Structural Funds and regional convergence across EU : Study of convergence patterns across European NUTS-2 regions between 2000 and 2016

Structural Funds and

regional convergence

across EU

MASTER’S DEGREE PROJECT THESIS WITHIN: Major in Economics NUMBER OF CREDITS: 30 credits / ECTS PROGRAMME OF STUDY: Economic Analysis AUTHOR: Jan Pokorný

JÖNKÖPING May 2020

Study of convergence patterns across European NUTS-2

regions between 2000 and 2016

Master Thesis in Economics

Title: Measuring regional convergence in EU Authors: Jan Pokorný

Tutor: Johan Klaesson, Emma Lappi Date: 2020-05-18

Key terms: convergence, regional, GDP, structural funds, V4, cohesion, ß-convergence

Abstract

The objective of this paper is to empirically analyse and measure the effects of Structural Funds on the speed of regional economic convergence for European regions at NUTS-2 regional level. This study is conducted using European regional data for 260 regions, over 17 years from 2000 to 2016. This paper utilizes existing econometric methods of panel data modelling, namely pool OLS and FE and accounts for the country and year fixed effects. In general, and in accordance to conditional ß-convergence approach, tests indicate on average pro-beta convergence results, meaning that poorer regions tend to exhibit faster growth rates than richer regions; apart from Luxembourg, which shows signs of economic divergence. On the other hand, Structural Funds demonstrate support for pro-growth tendencies in Visegrad Four regions, ceteris paribus, whilst equivalently undermining average per capita GDP growth in rest of EU, respectively. Theory of club convergence is also treated for; results are proven in one model specification, however, inconsistent in remaining variations and thus are conclusions are ambiguous.

Table of Contents

1.

Introduction ... 1

2.

Background information ... 1

2.1. EU funds and objectives ... 3

2.2. Historical and current outlooks of V4 ... 4

2.3. NUTS territorial classification ... 7

3.

Theoretical background ... 9

3.1. Neoclassical-convergence approach ... 10

3.2. Divergence-orientated theories ... 12

3.3. Convergence clubs ... 14

4.

Literature findings ... 15

4.1. Regional disparities studies ... 16

4.2. Evaluation of Structural Funds ... 17

4.3. Other findings ... 18

5.

Data, Methodology and Models ... 19

5.1. Model description ... 20

5.2. Variable description ... 21

5.3. Variable and data constraints ... 22

5.4. Empirical method ... 22

6.

Empirical analysis ... 24

7.

Discussion ... 30

7.1. Econometric analysis discussion ... 34

8.

Conclusion ... 35

References ... 36

Figures

Figure 1: Convergence goal funding (European Commission, 2015b). ... 4

Figure 2: Regions with highest GDP per capita, own computation (Eurostat, 2020) ... 5

Figure 3: Regions with lowest GDP per capita, own computation (Eurostat, 2020) ... 6

Figure 4: Map of studied NUTS-2 regions ... 7

Tables

Table 2: Summary table of used variables ... 19

Table 3: Descriptive statistics table of levelled parameters, own computation ... 24

Table 4: Pair-wise correlation matrix, own computation ... 25

Table 5: Preliminary regression results ... 26

Table 6: Lag optimization under pool OLS ... 27

Table 7: Lag optimization under FE ... 28

Table 8: Results output table ... 29

Appendix

Appendix 2: White´s heteroscedasticity, Wooldridge serial-correlation test – at level .... 42

Appendix 3: Hausman test result ... 43

Appendix 4: Breusch-Pagan LM test results ... 43

Appendix 5: White´s heteroscedasticity, Wooldridge serial-correlation test – logs and lagged var. ... 43

Appendix 6: Pool OLS robustness test ... 44

Appendix 7: FE model robustness test ... 46

Appendix 8: Country, region code and name list ... 48

Definitions list

(E)SF – (European) Structural Fund CEE – Central and Eastern Europe CF – Cohesion Fund

EAFRD – European Agricultural Fund for Regional Development EAGGF – European Agricultural Guidance and Guarantee Fund EGT – Endogenous Growth Theory

ERDF – European Regional Development Fund EU – European Union

FE(M) – Fixed Effects (model) GDP – Gross Domestic Product

GMM – Generalized Method of Moments NATO – North-Atlantic Treaty Organization NEG – New Economic Geography

NUTS – Nomenclature of territorial units for statistics

OECD - Organisation for Economic Co-operation and Development OLS – Ordinary Least Squared

R+D – Research and Development V4 – Visegrad Four / Visegrad Group WLS – Weighted Least Squares

1. Introduction

Structural Funds mainly exist to provide financial means by the EU to speed up and encourage convergence processes and tendencies, respectively. However, their overall effectiveness is often questioned, especially in a smaller, a more local context of individual countries or group of countries with similar economic, socio-political background – namely large parts of new member states situated in Eastern Europe. Since the latest EU-enlargements, this has become more questionable (Ederveen et al., 2006; Katsaitis and Doulos, 2009), particularly in Czech Rep., Slovakia and Hungary; where a relatively high rate of corruption or manipulation of SF, in economic and political context exists. However, policy makers take certain interest in knowing factors responsible for determining regional per capita income levels and catching up of poor regions, also known as regional convergence (Armstrong and Taylor, 2000). Since induction of new member states into ranks of EU, and EU´s primary goal of economic cohesion, convergence theories and subsequent studies are of high importance (Ederveen et al., 2003). Even if EU regions do not shown signs of convergence, Structural Funds play crucial role in monetary support through sets of regional policies to those lacking areas (Armstrong and Taylor, 2000).

Thus, convergence studies are still applicable and important, even more so since the expansions of EU memberships´ to countries, in 2004, 2007 and 20131 _{- therefore, providing} additional reasoning to keep analysing and reviewing Cohesion policy and economic and political convergence between countries and regions. Previous studies and literature findings suggest either positive (Barro and Sala-i-Martin, 1991; Mohl and Hagen, 2010; Becker et al., 2010;) or negative or ambiguous results (Borsi and Metiu, 2014; Dallerba and Gallo, 2008) with convergence and regional economic growth across EU and V4, alike. Visegrad Group countries have similar foundations; historical, cultural as well as economical (Lengyel and Kotosz, 2018). In comparison to other EU member states, their economic development; mainly GDP; has been steadier and less volatile to exogenous economic shocks over the years, as pinpointed by OECD´s statistics (OECD, 2020). Appendix 1 shows an economic development of all EU countries and demonstrates the catching-up pattern of V4 countries compared to the western countries of EU, often even surpassing old member states such as

