The Link between Global Value Chain Integration and Regional Inequality

(1)

The Link between Global Value Chain Integration and Regional Inequality

Master Thesis

MSc International Economics & Business

University of Groningen, Faculty of Economics and Business

DATE 18/06/2019

CLARA MARIE GOTTSCHALK | s3721191 Email: c.m.gottschalk@student.rug.nl

Supervisor, University of Groningen: Dr. Konstantin Wacker Co-Assessor, University of Groningen: Prof. Dr. Marcel Timmer

(2)

I Abstract

The aim of this thesis is to improve the understanding of potential consequences GVC integration might have apart from stimulating economic growth, by examining the relationship between GVC participation and regional inequality across 13 EU countries and 175 NUTS 2 regions over the period 2000-2010. Using a growth regression approach, the cross-sectional analysis indicates that the integration into GVCs is associated with increasing spatial disparities within countries and between regions of Europe. Looking at specific industries, this positive link is proved for GVC participation of the textiles and especially the machinery sector. The empirical results further suggest that the magnitude of the impact is marginally conditional on the position in GVCs.

(3)

II

1 Introduction

Global Value Chains (GVCs) are increasingly seen as potential growth engine for countries at different stages of economic development. GVCs provide substantial growth and development potentials, improving the income and employment levels as well as stimulating the productivity and the exporting activities of a country. Not only advanced and emerging economies can potentially benefit from GVCs, but especially developing countries are expected to gain from being integrated in the global economy, getting access to essential knowledge and entering new learning processes (OECD, WTO and World Bank, 2014). This is why policy makers of economically lagging countries or regions are increasingly told to aim for a take-off via a deepened integration into GVCs.

The question is if the theoretical thoughts behind this policy strategy are also empirically well-founded. Several research papers have proved a positive relationship between GVC participation and economic growth (e.g. Ignatenko et al., 2019). The term economic development, however, should not be equaled to that of economic growth. Economic development is a broad and complex concept which policy makers should have in mind when thinking of development strategies. The factor of interpersonal and spatial inequality is one critical element to be considered. Potential negative byproducts of economic growth which themselves could present barriers to a sustainable economic growth process, such as increasing regional inequality, have barely received attention from research so far. Considering the fact that increasing regional disparities are not only potentially hampering economic progress, but also threatening social cohesion and political stability in Europe (Iammarino et al., 2018), the topic is actually of major relevance. Recent political developments such as the Brexit, the election of Donald Trump or the general rise of populism are related to problems triggered by regional economic divergence (The Economist, 2016; Wishlade, 2019). In this context, several authors stress the importance of ‘geography of discontent’, resulting from the fact that people’s perceptions regarding their own quality of life and opportunities is dependent on how they perceive the situation in other regions in relative proximity (Los et al., 2017; McCann, 2018; Hendrickson et al., 2018).

(6)

2

1999) and the respective channels which are explained in the following theory section, GVCs might therefore lead to increasing clustering of firms in specific regions in a country. The model provides a strong theoretical argument for claiming that increasing GVC participation could increase regional inequality levels within countries or also across European regions.

The topic of economic integration, especially in form of trade openness, and spatial inequality has been addressed by several papers and one relatively novel finding is that there is a positive link between trade openness and regional inequality conditional on country-specific characteristics (Rodríguez-Pose, 2012). However, none of the existing papers dealing with this research area is considering the shift from conventional trade to an increasing global division of labor via GVCs. My research is meant to fill this gap by analyzing the link between GVC integration and regional inequality and therefore using a new approach that takes changing global trade patterns into account. As already mentioned, the link between GVCs and economic growth is theoretically and empirically proved. In order to implement effective policy plans, it is important to know all the facets increased GVC participation brings along. This paper wants to contribute to the research area by broadening the pool of available empiric facts for policy makers. Ultimately, policy decisions always have to be based on a weighting of the benefits and the downsides. They always involve winners and losers and it is crucial to know who the losers are and how they can be compensated for. The research question of this thesis is thus Is there a link between GVC integration and regional inequality and are there specific conditioning factors?

A first look at the data (Figure 1 and 2) shows that this research question is indeed justified. When comparing the two maps illustrating the development levels of EU NUTS 2 regions on the one hand and the degree of GVC integration on the other hand, one can see that regions which are lagging behind tend to engage less in GVCs than the leading regions of a country. This is for instance clearly visible in Italy where we observe a North-South divide in both maps or in Germany where the Southern part’s GVC participation is the strongest.

Figure 1: GDP p.c. (in PPS), EU regions, 2010

Source: Eurostat.

Figure 2: GVC Value Added as a share of regional GDP, EU regions, 2010

(7)

3

This research is meant to examine empirically via a growth regression approach whether there is a positive relationship between GVC integration and income levels of a region, which would simultaneously mean that there is a link to spatial inequality. To examine this question, a cross-sectional model consisting of 13 EU countries and 175 NUTS 2 regions is analyzed. The used data covers the time period from 2000 to 2010.

The key results of the empirical analysis indicate that GVC participation is associated with increasing regional spatial disparities in terms of per capita income as well as wage levels and Gross Value Added (GVA). When looking at specific industries participating in GVCs, this holds for the textiles and the machinery sector in particular. The magnitude of the impact seems to be to a certain extent conditional on the position in GVCs.

The rest of the paper is structured as follows: Firstly, findings from academic literature will be shortly presented and the theoretical framework the research question is based on will be developed. Secondly, the data and the selection of variables will be described. Then, the chosen econometric model will be explained, which is followed by the econometric analysis and a discussion of the results, including limitations of this research. Ultimately, conclusions will be drawn pointing out the most important findings of this paper.

2 Literature and Theory Section

In this section, a short overview of findings from previous scientific papers regarding the research area is given and the hypotheses this research is examining are identified. Afterwards, the theoretical framework for understanding the setting and estimation strategy of the empirical analysis is developed.

2.1 Literature Overview and Hypotheses

It is a fact that one can observe rising spatial inequality in several countries (OECD, 2016). This presents a relevant social and political challenge and the phenomenon of a more globalized world is considered to contribute to the observed changes regarding inequality. With changing global trade patterns (Figure 3), more precisely, the rise of so called GVCs, the question to what extent these processes are related arises. The concept of GVCs in the context of international trade literature was firstly introduced by Gereffi (1999). In the past decades, production chains have begun to spread across different countries, resulting in products no longer being produced in one single country (Baldwin, 2006) and this international production fragmentation has rapidly increased since the early 1990s (Feenstra, 1998), thus coinciding with the rising regional disparities within countries.

(8)

Cross-4

country panel data analyses covering a diverse set of countries have been rarer. One contribution that should be highlighted is a paper by Rodríguez-Pose (2012) which examines if the assumptions made in the World Bank’s World Development Report “Reshaping Economic Geography” (2009) hold (whether increases in trade lead to rising inequalities, whether these inequalities recede in time; whether increases in global trade affect the developed and developing world differently), using an unbalanced panel data set of 28 countries from 1975 to 2005. He finds that increases in trade per se do not lead to greater spatial disparities, but combined with certain country-specific conditions, trade openness has a significant effect on regional inequality. Besides, he reveals that trade has a more polarizing and persisting impact in low- and middle-income countries.

