Regional Convergence, Growth and Interpersonal Inequalities …ec.europa.eu/regional_policy/archive/policy/future/pdf/9_monfort_final... · Regional Convergence, Growth and Interpersonal

Regional Convergence, Growth and Interpersonal Inequalities across EU1

Report Working Paper of

Philippe Monfort

Directorate General Regional Policy European Commission

January 2009

1 This working paper has been written in the context of the report "An Agenda for a reformed Cohesion

Policy". It represents only the opinion of the expert and does not necessarily reflect the views of the European Commission.

2

TABLE OF CONTENTS

1. INTRODUCTION....................................................................................................... 3

2. EMPIRICAL EVIDENCE ON REGIONAL CONVERGENCE................................ 4

2.1. Summary measures of disparities ...................................................................... 4

2.2. Distribution dynamics ..................................................................................... 12

2.3. Growth, convergence and regional disparities ................................................ 23

2.3.1. Disparities, economic development and growth ............................... 24

2.3.2. Beta-convergence .............................................................................. 26

3. ANALYSING THE MACRO-EFFECTS OF COHESION POLICY....................... 29

4. THE RELATIONSHIP BETWEEN REGIONAL DISPARITIES AND INTERPERSONAL INEQUALITY ......................................................................... 33

5. CONCLUSIONS ....................................................................................................... 38

6. REFERENCES.......................................................................................................... 40

3

1. INTRODUCTION

The Treaty establishing the European Community sets economic and social cohesion as one of the main priorities of the Union. This priority is operationalised by the EU cohesion policy whose objectives are defined by Articles 2 and 4, and Title XVII of the Treaty. According to Article 2, cohesion policy should contribute to « promote economic and social progress as well as a high level of employment, and to achieve balanced and sustainable development ». Article 158 adds « in particular, the Community aims to reduce the disparities between the levels of development of the different regions and the backwardness of the least favoured regions or islands, including rural areas ».

Since its inception and the first programming period, the Treaty general mission has very often been interpreted as the promotion of convergence among regions, measured in terms of GDP per head. This type of convergence has also become a major aspect in assessing the effectiveness of European cohesion policy.

The perspective of per-capita income convergence is actually quite limited, for two reasons. First, income captures (especially from the point of view of measuring inequality) only one of the several dimensions of well-being. Second, and apart from that, regional convergence does not adequately capture the Treaty cohesion goal. This goal can in fact be interpreted as a combination of two general goals: an efficiency goal of offering to all regions the opportunity to more fully use their potential; an equity goal of promoting a reduction in the disparities of living standard of people living in different regions. The under-utilisation of capacity of a lagging region can successfully be reduced while a per-capita income gap with successful highly-performing regions increase; on the other hand, convergence of a lagging region can take place, but, due to rising interpersonal inequality within it, overall inequality might be on the rise.

These considerations do not suggest to disregards the analysis of convergence. Since we have very little tools to satisfactorily measure the productive potential of regions, the measure of GDP gap among regions and of how well is one region doing compared to other regions can offer some relevant information. It can also allow to explore the very complex relationship between growth, regional disparities and personal inequalities and to examine how it changes with time, place and/or context.

The objective of this paper is to take stock of the contributions that highlight the various dimensions of this relationship and present some additional findings. The paper is structured as follows. Section 2 provides an assessment of convergence among EU regions and Member States. Section 3 then examines the relationship between economic growth and convergence. Section 4 extends the analysis to the relationship between growth and interpersonal inequality. Very tentative conclusions are presented in section 5.

4

2. EMPIRICAL EVIDENCE ON REGIONAL CONVERGENCE

This section examines regional convergence among EU regions using different approaches and methodologies. The analysis is based on various measures capturing the evolution of regional disparities in order to verify the existence, a process through which disparities tend to decrease in time (or sigma-convergence2). The section also reviews methods allowing for an in-depth inspection of the distribution of regional performance.

2.1. Summary measures of disparities

One of the most frequently used summary measures of sigma-convergence is the coefficient of variation of regional GDP per head3. The following figure shows the evolution of the coefficient of variation computed on the EU-15 and EU-27 NUTS 2 regions for the period 1980-20054 and 1995-2005 respectively5.