Spain, Portugal, and Italy. As such, this provides researchers and politicians alike to effectively monitor and analyse necessary policies and tools, with aim of finding optimal conditions and factors, which affect either positively or negatively the entire process of integration. The necessity of integrational processes gave rise to concept convergence, though its origins are from neoclassical and endogenous growth theories (Solow, 1956; Swan, 1956; Romer, 1987; Blazek and Uhlir, 2002; Sala-i-Martin, 1994; Sala-i-Martin, 2004). Convergence may be explained as a process when two entities in either their socio-political, environmental or economical forms “converge” together, or in other words reduce their differences over time (Barro and Sala-i-Martin, 1991; 1992 and 2004). Their empirical studies and implemented models give us with means to quantitatively measure differences or disparities between individuals or groups. Beta-convergence is a concept when economies or countries tend to converge towards their long-run or some steady-state position – hence, those at lower initial levels should converge or grow at a faster rate than those at high initial levels, ceteris paribus (Barro and Sala-i-Martin, 2004). This concept is mainly based around the theoretical findings of Solow (1956) and Swan (1956), detailed in their Solow-Swan neoclassical model of economic growth. Sigma- convergence, on the other hand, occurs when the overall income dispersion reduces overtime (Barro and Sala-i-Martin, 2004). Therefore, this paper aims to empirically investigate, whether Structural Funds have any significant (positive or negative) impact on the rate of regional economic growth, namely the speed of convergence, or ultimately divergence, respectively. This paper relates to existing literature and studies the aspect of regional economic convergence on NUTS-2 level, considering all EU member states, with additional emphasis studying V4 or Visegrad Four group countries (Nevima and Melecky, 2011; Tvrdon and Skokan, 2011; Tvrdon, 2012; Zdrazil and Applova, 2016; Kuc, 2017; Lengyel and Kotosz, 2018; Zacek et al., 2019) – namely Czech Republic, Slovakia, Hungary and Poland. It will investigate and answer this through studying 2000 to 2016 period observations for over 200 NUTS-2 regions. Convergence (speed of) will be evaluated by allowing empirical testing of beta convergence. Henceforth, this paper will contribute to existing literature field by analysing and identifying and testing for larger time periods and observation panels (e.g. whole EU), using existing econometric methods - namely OLS panel data and FE models. As mentioned above, more emphasis will be focused on V4 countries.

Further background information on Visegrad Four and Structural funds will be provided in the following section. Other sections will delve into theoretical groundwork, with presented literature review and findings. Later parts will include a methodological approach to the paper, with faced limitations, followed by data description, empirical analysis, and an overall discussion of results. The last section will cover concluding remarks with potential policy implications.

2. Background information

The following section explains Structural Funds and EU funding, V4´s current and future outlooks with respect to Structural Funds, an overview of their competitiveness, economic outlook. Ending section will provide understanding into the territorial division of EU regions.

2.1. EU funds and objectives

EU operates its economic and regional policies through a series of funds. These funds effectively are allocated from common EU budget, covering the financial aspect of EU projects. Its national countries contributions are composed of measures via taxes, duties, levies etc (European Commission, (n.d.-a). Budget and complementary policies are then drafted and implemented in budgetary cycles that take a turn in 7-year term. Tested data will cover three latest budget cycles: 2000-2006, 2007-2013 and 2014-2020.

Individual programmes are then funded by the collected revenue from entire EU budget, and its main channels are then diverted into goals, such as regional cohesion, competitiveness in jobs, citizenship and security or sustainable growth (European Commission, (n.d.-b). However, in the first two-previously mentioned cycles, the majority of expenditure was funnelled into three main objectives – Objective 1 aimed at regional development and cohesion between lacking and developed regions (e.g. if GDP of that region was 75% of EU- average)2,3 _{(OECD, 2006), whilst Objective 2 sought to support economic and social} conversion in areas with difficult access and structural integrity. Finally, Objective 3 was

2 _{Objective 1 included further regional and policy division – for example, phasing-out and transition regions,}

which were above 75 % average-EU GDP or just crossed it, respectively.

3 _{SF are distributed based on the pre-programming period evaluation of the current state of individual regions}

implemented to support training and educational programmes and employment opportunities. Despite the relative complexity of the funds and its purposes, essentially, as shown by Figure 1, all three funds finance the convergence objective (previously Objective 1, later known as Economic, Social and Territorial Cohesion).

For example, 2007-2013 cycle allocated around 347 billion EUR (European Commission, 2015a), a similar percentage of EU funds towards convergence as 2014-2020 period, total monetary value however varies depending on year.

Figure 1: Convergence goal funding (European Commission, 2015b).

The actual funds that allow the proper funding of all EU´s current and long-term objectives are mainly ESIF (short for European Structural and Investment Funds). This group consists of sub-funds, such as ESF (European Social Fund), CF (Cohesion Fund) or ERDF (European Regional Development Fund), as well as others (DotaceEU, n.d.). Since CF and ERDF are the main forms of financing regional disparities, their distribution and recipient countries are determined at end of each budgetary period, where an overall evaluation of GDP (regional and national) in relation to EU average subsequently determines the funding for following periods. However, as Figure 1 shows, the convergence task was effectively funded by all three main sub-funds of Structural Funds, and therefore, later in our analysis it will not be differentiated between them and all will be considered/accounted for.

2.2. Historical and current outlooks of V4

Origins of their cooperation can be traced as back to as Austro-Hungarian empire. After series of geopolitical events throughout 20th century, V4 members (namely Czech Republic, Hungary, Slovakia and Poland) were initiated into two strong blocs of further development and coordination – Eastern bloc (mostly political) and subsequently EU (economical) and

NATO (political). From the end of the '80s and early '90s, all four countries began their economic transformation and transition to a more liberal, open market economies. And with it, came more apparent social and economic disparities both internally, but especially externally.

In terms of GDP per capita, V4 regions are relatively worse off than other western EU capital regions, whilst some of their non-capital regions are close to being described as the most lacking and less developed regions, in GDP per capita terms. Figures 2 and 3 show the 10 most and least developed regions in the EU, according to Eurostat and the later mentioned criteria. These regions have been ranked from highest to lowest depending on their 2016´s stance. From Figure 3 we can see than traditional Western Europe metropolitan regions dominate in GDP per capita terms, with London, Luxembourg, Stockholm, and Brussels in top 4, all with exception of West London well above 62 000 EUR per head. Even though GDP per capita is not an accurate representation of intranational as well as international economic disparities, it roughly illustrates these disparities, nonetheless. Bratislava and Prague, Slovakian and Czech capital regions have reached in 2016 figures of almost 37 000 and 35 000 EUR per head, respectively. Therefore, significant economic disparities are visible as these values are almost half of the most developed regions in Europe. On the other hand, Figure 3 shows the 10 least developed states ranked in accordance to their 2016 data.

Figure 2: Regions with highest GDP per capita, own computation (Eurostat, 2020)

0 50,000 100,000 150,000 200,000 250,000 in E U R

Highest 10 NUTS-2 regions (per capita GDP)

Though none of the V4 regions is listed there, as these are often occupied by rural regions of Bulgaria, Romania or former Yugoslavia with values around 4 000 to 7 000 EUR per head, some V4 regions are not that far behind – for example, majority of Hungarian regions are between 7 700 and 8 400 EUR, followed by Vychodne Slovensko region (i.e. Eastern Slovakia) with 10 400 EUR averaged per capita in year 2016 (Eurostat, 2020). Hence, country development figures are often driven by capital regions only, and thus overall country data is sometimes misleading, as is demonstrated by large variance in economic development in countries, especially in Central and Eastern Europe. However, these large variations may be also accounted for by the existence of misuse, mistreatment of SF, and present corruption schemes in the mentioned countries.

This may pose difficulty in later empirical testing and will be addressed later in discussion, as results of regional convergence or divergence may not be fully reflected by presence of the above mentioned.