In contrast, the understanding of the link between inequality and the participation in GVCs is at an early stage (OECD, 2013). Concerning this area, research has a relevant gap to fill as the analysis of GVCs seems to be essential in order to capture the mechanisms and consequences of global division of labor and income distribution (Lopez Gonzalez et al., 2015). Up to now, only Lopez Gonzalez et al. (2015) is addressing the link between GVC participation and inequality. Their paper focuses on the relationship between GVCs and wage inequality, thus interpersonal inequality, and uses data from the WIOD Release 2013 (40 countries, 1995 - 2009). The authors find that GVCs seem to be linked to wage inequality, although the effect appears to be relatively small. Moreover, a country’s position in GVCs matters, the findings show that offshoring low-skilled tasks leads to reductions in wage inequality while offshoring high-skill tasks results in increases in wage inequality (wages for higher skilled workers rise) in home countries.

Figure 3: Traditional and GVC Trade (Percent of nominal world GDP)

Source: Extracted from Ignatenko et al. (2019), based on Eora database.

(9)

5

The aim of this thesis is to contribute to the unexplored field of GVCs and inequality by examining the link between spatial inequality and integration into GVCs. It is based on the idea that GVC participation is linked to increased clustering of firms in specific regions and that clustering is further more likely in regions which are already characterized by higher GDP levels. These regions benefit from agglomeration effects and have consequently higher economic growth rates which promotes spatial inequality within a country. Additionally, it is analyzed if the effect of GVC participation on regional inequality depends on other conditioning variables. One question to look at is if it makes a difference in which stages of production a region engages. Ultimately, operating in the higher ends of supply chains and thus in higher value added activities implies higher economic growth potential. Another question is if the effect depends on the respective industry. According to economic theory, manufacturing industries would benefit the most from clustering, thus specific industries would tend more to cluster. The underlying theoretical model which justifies the hypotheses made is explained in detail in the next section. After having provided the necessary theoretical basis, the link from the theoretical foundations to the intended analysis is further deepened and potential measurement issues this research encounters are clarified.

2.2 Theoretical Framework

Production fragmentation is usually analyzed on a global level. However, it is often forgotten that GVCs also affect the production within countries and have therefore macroeconomic consequences for regions at a subnational level. Figure 4 presents a stylized value chain, showing the production process of a final product which is finished in Region 3. The regions being part of the GVC can be of the same country or of different countries.

Figure 4: A Stylized Value Chain

Source: Los and Chen 2016.

(10)

6

Final products are sold on the domestic market (to households, governments and firms) or are exported. Intermediate products such as raw materials, parts and components or business services, in contrast, are used in the production processes. During the production of final products and of intermediate goods, production factors such as capital and labor are used which in turn are remunerated with value added. Within the value chain, there are several complex linkages. Inputs might be stemming from the same region as they are further processed, from other regions of the same country or be imported from regions of other countries. In Figure 4 for example, Region 1 would act as second tier supplier, by selling intermediate products to Region 2, and consequently also contributing value added to the production of the final good in Region 3 (Los and Chen, 2016).

An important question is which mechanisms are driving locational decisions regarding the production of particular types of products. The mainstream theoretical framework used for explaining the link between market size and the location decision of firms is the New Economic Geography model introduced by Paul Krugman in the 1990s (Krugman, 1991; Fujita et al., 1999). According to this model, the industrial spatial distribution is resulting from specific agglomeration and dispersion forces at work.

Generally, locational outcomes and the mechanisms behind them have been the focus of research for a long time. The starting point is that firms are aiming at locating each stage of production in the lowest cost location possible. This decision is based on the balancing of two different types of costs – direct factor costs such as wages, capital costs and implicit or explicit subsidies and separation costs which include transmission and transportation costs as well as increased risk and managerial time. Further, the presence of potential spillovers is an important factor firms are considering when making their location decisions (Baldwin, 2013).

When thinking of GVCs and the spatial distribution consequences, one has thus to distinguish between two different steps. Firstly, there is a location decision. This decision is usually spanning different aggregation levels since a firm has first to decide in which country to locate their production stages (global level) and then also in which region to locate (country level). Secondly, this location decision has an effect on economic growth or income levels and consequently on spatial inequality.

2.2.1 Location Decision

(11)

7

On the other hand, Baldwin (2013) underlines two main agglomeration forces triggering production concentration on the global level, more precisely supply-side and demand-side linkages which are subject to a circular causality process. Demand-side linkages operate on an economy-wide basis and are related to the market size or market potential. Basically, they describe a potential domino effect, thus the fact that a country which is already characterized by high levels of economic activity is more attractive to firms that prefer being close to their customer base. Ultimately, customers attract suppliers whose employees become new customers. These demand-side linkages are the reason for manufacturing firms to continue their operations in high-wage economies. In contrast, supply-side linkages are relevant on a sectoral basis and linked to suppliers which in turn attract more suppliers. Due to the fact that firms source intermediate products from each other, spatial concentration seems intuitive considering the existing input costs.

The theory underlying industrial agglomeration on a more disaggregated level is the Core-Periphery Model launched by Krugman in his paper “Increasing Returns and Economic Geography” (1991). The model emphasizes the possible rational economic incentive for firms to concentrate in a region, based on the interaction of economies of scale and transport costs, and serves as a base when explaining why industrial clustering may contribute to economic growth.

In general, the Core-Periphery Model involves two opposing forces: the Home Market Effect and the Market Crowding Effect. The Home Market Effect describes the potential agglomeration forces. Increasing concentration implies a growing regional market and consequently also increasing demand which enables firms to pursue economies of scale. These efficiency gains in turn lead to a decline in the overall level of prices. As a consequence, real wages may rise, even though nominal wages may decline due to industry concentration. In contrast, the Market Crowding Effect captures potential dispersion mechanisms. In dispersed areas, nominal wages and prices are expected to increase which might incentivize manufacturing industries to disperse their production.

Which of the two described effects is dominating a firm’s location decision is basically determined by three factors: the share of income spent on manufactured products, the value of elasticity of substitution that ultimately determines a firm’s ability to pursue economies of scale and transport costs incurring to serve demand in remaining regions. An increasing share of income spent on manufactured goods and a higher elasticity of substitution encourages industrial agglomeration (Home Market Effect) as this results in a more cost-efficient production. By contrast, the effect of transport costs is level-dependent. While high transport costs induce firms to disperse (as they avoid competition on the dispersed market and are thus able to sell products without transport costs on the dispersed market), low transport costs reduce the cost advantage from dispersing and efficiency gains in production (Home Market Effect) become more relevant.

(12)

8

higher when an economic agglomeration becomes larger, consequently decreasing the efficiency level and therefore implying social welfare costs.

2.2.2 Agglomeration Economies

After having explained the forces triggering agglomeration or dispersion, a key question is why the clustering of firms should foster economic growth and hence potentially increase regional inequality. This link can be best explained by looking at the different effects resulting from industry concentration. The World Development Report (World Bank, 2009) builds on Kilkenny’s classification into Urbanization and Localization Effects (Appendix A1).