Figure 1: Coefficient variation: GDP per head, NUTS 2 regions, EU-12, EU-15 and EU-27

0.28

0.29

0.30

0.31

0.32

0.33

0.34

1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

EU-15

0.35

0.37

0.39

0.41

0.43

0.45

0.47

EU-15 EU-12

EU-27

EU-27 EU-12

Source: Cambridge Econometrics and EUROSTAT database. DG REGIO own calculation.

The results confirm the findings of previous analysis (see for instance Neven and Gouyette, 1995, Magrini 2004 or Ertur et al. 2006) that convergence among EU-15 regions has been strong up to the mid 90's but the process has since then lost momentum. From 1980 to 1996, the evolution of disparities among EU-15 regions indeed features a clear downward trend, the coefficient of variation decreasing from 0.33 to 0.28. On the contrary, from 1996 onwards, it remains relatively stable at around 0.29. Note that in the absence of counterfactual, these results are

2 The literature identifies two main concepts of convergence, beta and sigma convergence. Beta-

convergence mainly focuses on detecting possible catching-up processes. 3 Disparity measures can of course be computed on other dimensions than GDP per head. In particular,

the analysis of regional disparities can be completed with dimensions such as the employment rate, unemployment rate or education attainment which also convey information concerning social cohesion.

4 Series covering the period 1980 to 2005 in fact combine two databases, the Cambridge Econometrics database from 1980 to 1994 and EUROSTAT database from 1995 to 2005. Strictly speaking, the data have some differences. However, these are marginal and it has been checked that they do not produce significant breaks in the series.

5 Series have been smoothed by means of a moving average to eliminate possible influence of the business cycle on the extent of regional disparities.

5

uninformative concerning the effectiveness of cohesion policy: it is not based on such type of observation that Boldrin and Canova (2001) conclude to the relative failure of cohesion policy.

On the other hand, disparities decrease rapidly among EU-27 regions, the coefficient of variation falling from 0.43 in 1995 to 0.35 in 2005. This has led many observers to conclude that if convergence is still at work within the EU-27, it is due to the fact that the poorest regions in the new Member States catch-up on the Union's richest ones, while among EU-15 regions convergence is no longer taking place. Both conclusions require refinements.

First of all, in several Member States which have recently joined the Union regional disparities have generally increased, as underlined by the evolution of the coefficient of variation computed for the EU-12 (Figure 1). The following figure completes the analysis by displaying the evolution of the coefficient of variation calculated on regional GDP per head for each new Member States.

Figure 2: Coefficient variation: GDP per head, NUTS 2 regions, EU-12 countries

Coefficient variation: GDP/head, NUTS-2 regions, EU-12 countries

0

5

10

15

20

25

30

35

40

45

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

BG

CZ

HU

PL

RO

SI

Source: EUROSTAT database. DG REGIO own calculation.

For all countries considered, disparities have increased, sometimes dramatically like in Romania where the coefficient of variation rose from 0.15 in 1995 to almost 0.40 in 2005.

These results might indicate a phenomenon that is frequent in the early stages of economic catching up at country level: a few core regions drive the catching up since strong agglomeration effects are at work, while peripheral regions are more sluggish. On the other hand, these results might be strongly biased by a “commuting effect”.

6

The commuting effect tends to overestimate GDP in capital or big cities where many people produce and do not live and underestimate it for ‘commuter belt’ or ‘dormitory’ regions. Overall, it therefore rises measurable regional disparities. Correction for the commuting effect can lead to significant reduction in the extent of measured regional disparities. A recent study finds that the mean log deviation index is reduced by 15% across the EU-25 as a whole in 2005 and by around 30% in the EU-15 when adjusting for the commuting effect. The effect is even greater is some countries like Belgium, the UK, Slovakia and the Czech Republic6. This effect has increased in the new Member countries in recent years and contributes to the observed increased within-country regional disparities.

Differences in the evolution of regional disparities at the EU and country levels should of course be thoroughly taken into consideration when examining convergence of EU regions. This is particularly the case when analysing regions performance according to their level of development. The next figure is a scatter plot of regional growth rates (annual average) between 1995 and 2005 and regional GDP per head in 1995, both variables measures relative to the EU average.