Figure 3: Regions with lowest GDP per capita, own computation (Eurostat, 2020)

0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 in E U R

Lowest 10 NUTS-2 regions (per capita GDP)

Figure 4: Map of studied NUTS-2 regions

Figure 4 shows the map of all studied NUTS-2 regions included in latter empirical analysis. Blank areas, such as Turkey, Switzerland, Russia, or former Yugoslavia´s countries have been excluded from dataset, due to insufficient data observations. Regions in red (e.g. mostly capital regions, except Hamburg and Oberbayern – both Germany) represent the ten highest per capita GDP regions, whilst in blue, are the ten lowest per capita GDP regions – i.e. all of Bulgaria and most of Romania. Full list of all region’s names used, respective countries and ID codes will be provided in the Appendix.

2.3. NUTS territorial classification

As mentioned before, this paper will study regional economic growth and convergence of regions across the EU at NUTS-2 level. According to the EU (Eurostat, n.d.-a), a NUTS level 2 region has between 800 thousand to 3 million inhabitants. SF are distributed on NUTS-2 level, however unlike some studies contributing to convergence literature (Tvrdon and Skokan, 2011; Lengyel and Kotosz, 2018; Zacek et al., 2019), this paper´s analysis may result in less specific, and accurate results than NUTS-3 studies. Furthermore, it’s important to notice that few regions that are classified as NUTS-2, do no fulfil its criteria, and thus make overall NUT2-level data usage challenging – for example, some regions/countries,

such as in Baltics, are too small (e.g. population wise) and must be classified as both NUTS-1 and 2. On the other hand, regions like Oberbayern (Germany) or Andalucia (Spain), have population well exceeding this limit, but further analysis of actual NUTS territorial division and assigned ID´s reveal the same code. Hence, despite this, for remainder of this thesis, all regions with correct NUTS-2 ID and data availability will be treated regardless of EU conditions.

3. Theoretical background

Economic growth theory has been divided into two main channels, convergence, and divergence (Capello, 2007). Both channels represent the theoretical framework for understanding socio-political, but mainly economic disparities that may arise between and within countries alike. Beyond the scope of international convergence, inter-regional as well as intra-regional convergence patterns and motions; regional convergence tendencies and processes can happen due to similar reasonings as on a national level would. This is particularly the point made by Solow (1956), Swan (1956) and Barro and Sala-i-Martin (1991, 2004); whom we believe to be pioneers behind exploration and explanation of convergence theories, originating from neoclassical growth theories brought by Solow and Swan. Barro and Sala-i-Martin subsequently continue in finding both theoretical and empirical support for neoclassical growth and convergence theories, though they deviate a bit from core beliefs of the neoclassical stream.

On the other hand, more recent theories reveal a disagreement in the ideas of economic convergence and argue otherwise. For example, endogenous growth (Romer, 1986; Lucas, 1988) and new economic geography theories (Krugman, 1991; Fujita and Krugman, 2004) base their principles around the idea that regions or countries do in fact not converge, and it’s a matter of competitiveness, competitive advantage, and core-periphery ideals.

Therefore, both convergence and divergence streams will be further on discussed in relation to their depiction and understanding of economic disparities. As mentioned above, the core of convergence group has been formed around neoclassical Solow-Swan theory and its subsequent model, later followed by Barro and Sala-i-Martin´s beta and sigma convergence concepts. In this field, we may also categorize club convergence theory, as it incorporates both channels but projects them on common convergence characteristics and patterns that entities share when being subject to actual economic convergence itself (Alexiadis, 2013). Alternative theories will be presented, especially in the divergence field of the framework, as various authors and researchers have built-up on existing theories.

3.1. Neoclassical-convergence approach

Neoclassical growth theory and its subsequent model predicts that countries or less developed regions should catch up to more developed economies. Joining economic groups, unions, or communities, should promote and incentive their own economic growth and thus result in economic convergence overall (Solow, 1957; Swan, 1957). It theoretically assumes that factors of production are mobile, and trade is liberalized, and thus, convergence occurs (Martin et al., 2001). Economic growth can, therefore, be derived from a combination of capital, labour, and its level of technology, and is based around capital accumulation (Solow, 1956; Swan, 1956). Despite the growth of human capital, or labour, does not entirely lead to continuous growth as technology does, some argue that regional competitiveness may be perceived as a key factor for long-term convergence processes (Huggin et al., 2013; Banociova and Martinkova, 2017). Furthermore, neoclassical growth theory predicts, that regions with matching saving rate4_{, depreciation rate, population growth and technology will} converge.

Structural Funds mainly finance, as discussed in the background section, physical capital in less developed regions of EU regions, and hence regional growth can be temporarily fuelled above their normal steady-state growth levels. However, due to the assumption of decreasing marginal product of capital, such policy should therefore only contribute to faster convergence to pre-existing steady-state (Dallerba and Le Gallo, 2008). Hence, integration of countries leads to sharing and accessing various technological upgrades and innovations, thus encouraging convergence without the need of exogenous stimuli, in the form of for example SF.

Overall, however, neoclassical growth theory states that long-term growth in output may be reached by technological advancements only (e.g. technological advancements are vital in explaining different speeds of economic growth between regions), and that manipulation of required savings, population growth or the rate of capita depreciation may only lead to new steady-state (Solow, 1956; Swan, 1956; Blazek and Uhlir, 2002; Barro and Sala-i-Martin, 2004). In the neoclassical framework, SF would foster economic growth as they increase the actual as well as the rate of investments. However, since they are to specific projects, though

4 _{Due to aims of Cohesion Policy, conceptual understanding of SF and its nature, SF can be interchangeable}

mainly physical infrastructural projects, their nature can vary as much as environment or culture support investments, therefore not directly promoting economic growth.

However, there is a risk that using SF as additional public investment funding in less developed regions can crowd out governmental funding. This happens because national governments choose optimal budget allocation (Ederveen et al., 2003). Similarly, the presence of corruption on a national or regional level can defer SF from its original purpose, and thus leading to an inefficient fund allocation which may result in ambiguous results (Ederveen et al., 2006; Katsaitis and Doulos, 2009). But, due to relatively high risks of corruption in Central and Eastern European countries, this can have a significant effect – despite lacking later literature findings and empirical evidence (Mundgui-Pippidi, 2014). Recent modifications to neoclassical growth theory and further empirical development of convergence theories by Barro and Sala-i-Martin (1991,2004), implies that researchers and authors have means to measure the process of economic convergence using statistical, economic, and econometric methods. These pioneers of advanced conceptual understanding of convergence argue that rich countries would be caught up by poor countries, which are eventually growing at a faster rate. This is referred to as beta-convergence (Barro and Sala-i-Martin, 2004).

Beta- convergence tends to lead to sigma- convergence, on the other hand, sigma- convergence presence is also a presence of beta- convergence, and not vice versa (Barro and Sala-i-Martin, 2004; Blazek and Uhlir, 2002; 2011). Though not any significant, new theoretical concepts are provided (e.g. Barro and Sala-i-Martin were heavily influenced by neoclassical growth theory), consequently, it allows for empirical testing of convergence concepts. Similarly, to the Solow-Swan model, various adaptations of convergence, such as conditional and absolute convergence, can be derived from new convergence concepts. Whilst conditional convergence occurs when economies tend to grow faster, the further they are away from their steady-state, absolute, on the other hand, means that poor states tend to grow faster than rich ones, irrespective of any other characteristics of that economy. In practice, this implies that economies would grow unconditionally forever unless they are subject to other control variables, which may cause changes to steady-state level (Barro and Sala-i-Martin, 1991; Blazek and Uhlir, 2002; Barro and Sala-i-Martin, 2004).