Localization Effects refer to the effects stemming from the geographical concentration of a large number of firms of the same industry. The proximity results in attracting a larger customer base (Shopping), the process of workers specializing (Specialization) and the attraction of a skilled workforce (Marshall Labor Market Pooling). The “Marshall-Arrow-Romer learning by doing” effect is being described by Audretsch and Feldman (2003) as the concentration of an industry in a region facilitating knowledge spillovers across firms.

Apart from these four Localization Effects there are Urbanization Effects which refer to economies of scale between industries, arising from a variety of industries located in proximity to each other. The first Urbanization Effect refers to Jacobs (1969) which states that knowledge spillovers between industries result in innovation and are more likely to occur with different industries settling close to each other. In addition, “Adam Smith’s Division of Labor” effect describes a firm’s ability to specialize in the production of a certain good. In his paper “Increasing Returns and Long Run Growth”, Romer (1986) argues that knowledge or technological upgrading increases marginal economic productivity, delaying the process of diminishing marginal returns to production factors and the implied growth stagnation. In the context of his paper “Endogenous Technological Change”, Romer (1990a) additionally states that R&D costs decrease over time. Combining these two theories, firms ultimately have an incentive to further invest in technological upgrading or absorb new knowledge, which has a similar effect. Duranton and Puga (2004) summarize the reasons tempting firms in a particular industry to locate in geographic proximity and the agglomeration economies arising from within-industry or between-industry interactions as “sharing, matching and learning”. Whereas sharing refers to indivisible facilities, suppliers, customers and risk, matching describes the improved matching of labor and firms and learning the need for face-to-face contact as well as the accumulation and diffusion of knowledge.

(13)

9

economies to increasing spatial inequality is reasonable. The spatial concentration of economic activity fosters technological progress and increases the production efficiency and hence TFP in some regions. These more efficient production practices are slow to diffuse, even on the country level.

On the basis of the theoretical foundations provided previously, the hypothesis that GVC integration is linked to increased spatial inequality across regions within a country and simultaneously across European regions seems plausible. As stated in the World Development Report 2009, the concentration of economic activity increases with the development stage and economic integration of a country.

As mentioned before, value chains are complex and include the processing of many different intermediate products. Consequently, demand-side and supply-side linkages seem to be an important factor determining the location decision of firms. Clustering seems generally beneficial considering the potential agglomeration economies and efficiency gains which can be derived. Given the differences in comparative advantage across countries, developed economies are mainly operating in the high-end, intangible pre- and post-fabrication activities (R&D, design, services, marketing etc.), as shown in Appendix A2. Smile curves generally illustrate that the generation of value added and employment gains along GVCs are highly dependent on a country’s, region’s or sector’s position in GVCs and the magnitude of GVC participation (World Bank, 2017). Being active in the upstream or downstream parts of GVCs and thus in higher value-added activities is associated with larger economic benefits such as high-skilled employment and higher incomes. Operating in the higher ends of GVCs requires specialization as well as increasing productivity and quality and consequently also a pool of high-skilled labor (OECD, 2013). Ultimately, the engagement in the higher ends of the GVCs implies that continuous technological innovation and therefore technological spillovers are very important as economies want to defend their position in the higher ends of GVCs. This is exactly why industrial concentration, the related benefits derived from agglomeration economies and the three described mechanisms sharing, matching and learning are crucial. This provides a strong justification for the assumption that impact of GVC participation on spatial inequality is conditional on the position in GVCs. Additionally, one should consider that the degree of value added generation and requirements regarding the workforce and production conditions also depend on the respective industry. Having the forces at work in mind, not all sectors can benefit equally from agglomeration economies. Krugman’s NEG framework is particularly applicable to the manufacturing industry.

In brief, GVC integration should be linked to the concentration of economic activity in specific regions of a country which benefit from the agglomeration economies and therefore experience increased economic growth rates. At the same time, this process would potentially increase regional disparities within a country. Further, based on economic theory and as explained previously it seems reasonable to assume that the impact of GVC participation on spatial inequality might be dependent on the respective industry and also on the level of production stages a region mainly engages.

(14)

10

decision has on economic growth. It would go beyond the scope of this research to disentangle these two factors. However, it is important to emphasize this endogeneity problem and keeping it in mind when interpreting the estimation results. It cannot be known with certainty which effect is actually measured since there are factors which affect the level and stage of GVC participation as well as the magnitude of the impact of GVC participation on income levels. The estimate of the causal effect of GVC participation on income will therefore be automatically biased. However, the inclusion of country fixed effects allows to at least partly counter this endogeneity problem.

3 Data and Methodology

In this section, the data collection process, the selection of included variables and the econometric model used for the empirical analysis of the research question is described.

3.1 Data and Sample Construction

Data for this research was extracted from two main sources, the Eurostat Regional Database and an OECD Report (2016) by Los and Chen who computed Global Value Chain participation indicators based on a regionally extended version of the World Input-Output Database by Timmer et al. (2015). Data for two variables was further extracted from The Quality of Government Institute of the University of Gothenburg (Charron et al., 2016) and a research paper by Beugelsdijk et al. (2018). A detailed overview over the exact data sources for each variable can be found in Appendix A4.

Eurostat is the statistical office of the European Union located in Luxembourg and provides regional statistics by NUTS classification. The NUTS classification subdivides the EU territory into regions at 3 different levels, namely NUTS 1, NUTS 2 and NUTS 3. As the regional Input-Output tables are constructed on the basis of NUTS 2, this is the territorial unit of analysis for this research. The Eurostat database comprises different statistical domains, for this research the Eurostat Regional Economic Accounts, the Demographic Statistics, the Education Statistics, the Science and Technology Statistics, the Transport Statistics and the Labor Market Statistics are used.

(15)

11 Table 1: Overview of Final Sample

Country Number of spatial units

Austria 9 Czech Republic 8 Denmark 3 Finland 4 Germany 37 Greece 13 Hungary 7 Italy 21 Portugal 5 Slovakia 4 Spain 19 Sweden 8 United Kingdom 37

Total number of spatial units 175

Conditional on available data, the final sample covers the period from 2000 to 2010, comprising 13 European countries and 175 regions (Table 1, Appendix A3). Considering the relatively short time span, the average values of each variable are used so that the model finally counts 175 observations. The other 10 European Union countries included in the WIOD Database had to be excluded due to data limitations regarding the dependent variable or due to a small country size. As the definition of the NUTS 2 regions changed over time and the GVC participation indicator used in the analysis is based on the NUTS 2 2013 definition, some observations for the other indicators had to be left out. For some variables data was not available for the entire period of analysis. However, this is only a minor issue since the average values over the period 2000 to 2010 for each variable are used. According to the selection of included variables, the number of observations can vary between the analyzed models. For the purpose of this research, taking the averages is the preferred option as the data is more aggregated in this way. By using the average values, potential lagged effects can be accounted for, thus yearly fluctuations do not bias the estimates. Even though a panel model analysis is consequently not possible anymore, a cross-sectional model still facilitates the inclusion of country-fixed effects.

3.2 Variables

The selection of variables for the model meant to examine the link between spatial inequality and GVC integration as well as potential conditioning factors across EU countries and regions was inspired by related scientific papers referred to in the theory section (chapter 2) and in the following discussion.