Figure 3: GDP per head, growth and levels, NUTS 2 regions, EU-27, 1995-2005

y = -0,0001x + 0,014R2 = 0,2009

-4,0%

-3,0%

-2,0%

-1,0%

0,0%

1,0%

2,0%

3,0%

4,0%

5,0%

6,0%

0 50 100 150 200 250 300

GDP per head EU-27=100

GDP per head growth 1995-2005, EU-27=0


The graph clearly shows a catching-up process, poor regions (as measured by their GDP poor head level in 1995) tending to grow faster than rich ones. The regions located in North-West quadrant of the graph are those which at the same time were poorer but have grown faster than the average. They are usually referred as the catching-up regions in the literature (see for instance Ezcurra, 2007) and represent

6 Applica, Ismeri Europa and WiiW (2008), Ex Post Evaluation of Cohesion Policy Programmes 2000-

2006 financed by the European Regional Development Fund in Objective 1 and 2 Regions, on behalf of the European Commission.

7

32% of the regions considered here. This leads to the conclusion that convergence7 is taking place as such process indeed contributes to reduce regional disparities among EU regions.

However, results are quite different once regional economic performance is measured relative to Member States average instead of EU average, as shown by the following figure.

Figure 4: GDP per head, growth and levels, NUTS 2 regions, EU-27, 1995-2005

y = 4E-05x - 0,0042R2 = 0,0191

-3.0%

-2.0%

-1.0%

0.0%

1.0%

2.0%

3.0%

4.0%

5.0%

40 60 80 100 120 140 160 180 200 220 240GDP per head MS=100 1995

GDP per head growth 1995-2005, MS=0


There is no longer a significant relationship between the two variables and hence no signs of a general catching-up process. Interestingly, the proportion of catching-up regions is almost the same (31%). The min difference is that the proportion of regions located in the South-West quadrant, i.e. those that were relatively poor in 1995 and that has a negative relative growth rate (sometimes referred as losing regions) is much more important, 38% against 18% in case measures are taken against the EU-27 average. Convergence is therefore much less likely within country as in a number of cases (typically the ones observed in the new Member States), poorest regions seem to be left out of a rapid development process.

A part from these cautionary notes, the evidence in so far reviewed has led to the conclusion that in the last ten years disparities have diminished among countries and increased within countries. The Theil index, a frequently used dispersion indicator, allows to explore this issue by decomposing dispersion into within subgroups and among subgroups components. Applied to the context of EU regions, the Theil index can split disparities into between-countries (Bc) and within-countries (Wc) disparities.

7 This is in fact a so-called beta-convergence process, see section 3.2 below.

8

Figure 5: Theil index: GDP per head, NUTS 2 regions, EU-27 decomposed into Member States component (Bc) and regional component (Wc)

0.020

0.030

0.040

0.050

0.060

0.070

0.080

0.090

0.100

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Total

Betweeen

Within


The index clearly confirms an overall reduction of disparities among EU regions. It also confirms that this reduction is due to the fact that disparities among Members States are strongly decreasing. On the contrary, disparities among regions within Member States are slightly increasing.

The ratio Wc/Bc corresponds to the share of global disparities explained by regional disparities within countries while its complement measures the share explained by disparities among Member States. In 1995, 70% of regional disparities among EU regions reflected disparities among Member States, the remaining 30% being due to regional disparities within Member States. By 2006, disparities among Member States only account for 53% of regional disparities which for 47% are explained by regional disparities within Member States.

The same type of conclusion emerges when considering separately the EU-15 over the period 1980 to 2005. Regional disparities have decreased but this is mainly due to decrease in disparities between Members States, their share in global disparities falling from 55% in 1980 to 14% in 20058.

8 The jump observed for Wc in 1991 corresponds to the German reunification and the inclusion of East-

German Länders in the data.

9


0.000

0.010

0.020

0.030

0.040

0.050

0.060

1980

1981

1982

1983

1984

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

WcBcT


For the EU-12, regional disparities have increased between 1995 and 2005. This again mainly reflects a strong increase in disparities within the Member States but, compared to the EU-15, the reduction in disparities between Member States is much more modest, the index falling from 1995 to 2000 but increasing afterwards.