3.2. Divergence-orientated theories

In contrast, there is abundant theoretical framework opposing findings and beliefs of neoclassical growth theory, along with more recent development in convergence studies by Barro and Sala-i-Martin. Two fundamental theories contributing to this theoretical structure are endogenous growth (EGT) and new economic geography (NEG) theories. Whilst NEG emphasize market integration, scale economies, transport costs and home market effects (Krugman, 1991; Fujita and Krugman, 2004; Ottaviano and Puga, 1998), EGT focuses on human capital and innovation – the movement of the “technological frontier” and positioning with it (Romer, 1986; Lucas, 1988; Aghion et al., 2005). The endogenous growth model is not built to explain convergence, but rather give the reasoning for the existence of divergence (Mankiw et al., 1992). Since it assumes technology creates more tech, poorer countries will remain poor due to their inability to generate and invent (Ederveen et al., 2003). Innovation and thus endogenous research and development spending are main thrusters of economic growth (Romer, 1986; Lucas 1988). On other hand, Krugman in NEG states that economic activities are concentrated in the centralized clusters (Krugman, 1991; Fujita and Krugman, 2004; Ottaviano and Puga, 1998). Additionally, according to Krugman, regional competitiveness has more to do with absolute advantage than with comparative advantage. Hence, more productive regions should lead, as they attract more labour and capital. Economic integration brings about strong competitive position for many countries, however, particularly in less developed regions, the existence of monopolies, oligopolies and signs of spatial concentration can be harmful (Krugman, 1991; Fujita and Krugman, 2004; Ottaviano and Puga, 1998). Sub-branch of divergence theories, the neo-Marxist theories; are heavily based around the idea of inequality, that arises due to the existence of monopolies and oligopolies, which eventually create divergence processes. And therefore, an only viable solution is planning changes in society, that take place on a national, not local level (Blazek and Uhlir, 2002). Potential is unevenly distributed across space; hence specialization and establishment of core and periphery regions are unavoidable. However, should a region engage in accumulation and creation of old and new technology, respectively, their transaction costs will maintain low in comparison to non-engaged countries or regions and thus giving them a considerate advantage over the other. Therefore, to reinforce agglomeration of economic activity, setting up institutions facilitating innovation, research and development are vital for further growth (Lundvall, 1992). But inequalities may not be

harmful in the short run if their growth patterns are robust. Competing of slow and fast-growing countries can result in winners for both – even if the disparity level or change increases, both groups should be better than their initial positions. Nevertheless, we mainly observe agglomeration forces in strongly metropolitan areas, and since these are often space or expansion-potential limited, ultimately, they may suffer from diseconomies of scale (Duranton and Puga, 2000).

Neoclassical theories argue in the dispersion of economic activity, however as past literature findings suggest, the empirical evidence for this motion is sparse. Myrdal (1957) argues that economic activity is cumulative. His theory suggests any regions develop disproportionally, e.g. one region grows at a faster rate than others; then, as a result, both will continue developing at different speeds, divergence process will accumulate and culminate. This occurs if primary factors change, thus forcing on a change within secondary factors´ orientation and that snowballs again. Amongst primary factors, he lists agglomeration effects, the relative distance between individual markets, exogenous savings, uncontrollable productivity growth or fast changes in technologies. Hence, regions should invest in and plan future regional development strategies and layouts to cope with this (Myrdal, 1957).

Other divergence framework extensions include the following sub-theories - Perroux (1983) introduced a growth pole theory, based on assumption that economy contains faster growing and more important industries, with connections to other industry branches, thus keeping in check and influencing economic growth overall. Boudeville (1966) contributed to this with extension to growth pole theory by incorporating growth axis and midpoints, that dynamically interconnected industries are concentrated around a single “driving” force, a dominant industry. If we therefore assume and believe in the existence of such an industry in our region, the remaining firms will grow as a result, thus supporting and developing the region.

Harvey (2009) proposed a merge of the theory of uneven development with the theory of capitalist crisis, he believes that inter-regional disparities exist and deepens due to competition of local, or territorial alliances, their behaviour, technological advancement and investments into immobile infrastructure, which is characterized by economies of the scale who require a large concentration of people and capital (Harvey, 2009; Blazek and Uhlir, 2011). Emmanuel (1972) on the other hand claims that divergence happens due to unequal

international trading conditions, and regional policy should be focused on restricting trade between the two differently developed subjects.

3.3. Convergence clubs

Concept of club convergence has its main roots in empirical studies, rather than theoretical frameworks, nonetheless, they can be traced back to both channels of economic growth theories – convergence and divergence theories. Introduced by Baumol (1986), it is used to describe and reflect the relation between economic situations of subset of nations, with respect to world economy (Baumol, 1986). Clubs exists primarily due to exploitation of returns to scale, thanks to various specializations, their interactions and creation of ideas, which are group-limited and specific. Hence, countries may never catch up, leading to two extreme income-group poles overall. However, club convergence may exist as result of historical ties, economic dependency, and technological ties between nations (Cetorelli, 2002). So, mainly economies with similar conditions, such as level of income, structural characteristics and so forth, may exhibit likely convergence patterns (Alexiadis, 2013). Baumol et al. (1994) deem that convergence happens for regions after they are grouped, and as such can significantly spark their productivity and standards of living, should they believe in it. Hence, integration of countries into same trade blocs, unions or groups is beneficial, mainly for new members, since there is potential room for increased growth and opportunities to be “exploited” from such transition (Capello, 2007).

Table 1 demonstrates a summary of the studied theories with respect to convergence and divergence literature and research, following Gardiner et al. (2004). The table presents the summary of main studied theories, their authors and framework´s contributors, as well as core theoretical motives and developments are also described within the following table.

Table 1: Summary table of used literature and theories, own computation

Channel Theory Authors Reasoning Development

Convergence Neoclassical growth (Solow, 1957;

Swan, 1957) Technical homogeneity Assumed constant returns to scale, factor mobility, conditional convergence – i.e. low countries catch up Divergence Endogenous growth (Romer 1986,

Lucas, 1988) Technical heterogeneity Innovation, Knowledge, and technology are localized, hard to achieve and accumulate Divergence New Economic

Geography (Krugman and Fujita, 1991; Krugman, 2004) Clustering, agglomeration, spatial specialization Integration leads to spatial agglomeration of econometric activity, boosting developed regions first Divergence Neo-Marxist (Myrdal, 1957;

Boudeville, 1966; Perroux 1983; Harvey, 2009) Inequality, concentration of economic activity “snow-ball effect”, strongest regions win - attract more people, firms etc.

Both “Convergence club” (Baumol, 1986) Group or club

convergence Countries develop or grow in respective groups, based on similar characteristics and other bonds

4. Literature findings

Barro and Sala-i-Martin (1991) work gave rise to increasing focus and devotion to economic measures of both national and especially regional convergence. The original study is based on regional convergence of US states and incorporates data since the 1880s, up to 1988. It mainly analyses patterns and evolution of economic growth developments over decades, using a series of regional dummies. Though its focus lies within US data, the study also includes analysis of western European regions, with the exclusion of Spain and Portugal, for a relatively similar period. This study includes the application of beta- and sigma-convergence models, which are based predominantly on neoclassical growth theory (Barro and Sala-i-Martin, 2004, Blazek and Uhlir, 2002 and 2011). Pro-convergence results are gathered (e.g. average 2% annual speed of convergence is concluded), and both results and approach form a foundation for future convergence studies. The presented literature findings will be distinguished into three themes, where individual past literature contributions will be

presented: studies on regional disparities and evaluation of Structural Funds studies and other remaining findings.