(16)

12

unit of choice for GDP p.c. is Purchasing Power Standard.1 Additionally, the average annual

GVA growth rate, the average annual wage growth rate and the average annual employment growth rate are used as dependent variables to check for robustness of the results.

The most relevant explanatory variable for the purpose of this research is GVC participation. In their paper “Global Value Chain Participation Indicators for European Regions” (2016), Los and Chen propose different indicators to measure GVC integration and discuss the respective advantages and disadvantages. By choosing Global Value Chain Value Added as a share of regional GDP as indicator for GVC participation, this thesis follows the suggestions made by Los and Chen. The computation of this indicator basically involves two steps. First, it is analyzed which industries out of the 14 in total can be considered as being part of truly global value chains. In order to determine this, the authors compute the extent to which non-EU countries contribute value added to the final products delivered by NUTS 2 regions. Subsequently, they add the calculated contributions by countries outside the EU over the different industries across regions. All the industries which are characterized by a minimum non-EU value added share of 10% in total final output are considered to be part of GVCs. According to Los’ and Chen’s calculations, this threshold is reached by the following industries: (i) Mining products; (ii) Textiles; (iii) Fuels; (iv) Machinery; and (v) Other manufactures. In contrast, the value chains for agricultural products, food manufacturing products, construction and all types of services are not considered to be truly global. In a second step, the value added that each NUTS 2 region generates within the production process of the five product types mentioned is calculated. Therefore, it does not matter in which stage of production (final stage or intermediate stages) the value added is generated or in which region and country the final

product is ultimately produced.2_{The final indicator is then GVC Value Added scaled to the}

level of regional GDP to account for different sizes of regions.

Although Los and Chen (2016) stress the drawbacks of other existing indicators for regional GVC participation, three other GVC participation indicators are used to test for the robustness of the results. The first one is Domestic Value Added in Exports as a Share of Regional GDP which is measuring the extent to which exports are contributing to regional GDP. The second one is Domestic Value Added in Exports as a Share of Regional Gross Exports which is basically measuring the degree of vertical specialization. The last indicator is the Share of Regional Domestic Labor Income in Exports in GDP.

Los and Chen (2016) also computed Global Value Chain Value Added as a share of regional GDP per sector. This data is used to examine if the effect of GVC participation is industry-dependent. It seems reasonable to use this data even though the approach does not take into account that the extent to which industries are producing output for GVCs might differ. It can be argued that it does not matter for the channels of industry clustering the hypotheses are based on if the output is part of GVCs in the end or not. However, in order to account for the importance out of total GVC participation, GVC VA per sector is used as the share of total GVC VA.

1_{It is important to note that PPP are calculated at national level only; this means that for a given country and a} given year the same PPP is applied for all regions of that country.

(17)

13

Regarding the position in GVCs, an indicator which is based on the share of Regional GVC VA attributable to GVCs with a different Country-of-Completion is used. The higher the respective share, the more a region is considered to be engaging in the higher ends of GVCs. This indicator assuming that the final stage of production is completed in another country is preferred to an indicator which only assumes that the final stage of production is in a different region as this would not clearly communicate if the region-of-completion is mainly in the same country or in another country.

By including different interaction terms, the hypotheses that GVC participation increases regional economic inequality and that the impact might be dependent on the respective industry as well as on the level of production stages a region mainly engages can be tested.

For the empirical analysis of this research, other factors next to the main variables of interest potentially affecting the regional level of economic growth have to be taken into consideration. The list of relevant variables stressed by the literature is quite extensive and the following selection of determinants of regional economic growth is not meant to be complete. The selection is based on the aim of including variables of importance in terms of the NEG framework as well as economic, socioeconomic and institutional factors.

Economic Geography

Generally, factors concerning the economic geography of the different regions have to be considered. The background of the respective channels were already explained extensively in section 2.2.2. Also the EU Cohesion Policy stresses the relevance of technological progress, innovation and knowledge externalities for regional economic performance (McCann and Ortega-Argilés, 2015) and therefore recognizes the role of economic geography factors. To capture this field, population density and the degree of urbanization are computed. These two factors are linked to positive agglomeration economies (Brakman et al., 2009; Boschma and Iammarino, 2009). Due to a high number of missing observations, the urbanization degree is finally left out in the regressions.

In addition, the proximity to the economic center is taken into account. According to Rice et al. (2006), this is a relevant indicator to capture potential spillover effects from the economic center to peripheral regions. A dummy variable additionally indicates if the capital of the respective country is located in the region or not.

The knowledge-related externalities and innovative capacity of each region is captured by the share of workers employed in science and technology (Anselin et al., 1997) and the regional expenditures on Research and Development as share of regional GDP.

Economic Factors

(18)

14

As stated by Rodríguez-Pose (2012), also differences in infrastructural endowments which determine the degree of market access may contribute to differences in regional economic growth. According to Faber (2007), regions that are characterized by a better access to foreign markets attract the winners of global economic integration and drive out weaker sectors, leading to higher regional growth rates than in regions with limited access. Infrastructural endowments are proxied by the total railway lines of a region, adjusted for population density. However, this variable finally cannot be used in the regressions due to missing observations.

Socioeconomic Factors

One important socioeconomic factor considering the knowledge-based economies of the EU is human capital. Many authors have emphasized the key role of human capital as attained through education in fostering economic growth and development (Lukas, 1988; Mankiw, Romer and Weil, 1992). Firstly, there is a strong link between the education level and labor productivity (Romer, 1990b) and secondly, human capital leads to increased competitive advantage through innovation and the diffusion of technology (Pelinescu, 2015). Rodriguez-Pose and Vilalta-Bufi (2005) found a strong relationship between differences regarding human capital endowments and regional economic growth performance in Europe. Thus, human capital in terms of qualification of the regional labor force is included as control variable.

When thinking about the factor of human capital, it is important to consider that labor is mobile, thus people move for the purpose of work. Labor mobility affects a region’s economic growth rate by determining the availability of human capital (Rodriguez-Pose and Vilalta-Bufi, 2005). Depending on country-specific characteristics, labor mobility on the country level may trigger agglomeration, as workers are attracted by higher wages of economic core areas, or dispersion, if workers follow firms that are seeking lower costs in peripheral areas (Puga, 1999). The measurement of labor mobility is generally difficult as it encompasses several statistical domains. In the empirical analysis, the crude rate of net migration is included as vague proxy for interregional labor mobility.

Socioeconomic factors also relate to the demographic structure of the population. The share of elderly is expected to increase in most countries. People aged 65 and over tend to lower the labor force participation and savings rates and are therefore considered to negatively influence economic growth (Bloom et al., 2010). The proportion of population aged 65 and more is thus included as additional explanatory variable for differences in regional economic performance.

Institutions

(19)

15

Table 2 shows all variables included in the baseline model and the extended models. All variables are computed as mean values over the period 2000-2010. A detailed description of the computation of all variables used in the analysis or which were initially computed is provided in Appendix A4. Some variables which were initially computed had to be left out since their inclusion would have decreased the number of observations significantly. Generally, it is important to note that the possibilities of computation or choice of indicators were restricted due to data availability on the NUTS 2 regional level. Thus the measurement of indicators might not be the best possible choice and indicators should be considered as proxies.

