Convergence and growth of the
different NUTS 2 regions of the
European Union
M.Sc. International Economics and Business - University of Groningen
M.A. International Economy and Business - Corvinus University of Budapest
Double Degree Program
Master Thesis
Gábor Kun
S2824582
g.p.kun@student.rug.nl
2
Abstract
The current thesis is an investigation of unconditional beta-convergence and GDP per capita growth of the EU 27 countries and its NUTS 2 regions, distinguishing them according to their urbanisation level. To achieve this analysis a panel dataset has been used for the period 2001-2011 stems from the Eurostat website. The results of the pooled OLS regressions confirm the varied economic performance of the differently urbanised regions of the new EU countries. The different influence on economic growth of certain driving forces of the EU countries’ regions has been proved by the analysis.
3
Table of content
I. Introduction ... 4 II. Hypotheses ... 8 III. Methodology ... 10 1. Data ... 10 2. Model ... 14 IV. Results ... 18 V. Conclusion ... 27Limitations and future research ... 27
References ... 29
4
I. Introduction
The comparison of economies in respect of their economic and/or social performance has a long history. In the European Union the desire for economic and social cohesion has been implemented and defined in the Single European Act through the creation of the Single Market (20 Years of the European Single Market, 2012). The process is achieved through the cohesion policy. One of the main aims of the policy is to help the less developed regions to catch-up, to encourage the convergence within the European Union, to promote a more balanced and a more sustainable territorial development to decrease gaps between the different economies of the European Union (Eurostat Regional Yearbook 2014, 2014).
The main purpose of the current thesis is the investigation of the differently urbanised NUTS 2 regions’ convergence and growth prospect in the period 2001-2011. The research consists of three main parts. The first part, investigates the presence of the trade-off convergence (Kertész, 2004) of the Visegrád Group countries. That includes the examination of unconditional beta-convergence at both country and regional level. The second section, analyses whether the differently urbanised regions have significantly different growth performance as stated by Dijkstra et al (2013). The NUTS 2 regions are labelled on a scale from the least urbanised to the most urbanised within the categories of predominantly rural, intermediate or predominantly urban. The third section examines whether certain determinants have a different effect on the economic performance of the old EU countries meanwhile not influencing the new EU countries’ economic growth.
5 absolute or unconditional convergence and conditional convergence. Vojinovic et al. (2009) claimed if regions converge to the same steady-state level absolute convergence is present, while if regions converge to different steady-state levels because of their different characteristics, conditional convergence is present.
Kertész (2004) emphasised the possible contrary realization of the convergence effect on national and on regional level. He states if a country starts to converge on national level it is very likely that leads to divergence (at least slower convergence) between the certain nation regions. Because, the most developed regions start to develop faster than the less developed regions. This called the trade-off theory. Herz and Vogel (2003) did not find empirical evidence of the existence of trade-off between regional and country level convergence in their paper, which focused on Central and Eastern Europe for the period 1991-2002.
6 (2011) summarised the impact of population growth as follows. Population growth can have a negative influence on the economy as it has effects on investments and savings. Furthermore, it can negatively influence human capital by decreasing the availability of it. Although, she stressed that population density and high number of available working age population is usually appreciated as an advantage. Wagner and Hlouskova (2005) in their Central and Eastern European country level analysis claimed negative effects of population growth on GDP per capita growth in the period 1991-2001. They argued that ceteris paribus with increasing population the same GDP has to be divided between more people.
After the biggest enlargement of the European Union in 2004 the cohesion policy has become an even more serious topic in the organization as the amount of less developed territories increased rapidly. Besides many others the so-called Visegrád Group (V4) countries namely: the Czech Republic, Hungary, Poland and Slovakia achieved membership. These countries have many similarities as they went through similar economic, social and political development in the recent past. At first look it seems the V4 countries developed and catch-up quite well in the last decade, however if we take a closer look it can be seen the capitals and their regions developed a lot, but other regions’ development is less amazing. According to Nagy and Kuttor (2008) analysis, in the period 1995-2005 the Visegrád Group countries could converge to Western Europe, but territorial disparities on regional level increased dramatically. They conclude that the V4 countries’ convergence in most cases happened because of the growth of the four capital regions.
The impact of cities on national economic growth has attracted attention since the famous work of Paul Krugman (Geography and Trade, 1991) and did not get lower neither since then. The case of the V4 countries seems parallel with the work of Sassen (2002). Who highlighted the important role of capital cities, which highly attracts well-trained human capital, also mentioned by Glaeser et al. (2010). Thereby, cities are usually favoured places of investments and witnesses of the so-called spillover effects (Ollé and Marsal, 2003).
7 countries’ improvement strongly depends on the performance of the capitals (Poledníková, 2014).
Dijkstra, Garcilazo and Mccann (2013) achieved a research that compared the economic performance of NUTS 3 regions of the European Union member states, with focus on the area of ”uniquely polycentric urban structure” (page 334). Three different regions were distinguished in their analysis; predominantly urban (PU), predominantly rural (PR) and intermediate (IN) regions, based on their characteristics guided by the OECD categorization. The different performance of different regions between the EU15 and the CEECs (12 Central and Eastern European Countries, which joined the EU after 1995) was noticed. They stated that between 2001 and 2007 rural areas have demonstrated the strongest GDP per capita growth in the EU15 regions, while in the case of the new members; in the CEECs the predominantly urban regions have the strongest growth.
8
II. Hypotheses
The hypotheses of the current thesis can be categorized into two main groups. The first hypothesis examines the existence of the so-called trade-off convergence of the V4 countries. The second group of hypotheses investigate the differently urbanized NUTS 2 regions’ growth performance, focusing on the V4 and new EU member states.