4.1. Regional disparities studies

The following sample of regional disparities studies differentiates from the second theme by failing to directly address the issue of economic convergence concerning Structural Funds effects and economic relationships amongst them, as is the aim of this paper.

Nevima and Melecky (2011) study regional convergence of NUTS-2 V4 regions from 1995 to 2008 using panel data models, such as pooled OLS and LSDV (i.e. Least Squares Dummy Variables). Their approach involves a simple use of dependant and independent variables in an unconditional convergence study. Hence, though accounting for convergence study, they, indifference to this study, fail to address other control variables which may affect the overall significance of regional convergence speeds. Nonetheless, results drawn show an inconclusive conclusion towards economic convergence. Tvrdon and Skokan (2011) on the other hand implement a NUTS-3 V4 data approach. To comply with sigma-convergence, authors use a series of indices, such as Theil and Herfindahl index or Gini coefficient, similarly to Tvrdon (2012). However, his later work in based around NUTS-2 regions only. Nevertheless, Tvrdon (2012) concludes that economic convergence is effectively taking place between 1996 to 2007, but not within new member states due to increasing disparities. Tvrdon and Skokan (2011), Tvrdon (2012), and Lengyel and Kotosz (2018) apply the similar methodological approach, involving measures like Theil index for V4 and EU-15 initial members, to test for economic convergence overall. In between 2000 and 2014, their contextual data-setting, regional convergence based on NUTS-3 level is claimed, however, countries like the Czech Republic and Hungary experience drawbacks and thus increasing disparities between 2006 and 2008. Zdrazil and Applova (2013) also address V4 regions throughout 2000 to 2013 and conclude general convergence tendencies using a sigma-convergence method on NUTS-2 level, however, the core of this paper´s empirical testing utilizes beta-convergence methodology, hence the difference.

Smetkowski and Wojcik (2012) and Kuc (2017) incorporate mode recent econometric evaluation techniques by utilizing spatial autocorrelation models to explain regional economic convergence. Smetkowski and Wojcik (2012) use NUTS-3 regional data of new

member states for the period of 1998 to 2005, whilst Kuc (2017) focuses on both EU-27 and more-so on V4 regions between 2004 and 2014. Both studies conclude convergence, however, earlier paper´s results demonstrate weak economic convergence using augmented neoclassical growth convergence set-up (e.g. endogenous technology augmentation and log-t-test), whilst later study´s results infer social convergence rather than economic. Both also use beta- and sigma-convergence approaches but incline later to spatial models.

4.2. Evaluation of Structural Funds

The following sample of past studies consists of those, which to some extent evaluated the effects and statistical relationships of Structural Funds granted by EU, on regional economic growth. For example, Cappelen et al. (2003) conclude a significant and positive impact of EU regional support on growth, given their use of Objective 1 framework. Similarly, to this paper, authors used a comparable set of exogenous variables but add on infrastructure, industry sectors and unemployment as well. Their study is set between the 1980´s and 1990´s western Europe regions, however. Dallerba and Gallo (2008) on the other hand find no statistical impact of SF on regional convergence between 1989 and 1999. Their use of 145 NUTS-1/2 regions and using cross-sectional and spatial lag with IV (e.g. instrumental variables) study brings about interesting results, nonetheless. Esposti and Bussoletti (2008) through GMM method applied on 206 EU-15 NUTS-2 regions reach inverse results however, as they conclude their study with a positive impact of SF on Objective 1 regions across all EU. In relation to Dallerba and Gallo (2008), their study also includes series of variables such as infrastructure indices, human capital and research and development, but due to its almost identical test period coverage of 1989 to 2000, both sets of results are rather unexpected.

Becker et al. (2010) and Mohl and Hagen (2010) implemented different empirical strategies, but with similar results. Becker et al. (2010) studied the impact of growth effects on Objective 1 regions over 1989 to 2006 period, accounting for over 1000 NUTS-3 regions using regression discontinuity methodological approach. Their main variable of testing consisted of a dummy, with a value of 1 if the region received Objective 1 funding or not, 0 otherwise. Results were positive and significant, just as by Mohl and Hagen´s (2010) study. In this, however, variables have been studied in per capita form, with logged relationships and combined method of both panel data models such as pool OLS and FE, with more advanced

Objective 1 funding, whilst combined Objective 1-3 funding was insignificant. At last, Zacek et al. (2019) applied similar methodological approach to that of Mohl and Hagen (2010) using a combination of models to achieve similar findings, however, only for Czech Republic regions in-between 2004 to 2015. Both studies also confirmed strong regional spillovers effects.

4.3. Other findings

Regarding the remaining literature findings, most of these, though sharing either their territorial data focus or methodological approach, stand out from previously addressed findings with their focus being steered towards finding evidence of club convergence. Smetkowski and Wojcik (2012), as mentioned above, use spatial econometric models and Kernel density to find the presence of group convergence. Similarly to Bartkowska and Reidl (2012), both studies incorporate log t-test to determine the presence of club convergence, however, whilst evidence of weak convergence is found by Smetkowski and Wojcik (2012), the latter authors find strong support for club convergence, especially in western Europe NUTS-2 regions between 1990 to 2002. Borsi and Metiu (2014) study EU regions for 1970 to 2010 period and conclude no significant economic convergence, but the overall presence of club convergence – which authors base on conditions of geographical locations, similar historical ties, but not membership within Eurozone. Findings by Horridge and Rokicki (2017) have mainly theoretical contributions, as they study V4 regions and specifically the effects of convergence have they not joined the EU, using 2000 to 2013 data. Their outcomes are in favour of joining as if not, regions would develop a slower convergence tendency. Given the theoretical and literature frameworks, this paper will incorporate and answer the following hypothesis questions to strengthen further the research question of regional economic convergence and potential impacts of Structural Funds upon it. Therefore, in this paper, we would like to reject the following hypotheses:

1) H0: Structural Funds do not stimulate (speed of) regional convergence 2) H0: Structural Funds in V4 regions exhibit no signs of convergence

a. H0: No evidence of club convergence

Previous theories and particularly past literature findings are somewhat conflicted as to the effect of Structural Funds on regional convergence, as was clear in earlier sections. Despite

statistical significance of convergence is expected, same cannot be so far said for Structural Funds as such. This is even more so the case with V4 regions, to which this paper pays a greater emphasis. Personal motivation behind this paper is based around finding the localized empirical effects of convergence within Visegrad Group. And lastly, convergence club theory, though commonly accounted for in existing literature reviews and findings, the nature of detecting, testing, and measuring club convergence presence varies across literature – thus, not as important in current context, an empirical attempt will be made.

5. Data, Methodology and Models

Empirical section of this paper utilizes secondary data collected from one source. Since the aim of this paper is to study regional convergence across EU, only source with such data is

Eurostat (Eurostat, n.d.-b), which provides regional data based on typology or NUTS-level

specification.

