(2), 339 – 350. Bathelt, H., Malmberg, A., Maskell, P. (2004) Clusters and knowledge: local buzz, global pipelines and the process of knowledge creation. Progress in Human Geography, 28(1): 31– 56.

Berk, R. (1990). A primer on robust regression. In Modern Methods of Data Analysis, ed. J. Fox and J. S. Long, 292–324. Newbury Park, CA: Sage.

Boschma R., Frenken K., (2006). Why is economic geography not an evolutionary science?

Journal of Economic Geography, 6(3), 273-302.

Boschma, R. A., Wenting, R. (2007). The spatial evolution of the British automobile industry. Does location matter? Industrial and Corporate Change ,16, 213–238.

Boschma, R., Frenken K. (2011). The Emerging Empirics of Evolutionary Economic Geography. Journal of Economic Geography, 11, 295-307.

Brakman, S., & van Marrewijk, C. (2013). Reflections on cluster policies. Cambridge journal

of regions economy and society, 6(2), 217-231.

Brakman, S., Garretsen H. & van Marrewijk, C. (2009). The New Introduction to

Geographical Economics. Cambridge: Cambridge University Press,

Brülhart, M. & Mathys, N. (2007)., Sectoral Agglomeration Economies in a Panel of European Regions

(28)

28 Carlino, G. &Kerr, W. (2015) Agglomeration and Innovation, Bank of Finland Research

Discussion Paper, 27

Ciccone, A. & Hall, R. (1996), Productivity and the Density of Economic Activity, American

Economic Review, 86(1), 54-7

Ciccone, A., (2002), Agglomeration Effects in Europe, European Economic Review, 46(2), 213-227.

Coe N. (2011). Geographies of production I: An evolutionary revolution?, Progress in

Human Geography, 35 (1) , 81-91.

Combes, P. & Gobillion (2015) The empirics of agglomeration economies, in: Duranton, G., Henderson, V. and Strange, W. Handbook of Urban and Regional Economics : 247-348. Amsterdam: Elsevier.

De Jong, G., Phan, T.B. and van Ees, H. 2011. Does the meta-environment determine firm performance? Theory and evidence from European multinational enterprises. International

Business Review, 20: 454–465.

Dixit, A. & Stiglitz, J. (1977). Monopolistic Competition and Optimum Product Diversity.

American Economic Review, 67 (3), 297-308.

Dreher, A., Mikosch, H., Voigt, S.(2015). Membership has its Privileges — The Effect of Membership in International Organizations on FDI. World Development, 66, 346 – 358. Duranton, G. (2011) California Dreamin‘: The Feeble Case for Cluster Policies. Review of

Economic Analysis ,3 , 3–45

Duranton, G. and Puga, D.(2004). Micro-foundations of urban agglomeration economies, Handbook of Regional and Urban Economics, in: Henderson J. & Thisse J. (ed.), Handbook

of Regional and Urban Economics :2063-2117. Amsterdam: Elsevier.

Egeln, J.,Gottschalk, S.,Rammer, C. (2004). Location decisions of spin-offs from public

research institutions. Industry and Innovation, 11(3): 207–223.

(29)

29

http://epp.eurostat.ec.europa.eu/cache/ITY_SDDS/en/htec_esms.html. Accessed on

01-04-2016

Eurostat; Eurostat indicators on High-tech industry and Knowledge - intensive services

http://ec.europa.eu/eurostat/cache/metadata/Annexes/htec_esms_an8.pdf. Accessed on 01-04-2016

Florida, R. (2005). Cities and the Creative Class. New York: Routledge

Frost, T.(2001). The geographic sources of foreign subsidiaries‘ innovations Strategic

Management Journal, 22, 101–123

Glückler J (2007) Economic geography and the evolution of networks. Journal of Economic

Geography, 7, 619-34

Gupta, A.K., Govindarajan, V. (2000). Knowledge flows within multinational corporations.

Strategic Management Journal, 21 (4), 473–496

Hair Jr., J., Anderson, R., Tatham, R., and Black, W. (1995). Multivariate data analysis, 4th Ed. New Jersey: Prentice Hall.