0.000

0.020

0.040

0.060

0.080

0.100

0.120

0.140

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Wc

Bc

T


Notice finally that other summary measures, like for instance the Gini index, the Atkinson index or the MLD, qualitatively yield the same type of results.

Disparity measures can of course be computed on other dimensions than GDP per head. In particular, the analysis of regional disparities on the dimension of GDP per head is often completed with one on dimensions such as the unemployment and the employment rates which also convey information concerning social cohesion. The following figures display the coefficient of variation calculated on these dimensions.

10

Figure 8: Coefficient variation: Unemployment rate, NUTS 2 regions, EU-15 and EU-27

40.00

45.00

50.00

55.00

60.00

65.00

70.00

1999 2000 2001 2002 2003 2004 2005 2006 2007

EU-15

EU-27


Figure 9: Coefficient variation: Employment rate, NUTS 2 regions, EU-15 and EU-27

10.0

10.5

11.0

11.5

12.0

12.5

13.0

13.5

14.0

2000 2001 2002 2003 2004 2005 2006

EU-15EU-27


Regional disparities are declining, both for the EU-15 and the EU-27, which points to the fact that the convergence process is at work on these dimensions, even in the EU-15 where it seemed less vivid on the dimension of GDP per head. These variables are likely to be linked to one another and one would like to check if low levels of development are systematically associated with low performance on the labour market. The following figures illustrate the correlation between GDP per head, unemployment rate and employment rate for a panel data covering EU-27 NUTS 2 regions between 2000 and 2005.

11

Figure 10: Correlation between GDP per head and unemployment rate (UR), NUTS 2 regions, EU-27, 2000-2005

y = -0.0715x + 15.952R2 = 0.242

0

5

10

15

20

25

30

35

0 50 100 150 200 250 300

GDP per head

UR


Figure 11: Correlation between GDP per head and employment rate, NUTS 2 regions, EU-27, 2000-2005

y = 0.1146x + 51.784R2 = 0.289

35

45

55

65

75

85

95

0 50 100 150 200 250 300

GDP per head

Employment rate


GDP per head is negatively correlated with the unemployment rate (coefficient of correlation is -0.49) and positively correlated with the employment rate (coefficient of correlation is 0.54). Lagging regions therefore often feature relatively lower rates of the labour force utilisation. The scope for tapping on unused resources may therefore be quite large in such regions.

12

Finally, GDP per head, defined as the ratio of GDP to population, can be decomposed as follows:

GDP/Population = (GDP/Employment) * (Employment/Working age population) * (Working age population/Population)

The first element of the right hand side of the equality is labour productivity (L Pr), the second is the employment rate (ER) and the third is the share of the population in working age (WA). Convergence of GDP per head could therefore stem from convergence in each of these dimensions9. The next table reports the coefficient of variation for the variables at stake from 2000 to 200510.

Table 1: Coefficient of variation, GDP/head, L Pr, ER and WA, EU-27, EU-15 and EU-12, 2000-2005

2000 2001 2002 2003 2004 2005 GDP/head 0.42 0.42 0.41 0.40 0.40 0.39

L Pr 0.45 0.47 0.46 0.46 0.46 0.47 ER 0.13 0.13 0.14 0.13 0.12 0.12

EU-27

WA 0.07 0.07 0.07 0.07 0.08 0.08 GDP/head 0.28 0.29 0.29 0.29 0.29 0.29

L Pr 0.22 0.22 0.22 0.22 0.23 0.25 ER 0.13 0.13 0.12 0.12 0.11 0.11

EU-15

WA 0.03 0.03 0.03 0.03 0.05 0.05 GDP/head 0.45 0.46 0.45 0.44 0.42 0.44

L Pr 0.52 0.51 0.48 0.47 0.46 0.46 ER 0.22 0.22 0.20 0.21 0.21 0.20

EU-12

WA 0.14 0.14 0.14 0.14 0.14 0.14 Source: EUROSTAT database. DG REGIO own calculation.

For the EU-15, increasing disparities in terms of labour productivity and the share of the population in working age have been somewhat compensated by a reduction in disparities in the employment rate. For the EU-12, the evolution of GDP per head disparities reflects a coincidence of movements in disparities on the dimensions of labour productivity and the employment rate.