Hence, this paper collected observations on the following variables: Table 2: Summary table of used variables

Dependent Variable Expected coefficient sign

Description Regional economic growth =

𝑮𝑫𝑷𝒊,𝒕

𝑮𝑫𝑷𝒊,𝒕−𝟏

Region i (per capita) growth at time t, with respect previous period

Main variables of interest

GDP per capita i, t negative GDP per capita in a region i at

time t

V4 dummy variable unknown (1 if region belongs to V4

country, 0 otherwise)

Structural Funds per capita i, t positive Structural Funds per capita in a

region i at time t

V4-SF Interaction term t unknown Interaction term of being V4 and

receiving per capita SF at time t

Control variables

Innovation (R+D proxy) per capita i, t

Positive/negative Intermural R&D expenditure per capita

Population growth i, t positive Population growth (t – t-1) in a

region Human capital (education proxy) i,

t

Positive/negative Percentage of 25-64 yrs, with tertiary education in region

The data for the following variables is thence collected for 2000 to 2016 period, with annual observations. Whole sample consists of 260 NUTS-2 regions, accounting for almost 29 000 observations in an unbalanced panel. Analysis will account this by using FE and Pooled OLS modelling. Some NUTS-2 regions have been excluded (ex. Slovenian and Croatian regions, London) due to too many missing observations, and likely presence of data outliers, respectively. In addition, French overseas territories have been excluded to keep a constrain to continental-Europe regional convergence study.

5.1. Model description

For this paper, the following models will be presented and analysed furthermore. Standardized ß-convergence equation based on neoclassical model as presented by Barro and Sala-i-Martin (1991,2004) and Blazek and Uhlir (2002, 2011), is transformed similarly, to Ederveen et al., (2006), Mohl and Hagen (2010) and Zacek et al. (2019). The resulting empirical model is presented as follows

𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡= 𝛼𝑖,𝑡+ ß𝟏 ∗ 𝐥𝐨𝐠 𝑮𝑫𝑷𝒊,𝒕+ ß𝟐 ∗ 𝑽𝟒𝒕

+ ß𝟑 ∗ 𝑰𝒏𝒕𝒆𝒓𝒂𝒄𝒕𝒊𝒐𝒏 𝒕+ ß𝟒 ∗ 𝐥𝐨𝐠 𝑺𝒕𝒓𝒖𝒄𝒕𝒖𝒓𝒂𝒍 𝑭𝒖𝒏𝒅𝒔 𝒊,𝒕+ ß5 ∗ log 𝐼𝑛𝑛𝑜𝑣𝑎𝑡𝑖𝑜𝑛 𝑖,𝑡

+ ß6 ∗ 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 𝑖,𝑡+ ß7 ∗ 𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑖,𝑡+ 𝑒 𝑖,𝑡(1)

Where, growth uses GDP values of region(s) i at time t and t-1; ßj = 1,2…7 represents the slope parameters for respective independent and control variables, where special notice should be made towards ß1 to ß4. T is referring to study period of 2000-2016, and ei,t stands for a random error term.

This paper considers conditional convergence, in line with ß-convergence approach and thus, series of control variables are applied additionally. An explanatory variable partially equivalent to the dependant one, must be regressed as independent variable, therefore allowing for empirical measurements of convergence or divergence to take place. The following model as seen by Equation 1 will be modified accordingly to test for its different variations. Motivation for these changes will be addressed in the upcoming sections of the paper.

5.2. Variable description

Structural Funds

This, main variable of interest, represents the combined sum of all EU Structural funds inflows received by any country at a given year, in Euros in current prices of that year. The sum is then divided by the population of each respective region i at time T, also obtain from Eurostat. Hence combined EU investments into physical and non-physical infrastructure, and subsequent transform to per capita form, are explained under this single variable, irrespective of the origin fund (e.g.EAFRD/EAGGF, ESF, CF, ERDF).

GDP per capita

Economic convergence is relatively easy concept to study and measure empirically. Often it is explained by changes in level or growth of GDP per capita figures. This variable is capturing the income per head, denoted in Euros (PPS). However, further methodological approach and nature of empirically understanding the concept of convergence varies across studies.

Education

Tertiary education variable serves as a proxy for human capital. Ultimately, this figure tells us the percentage rate or share of population, aged between 25-64 years (e.g. working class) with tertiary education – i.e. short-cycle tertiary education, bachelor´s, master´s and doctoral or equivalent degrees, as in accordance to Eurostat database. Nature of choosing optimal proxy used to explain human capital is difficult and varies between studies, however, education is often considered as standard approach when evaluation economic effects of human capital.

Population

Populating growth variable may be understood as rate of population growth between year T and T-1. Changes in population growth, positive or negative, should reflect any potential economic impacts of growth in inhabitants on state of economy and its convergence abilities.

Innovation

Innovation can be regarded as additional economic boost to convergence patterns in between countries and regions alike. It accounts for additional research and development spending to the regional and national economy. Innovation is proxied by R&D per capita spending, denoted in Euros (PPS). According to Eurostat´s database, this represents total intramural R&D expenditure from all economic sectors (e.g. business enterprise, government, higher education, private non-profit – irrespective of source of funds) in Euros per inhabitant.

5.3. Variable and data constraints

The presented data, gathered from Eurostat suffers from shortages in reporting and thus contains a significant amount of missing observations, especially in R&D per capita variable. This, along with other variables may be due to individual statistical offices reporting mechanisms, and general difficulty in collecting such data (e.g. regional) in the first place. Nevertheless, regarding the main variable of interest; SF per capita; this variable is based on historical EU-fund expenditure data and thus, is challenging to pinpoint in combination with previously stated EU Objectives and Convergence goals, as to how much these readings are reliable. On the other hand, if we assume that SF shares similar characteristics of FDI inflows, this results in further confusion because FDI inflows and investments, in general, take years before its economic impacts can be empirically measured. Moreover, according to EU, Structural Funds maybe, due to complex processes, administered for additional 2-years after the end of last programming period (European Commission, n.d.), which makes their economic backtrack even more complicated.

5.4. Empirical method

For the empirical analysis of this paper, a panel data analysis using OLS is applied. Advantage of using panel data lies within maximization of observations, by combining time-series and cross-sectional data, resulting in more observations in econometric analysis. Increased degrees of freedom are also carried with it, as well as a reduction in collinearity problems (Gujarati and Porter, 2009). However, along with it come numerous assumptions of both cross-sectional and time-series data and more.

Furthermore, a combination of pooled OLS with FE models is undergone to observe for the difference in results, as well as to determine to a most suitable and efficient econometric model for this paper´s given data set. FEM´s strength lies in several attributes – it accounts for unobservable heterogeneity problems, and its results are often more robust than efficient, as is the case of the RE model. However, due to the nature of this paper´s data, the number of cross-sections N outnumbers the amount of periods T. As such, the REM model should be used. Nevertheless, since this paper is set as a regional study of economic effects, a FE and pooled OLS, with use of country and year fixed effects, provide a strong case and basis for empirical interpretation and discussion. Other means of empirical testing of regional effects may be done by uses of more advanced econometric models, specifically tailored to spatial effects such as GMM or spatial error and autoregressive models. This, however, is not the focus of this study, and though its potential future benefits are measurable, they will not be accounted for in the later empirical analysis. Still, negligence of spatial heterogeneity and spatial autocorrelation can lead to biased results (Arbia et al., 2008; Badinger et al., 2004) Nonetheless, related to FEM, use of fixed effects provides a degree of consistency under both scenarios – e.g. should Hausman test suggest the use of either FEM or REM. If, however, Hausman test cannot reject its null hypothesis, simultaneously, an alternative Breusch-Pagan Godfrey LM test is run to decide between pooled OLS and REM – depending on the variance of random effects (Gujarati and Porter, 2009). In addition to standard diagnostic tests for panel data and associated potential problems of autocorrelation, heterogeneity, or multicollinearity – with trending data, arises a new problem. GDP and its alternative measures are often accompanied by developing trends over time, and thus stationarity-related problems may have to be controlled for, otherwise, spurious regressions may incriminate overall result interpretation.