Henderson. J.V. (2003). Marshall´s scale economies. Journal of Urban Economics, 53, 1–28. Hill, R.C., Griffith, W.E.,Lim, G.C. (2011). Principles of Econometrics, 4th edition,

International Student Edition, New York: John Wiley & Sons

Jackson, G, Deeg, R. (2008). Comparing Capitalisms: Understanding Institutional Diversity and its Implications for International Business. Journal of International Business Studies, 39, 540–561.

Jacobs, J. (1969). The economy of cities. New York: Random House.

Klepper, S. (2007) Disagreements, spinoffs, and the evolution of Detroit as the capital of the U.S. automobile industry, Management Science, 53(4),616-631.

Krugman (1998) What's new about the new economic geography? Oxford Rev Econ Policy 14 (2): 7-17

(30)

30 Li, L. (2007). Multinationality and performance: a synthetic review and research agenda.

International Journal of Management Reviews, 9(2), 117-139.

MacKinnon, D.; Cumbers, A.; Pike, A.; Birch, K.; and McMaster, R. (2009). Evolution in geography: Institutions, political economy, and adaptation. Economic Geography 85,129– 150.

Marshall, A. (1890) Principles of Economics: An introductory volume, London: Macmillan and Co.

Moretti, E., Kline, P. (2014). Local Economic Development, Agglomeration Economies, and the Big Push: 100 Years of Evidence from the Tennessee Valley Authority, The Quarterly

Journal of Economics, Oxford University Press, 129(1), 275-331

Nunn, N., Trefler, D. (2014). Domestic Institutions as a Source of Comparative Advantage. In Rogoff,K., Helpman ,E. Gopinath,G., (Eds.), Handbook of International Economics ,Vol. 4

of Handbook of International Economics: 263-315. Amsterdam: Elsevier.

Porter (2000). Locations, clusters and company strategy. In Clark, G.L., Feldman, M.P., Gertler, M.S. ,(Eds.) (2000), The Oxford handbook of Economic Geography : 49-61. New York : Oxford University Press.

Puga, D. (2010), The magnitude and causes of agglomeration economies. Journal of

Regional Science, 50: 203–219

Rosenthal, S. & Strange W. (2001), The Determinants of Agglomeration, Journal of Urban

Economics 50, 191-229.

Selling, T. and Stickney, C. (1989). The effects of business environment and strategy on firms rate of return on assets. Financial Analysts Journal, 45(1), 43-68.

Sorenson, J. and Stuart, T. (2000). Aging, obsolescence and organizational innovation.

Administrative Science Quarterly, 45, 81–112.

Tomohara, A. and Yokota, K. (2013). Who gains from foreign direct

investment-inducedtechnology spillovers? Horizontal and vertical linkages. The Australian Economic

(31)

31 Visser E.J., Atzema O. (2008). With or without clusters: facilitating innovation through a differentiated and combined network approach. European Planning Studies, 16 (9), 243-263.

White, H. (1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica, 48, 817-838.

World Bank; Knowledge Economy Index (KEI) 2012 Ranking ,

http://siteresources.worldbank.org/INTUNIKAM/Resources/2012.pdf. Accessed on 01-04-2016

World Bank; Urbanization degree (2016)

(32)

32

Appendix

(33)

33

Table 1: Knowledge intensive-countries with KEI-Rank, from World Bank

Sweden 9.43

Finland 9.33

Denmark 9.16

Netherlands 9.11

Germany 8.91

Table 2: Urbanized countries in EU & percentage of urban population, from World Bank Belgium 97.5 Malta 94.8 Denmark 86.9 France 85.8 Luxembourg 85.4

Table 3: Explanation variables and abbreviations

Variable name Variable label

lnrevenue2015 Natural logaritm of Revenue in 2015

large Dummy variable for a large or very large companies

age Age of the Company

industryC Knowledge-intensive activities in section C :

Manufacturing

industryH Knowledge-intensive activities in section H :

Transporting and Storage

industryJ Knowledge-intensive activities in section J :