Panel data is used to achieve the investigation. The regressions are used standard pooled OLS and fixed effects models approaches completed with dummy variables. Both groups of hypotheses are concerned to the period 2001-2011.
Trade-off effect
Hypothesis 1: Trade-off effect exists in case of the Visegrád Group countries’ convergence
process.
To check the presence of the trade-off effect, two conditions must be met at the same time. Firstly, unconditional beta-convergence must exist between the V4 and old EU countries. Secondly, unconditional beta-convergence should not be presenct between the V4 countries’ regions. Thus, the following two minor hypotheses are formulated to accept or reject hypothesis 1.
H1a: Beta-convergence exists between the V4 countries and the OLD EU members. H1b: Beta-convergence exists between the NUTS 2 regions of the V4 countries.
Urban-rural performance
Hypothesis 2: The GDP per capita growth performance of the PU, IN and PR regions are
different.
9 These are compared at first to the intermediate (IN) and then to the predominantly urban (PU) regions’ GDP per capita growth.
The first two (a, b) minor hypotheses compare all of the differently urbanised NUTS 2 regions of the whole EU.
H2a: The IN regions of the EU 27 countries are expected to experience higher GDP per capita growth than the PR regions of the EU 27 countries.
H2b: The PU regions of the EU 27 countries are expected to experience higher GDP per capita growth than the PR regions of the EU 27 countries.
The second two (c, d) minor hypotheses compare the 35 differently urbanised NUTS 2 regions of the Visegrád Group countries.
H2c: The IN regions of the V4 countries are expected to experience higher GDP per capita growth than the PR regions of the V4 countries.
H2d: The PU regions of the V4 countries are expected to experience higher GDP per capita growth than the PR regions of the V4 countries
The third pair (e, f) of minor hypotheses compares the differently urbanised NUTS 2 regions of the 12 EU countries that are involved into a group called new EU.
H2e: The IN regions of the NEW EU countries are expected to experience higher GDP per capita growth than the PR regions of the NEW EU countries.
10
III. Methodology
1. Data
All the data stem from the European Commission’s online statistical database (Eurostat) covering the period 2001-2011It includes country and mainly regional level dataset of the European Union 27 countries (Croatia is excluded) and its 270 NUTS 2 regions.
Dependent variable
The dependent variable is the growth of gross domestic product per capita in purchasing power standards (GDP per capita PPS) specified in euro. The advantages of using per capita values in PPS instead of simple GDP growth numbers are the following. It takes into consideration the certain territories’ different population sizes. Secondly, it eliminates the differences in price levels between the NUTS 2 regions. GDP per capita is a commonly used (Vojinovic et al., 2009; Raileanu Szeles and Marinescu, 2010; Mohl and Hagen, 2010) variable to measure convergence and growth. As GDP per capita values are known as a proxy of measuring income differences that is achieved by the beta-convergence analysis. Moreover, the European Union uses this specific value to rank economies.
Independent variables
In the first set of unconditional beta-convergence calculations (models 1-6.) only one independent variable is used, the first year (2001) GDP per capita (GDPper1) value, following the work of Raileanu Szeles and Marinescu (2010).
The NUTS 2 territories are characterized into three groups, these are predominantly urban (PU), intermediate (IN) and predominantly rural (PR). The expected different economic performance of these regions is tested by the models 7-9.
11 In the model 9 the urban-rural categories are intercepted with the group of new countries. The received three variables PU*new, IN*new and PR*new are used as independent variables to test the hypotheses 2e and 2f.
For detecting the growth performance of the NUTS2 regions, another set of independent variables are used in model 7 and models 10-14. The choice of these variables was driven by the aim to capture the determinants of the growth rates of the differently urbanised regions. The limited availability of data on NUTS 2 level also influenced the choice of the variables. The first set of the independent variables covers or proxies of the following categories: employment, infrastructure, population, investment and education.
The first variable is female employment (FEmp). As mentioned before, different employment variables play an important role in the catch-up process (Vogel 2003). The usually chosen employment rates are related to sectors, like electronic-sector (Vogel 2003) or agriculture and industry (Cappelen et al. 2003) employment share. Unfortunately, the available sector related data is limited that would lead to the shortening of the examined period by three years. These facts lead me to choose female employment that is expected to have a different presence for the differently urbanised regions similarly to sectorial employment. The variable is given in percentage for persons aged 15 to 64 years.
The second variable is the length of motorways per 1000 km2 (MwaysKm2lag) used as a proxy of infrastructure, given in the unit of kilometres. The reason to use an infrastructure related variable is triggered by the fact that it probably has different effects on growth in case of more and less developed economies as highlighted by Oosterhaven and Knaap (2003).
As the effect of population growth on economic performance has been well-known since Solow (1956), the third variable is the population change (PopC). Population growth impact on growth has often been investigated just like by Wagner and Hlouskova (2005). The currently used variable is a calculated value, used in percentages. The variable shows the change of the population on 1 January from the previous year to the certain year.
12 For the variables FDI and motorways per 1000 km2 t-1 lagged values have been used. Following the work of Delgado et al. (2000) and Dreher (2000) in order to avoid the problem of reverse casuality bias and to capture the delayed effect on growth. Canning and Pedroni (1999) also highlighted the mutual casuality between growth and infratructure, that stems from the fact that growth strenghtens the demand for infrastructure meanwhile infrastructure increases productivity.
The last variable is the early leavers of education and training (ErlEd) in age 18-24, which is used as a proxy of education. The reason for choosing this unusual variable is that other potential education variables, like tertiary education, secondary or even primary education are showing correlation with the female employment variable. The variable is given in percentages.