Table 2: Overview of Variables Dependent variable

Variable Name Description

Average annual GDP p.c. growth

GDP_AVGR GDP per capita growth rate, in %

Alternative dependent variables

Average annual GVA growth

GVA_AVGR GVA growth rate, in %

Average annual wage growth

WAGE_AVGR Wage growth rate, in %

Average annual employment growth

EMPL_AVGR Employment growth rate, in %

Key explanatory variables

Initial GDP p.c. LNGDP2000 GDP p.c. level in year 2000 (natural logarithm used), in

PPS

GVC participation GVC_VA Global Value Chain Value Added as a share of regional

GDP, in %

GVC position POSITION Based on share of Regional GVC Value Added

attributable to GVCs with a different Country-of-Completion, in %

GVC participation sector 1 (Mining)

GVC_S1 Mining GVC Value Added as a share of regional GDP,

computed as share of total GVC participation, in % GVC participation sector

2 (Textiles)

GVC_S2 Textiles GVC Value Added as a share of regional GDP,

3 (Fuels)

GVC_S3 Fuels GVC Value Added as a share of regional GDP,

4 (Machinery)

GVC_S4 Machinery GVC Value Added as a share of regional GDP,

5 (Other Manufacturers)

GVC_S5 Other Manufacturers GVC Value Added as a share of

(20)

16

Alternative GVC participation indicators

GVC participation (Alternative 1)

GVC_DVA Share of Regional Domestic Value Added in Exports in

GDP, in % GVC participation

(Alternative 2)

GVC_VS Vertical Specialization (Regional Domestic Value Added

as a share of Regional Gross Exports), in % GVC participation

(Alternative 3)

GVC_DLI Share of Regional Domestic Labor Income in Exports in

GDP, in % Additional explanatory variables

Distance to economic center

DISTANCE Areal distance between each region’s largest city and the

economic center of the respective country

Proximity to capital CAPITAL Indicator variable:

1 = capital is located in region 0 = capital is not located in region

Population density POP_ DENS Population density per NUTS 2 region, inhabitants per

square kilometer

Human capital HC Population aged 25-64 with tertiary education (levels 5-8),

in % Workers in science and

technology

SCIENCE Share of workers employed in science and technology out

of total active population, in %

R&D expenditures RD Intramural R&D expenditure (GERD), in % of GDP

Sectoral structure AGRI Deviation of employment share in agriculture from

country specific mean

Labor mobility MOBIL Crude rate of net migration plus statistical adjustment,

expressed per 1000 inhabitants

Share of old population POP_OLD Proportion of population aged 65 years and more, in %

Average annual population growth

POP_AVGR Population growth rate, in %

Initial GVA LNGVA2000 GVA level in year 2000 (natural logarithm used), in

Million Euro

Initial wage level LNWAGE2000 Compensation of employees in year 2000 (natural

logarithm used), in Million Euro

Initial employment LNEMPL2000 Employment in year 2000 (natural logarithm used), in

thousand persons

3.3 Descriptive Statistics

(21)

17

GDP p.c. is approximately on the same level with the poorest Swedish region. Simultaneously, Figure 5 illustrates that within-country inequalities also matter since there are large disparities between the regions of a country. Comparing GDP p.c. levels in 2000 and 2010 (Appendix A7), one can see that spatial inequality has increased in most of the analyzed EU countries. This trend of increasing polarization becomes also visible when looking at the difference between the top and bottom regions (Figure 6). In more than half of the sample countries, the top region has grown faster between 2000 and 2010 than the bottom region. To sum up, for the standard of living of an individual it seems to play a role where to live within a country, but also where to live in Europe.

GVC Participation (GVC VA) ranges between approximately 5.43% and 31.98% of regional GDP with a mean of 19.28% across the 175 sample regions. The boxplots in Appendix A8 illustrate that the Czech, German, Hungarian and Slovak regions are generally engaging the most in GVCs. In contrast, GVC participation of Greek, Portuguese or Spanish regions is on a much lower level. The boxplots of GVC participation per sector by countries (Appendix A9) additionally show which sectors are accounting for most of the total GVC participation. In most countries, the machinery and the other manufacturers sector are clearly dominating, accounting for the largest share of GVC participation. The mining sector plays generally rather a minor role. The textiles sector is of certain importance in Greece, Italy and Portugal whereas the fuels industry accounts on average for approximately 5% to 20% of total GVC participation.

0 10.000 20.000 30.000 40.000 50.000 60.000 70.000 80.000 90.000 PPS

Minimum Maximum Country average

Figure 5: Regional disparities in GDP p.c., 2010, NUTS 2 regions

(22)

18

Having a look at the relation between regional economic growth and GVC participation by plotting the average annual GDP p.c. growth rate and GVC participation to obtain a scatter graph (Figure 7), one can see that there is generally a positive, albeit not perfect, correlation between the two variables. Certainly, this does not allow to draw conclusions regarding the causal relationship, but it provides a reasonable starting point for the intended empirical analysis. -2 -1,5 -1 -0,5 0 0,5 1 1,5 2 2,5 3 Germany Spain Austria Finland Portugal Italy Sweden Czech Republic Greece Denmark UK Hungary Slovakia

Percentage point difference Bottom grew faster Top grew faster

Source: Based on Eurostat data.

Figure 6: Average annual GDP p.c. growth, difference between top and bottom regions

Figure 7: Scatter Plot Average annual GDP p.c. growth and GVC VA

(23)

19 3.4 Econometric Model

This research is based on a cross-sectional analysis. By including country fixed effects, it is possible to control for unobserved heterogeneity, that is for unobserved sources of individual heterogeneity that vary across entities but not over time (Stock and Watson, 2011). The inclusion of country fixed effects seems intuitive in this case as they absorb differences in terms of for example level of technology, institutional and trade openness or the political system. These factors play a relevant role for the location decision of firms. In this way, certain endogeneity issues and omitted variable biases can be dealt with.

The estimation is based on a multiple linear regression model using the ordinary least squares (OLS) estimator. Several tests are executed in order to test for the validity of the basic assumptions of OLS. One essential assumption is homoskedasticity which describes a situation in which the error term does not vary across observations of the explanatory variables. In case of presence of heteroskedasticity, the assumption of homoskedasticity of OLS is violated and the variance of the estimated coefficients is consequently underestimated. This means that the t-values are too high which may lead to erroneous conclusion of significance and acceptance of the null hypothesis (Hill et al., 2018). Testing for the presence of heteroskedasticity suggests

that robust standard errors should be used.3 Another relevant aspect to consider is the

assumption of normality which presumes that the errors, and hence the dependent variable y, are normally distributed (Hill et al., 2018). Tests for the fulfillment of this assumption show that the log transformation of the variable initial GDP p.c. should be used. This is also an intuitive choice as GDP levels are usually unnormal distributed. The log transformation of initial GDP p.c. allows to correct for this to a certain extent (Appendix A11). Furthermore, OLS assumes no collinearity which means that the values of each explanatory variable should not be exact linear functions of the other explanatory variables in order to obtain reliable estimates (Hill et al., 2018). The correlation matrix provided in the Appendix A12 shows that the

correlation degree of all used explanatory variables is on acceptable levels.4

Most papers dealing with the topic of spatial inequality are using the Gini Index or similar indexes measuring income inequality to examine their research question. This research is not using this empirical method as indexes such as the Gini have considerable downsides and can move for several reasons, decreasing the informativeness.