Information and Communication

industryK Knowledge-intensive activities in section K : Finance and Insurance activities

industryM Knowledge-intensive activities in section M :

Professional, Scientific & technical activities

industryN Knowledge-intensive activities in section N :

Administrative and support services

IndustryO Knowledge-intensive activities in section O : Public administration

industryP Knowledge-intensive activities in section P :

Education

industryQ Knowledge-intensive activities in section Q: Human

health and social work activities

industryR Knowledge-intensive activities in section R : Arts, Entertainment and Recreation

industryS Knowledge-intensive activities in section S: Other

Service Activities

MLT Companies located in Malta

NLD Companies located in the Netherlands

FRA Companies located in France

GER Companies located in Germany

BEL Companies located in Belgium

DEN Companies located in Denmark

SWE Companies located in Sweden

(34)

34

Time2007 Dummy variable for data on year 2007

countryXage Cross effect between country of location and age

countryXsize Cross effect between country of location and size

DGUR Dummy variable for location in an urbanized

community

hightech Dummy variable for operating in a high-tech sector

KIC Dummy variable for operating in a Knowledge

intensive sector

KIS Dummy variable for location in a knowledge intensive

country

urbancountries Dummy variable for location in an urbanized country

DGURandKIC Dummy variable for location in an urbanized

community in a knowledge intensive country

Table 4: Descriptive statistics

(1) (2) (3) (4) (5)

VARIABLES N mean sd min max

large 153,761 0.670 0.470 0 1 age 153,761 23.58 16.06 1 100 NLD 153,761 0.0510 0.220 0 1 MLT 153,761 0.00206 0.0454 0 1 FRA 153,761 0.158 0.364 0 1 GER 153,761 0.0811 0.273 0 1 SWE 153,761 0.114 0.318 0 1 FIN 153,761 0.0331 0.179 0 1 BEL 153,761 0.0455 0.208 0 1 DEN 153,761 0.0168 0.128 0 1 time2014 153,761 0.156 0.363 0 1 time2013 153,761 0.144 0.351 0 1 time2012 153,761 0.134 0.340 0 1 time2011 153,761 0.123 0.329 0 1 time2010 153,761 0.109 0.311 0 1 time2009 153,761 0.0980 0.297 0 1 time2008 153,761 0.0884 0.284 0 1 time2007 153,761 0.0792 0.270 0 1 time2006 153,761 0.0687 0.253 0 1 DGUR 153,761 0.729 0.445 0 1 hightech 153,761 0.0732 0.260 0 1 KIS 153,761 0.900 0.300 0 1

revenue2015 153,761 402,095 3.740e+06 0.000731 2.699e+08

lnrevenue2015 153,761 9.430 2.982 -7.221 19.41

industryC 153,761 0.0732 0.260 0 1

industryH 153,761 0.00421 0.0647 0 1

(35)

35 industryK 153,761 0.274 0.446 0 1 industryP 153,761 0.00845 0.0915 0 1 industryS 153,761 1.95e-05 0.00442 0 1 industryM 153,761 0.397 0.489 0 1 industryO 153,761 0.000403 0.0201 0 1 industryQ 153,761 0.00942 0.0966 0 1 industryN 153,761 0.0147 0.120 0 1 KIC 153,761 0.611 0.487 0 1 countryXsize 153,761 53.86 39.66 1 140 countryXage 153,761 53.86 39.66 1 140 urbancountries 153,761 0.227 0.419 0 1 DGURandKIC 153,761 0.436 0.496 0 1

Table 5: Regression hypothesis 1

(1) (2) (3) (4) (5)

VARIABLES Model Model Model Model Model

(36)