Control variables
Five of the above listed independent variables: female employment, length of motorways per 1000 km2,population change, foreign direct investment and early leavers of education and training are used as control variables in case of the models 7, 8 and 9. As those models are focusing on the urban-rural regions’ different performance.
Countries
13
NUTS 2 regions
The Nomenclature of territorial units for statistics (NUTS) is a hierarchical classification system used by the European Union. The classification is essentially based on population size and subdivides every country of the European Union into three groups: NUTS 1, 2 and 3 from larger to smaller (Eurostat Regional Yearbook 2014). The NUTS classification has been changed a lot during the years. In my thesis I use the classification version 2006. Those regions that have merged or split since then are treated as they have had before. To summarize, the dataset includes 270 NUTS 2 regions of the EU27 countries. Thus, I could include all regions of the current member states except the two NUTS 2 regions of Croatia.
Urban-rural typology on NUTS 2 level
As there is no urban-rural classification on a NUTS 2 regional level available (of the European Union website) I prepared my own classification based on the available NUTS 3 level classification of the Eurostat website.
The European Union NUTS 3 level classification, that is similar to the OECD approach, classifies regions as predominantly rural (PR), intermediate (IN) or predominantly urban (PU). I used the same names for my NUTS 2 level classification.
The created classification is bases on the average number of the certain NUTS 2 region's NUTS 3 regions and its classification. The following number-characters matching process is used: PR=1, IN=2, PU=3.
14 2. Model
General
First and foremost, basic checks of the database have been done. There is no support of the presence of correlation (related tables are in the appendix) or multicollinearity. Outlier detection and normality tests show appropriate results for the given database.
In the current thesis OLS pooled and fixed effects models are used. Fixed effects models allow for different intercepts for each individual and control for the effects of time-invariant variables, omitted variables. However, fixed effects are the most commonly used to deal with panel data random effects models could be more efficient. To test between random and fixed effects models the Hausman test is used, which supported the use of fixed effects models in case of the current analysis.
Presence of heteroskedasticity is shown in case of all the regression models. To control for it robust standard errors are used in all case.
In all regressions the same dependent variable, GDP per capita growth is used. Equation 1 is used for country and regional level analysis, as well. All the other equations are used only for NUTS 2 regional level analysis. The reason of regional level investigation is stems from the purpose to compare the differently urbanised regions’ performance. Furthermore, as the countries of the European Union spread from really small like Slovenia or Estonia to relatively huge countries like Germany, a regional analysis is usually more meaningful (Eurostat regional yearbook 2014). Besides the above described independent and control variables different dummy variables are used. In every regression (
ε
it) describes theerror term.
Trade-off effect
15 existence of the trade-off effect formulated by hypothesis 1 the following standard pooled OLS regression is used completed with dummy variables for years.
GDP per capita gr. 𝑖t = 𝛽0 + 𝛽1(GDP per capita𝑖2001) + year_ dummy + εit Eq. (1)
To accept the first hyphotesis two conditions must be satisfied at the same time. Firstly, the equation needs to be run on a country level, to check the existence of beta-convergence between the Visegrád Group and the old EU countries. If beta-convergence exists the coefficient 𝛽1 needs to be negative and significant (hypothesis 1a) following the analysis of Raileanu Szeles and Marinescu (2010). This would prove the V4 countries’ faster growth compared to the old countries.
Secondly, the equation needs to be run once more, on a regional level focusing only on the V4 countries’ NUTS 2 regions. In this case, to accept the existence of the trade-off effect beta-convergence should not be shown (hypothesis 1b). Thus, the coefficient 𝛽1 should not be negative and significant. That would not prove the less developed, mostly rural regions’ faster GDP per capita growth.
Urban-rural performance
In order to investigate the different NUTS 2 regions’ different GDP per capita growth, the following three (Eq. 2, Eq. 3, Eq 4.) pooled OLS regressions are used.
GDP per capita gr. 𝑖t = 𝛽0 + 𝛽1 FEmp 𝑖t + 𝛽2 MwaysKm2lag 𝑖t + 𝛽3 PopC 𝑖t + 𝛽4 lnFDIlag 𝑖t + 𝛽5ErlEd 𝑖t + country_ dummy + year_ dummy + urban_ dummy + εit Eq. (2)
If the GDP per capita growth is different between the old and new countries’ differently urbanised regions the dummy variable urban_ dummy needs to be significant.
GDP per capita gr. 𝑖t = 𝛽0 + 𝛽1 FEmp 𝑖t + 𝛽2 MwaysKm2lag 𝑖t + 𝛽3 PopC 𝑖t + 𝛽4 lnFDIlag 𝑖t + 𝛽5 ErlEd 𝑖t + country_ dummy + year_ dummy + urban_ dummy + V4_dummy +
16 If the GDP per capita growth is different in case of the V4 countries’ NUTS 2 regions the variables urban_ dummy *V4_dummy need to be significant.
GDP per capita gr. 𝑖t = 𝛽0 + 𝛽1 FEmp 𝑖t + 𝛽2 MwaysKm2lag 𝑖t + 𝛽3 PopC 𝑖t + 𝛽4 lnFDIlag 𝑖t + 𝛽5 ErlEd 𝑖t + country_ dummy + year_ dummy + urban_ dummy + EU15_dummy +
+ urban_ dummy *EU15_dummy + εit Eq. (4)
If the GDP per capita growth is different in case of the new and old countries differently urbanised NUTS 2 regions, the variables urban_ dummy*EU15_dummy need to be significant.