In order to examine the research question, a growth regression within the neoclassical growth framework (Mankiw et al., 1992) is chosen. Equation (1) presents the basic specification followed in the regressions.

(1) ln(𝑌_𝑖𝑗𝑡) − ln(𝑌_{𝑖𝑗𝑡−1}) = ∆ ln(𝑌_𝑖𝑗𝑡) = (∅ − 1) ln(𝑌_{𝑖𝑗𝑡−1}) + 𝛽𝑋_𝑖𝑗𝑡+ 𝛼_𝑖 + 𝑢_𝑖𝑗𝑡

i = country, j = region, t = year X = Set of variables

αi (i = 1...13) = country fixed effects

3_{Note that cluster robust standard errors would be the preferred option, but cannot be used due to statistical} issues.

(24)

20

The modelling is divided into three parts. In this way, the different hypotheses can be examined stepwise and testing for robustness is more structured.

Question 1: Is there a link between GVC participation and regional economic growth? The foregoing discussion in chapter 2 and 3.2 leads to the transformation of equation (1) into the following empirical specification (2).

(2) GDPAVGR = 𝛽1+ 𝛽2𝐿𝑁𝐺𝐷𝑃2000𝑖𝑗𝑡+ 𝛽3𝐺𝑉𝐶𝑖𝑗𝑡+ 𝛽4𝐷𝐼𝑆𝑇𝐴𝑁𝐶𝐸𝑖𝑗𝑡+

𝛽₅𝐶𝐴𝑃𝐼𝑇𝐴𝐿_𝑖𝑗𝑡+ 𝛽₆𝑃𝑂𝑃_{𝐷𝐸𝑁𝑆𝑖𝑗𝑡}+ 𝛽₇𝐻𝐶_𝑖𝑗𝑡 + 𝛽₈𝑆𝐶𝐼𝐸𝑁𝐶𝐸_𝑖𝑗𝑡+ 𝛽₉𝑅𝐷_𝑖𝑗𝑡+

𝛽₁₀𝐴𝐺𝑅𝐼_𝑖𝑗𝑡+ 𝛽₁₁𝑀𝑂𝐵𝐼𝐿_𝑖𝑗𝑡 + 𝛽₁₂𝑃𝑂𝑃_𝑂𝐿𝐷_𝑖𝑗𝑡+ 𝛼_𝑖+ 𝑢_𝑖𝑗𝑡

GVC = GVC_VA, GVC_DVA, GVC_VS or GVC_DLI

In a first step, it is examined if there is a general positive relationship between GVC participation and regional economic growth. Therefore, the average annual GDP p.c. growth rate is regressed on GVC participation and the additional explanatory variables distance to the economic center, proximity to capital, population density, human capital, workers in science and technology, R&D expenditures, sectoral structure, labor mobility and share of old population. For GVC participation to have a positive impact on regional economic growth, the coefficient has to be positive and significant.

Question 2: Is there a link between total GVC participation and regional inequality?

(3) GDP_AVGR = 𝛽₁+ 𝛽₂[𝐺𝑉𝐶 ∙ 𝐿𝑁𝐺𝐷𝑃2000]_𝑖𝑗𝑡+ 𝛽₃𝐷𝐼𝑆𝑇𝐴𝑁𝐶𝐸_𝑖𝑗𝑡+ 𝛽₄𝐶𝐴𝑃𝐼𝑇𝐴𝐿_𝑖𝑗𝑡+

𝛽₅𝑃𝑂𝑃_{𝐷𝐸𝑁𝑆𝑖𝑗𝑡}+ 𝛽₆𝐻𝐶_𝑖𝑗𝑡+ 𝛽₇𝑆𝐶𝐼𝐸𝑁𝐶𝐸_𝑖𝑗𝑡+ 𝛽₈𝑅𝐷_𝑖𝑗𝑡+ 𝛽₉𝐴𝐺𝑅𝐼_𝑖𝑗𝑡 + 𝛽₁₀𝑀𝑂𝐵𝐼𝐿_𝑖𝑗𝑡+ 𝛼_𝑖 + 𝑢_𝑖𝑗𝑡

GVC = GVC_VA, GVC_DVA, GVC_VS or GVC_DLI LNGDP2000 = indicator or continuous variable

(25)

21

marginal effect of GVC participation on the linear prediction, conditional on the initial GDP p.c. level, the same regressions are repeated with initial income as continuous variable. In specification (4), two control interactions are added, more precisely an interaction of GVC participation and the number of workers employed in science and technology and an interaction of GVC participation and population density. These control for the fact that the impact of GVC participation on economic growth might be higher or lower if the differences regarding the endowment with economic geography factors are already higher across regions.

(4) GDPAVGR = 𝛽1+ 𝛽2[𝐺𝑉𝐶 ∙ 𝐿𝑁𝐺𝐷𝑃2000]𝑖𝑗𝑡+ 𝛽3[𝐺𝑉𝐶 ∙ 𝑆𝐶𝐼𝐸𝑁𝐶𝐸]𝑖𝑗𝑡+ 𝛽4[𝐺𝑉𝐶 ∙

𝑃𝑂𝑃_{𝐷𝐸𝑁𝑆}]_𝑖𝑗𝑡+ 𝛽₅𝐷𝐼𝑆𝑇𝐴𝑁𝐶𝐸_𝑖𝑗𝑡+ 𝛽₆𝐶𝐴𝑃𝐼𝑇𝐴𝐿_𝑖𝑗𝑡 + 𝛽₇𝐻𝐶_𝑖𝑗𝑡+ 𝛽₈𝑅𝐷_𝑖𝑗𝑡+

𝛽9𝐴𝐺𝑅𝐼𝑖𝑗𝑡+ 𝛽10𝑀𝑂𝐵𝐼𝐿𝑖𝑗𝑡+ 𝛼𝑖 + 𝑢𝑖𝑗𝑡

GVC = GVC_VA, GVC_DVA, GVC_VS or GVC_DLI LNGDP2000 = continuous variable

Note that in specification (3) and (4) and in all the following models, the variables constituting the interaction terms are also individually included. The exact specification (choice of included explanatory variables and control interaction terms) is selected on the basis of the results of part 1.

In order to test for the robustness of the results, specification (3)5 is further repeated with the

three alternative dependent variables average annual GVA, wage and employment growth. Ultimately, an important aspect to highlight is that GDP per capita and thus the choice of the average annual GDP p.c. growth rate as dependent variable of the model obviously has some disadvantages. Value added can be shifted quite arbitrarily in GVCs and the generation of value added is therefore not necessarily increasing the income level of a region and benefitting their population. Especially due to tax considerations, profits are often shifted to other regions or even countries. For these reasons, it is useful to not only analyze the research question using the growth rate in terms of GDP p.c., but to additionally look at GVA, wages and employment. These three factors are likewise expected to increase with GVC participation when thinking of the theoretical base of agglomeration economies. By looking at the interaction effect of GVC VA and initial GDP, it is possible to analyze if the impact of GVC participation on growth in GVA, wage levels and employment is higher for regions which are already higher developed in terms of GDP p.c. In this case, GVC integration would foster spatial polarization.