36 industryJ -0.10*** (0.027) industryK -0.60*** (0.026) industryP -0.50*** (0.061) industryS 0.45 (1.163) industryM -0.47*** (0.026) industryO 0.57* (0.257) industryQ -0.32*** (0.058) BEL -0.85*** -0.85*** -0.85*** -0.02 -0.02 (0.080) (0.080) (0.080) (0.075) (0.075) MLT -0.60*** -0.70*** -0.71*** -0.85*** -0.85*** (0.136) (0.136) (0.136) (0.128) (0.128) DEN 0.07 0.03 0.02 0.64*** 0.64*** (0.086) (0.086) (0.086) (0.081) (0.081) FRA -0.71*** -0.74*** -0.74*** -0.28*** -0.28*** (0.077) (0.077) (0.077) (0.073) (0.073) hightech 0.40*** (0.020) KIS -0.45*** (0.017) countryXage 0.01*** (0.000) countryXsize 0.01*** (0.000) Constant 7.10*** 6.75*** 7.20*** 6.02*** 6.02*** (0.033) (0.022) (0.028) (0.022) (0.022) Observations 153,761 153,761 153,761 153,761 153,761 R-squared 0.47 0.47 0.47 0.52 0.52 Adj. R-squared 0.471 0.468 0.469 0.516 0.516

(37)

37

(1) (2) (3) (4) (5)

(38)

38 DEN 0.60*** 0.53*** 0.52*** 0.90*** 0.90*** (0.041) (0.041) (0.041) (0.038) (0.038) NLD 0.68*** 0.51*** 0.52*** 0.13*** 0.13*** (0.025) (0.024) (0.024) (0.023) (0.023) hightech 0.41*** (0.020) KIS -0.47*** (0.017) countryXage 0.02*** (0.000) countryXsize 0.02*** (0.000) Constant 7.11*** 6.73*** 7.19*** 6.00*** 6.00*** (0.033) (0.023) (0.028) (0.023) (0.023) Observations 153,761 153,761 153,761 153,761 153,761 R-squared 0.48 0.47 0.47 0.53 0.53 Adj. R-squared 0.477 0.473 0.474 0.526 0.526

Standard errors in parentheses *** p<0.001, ** p<0.01, * p<0.05

(1) (2) (3) (4) (5)

(39)

39 industryH 1.07*** (0.083) industryJ -0.10*** (0.027) industryK -0.59*** (0.026) industryP -0.50*** (0.061) industryS 0.43 (1.163) industryM -0.46*** (0.026) industryO 0.55* (0.257) industryQ -0.32*** (0.058) BEL -0.33*** -0.34*** -0.33*** 0.48*** 0.48*** (0.025) (0.025) (0.025) (0.025) (0.025) MLT -0.10 -0.22 -0.22 -0.35*** -0.35*** (0.113) (0.113) (0.113) (0.106) (0.106) DEN 0.59*** 0.55*** 0.54*** 1.14*** 1.14*** (0.040) (0.040) (0.040) (0.038) (0.038) FRA -0.21*** -0.25*** -0.24*** 0.22*** 0.22*** (0.014) (0.014) (0.014) (0.014) (0.014) hightech 0.43*** (0.020) KIS -0.47*** (0.017) countryXage 0.01*** (0.000) countryXsize 0.01*** (0.000) Constant 7.02*** 6.64*** 7.11*** 5.99*** 5.99*** (0.034) (0.024) (0.029) (0.024) (0.024) Observations 153,761 153,761 153,761 153,761 153,761 R-squared 0.47 0.47 0.47 0.52 0.52 Adj. R-squared 0.471 0.469 0.469 0.516 0.516

(40)

40

(1) (2) (3) (4) (5)

(41)

41 DEN 0.64*** 0.55*** 0.54*** 1.01*** 1.01*** (0.040) (0.040) (0.040) (0.038) (0.038) NLD 0.71*** 0.53*** 0.54*** 0.25*** 0.25*** (0.024) (0.024) (0.024) (0.023) (0.023) hightech 0.42*** (0.020) KIS -0.48*** (0.017) countryXage 0.01*** (0.000) countryXsize 0.01*** (0.000) Constant 7.12*** 6.73*** 7.20*** 6.10*** 6.10*** (0.033) (0.023) (0.028) (0.022) (0.022) Observations 153,761 153,761 153,761 153,761 153,761 R-squared 0.48 0.47 0.47 0.52 0.52 Adj. R-squared 0.476 0.474 0.474 0.524 0.524