Comparison of growth determinants of the old and new EU countries
One of the initial triggers of the recent research was to investigate the different economic performance of the NUTS 2 regions of the Visegrád Group countries. Unfortunately, the used database collinearity check proved correlation of the variables within the V4 countries’ sample. This probably affected by the relatively low number of NUTS 2 regions of the certain group, accurately 35. To handle the problem and be able to investigate the different regions’ performance the group became expanded to a group called new. That group includes all the NUTS 2 regions of those 12 countries of the European Union that joined after 1995. The distribution and performance of the expanded group’s differently urbanised regions is similar to the V4 group regions as can be seen in Table 4. The group called new involves 55 countries, the number of NUTS 2 regions increased from 35 to 55. Within this group the collinearity numbers are acceptably low.
To investigate the different variables’ effects on the old and new countries’ NUTS 2 regions’ GDP per capita growth, the following pooled OLS and fixed effect regression are used. The first equation of this section is identical with equation 2, but this time the focus is on the coefficients of the five growth determinants variables not on the urban dummies.
In both cases year dummies are used, furthermore in case of the mentioned equation 2 urban and country dummies are used, as well.
Pooled OLS model:
17 The main difference of equation 5 compared to equation 2 is that the former does not include country and urban dummies. As the fixed effects model distinguishes between all the NUTS 2 regions of the samples including country and urban dummies would not change the results and many dummy variables would be omitted.
Fixed effect model:
18
IV. Results
Trade-off effect
Following the work of Raileanu Szeles and Marinescu (2010) to measure unconditional beta-convergence GDP per capita values are used, as proxy of income. The dependent variable in my pooled OLS regression is the annual GDP per capita growth for the period 2001-2011. The first and only independent variable of the regression is the first period (2001) GDP per capita value. If beta-convergence exists the first period GDP per capita coefficient needs to have a negative and significant value.
Firstly, beta-convergence is examined on a country level. Table 1 shows the results. To get a better and wider perspective I ran the same pooled Ordinary Least Square (OLS) regression three times for different groups of countries. The first column includes all the 27 countries of the European Union; the second includes the old or EU15 countries; while the third group involves the EU15 plus the V4 countries together to test the first hypothesis. To get more sophisticated results year dummies are used.
Table 1: Trade-off effect - country level models Groups Model 1 (EU27) Model 2 (OLD) Model 3 (OLD+V4) VARIABLES GDPpercGr GDPpercGr GDPpercGr
GDPperc1 -7.75e-05*** 1.03e-05 -4.39e-05*** (1.20e-05) (1.13e-05) (1.08e-05)
year Dummy yes yes yes
Constant 4.520*** 1.369** 3.139***
(0.496) (0.450) (0.363)
Observations 286 165 209
R-squared 0.563 0.639 0.586
19 The results proved the existence of beta-convergence for the whole (EU27) dataset and in case of the EU15+V4 group, as in these columns the indepdendent variables are negative and highly significant. Thus, the first condition of beta-convergence is met. For the EU15 countries positive and insignificant results are shown that donot prove the existence of beta-convergence.
Secondly, the existence of regional beta-convergence is examined again, but on a regional level. The same pooled OLS regressions are used (equation 1). The first two colums are equvivalent with the first 2 columns of the first table, while the third column only includes the NUTS 2 regions of the V4 countries, to check the relevance of the so called trade-off effect. Table 2 shows the results.
Table 2: Trade-off effect - regional level models Groups Model 4 EU27 Model 5 OLD Model 6 V4 VARIABLES GDPpercGr GDPpercGr GDPpercGr
GDPperc1 -0.000118*** -1.58e-05 2.15e-05 (9.72e-06) (9.96e-06) (3.89e-05)
year Dummy yes yes yes
Constant 6.502*** 3.997*** 5.296*** (0.296) (0.310) (0.852)
Observations 2,970 2,365 385
R-squared 0.478 0.508 0.434
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
20 hypothesis 1, the existence of trade-off effect is accepted. Thus, the hypothesis 1a is accepted while the hypothesis 2b is rejected.
Urban-rural performance
Dijkstra et al. (2013) stated in the period 2001 and 2007 the predominantly rural (PR) areas demonstrated the strongest GDP per capita growth in the old, EU15 countries’ regions; while in the case of the new members, the CEECs, the predominantly urban (PU) regions have stronger growth performance. Table 3 shows a similar, simple unweighted average calculation of the current database. The difference between the used databases and the one used by Dijkstra et al. (2013) is twofold. Firstly, the authors investigated the growth prospects of NUTS 3, not NUTS 2 regions. Secondly, their period covers the ages from 2001 to 2007, while the period of the current analysis is four years longer, spread from 2001 until 2011. Instead of the above mentioned differences the received results are more or less parallel with the findings of Dijkstra et al. (2013). The main difference in case of both the old and new groups is the relatively strong performance of the intermediate regions. For the old member states, intermediate regions have the highest growth, and predominantly urban regions have underperformed compared to the predominantly rural regions. While, in case of the new countries predominantly urban and intermediate regions have better performance than predominantly rural areas. Thus, in case the group of old member states PR regions overperform PU regions, while in case of the new member states PU regions overperform the PR regions. In the third row the performance of the 35 NUTS 2 regions of the V4 countries can be seen. In this case, the GDP per capita growth of predominantly urban (PU) areas is especially strong, stronger than the performance of the intermediate (IN) regions and predominantly rural (PR) territories, that have the lowest growth rates, similarly to the group of the new countries.
21 group of the new countries. The low number of predominantly urban areas in the new and V4 countries is striking. To check whether the differently urbanised regions’ growth performance differs significantly, models 7, 8 and 9 are run. The results can be seen below in Table 5.