(26)

22

Question 3: Is there a link between sector-specific GVC participation and regional inequality and is the impact of GVC participation dependent on the respective position in GVC?

(5) GDPAVGR = 𝛽1+𝛽2[𝐺𝑉𝐶 ∙ 𝐿𝑁𝐺𝐷𝑃2000]𝑖𝑗𝑡+ 𝛽3[𝐺𝑉𝐶𝑆2∙ 𝐿𝑁𝐺𝐷𝑃2000]𝑖𝑗𝑡 +

𝛽₄[𝐺𝑉𝐶𝑆3∙ 𝐿𝑁𝐺𝐷𝑃2000]𝑖𝑗𝑡+ 𝛽5[𝐺𝑉𝐶𝑆4∙ 𝐿𝑁𝐺𝐷𝑃2000]𝑖𝑗𝑡 + 𝛽6[𝐺𝑉𝐶𝑆5∙

𝐿𝑁𝐺𝐷𝑃2000]𝑖𝑗𝑡 +𝛽7𝐶𝐴𝑃𝐼𝑇𝐴𝐿𝑖𝑗𝑡+ 𝛽8𝑃𝑂𝑃_{𝐷𝐸𝑁𝑆𝑖𝑗𝑡}+ 𝛽9𝐻𝐶𝑖𝑗𝑡+ 𝛽10𝑆𝐶𝐼𝐸𝑁𝐶𝐸𝑖𝑗𝑡 +

𝛽₁₁𝑅𝐷_𝑖𝑗𝑡 + 𝛼_𝑖+ 𝑢_𝑖𝑗𝑡

GVC = GVC_VA, GVC_DVA, GVC_VS or GVC_DLI LNGDP2000 = continuous variable

Likewise, it is tested if the impact of GVC participation on spatial inequality depends on the specific GVC (industry). Therefore, the first interaction term with an specific GVC participation indicator per industry (textiles, fuels, machinery, other manufactures) is used. The mining industry (sector 1) is being left out on purpose. This seems economically reasonable since the location of mining activities is logically linked to the distribution of mining opportunities in a country. Mining activities are usually uneven distributed and the clustering and agglomeration economies theory this research is based on (chapter 2) does not apply to this sector. Additionally, a selection of control variables in included. To check the robustness of the results, the estimation is repeated including the two control interactions as in specification (4).

In a last step, in case that there is an effect of GVC participation or sector-specific GVC participation on regional inequality, it is analyzed if the impact depends on the respective stage of production a region is mainly engaging in GVCs. In order to capture this, an indicator variable for the position is added to the interaction terms. A value of 1 means that a region is mainly operating in the higher ends of GVCs (share of GVC participation attributable to GVCs with a different Country-of-Completion is above the mean). Having the theory in mind, one would expect that operating primarily in the upstream or downstream parts would further amplify the effect GVC integration has on spatial inequality. One would thus expect the 3-way-continuous interaction to have a positive sign and to be higher for a position equaling 1 compared to the position indicator variable equaling 0.

(6) GDP_AVGR = 𝛽₁+ 𝛽₂[𝐺𝑉𝐶 ∙ 𝑃𝑂𝑆𝐼𝑇𝐼𝑂𝑁]_𝑖𝑗𝑡+ 𝛽₃[𝐺𝑉𝐶 ∙ 𝐿𝑁𝐺𝐷𝑃2000 ∙

𝑃𝑂𝑆𝐼𝑇𝐼𝑂𝑁]_𝑖𝑗𝑡+ 𝛽₄[𝐺𝑉𝐶𝑆2∙ 𝐿𝑁𝐺𝐷𝑃2000 ∙ 𝑃𝑂𝑆𝐼𝑇𝐼𝑂𝑁]𝑖𝑗𝑡+ 𝛽5[𝐺𝑉𝐶𝑆3∙

𝐿𝑁𝐺𝐷𝑃2000 ∙ 𝑃𝑂𝑆𝐼𝑇𝐼𝑂𝑁]𝑖𝑗𝑡+ 𝛽6[𝐺𝑉𝐶𝑆4∙ 𝐿𝑁𝐺𝐷𝑃2000 ∙ 𝑃𝑂𝑆𝐼𝑇𝐼𝑂𝑁]𝑖𝑗𝑡+

𝛽₇[𝐺𝑉𝐶_𝑆5∙ 𝐿𝑁𝐺𝐷𝑃2000 ∙ 𝑃𝑂𝑆𝐼𝑇𝐼𝑂𝑁]_𝑖𝑗𝑡+ 𝛽₈𝐶𝐴𝑃𝐼𝑇𝐴𝐿_𝑖𝑗𝑡+ 𝛽₉𝑃𝑂𝑃_{𝐷𝐸𝑁𝑆𝑖𝑗𝑡} +

𝛽₁₀𝐻𝐶_𝑖𝑗𝑡 + 𝛽₁₁𝑆𝐶𝐼𝐸𝑁𝐶𝐸_𝑖𝑗𝑡+ 𝛽₁₂𝑅𝐷_𝑖𝑗𝑡 + 𝛼_𝑖+ 𝑢_𝑖𝑗𝑡

(27)

23

Throughout the empirical analysis, the robustness of the results is tested by the stepwise development of the models, the inclusion of additional explanatory variables as well as control interaction terms. To additionally test for the robustness of the results, different GVC indicators are used for general GVC participation, although GVC Value Added as a share of regional GDP should be considered to be the most accurate (Los and Chen, 2016).

In general, the interaction terms present the core of the used methodology to examine the research question. When model specifications include interactions between variables, the coefficients cannot be interpreted directly as the relationship between the covariates and the outcome is determined by several coefficients. By making use of the margins command in Stata, the interaction terms can be easily interpreted. Marginal effects reflect the slope of the regression surface regarding a certain covariate. They thus give the rate at which the dependent variable changes at a specific point in the covariate space, holding all explanatory variable values constant. In case of an OLS estimation, marginal effects therefore present the marginal contribution of a covariate x to the outcome. In general, marginal effects can be examined in three different ways, the effects can be derived at representative values (MERs), at means (MEMs) or as average values (AMEs). MERs allow to set a specific combination of values of a covariate that is theoretically interesting to answer the research question and compute the marginal effect at these values. AMEs calculate the marginal effects at every observed value of the covariate and average across the corresponding effect estimates. In the context of this research, AMEs are particularly useful as they contain a lot of information about the impact that a covariate has on the outcome. (Leeper, 2017).

4 Empirical Results

In this section, the results of the different empirical estimations are analyzed and subsequently discussed from a social welfare perspective.

4.1 Global Value Chain Participation and Regional Economic Growth

(28)

24

percentage points in the average annual GDP p.c. growth rate of a region. As can be seen in Table 3, also the fact of being a region in which the country’s capital is located as well as human capital and the number of workers employed in science and technology have a positive and (highly) significant impact on the regional economic growth performance. Further the sectoral structure seems to play a role, a higher share of employment in the less productive agricultural sector has a negative effect on income growth. These findings are in line with previous scientific papers referred to in chapter 3.2.