2.2. Distribution dynamics

Summary measures of disparities have the interest of greatly synthesising the information but they are not capable of capturing changes in the positions of individual regions in the ranking of regions when this change does not affect the distribution. Furthermore these summary measures are left unaffected by changes of the distribution that compensate each other. The examination of changes in the ranking of individual regions and of the different parts of the distribution can add considerable insights to the analysis of regional disparities by providing more details about the mechanisms at work in the convergence process. We therefore conduct a visual inspection of the distribution and its dynamics by relying on non-parametric estimation of the density functions and Salter graphs. We then turn to

9 The covariance of the components could also partly explain the convergence of the GDP per head. 10 Note that the coefficient of variation of GDP per head cannot be reconstructed from the coefficients of

variation of its components. In particular, it does not correspond to the product the coefficients of variation of its components.

13

Markov chain analysis which allows for a statistical characterisation of the distribution dynamics (see for instance Quah, 1996, Fingleton, 1997 or Pellegrini, 2002).

The non-parametric estimation of the density functions is based on Gaussian kernels. The next figures show the estimation of the GDP per head distributions for the EU-27 and EU-15 NUTS 2 regions for the years 1995 and 2005.

Figure 12: GDP/head (EU-27=100): Distribution EU-27 NUTS 2 regions, 1995-2005, Gaussian kernel estimation

0

0.002

0.004

0.006

0.008

0.01

0.012

0.014

0.016

0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 320 340 360 380

1995

2005

Frequency

GDP/head


Figure 13: GDP/head (EU-15=100): Distribution EU-15 NUTS 2 regions, 1995-2005, Gaussian kernel estimation

0

0.005

0.01

0.015

0.02

0.025

0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 320 340 360 380

1995

2005

Frequency

GDP/head


14

The estimated distributions reveal a convergence process at work both for the EU-15 and the EU-27. Frequencies around the mean increase while they tend to decrease for values below 80% and above 120% of the EU average. For the EU-27, the distribution also evolves from bimodal to unimodal. This is particularly interesting as most analysis had indeed detected a bimodal distribution from the 1980's up to the end of the 1990's, leading to the conclusion that a polarisation process was taking place in Europe, with a "club" of poor regions converging towards a low steady state (around 40% of the EU average in Figure 12) and another club of richer regions converging towards a high steady state (around 110% of the EU average in figure 13). The shape of the distribution in 2005 is much less suggestive of polarisation, making the scenario of various convergence clubs among EU regions less likely. Note that the visual examination of the distribution does not allow identifying the regions that actually move in the distribution. In particular, there is no way to know if this evolution towards a unimodal distribution is due to the catching-up of poor regions all over the EU or to the catching-up of some Member States.

Salter's graph is another method which allows for a visual examination of the distribution dynamics. It consists in ranking regions along the horizontal axis according to their GDP per head and report the corresponding level of GDP per head on the vertical axis for a base year. Then holding the base year rank positions of regions constant on the horizontal axis, new series show the regions' GDP per head for subsequent years. As a result, any significant changes in the regional distribution of GDP per head become visible. In addition, regions can be identified and their performance compared.

Such graphs can be used to detect patterns of persistence or gradual change in the regional distribution on GDP per head. In particular, the more the series is horizontal, the more it reflects a distribution where disparities are limited. The following figure reports the Salter graph for the EU-27 NUTS 2 regions, comparing the distributions of their GDP per head in 1995 and 2005.

Figure 14: GDP/head (EU-27=100): Salter graph, NUTS 2 regions, 1995-2005

0

50

100

150

200

250

300

350

1995

2005

Poly. (2005)

Latvija

Eesti

Bucureşti - Ilfov

Mazowieckie

Közép-Magyarország

Border, Midland and Western

Attiki

Bratislavský kraj

Southern and Eastern

PrahaBerkshire, Buckinghamshire and Oxfordshire

Inner London

GuyaneDytiki Ellada

Sicilia Malta Molise LüneburgMünster Berlin

Région de Bruxelles-Capitale / Brussels Hoofdstedelijk Gewest

Prov. HainautEU regions

GDP/head

Poor regions Rich regions Source: EUROSTAT database. DG REGIO own calculation.