Table 5: Urban-rural performance (EU27) models focus Model 7 (Urban D) Model 8 (Urban*V4) Model 9 (Urban*New)
VARIABLES GDPpercGr GDPpercGr GDPpercGr
_IUrban_2 (IN) 0.146 0.0565 -0.103 (0.169) (0.193) (0.187) _IUrban_3 (PU) 0.357 0.160 -0.0355 (0.260) (0.290) (0.273) _IV4XUrb_2_2 - 0.277 - (0.417) _IV4XUrb_2_3 - 0.949 - (0.603) _IEU1XUrb_2_2 - - 0.695* (0.407) _IEU1XUrb_2_3 - - 1.472*** (0.669) FEmp 0.0508*** 0.0499*** 0.0505*** (0.0109) (0.0109) (0.0109) MwaysKm2lag -0.00740** -0.00653* -0.00609* (0.00336) (0.00342) (0.00341) PopC -0.568*** -0.575*** -0.593*** (0.107) (0.108) (0.107) lnFDIlag 0.503** 0.506** 0.507** (0.198) (0.198) (0.198) ErlEd 0.0126* 0.0142* 0.0162** (0.00730) (0.00738) (0.00721) V4_Dummy - yes - NEW_Dummy - - yes
year_Dummy yes yes yes
country_Dummy yes yes yes
Constant -2.671 2.187 -2.439
(1.792) (1.715) (2.295)
Observations 2,357 2,357 2,357
R-squared 0.616 0.617 0.618
22 As Table 5 shows there is no support of higher growth of intermediate or predominantly urban compare to the predominantly rural regions within the all NUTS 2 regions of the EU27 countries (model 7). Thus, both hypothesis 2a and hypothesis 2b can be rejected.
The second column of Table 5 does not stress better growth prospects of the urban regions (PU) neither of the intermediate (IN) regions of the V4 countries compared to the rural (PR) regions of the V4 subgroup. Thus, hypotheses 2c and 2d are rejected as well.
Despite of these findings, comparing the different regions within the new countries, different performances become significantly different. Model 9 shows evidence of the better performance of rural regions on a 10% significance level, in the group of new EU countries. The economic performance of urban (PU) regions is even stronger within the same new subgroup, compare to rural regions. The predominantly urban regions’ better performance is significant at 1% significance level compare to the predominantly rural regions’ performance of the same subgroup. Thus, hypotheses 2e and 2f are accepted. These regression results also support the findings of Dijkstra et al. (2013), who claimed the better economic performance of predominantly urban and intermediate regions compare to predominantly rural areas within the new EU countries.
To conclude the results briefly, the hypothesis 2, higher GDP per capita growth of certain type of NUTS 2 regions (in case of the new countries) can be only partially accepted.
The five control variables’ effects seem to be significant in case of all three models. The employment, the investment and the education variables’ impacts are positive, while the population and the infrastructural variables’ effects are negative. A more accurate investigation of these variables is proceeding in the following. As the equations of Table 6 are focusing on these variables’ effects, besides the pooled models fixed effects models are used as well.
Comparison of growth determinants of the old and new EU countries
23 groups. The R-squared of the models fluctuate around 0.6 in all cases. The new group’s regressions have lower R-squared values than the old group.
Secondly, comparing the OLS and fixed effects models, the variables’ coefficients are stronger in cases of the fixed effects models within the same group. The only exception of this is the investment variable (lnFDIlag) that shows stronger effects on economic growth in case of the pooled OLS regression compare to the fixed effect model for the group old. This probably stems from the fact that the FDI variable is a country level variable not regional level opposed to the other variables, so using fixed effects model for regions has no effect on it. Besides the volume the fixed effects models’ significance level is higher compare to the pooled OLS models’ results as well, except the population variable that significance level decrease from 1% to a 5% level comparing model 13 to model 14.
Table 6: Comparison of growth determinants of the old and new EU countries models regressions Model 7 P.OLS Model 10 FE Model 11 P.OLS Model 12 FE Model 13 P.OLS Model 14 FE Groups VARIABLES EU27 GDPpercGr OLD GDPpercGr NEW GDPpercGr FEmp 0.0508*** 0.210*** 0.0325*** 0.195*** 0.170*** 0.347*** (0.0109) (0.0318) (0.0112) (0.0338) (0.0507) (0.0942) MwaysKm2lag -0.00740** -0.0665** -0.00638* -0.0775** -0.0199 -0.0359 (0.00336) (0.0288) (0.00340) (0.0337) (0.0325) (0.0597) PopC -0.568*** -0.793*** -0.464*** -0.569*** -1.033*** -1.220** (0.107) (0.181) (0.113) (0.207) (0.377) (0.553) lnFDIlag 0.503** 0.537** 1.695*** 1.466*** -0.365 -1.029 (0.198) (0.258) (0.502) (0.540) (0.584) (0.640) ErlEd 0.0126* 0.115*** 0.0195*** 0.151*** -0.107* -0.335** (0.00730) (0.0374) (0.00703) (0.0369) (0.0618) (0.154)
urban_Dummy yes - yes - yes -
year_Dummy yes yes yes yes yes yes
country_Dummy yes - yes - yes -
Constant -2.671 -15.44*** -16.94*** -28.41*** -0.123 0.171 (1.792) (3.917) (6.197) (7.224) (4.833) (5.975)
Observations 2,357 2,357 1,844 1,844 513 513
R-squared 0.616 0.594 0.614 0.625 0.595 0.572
24 The impact of female employment seems to be positive and significant at 1% significance level in all models of Table 6. The impact of female employment is stronger in case of the new countries. Based on these results hypothesis 3a can be rejected, as in both cases the impact of the variable is positive and significant.
The second variable is used as a proxy of infrastructure. The results showed different effects on growth for the two groups. In both cases the coefficients are negative but not significant for the new countries. According to these results, the hypothesis 3b can be accepted as the effect of infrastructure is significant for the old EU countries but insignificant for the group called new. The different effect of infrastructure was detected by Oosterhaven and Knaap (2003) between the old and new EU countries. However, the negative effect is surprising and the expected effect would be the stronger positive growth impact in case of the new countries as they have rarer infrastructural endowment, but the results do not meet these expectations.