However, as already explained it seems reasonable to add country fixed effects to the specification and it is interesting to see how their inclusion changes the estimation results. The third estimation has clearly the highest fit compared to (1) and (2). As can be seen in column (3), the inclusion of country fixed effects makes the coefficient of GVC VA insignificant. The only variable which has a significant effect as before in (2) is the number of workers in science and technology. Additionally, population density and labor mobility have a slightly negative

Table 3: GVC Participation and Regional Economic Growth

(1) (2) (3) VARIABLES GVC VA 1 GVC VA 2 GVC VA 3 Initial GDP p.c. -2.635*** -2.677*** -0.604 (0.409) (0.339) (0.464) GVC VA 0.0228 0.0331** -0.00151 (0.0169) (0.0166) (0.0188)

Distance to Economic Center -0.000588** -0.000411 -6.52e-05

(0.000291) (0.000284) (0.000257) i. Capital 0.673 0.900** 0.255 (0.423) (0.445) (0.238) Population Density -0.000169 -0.000144 -0.000150* (0.000147) (0.000147) (8.92e-05) Human Capital 0.0231** 0.0299*** 0.0249 (0.0105) (0.0107) (0.0153)

Science & Technology 0.0632*** 0.0547** 0.0498*

(0.0219) (0.0213) (0.0274) R&D Expenditures -0.0779 -0.0690 -0.0424 (0.0802) (0.0763) (0.0381) Sectoral Structure -1.821 -3.603** 0.285 (1.846) (1.728) (1.524) Labor Mobility -0.0110 -0.00746 -0.0746*** (0.0171) (0.0175) (0.0162) Old Population -0.0813** (0.0330) Observations 156 163 163 R-squared 0.524 0.530 0.830 r2_a 0.487 0.499 0.803

Robust S.E. YES YES YES

Country FE NO NO YES

(29)

25

effect. These results are plausible since population density is not only linked to positive agglomeration economies but possibly also increases congestion costs (Brakman et al., 2019). Depending on the movement patterns of workers, labor mobility can have different consequences which means that a negative effect on growth is also possible. When doing the regression as in (2) and (3) with the three alternative GVC participation indicators (Appendix A13), the results provide a similar picture. The effect of GVC participation using the Domestic VA in exports as a share of regional GDP (GVC DVA) or the share of regional Domestic Labor Income in exports in GDP (GVC DLI) indicator is only significant before country fixed effects are included. Using the Domestic VA in exports as a share of regional gross exports (GVC VS) indicator, the effect is insignificant even before adding fixed effects.

That the effect GVC participation has in regression (2) is eliminated when country fixed effects are included indicates that they absorb other factors not accounted for in the model as for example differences in technology levels, institutional and trade openness or the political system. However, the fact that the coefficient of GVC participation is not significant for all the different indicators once country fixed effects are added does not automatically imply that there is no positive relation between GVC participation and regional inequality. The impact might be dependent on other factors which are not included in the first model. In the following section, it is analyzed if the impact of GVC participation might be dependent on a region’s initial per capita income level.

4.2 Global Value Chain Participation and Regional Inequality

After having provided some basic regressions in section 1, the main hypothesis of this research is now analyzed. In order to do so, GVC participation is interacted with the initial GDP p.c. level. As specified in chapter 3.4, this is first done by using an indicator variable for the initial GDP p.c. level and subsequently with initial GDP p.c. as continuous variable.

(30)

26

In order to see if this is the case, the same estimation is repeated with initial GDP p.c. as continuous variable. The results in column (2) reveal that the interaction is positive and highly significant. The positive coefficient of the interaction indicates that the effect of GVC participation on regional economic growth is higher for regions which are characterized by a higher initial GDP p.c. level. To test for the robustness of this result, two control interactions are added in a second estimation. These control for the fact that the impact of GVC participation on growth might be dependent on the number of workers employed in science and technology and the degree of population density. As can be seen in column (3), the inclusion of these interactions does not change the main result, although decreasing the magnitude of the positive coefficient of the interaction between GVC participation and initial GDP p.c. as well as the significance level. Both control interactions do not have a significant effect. Generally, specification (2) and (3) have a high explanatory power when comparing R-squared levels. As the adjusted R-squared value is slightly higher for (2), this is the preferred specification which is used to analyze the interaction effect more detailed.

As in section 4.1, the equivalent regressions are subsequently done with the alternative GVC

participation indicators (Appendix A15).6_{The interaction of GVC participation and the}

indicator variable for initial GDP p.c. is only positive and significant at a 10% level when the share of regional Domestic Labor Income in exports in GDP is used. However, the average marginal effects of GVC participation conditional on the initial income dummy variable are also insignificant. Using initial GDP p.c. as continuous variable, none of the interactions is significant. However, when evaluating the robustness of the results it should be considered that

6_{Domestic VA in exports as a share of regional gross exports is left out as the results for this GVC participation} indicator were already insignificant in section 1.

Table 4: GVC Participation and Regional Inequality

(1) (2) (3) VARIABLES Interaction GVC VA 1 Interaction GVC VA 2 With control interactions GVC VA # i. Initial GDP 0.0349** (0.0171) GVC VA # Initial GDP 0.121*** 0.110** (0.0434) (0.0543)

GVC VA # Science & Technology 0.00123

(0.00224)

GVC VA # Population Density -1.33e-05

(2.10e-05)

Observations 163 163 163

R-squared 0.836 0.843 0.843

r2_a 0.807 0.817 0.815

Robust S.E. YES YES YES

Country FE YES YES YES

Control Variables YES YES YES

(31)

27

the different indicators capture slightly different things and that GVC VA comes closest to actual GVC participation.

In order to provide an additional picture and check for robustness, the regressions as specified in column (2) are repeated using different dependent variables, more precisely GVA growth, wage growth and employment growth apart from GDP p.c. growth. This is interesting to look at since GVC participation should affect different dimensions of regional inequality according to the theoretical base provided in chapter 2.2.2. A comparison of the results can be found in Appendix A16. The interaction of initial GDP p.c. and GVC participation is positive and significant for all the four dependent variables. While the interaction term is significant at a 1% level when the average annual GDP p.c. growth rate, GVA growth rate and wage growth rate is used, it is significant at a 10% level when employment growth is used. The magnitude of the effect is similar when GDP p.c. growth and GVA growth is used, slightly higher for wage growth and lower for employment growth.

Having a closer look at the interaction effect via average marginal effects, the magnitude of the interactions becomes visible. In Figure 8, the average marginal effect of GVC participation according to different levels of initial GDP p.c. is presented. One can see that the average marginal effect of GVC participation on economic growth is increasing with higher levels of initial GDP p.c. and this trend holds for every dependent variable used.

Figure 8: Average marginal effects of GVC participation, dependent on the initial GDP p.c. level

Source: Own Stata results.

The Link between Global Value Chain Integration and Regional Inequality

Table of Contents

1 Introduction

2 Literature and Theory Section

3 Data and Methodology

4 Empirical Results