6. Empirical analysis

Empirical analysis section is devoted to the presentation of the results of this paper. Firstly, descriptive and correlation tables present the nature of the data, with mean, min and max values and standard deviation. Secondly, regression output tables are analysed, and results latter presented and discussed afterwards.

Table 3: Descriptive statistics table of levelled parameters, own computation

Variable Obs. Mean Std. Dev. Min Max

Growth 3732 1.03 0.06 0.79 1.46 GDP 3992 24151.60 12074.69 1300.00 93900.00 V4 4420 0.13 0.34 0.00 1.00 Interaction term 4223 0.46 1.68 -11.97 6.60 Structural Funds 4322 102.96 153.84 0.00 1298.68 Innovation 3070 367.44 450.38 0.00 3737.30 Education 4296 0.24 0.09 0.04 0.56 Population 4420 0.00 0.01 -0.09 0.06

Table 3 presents the descriptive statistics of the paper´s data set. As seen above, analysis will be working with unbalanced panel with variables of varying observations. Each variable consists well over 4 000 individual observations (e.g. for 17 years and 260 regions), with exception of Innovation, which has just over 3 000, due to missing values obtained from Eurostat. Ultimately however, due to the above and following correlation matrix analysis, this parameter will be dropped throughout regression testing. Data set it mainly limited to its main and control variables, as only historical SF payments can be used, and regional data acquired from Eurostat often date to only 2000, or late 1990s. Largest SD is observable in GDP per capita variable, GDP per capita, as for reasons mentioned in background section. Average per capita income of all regions is 24 151 EUR. To establish and test for convergence using ß-convergence method, we cannot change or dispose of this variable. Table 4 presents a summary of correlations between selected variables. In difference to standard Pearson correlation, the table above shows pair-wise correlation matrix. This better shows the relations between variables individual observations only and excludes missing values. In a way, this is more applicable to this paper, given the nature of the data, though, the correlation coefficients changes between both approaches are minor.

Table 4: Pair-wise correlation matrix, own computation

In addition, the table shows all variable links to be statistically significant at a 5% level of significance. In combination with presented theories, we can establish a prediction of what sign (e.g. positive or negative) will individual variables likely bear once regressed. Despite its insignificance, GDP is demonstrating its expected coefficient sign. Conflicting results are seen between Structural Funds and Interaction term. Regarding past literature findings, both should be positively related, and are expected to be positive in accordance to economic reasoning. However, SF demonstrates a negative relationship, as well as Innovation parameter does. Here, however, economic expectations clash with theoretical background as pinpointed earlier in the paper by endogenous growth theory. Relatively strong correlations can be witnessed between Education and GDP and Innovation parameters, respectively. Thus, later analysis will sequentially drop these parameters, and monitor any significant changes to the regression output results. Remaining variables of interest demonstrate expected correlation signs, as predicted by both theory and literature.

Appendix 2-5 illustrates series of tests conducted on the dataset to determine another modifications and model behaviour. Hausman test value indicates that FE is more feasible over RE since we can reject the tests null hypothesis (p-value 0.000). Alternatively, Breusch-Pagan-LM test shows support for pooled OLS (p-value 1.000), and thence, both pool OLS and FE will be used in empirical analysis and testing. Wooldridge autocorrelation and Breusch-Pagan/White´s heteroscedasticity tests show problems with present heteroscedasticity and (serial) autocorrelation problems. Use of country and year-fixed effects, clustered SEs and overall panel data models should mitigate these problems, however, due to the nature of this study being based on regional analysis, these problems are

Growth GDP V4 Interaction Structural

Funds Innovation Education Population

Growth 1.00 GDP -0.2051* 1.00 V4 0.1529* -0.3883* 1.00 Interaction 0.0433* -0.2750* 0.7323* 1.00 Structural Funds -0.0623* -0.3569* 0.1429* 0.2661* 1.00 Innovation -0.1170* 0.6975* -0.2876* -0.1977* -0.3026* 1.00 Education -0.1322* 0.5065* -0.2254* -0.1172* -0.2147* 0.5135* 1.00 Population -0.1387* 0.3666* -0.1760* -0.1008* -0.0739* 0.2626* 0.2299* 1.00

to a degree expected and may only be efficiently resolved by more sophisticated estimation methods, such as WLS, GMM or spatial models.

Table 5: Preliminary regression results

Pool OLS Pool OLS FE

VARIABLES Growth Growth Growth

Log GDP (T-1) -0.036*** -0.034*** -0.173*** (0.003) (0.004) (0.014) V4 -0.008 -0.011 (0.013) (0.009) Interaction (T-1) 0.003* 0.003** -0.001 (0.002) (0.002) (0.002)

Log Structural Funds (T-1) -0.006*** -0.007*** 0.002

(0.001) (0.001) (0.002) Log Innovation (T-1) -0.005*** -0.001 -0.024*** (0.001) (0.002) (0.005) Education (T-1) 0.158*** 0.063*** 0.012 (0.022) (0.016) (0.065) Population (T-1) 0.041 -0.164 0.525** (0.137) (0.143) (0.220) Constant 1.458*** 1.407*** 2.838*** (0.032) (0.034) (0.129) Observations 2,411 2,411 2,411 R-squared 0.503 0.458 0.498

Country FE Yes No Yes

Year FE Yes Yes Yes

Clustered SE Yes Yes Yes

Number of ID 249

Robust standard errors in

parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 5 shows the preliminary regression output results for both FE and pool OLS. All variables are appropriately transformed i.e. lagged and logged, to normalize and level out the testing parameters to that of GDP variable. As seen above, all variables when

significant; respective coefficients show expected signs, as predicted by theory and literature findings. However, neither V4 nor interaction term are significant at any level, except under pooled OLS (10% significance). Education turns significant to insignificant from pooled OLS to FE, whilst Population vice versa. Relatively low R-squared figure is reported in both models, around 0,5. Both models have accounted for country and year fixed effects, and have clustered standard errors included.

Tables 6 and 7 present the lag optimization i.e. selecting appropriate lags based on economic reasoning. As discussed earlier, if we assume that Structural Funds and

Innovation are both forms of investments like FDI, economically, we would in theory except them to demonstrate their empirical effects with time delay. Therefore, in line with the nature of these variables and according to EU, the following tables will show lagged versions of these variables and their respective outcomes. Result consistency will determine the optimal choice of a lag.

Under pooled OLS, lag selection for Structural Funds does not matter, as all three are significantly affecting the dependant variable. Interaction term is only significant with two-lagged period, and one-lag period, but under 10% significance testing. Innovation lag is only lagged twice, first-lag is only significant under all p-values. Relative changes between other variables in terms of coefficient magnitudes, given their statistical significance and coefficient signs, are mainly.

However, regarding FE, all lags of Structural Funds and Interaction term are insignificant, and both lags of innovation parameter are significant, respectively. Education is no longer

significant under FE, but Population is under first two lags of SF. Therefore, regarding both FE and pooled OLS outputs, the optimal lags for Structural Funds are uncertain5_{, whilst}

Innovation has first-lag consistency under both, respectively.