The results of population change are identical to the findings of Wagner and Hlouskova (2005), who also found negative effects of population growth on economic performance focusing on Central and Eastern European countries. The impact is clearly identical for all cases, negative and highly significant at 1% significance level. The effect of the variable is stronger for the new countries. To sum up, hypothesis 3c can be rejected.
The impact of FDI on growth is different within the two groups. For the old countries the effect is positive and significant on 1% that is parallel with most of the literature findings of the issue (Timakova, 2011). Surprisingly, the effect for the new group is negative but not significant. Thus, hypothesis 3d can be accepted.
The last variable’s impact on growth is significant for every group, contrary for the groups old and new. Thus, it can be concluded, hypothesis 3e can be rejected. The effect is positive and significant at 1% significance level in case of the old countries, while negative and significant only at 5% significance level for the new countries. The negative influence of higher number of early leavers seems to be more relevant, as education improves the quality of human capital that is known as a trigger of economic growth (Heller Sahlgren, 2014). I could not find any research that would detect positive influence of higher early leavers on growth.
25 developed compare to other old EU countries. In 2011 the PIGS countries’ average development level based on GDP per capita comparison was below the EU-27 average (Sopek, 2013). In case of the new EU countries I excluded the countries that joined in 2007 to the organization, namely Romania and Bulgaria, as these countries are less developed compare to the other ten new countries (Sopek, 2013).
To test the results of the trade-off effects I run model 3 again without the PIGS countries. The received results are identical to the original ones. To test the urban-rural regions’ expected different growth I run the models 7, 8 and 9 again excluding the countries listed above. The regressions’ values seem identical to the original regression in case of the urban-rural variables. Checks of the main models in Tables 1 and 2 lead to the same results. Support of the existence of the trade-off theory and emphasized the higher growth of the urban and intermediate regions of the new EU countries.
Contrary, models 14 do not lead to the same results in case of all variables. For models 11-12 the number of observations decreased from 1844 to 1365 but the R-squared of the regressions stays above 0.6. Only the infrastructural variable results are changed. In case of the OLS regression the variable’s significance level decreased from 5% to 10% and in case of the fixed effects models the variable’s impact become insignificant. A similar change is achieved by excluding Romania and Bulgaria for the new countries, as mentioned above. The R-squared stays around the same values, slightly under 0.6 despite the decrease in the number of the observations. The only variable change is the early education and training leavers, which variable become insignificant in case of both the OLS and the fixed effects models. According to the database the reason of the change is probably the higher level of education and training leavers in these two countries.
One other solution could be to extend the regression with other variables and test whether the variables change or not. Unfortunately, this process could not be achieved as the other available variables have many missing values.
26 The variables foreign direct investment and motorways per km2 both have a significant effect for the old countries, while having no significant effect on the new countries. The negative effect of the infrastructural variable in the case of the old countries is unexpected. Thus, it is not surprising the checks do not support these results.
The variable early leavers of education and training has a significant influence on growth for both groups. But the tests change the results, as the tests make the values insignificant for the new countries.
27
V. Conclusion
The main purpose of the current thesis is the investigation of the differently urbanised NUTS 2 regions’ convergence and growth prospects in the period 2001-2011.
As a first step the detection of trade-off effect within the V4 countries’ convergence process has been achieved. The results accepted the existence of the trade-off theory. While the V4 countries seem to catch-up to the old member states, the convergence could not be detected within the four countries’ NUTS 2 regions. According to Nagy and Kuttor (2008), in the period 1995-2005 the convergence happened because of the growth of the four capital regions, which are the most heavily urbanised areas of the four countries. To detect their findings as a next step, a regional level analysis has undertaken the task of comparing regions based on their urbanisation level. The NUTS 2 regions are labelled on a scale from the least urbanised to the most urbanised within the categories of predominantly rural, intermediate or predominantly urban. The results do not show significantly higher growth prospects of urban regions of the V4 countries in the given period. A similar investigation was repeated, but the focus group was expanded to all the 12 new member states of the European Union. Within these countries the urban regions’ significantly higher GDP per capita growth has been accepted, supporting the research of Dijkstra et al. (2013).
As a last additional step the investigation compares the effects of five growth determinants on the growth between the old and new countries’ regions. The positive effect of the employment variable and the negative effect of the population growth, providing the core results of the investigation.
Limitations and future research
28 variables is also influenced by the fact that the analysis focuses on a longer period, as many more variables are available only for limited periods between 2001 and 2011. Despite these difficulties, I successfully used proxies and variables of infrastructure, employment, education, population growth and investment. Preferably, variables that capture the effects of technology and the business environment would be added. Analysis of convergence and growth also commonly use different structural fund variables. Furthermore, use of another education variable and a regional, not country, level investment variable would improve the quality of the research.
29 References
20 Years of the European Single Market (2012). Publications Office of the European Union, Luxembourg
Barro, Robert J. and Sala-i-Martin X. (1992). Convergence, Journal of Political Economy, Volume 100, 223–251.
Calderon C. and Serven L. (2004). The effects of infrastructure development on growth and income distribution. World Bank Policy Research Working Paper, no. 3400.
Canning D. and Pedroni P. (1999). Infrastructure and long run economic growth. Paper presented at the 1999 Econometric Society Summer Meeting, Madison, Wisconsin.