15

The graph indicates a general tendency towards increased horizontality of the series, which is the sign of convergence among EU regions. This is emphasised by comparing the 1995 series with the dashed line which is a polynomial approximation of the series for 2005. The graph also shows that this evolution is clearly due to a process where poor regions catch up on the richest ones. The frequency of upward movements in the distribution is indeed higher in the low end of the distribution compared to that of downward movements in the high end of the distribution. Such movements are of course not uniform among poor and rich regions. Some poor regions see their relative GDP per head decline during the period of observation (this is for instance the case for Dytiki in Greece or Hainaut in Belgium) while for some rich regions, relative GDP per head increases (e.g. Inner London).

Figure 15: GDP/head (EU-15=100): Salter graph, NUTS 2 regions, 1995-2005

0

50

100

150

200

250

300

1995

2005

Poly. (2005)

Andalucia

Border, Midland and Western

Cantabria

Attiki

Aragón

Bratislavský kraj

País Vasco

Southern and EasternBerkshire, Buckinghamshire

and Oxfordshire

Schleswig-HolsteinGuyane

AbruzzoSicilia

Molise UmbriaKoblenz Berlin

Inner London

EU regions

GDP/head

Poor regions Rich regions

Região Autónoma da Madeira

Köln

Valle d'Aosta


The same type of conclusion can be drawn at the EU-15 level. The series clearly has a tendency to become more horizontal, with frequent upward movements at the low end of the distribution ad downward movements at the high end of the distribution.

This information is conveniently complemented by mapping changes in GDP per head between two particular dates, as displayed in the following figure for the period 1995 to 2005.

16

Figure 16: GDP/head (EU-27=100): Change in GDP per head, EU-27 NUTS 2 regions, 1995-2005

The issue of how persistent is the distribution, whether and how quickly regions with lower-than-average income move up and the convergence to a long term stationary distribution can be analysed by the Markov chain analysis. If the Markov process which is supposed to underlie the dynamics of the distribution is ergodic11, there exist a stationary distribution corresponding to a steady state towards which the distribution will converge in time and which can be interpreted as a projection of the distribution in the future given the transition process described today by the data.

11 A Markov chain is ergodic if it is possible for the variable under examination to move from every state

of the distribution to any other state in a finite number of steps. Ergodicity and the existence of a stationary distribution is ensured when the modulus of the second eigenvalue of the transition matrix is strictly smaller than 1.

17

The pace of convergence towards the steady state can be assessed by computing the half-life of the process, i.e. the amount of time it will take to cover half the distance separating the current distribution from the stationary distribution. The speed of the transition process may also be examined by assessing how long it takes to transit from one state to another. In Markovian terminology, this is called the mean first passage time.

Finally, it is possible to characterise the stability of the process, i.e. the probability of having no movements within the distribution (stability index) or the extent to which it is driven by convergence forces, i.e. the probability that movement from the original distribution to the final one is in the direction of an increasing convergence toward the mode of the stationary distribution (convergence index).

Obviously this analysis just allows to project and appreciate the long term effects of past trends. They are mute on whether the past trends are actually likely to persist, as this statistical analysis assumes.

Applying these concepts to our example yields the following results:

Table 2: GDP/head (EU-27=100): Transition probability matrix, EU-27 NUTS 2 regions, 1995-200512

Transition probability matrix

2005

GDP/head Percentage of regions 0-50 50-75 75-100 100-150 150-

0-50 13% 75.8% 24.2% 0.0% 0.0% 0.0%

50-75 12% 0.0% 80.6% 19.4% 0.0% 0.0%

75-100 25% 0.0% 7.7% 76.9% 15.4% 0.0%

100-150 43% 0.0% 0.0% 14.4% 80.2% 5.4%

1995

150- 7% 0.0% 0.0% 0.0% 44.4% 55.6% Summary statistics

0-50 50-75 75-100 100-150 150-

Stationary distribution 0% 15% 39% 41% 5%

Half-life 3.9 periods

S 0.74

C100-150 0.40

C75-100 0.36

Mean first passage time 0-50 50-75 75-100 100-150 150-

0-50 9.01* 1015 4.13 9.29 18.38 61.10

50-75 3.72* 1016 6.53 5.17 14.25 56.97

75-100 3.72* 1016 28.56 2.59 9.08 51.81

100-150 3.72* 1016 36.34 7.78 2.43 42.72

150- 3.72* 1016 38.59 10.03 2.25 19.99 Source: EUROSTAT database. DG REGIO own calculation.