Table 6: Lag optimization under pool OLS

(1) (2) (3) (5) (6) (7)

VARIABLES Growth Growth Growth Growth Growth Growth

Log GDP (T-1) -0.036*** -0.035*** -0.040*** -0.037*** -0.036*** -0.038*** (0.003) (0.003) (0.004) (0.004) (0.004) (0.004) V4 -0.008 -0.021* -0.009 0.008 -0.018 -0.005 (0.013) (0.012) (0.012) (0.014) (0.014) (0.013) Interaction (T-1) 0.003* 0.000 (0.002) (0.002) Log Structural Funds (T-1) -0.006*** -0.007*** (0.001) (0.001) Interaction (T-2) 0.005*** 0.005*** (0.001) (0.002) Log Structural Funds (T-2) -0.007*** -0.008*** (0.001) (0.001) Interaction (T-3) 0.001 0.002 (0.001) (0.002) Log Structural Funds (T-3) -0.004*** -0.005*** (0.001) (0.001)

Log Innovation (T-1) -0.005*** -0.005*** -0.004*** (0.001) (0.001) (0.002) Log Innovation (T-2) -0.003 -0.003* -0.002 (0.002) (0.002) (0.002) Education (T-1) 0.158*** 0.159*** 0.188*** 0.132*** 0.128*** 0.144*** (0.022) (0.023) (0.025) (0.028) (0.028) (0.029) Population (T-1) 0.041 0.061 -0.009 -0.120 -0.152 -0.043 (0.137) (0.138) (0.128) (0.108) (0.107) (0.113) Constant 1.458*** 1.451*** 1.479*** 1.450*** 1.441*** 1.454*** (0.032) (0.034) (0.043) (0.037) (0.039) (0.042) Observations 2,411 2,293 2,200 2,286 2,267 2,149 R-squared 0.503 0.508 0.507 0.499 0.498 0.495

Country FE Yes Yes Yes Yes Yes Yes

Year FE Yes Yes Yes Yes Yes Yes

Clustered SE Yes Yes Yes Yes Yes Yes

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

Robustness tests can be found within Appendix 6-7, where variables are sequentially removed from the regression testing, based on their correlation output results shown earlier. Both robustness tests further support the findings from Tables 5-7, showing identical coefficient sign behaviour, changes in statistical significance of selected variables between the two models, relatively significant country, and year fixed-effects for majority of the countries and years, respectively.

Table 7: Lag optimization under FE

(1) (2) (3) (4) (5) (6)

VARIABLES Growth Growth Growth Growth Growth Growth

Log GDP (T-1) -0.173*** -0.170*** -0.219*** -0.204*** -0.203*** -0.236***

(0.014) (0.012) (0.012) (0.018) (0.014) (0.014)

V4 - - - -

Interaction (T-1) -0.001 -0.003*

(0.002) (0.002)

Log Structural Funds (T-1) 0.002 0.002

(0.002) (0.001)

Interaction (T-2) 0.002 0.002*

(0.001) (0.001)

Log Structural Funds (T-2) -0.001 -0.001

(0.002) (0.001)

Interaction (T-3) -0.002 -0.002

(0.001) (0.001)

Log Structural Funds (T-3) 0.002 0.002

(0.001) (0.001)

Log Innovation (T-1) -0.024*** -0.025*** -0.026***

Log Innovation (T-2) -0.019*** -0.018*** -0.022*** (0.005) (0.005) (0.006) Education (T-1) 0.012 0.011 0.083 -0.010 -0.034 0.047 (0.065) (0.073) (0.072) (0.070) (0.071) (0.078) Population (T-1) 0.525** 0.562** 0.340 0.254 0.233 0.370* (0.220) (0.226) (0.220) (0.190) (0.191) (0.209) Constant 2.838*** 2.812*** 3.265*** 3.100*** 3.106*** 3.429*** (0.129) (0.104) (0.110) (0.167) (0.132) (0.129) Observations 2,411 2,293 2,200 2,286 2,267 2,149 R-squared 0.498 0.497 0.532 0.510 0.510 0.528 Number of ID 249 249 246 254 254 251

Country FE Yes Yes Yes Yes Yes Yes

Year FE Yes Yes Yes Yes Yes Yes

Clustered SE Yes Yes Yes Yes Yes Yes

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

Table 8 therefore, similarly to Table 5, shows the last output from regression testing, using one-period lags for all explanatory variables. V4 is insignificant at all levels, whilst Interaction term is significant only under 10% level of significance (and 5% significance in Table 5) (0.003 coef. magnitude). All coefficient signs for statistically significant relations are in accordance to the correlation matrix as well as theories. Again, change in significance can be seen between each model for Education and Population parameters. Magnitude of log GDP is significant and higher, in both cases and in case of FE, respectively.

Table 8: Results output table

Pool OLS FE

VARIABLES Growth Growth

Log GDP (T-1) -0.036*** -0.173*** (0.003) (0.014) V4 -0.008 (0.013) Interaction (T-1) 0.003* -0.001 (0.002) (0.002)

Log Structural Funds

(T-1) -0.006*** 0.002 (0.001) (0.002) Log Innovation (T-1) -0.005*** -0.024*** (0.001) (0.005) Education (T-1) 0.158*** 0.012 (0.022) (0.065) Population (T-1) 0.041 0.525** (0.137) (0.220) Constant 1.458*** 2.838*** (0.032) (0.129) Observations 2,411 2,411

Country FE Yes Yes

Year FE Yes Yes

Clustered SE Yes Yes

Number of ID 249

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

7. Discussion

First and foremost, we need to address the regression results and various robustness tests and diagnostic check themselves. Throughout this paper, either by using pooled OLS or FEM estimation techniques, the findings indicated to a degree, strong, statistically significant results supporting beta-convergence theory and therefore neoclassical growth theory, on which it's based around. Models have to a great extent confirmed convergence with respect to ß-convergence concept, such that poorer regions tend to grow faster than richer regions. This claim is also generally valid for many of the countries and individual years within the tested time span. Robustness tests confirm these findings further, however, certain anomalies or outliers must be pointed out.

Poorer countries tend to grow faster on average, but, closer analysis reveals that this may not fully be the case. All countries with exception of Luxembourg shown tendencies to converge; Luxembourg is the only country exhibiting divergence, especially at 1%

significance level (e.g. 0.008 and 0.024, respectively). The absolute value of the coefficient of “convergence” illustrates the speed at which countries and their regions, on average, converge to or diverge from. On average, however, NUTS-2 EU regions should convergence at speeds between 0.00031 and 0.00036, if their GDP per capita were to increase by 1% (e.g. a one percent increase in GDP per capita will lead to ßi/100 change in our dependant variable, thus growth). Nevertheless, this study´s aim is to examine and analyse the statistical and economic effects of Structural Funds, with respect to the EU and V4.

Throughout the empirical study, we find no evidence supporting the club convergence theory. Although this is only tested on the V4 members' group as such, partial evidence may be only observed when Structural Funds are lagged by two-term period, under pool OLS model and 10% significance level. However, since this variable remains statistically insignificant for the remaining model variations, we address this further and conclude that there is no tendency for V4 countries to exhibit common degree or extent of regional economic convergence patterns. This is later confirmed in robustness tests when Slovakia