Cappelen A., Castellacci F., Fagerberg J. and Verspagen B. (2003). The Impact of EU Regional Support on Growth and Convergence in the European Union. JCMS 2003 Volume 41. Number 4. pp. 621–44
Crescenzi, R. and Rodríguez-Pose, A. (2008). Infrastructure endowment and investment as determinants of regional growth in the European Union In: Infrastructure investment, growth and cohesion. The economics of regional transport investment. European Investment Bank Papers, Vol. 13, No. 2, pp. 62-101.
Csendes G. (2013). The Impact of Infrastructure Development on the Economic Growth in Europe, Master Thesis. International Economics and Business, Rijksuniversiteit Groningen and International Economy and Business, Corvinus University of Budapest
Darvas Z. (2014). 10 years EU enlargement anniversary: Waltzing past Vienna. Available at: http://www.bruegel.org/nc/blog/detail/article/1319-10-years-eu-enlargement-anniversary-waltzing-past-vienna/#.U2IGNFcGtbs.twitter [downloaded: 4 February 2015]
30 Dijkstra L., Garcilazo E. and McCann P. (2013). The Economic Performance of European Cities and City Regions: Myths and Realities, European Planning Studies, Vol. 21, No. 3, 334–354,
Dreher A. (2006). The IMF and Economic Growth: The Effects of Programs, Loans, and Compliance with Conditionality, World Development, vol. 34, no. 5, pp. 769-788.
Ederveen S., de Groot Henri L.F. and Nahuis R. (2006). Fertile Soil for Structural Funds? A Panel Data Analysis of the Conditional Effectiveness of European Cohesion Policy, KYKLOS, Vol. 59 – 2006 – No. 1, 17–42
ESPON Evidence Brief (2013). Secondary cities are important growth poles Available at: http://www.espon.eu/export/sites/default/Documents/Publications/EvidenceBriefs/EEB5-Secondarycitiesasimportantgrowthpoles/196728_ESPON_EVIDENCE_BRIEF_5_Final.pdf [downloaded: 2015 February 4]
Eurostat Statistical books (2014). Eurostat Regional Yearbook 2014, Publications Office of the European Union, Luxembourg
Glaeser E. L. and Resseger M. G. (2010). The Complementary Between Cities and Skills, Journal of Regional Science, Volume 50, Issue 1, pages 221–244, February 2010
Heller Sahlgren G. (2014). Incentive to invest? How education affects economic growth, Adam Smith institute, Available at: http://www.adamsmith.org/research/reports/incentive-to-invest-how-education-affects-economic-growth/, [downloaded: 2015 May 16]
Herz B. and Vogel L. (2003), Regional Convergence in Central and Eastern Europe: Evidence from a Decade of Transition, in Hausen, C., Resinek, M., Schürmann, N. and Stierle, M. (eds.): Determinants of Growth and Business Cycles: Theory, Empirical Evidence and Policy Implications, Berlin: Springer, 155-179.
31 Kertész K. (2004). A nemzeti és a regionális felzárkózás váltómozgása az EU-országokban és Magyarországon. Külgazdaság, 48(2), 65-76.
Krugman, P. (1991). Geography and Trade, Cambridge, MA: MIT Press.
Mohl P. and Hagen T. (2010). Do EU structural funds promote regional growth? New evidence from various panel data approaches, Regional Science and Urban Economics 40 (2010) 353–365
Monfort P. (2008). Convergence of EU regions Measures and evolution, Working papers A series of short papers on regional research and indicators produced by the Directorate-General for Regional Policy
Nagy Z. and Kuttor D. (2008), Analysis of the Territorial Disparities in the Visegrad Four Countries: Measurement and Visualisation of Territorial Processes at Regional Level in Central Europe, Regions: The Dilemmas of Integration and Competition: Regional Studies Association Annual International Conference, 27–29 May 2008, Seaford, p. 1–13.
Nevima J. and Melecký L. (2011). The β-Convergence analysis of the visegrad four NUTS 2 regions, Mathematical Models and Methods in Modern Science, 2011/12/10, pp 48-53
News release - Eurostat – European Commission (2013). Regional GDP, Regional GDP per capita in the EU in 2010: eight capital regions in the ten first places, STAT/13/46, 21 March
2013, Available at:
file:///C:/Users/G%C3%A1bor/Desktop/RUG/Thesis/Forr%C3%A1sok%20m%C3%A1jus% 20v%C3%A9ge/STAT-13-46_EN.pdf [downloaded: 16 February 2015]
32 Ollé, A. S. and Marsal, E. V. (2003). Fiscal and growth spillovers in large urban areas, Working Papers / Institut d'Economia de Barcelona (IEB), Facultat d'Economia i Empresa, Number 2003/1 39 pages
Oosterhaven, J. and Knaap, T. (2003). Spatial Economic Impacts of Transport Infrastructure Investments. Appeared in: Pearman A., Mackie P. and Nellthorp J. (eds) Transport Projects, Programmes and Policies: Evaluation Needs and Capabilities, pp. 87-1.
Poledníková I. E. (2014). Multicriteria analysis of regional disparities in the context of the EU cohesion, VŠB – Technical University of Ostrava, Faculty of Economics, Hustopeče 18.–20. 6. 2014, XVII. mezinárodní kolokvium o regionálních vědách
Popescu G. H. (2014). FDI and Economic Growth in Central and Eastern Europe, Sustainability 2014, 6, 8149-8163
Ralieanu Szeles M. and Marinescu N. (2010). Real convergence in the CEECs, euro area accession and the role of Romania, The European Journal of Comparative Economics, Volume 7, n. 1, pp. 181-202
Solow R. M. (1956). A Contribution to the Theory of Economic Growth, Quarterly Journal of Economics, 70, pp. 65-94.