12 Quah (1993) or Legallo (2004) rely on a different method for computing the transition probability

matrix where cell ij is the number of occurrence of passages from class i to class j during the whole period of observation. This has the advantage of exploiting the panel dimension of the data and of giving a more precise estimation of the true transition probabilities. Adopting this approach did not led to substantial differences in the results presented here and we therefore chose to measure transition between the two end dates of the period of observation as it makes it easier to interpret the transition probability matrix.

18

Table 2 presents results when the classes of GDP per head are predefined with thresholds at 50%, 75%, 100% and 150% of average values. The transition probability matrix measures movements within the distribution. Each element ij represents the percentage of regions in GDP per head class i moving to class j between 1995 and 2005. For instance, the transition probability matrix above indicates that 24.2% of the regions which were in the class [0, 50] in 1995 moved to the class ]50, 75] in 2005. On the other hand, all the remaining 75.8% remained in the same category.

The transition probability matrix indicates a relative persistence of the distribution. The values on the diagonal are quite high, suggesting a high probability to remain in the same class of GDP per head. Following Pellegrini (2002), we compute a stability index based on the trace of transition probability matrix, i.e. the sum of the elements of the main diagonal. Its value is 0.74 which confirms a relatively high persistence of the distribution.

However, persistence is less pronounced at the end classes of the distribution. In particular, as we saw, 24.2% of the poorest regions in 1995 moved up to the next category in 2005, while 19.4% of the regions in the class ]50, 75] in 1995 moved to the class ]75, 100] in 2005, and none to the lower class. In general, for regions with GDP per head lower than 100% of the EU average, movements towards upper categories are much more frequent than movements down, the reverse being true for regions with GDP per head above this threshold.

This suggests a convergence process where poorer regions catch-up on the other ones, the distribution evolving towards one with lower frequencies at the tails, as clearly indicated by the stationary distribution. The distribution is therefore likely to feature fewer disparities in the long-run with a concentration of observations in the central categories. The underlying convergence process can be summarised by a convergence index proposed by Pellegrini (2002), measuring the probability to stay or move to a cell that increases convergence. The convergence index is 0.40 for convergence towards the class ]100, 150] and 0.36 for convergence towards the class ]75, 100] which confirms a transition dynamics containing strong convergence forces.

Convergence towards the stationary distribution is rather slow with a half-life of 3.9 periods of 10 years, i.e. 39 years. The thickness of the system is also well captured by the mean first passage time matrix (MP) whose element ij measures the time necessary on average to move from class i to class j. The elements outside the main diagonal of MP indicate that the transitions to other categories are relatively slow, the lower passage time being 4.13 periods from the class ]0, 50] to the class ]50, 75]. However, the pace of transition is systematically higher for low GDP per head classes while in general movements up are faster than movements down.

The same analysis conducted on the EU-15 regions lead to similar conclusions as summarised in the following table.

19


Transition probability matrix 2005

GDP/head Percentage of

regions 0-50 50-75 75-100 100-150 150- 0-50 5% 54.5% 45.5% 0.0% 0.0% 0.0% 50-75 17% 8.3% 52.8% 38.9% 0.0% 0.0% 75-100 43% 0.0% 6.7% 76.4% 16.9% 0.0% 100-150 31% 0.0% 0.0% 24.6% 73.8% 1.5%

1995


0-50 50-75 75-100 100-150 150- Stationary distribution 2% 9% 51% 35% 4% Half-life 4.5 periods S 0.69 C100-150 0.43 C75-100 0.40


0-50 61.78 2.20 5.24 12.39 191.80 50-75 133.72 11.33 3.04 10.19 189.60 75-100 159.81 26.08 1.96 7.15 186.56 100-150 164.31 30.58 4.50 2.87 179.41 150- 171.31 37.58 11.50 7.00 26.63