Sopek P. (2013). Real Convergence of EU-27 and Croatia in the Period 1995-2017, The 8th Young Economists’ seminar to 19th Dubrovnik Economic Conference, June 12, 2013. Draft version, Dubrovnik, Available at: http://www.hnb.hr/dub-konf/19-konferencija/yes/sopek.pdf, [downloaded: 2015 June 2]
Timakova M.V. (2011). Conditional Convergence and the Solow Model: an Empirical Study, Erasmus University Rotterdam, Rotterdam School of Economics, Department of Economics
33 Urban-rural classification, Eurostat, Available at: http://ec.europa.eu/eurostat/statistics-explained/index.php/Urban-rural_typology#Publications, [downloaded: 25 April 2015]
Vogel L. (2003). Economic Integration and Regional Development: Are European Regions Converging? Department of Economics, University of Bayreuth, Germany, Available at: http://ftp.zew.de/pub/zew-docs/sw_2003/vogel.pdf [downloaded: 15 March 2015]
Vojinovic B., Acharya S. and Prochniak M. (2009). Convergence Analysis Among Ten European Transition Economies, Hitotsubashi Journal of Economics 50, pp.123-141.
Wagner M. and Hlouskova J. (2005). CEEC growth projections: Certainly necessary and necessarily uncertain, Economics of Transition Volume 13 (2) 2005, 341–372
Young, A.T., Higgins, M.J., Levy, D., 2008. Sigma convergence versus beta convergence: evidence from U.S. county-level data. Journal of Money, Credit and Banking 40(5), 1083-1093.
Appendix
1. Summary tables about the predominantly urban, intermediate and predominantly rural NUTS 2 regions:
Table 3: NUTS 2 regions GDP per capita growth
2001-2011 groups PU IN PR all OLD 1,98 2,39 2,01 2,19 NEW 5,59 5,59 5,15 5,41 V4 5,29 4,92 4,40 4,77 EU27 2,4 3 2,93 2,85
34 Table 4: Number of NUTS 2 regions
2. Correlation Matrices
A. Country level
Table A.1 – Correlation Matrix, EU27 Correlation matrix country level Beta-conv.
GDPpercGr GDPperc1
GDPpercGr 1.0000
GDPperc1 -0.2971 1.0000
Table A.2 – Correlation Matrix, OLD Correlation matrix country level Beta-conv.
GDPpercGr GDPperc1
GDPpercGr 1.0000
GDPperc1 0.0464 1.0000
Table A.3 – Correlation Matrix, OLD+V4 Correlation matrix country level Beta-conv.
GDPpercGr GDPperc1
GDPpercGr 1.0000
GDPperc1 -0.2114 1.0000
B. Regional level
Table B.1 – Correlation Matrix, V4 Correlation matrix country level Beta-conv.
GDPpercGr GDPperc1 GDPpercGr 1.0000 GDPperc1 0.0234 1.0000 PU IN PR all OLD 53 109 53 215 NEW 7 26 22 55 V4 5 16 14 35 EU27 60 135 75 270
Number of NUTS 2 regions
35 NOTE: As the variable GDPperc1 is not used simultaneously with the other 5 independent variables (FEmp Mwayskm2, PopC, FDI and ErlEd) the correlation numbers of GDPperc1 with those are removed.
Table B.2 – Correlation Matrix, EU27
GDPpercGr GDPperc1 FEmp MwaysKm2 PopC FDI ErlEd
GDPpercGr 1.0000 GDPperc1 -0.1985 1.0000 FEmp -0.1087 - 1.0000 MwaysKm2 -0.1098 - 0.2543 1.0000 PopC -0.1751 - 0.0914 0.1301 1.0000 FDI -0.2478 - 0.4511 0.2672 0.1594 1.0000 ErlEd -0.0658 - -0.3180 0.0136 0.3359 -0.0106 1.0000
Table B.3 – Correlation Matrix, OLD
GDPpercGr GDPperc1 FEmp MwaysKm2 PopC FDI ErlEd
GDPpercGr 1.0000 GDPperc1 -0.0285 1.0000 FEmp -0.0409 - 1.0000 MwaysKm2 -0.0077 - 0.1442 1.0000 PopC -0.0556 - -0.0612 -0.0435 1.0000 FDI -0.1235 - 0.3744 0.0586 -0.1099 1.0000 ErlEd 0.0411 - -0.4834 -0.1290 0.3039 -0.3179 1.0000
Table B.4 – Correlation Matrix, NEW
GDPpercGr GDPperc1 FEmp MwaysKm2 PopC FDI ErlEd
36
3. Descriptive statistics
Table C.1 – Descriptive statistics, EU27
EU27
Variable Obs Mean Std. Dev. Min Max GDPpercGr 2970 2.848875 4.73485 -15.56017 21.75732 GDPperc1 2970 19098.89 7788.92 3900 65100 FEmp 2889 57.1297 9.913288 23,8 78,7 MwaysKm2 2587 27.49208 29.51769 0 225 PopC 2970 .3473849 .803453 -5.590974 4.758798 FDI 2843 446121.1 407947.7 -87 1307705 ErlEd 2815 27.73794 14.78186 3 85,9
Table C.2 – Descriptive statistics, OLD
OLD
Variable Obs Mean Std. Dev. Min Max GDPpercGr 2365 2.192637 4.364695 -15.56017 21.25984 GDPperc1 2365 21455.35 6573.555 3900 65100 FEmp 2294 58.35654 10.29421 23,8 78,7 MwaysKm2 2011 33.09796 30.70226 0 225 PopC 2365 .4960961 .7227392 -2.3324 4.758798 FDI 2301 554823.9 384726.4 7414 1307705 ErlEd 2227 30.02519 14.85645 3 85,9
Table C.3 – Descriptive statistics, NEW
NEW