In particular, the analysis indicates that, although at a much slower pace than for the EU-27, a convergence process has definitely taken place among EU-15 regions for the period considered. The stability index is only 0.69 which is mainly explained by the low values observed on the diagonal for the lower GDP per head classes and important movements to upper categories. Indeed, 45.5% (respectively 38.9%) of the regions in the class ]0, 50] (respectively ]50, 75]) moved to the next class between 1995 and 2005. The stationary distribution shows that if most of the convergence has already taken place for the classes of GDP per head above 75% of the EU average, the process remains vivid for the lower classes and is expected to continue in the future. The convergence index is 0.43 for the class ]100, 150] and 0.40 for the class ]75, 100].

This tendency is however not captured by the aggregate inequality measures reviewed in the preceding section. The explanation is that the number of regions in the lower categories is relatively small and even if within the EU-15, poor regions are rapidly catching-up, their weigh is too small for this movement to be reflected in summary measures. This calls for a prudent attitude when drawing conclusions based on this type of instruments. By summarising the dispersion of the distribution into one measure, they provide convenient and helpful indicators. However, they fail to capture movements that may be relatively small in statistical terms but are nevertheless of importance from a policy point of view.

20

Transition probability matrices allow to identify which regions are moving and in which categories. For instance, the EU-27 regions moving from the [0, 50] class to the ]50, 75] class between 1995 and 2005 are the following: Estonia, Lithuania, Közép-Dunántúl (Hungary), Wielkopolskie (Poland), Dolnośląskie (Poland), Pomorskie (Poland), Bucureşti – Ilfov (Romania) and Západné Slovensko (Slovakia).

Figures 17 and 18 respectively map movements within EU-27 and EU-15 regions between 1995 and 2005. Regions are identified according to three fundamental transition regimes: upwardly mobile (green), stationary (orange) and downwardly mobile (red). The category of GDP per head in 1995 is also represented, darker tones indicating lower categories of GDP per head in 1995.

Figure 17: GDP/head (EU-27=100): Mapping of transition, EU-27 NUTS 2 regions, 1995-2005

21

Figure 18: GDP/head (EU-15=100): Mapping of transition, EU-15 NUTS 2 regions, 1995-2005

The maps reveal the predominance of regions in the stationary regime. It also shows that regions in the upwardly (respectively downwardly) mobile regime are mainly poor (respectively rich) regions. Finally, the map reflects a strong country component in regional evolutions, with a tendency for regions belonging to a same country to move in relatively similar directions.

Results of Markov chain analysis are strongly dependent on the discrete approximation of the range of values into non-overlapping classes. The choice of the classes indeed uniquely determines the transition matrix and hence the whole set of results. In order to overcome such problem, it is sometimes preferred to define the classes using quantiles so that each class includes the same number of observations. The definition of the classes is then less arbitrary. As a matter of example, the following tables report the results obtained following this approach for the EU-15.

22




regions 0-72.8 72.8-87.4 87.4-96.7 96.7-109.7 109.7- 0-72.8 20% 66.7% 31.0% 2.4% 0.0% 0.0% 72.8-87.4 20% 12.2% 63.4% 19.5% 2.4% 2.4% 87.4-96.7 20% 0.0% 23.8% 35.7% 35.7% 4.8% 96.7-109.7 20% 0.0% 2.4% 31.7% 43.9% 22.0%

1995

109.7- 20% 0.0% 0.0% 2.4% 26.2% 71.4% Summary statistics

0-72.8 72.8-87.4 87.4-96.7 96.7-109.7 109.7- Stationary distribution 8% 22% 20% 25% 25% Half-life 3.8 periods S 0.56 C96.7-109.7 0.53

Mean first passage time 0-72.8 72.8-87.4 87.4-96.7 96.7-109.7 109.7- 0-72.8 12.66 3.84 9.26 13.05 20.27 72.8-87.4 34.13 4.63 6.74 10.39 17.54 87.4-96.7 45.95 11.82 4.92 5.70 13.80 96.7-109.7 50.06 15.94 5.37 3.93 10.34 109.7- 53.22 19.10 8.42 3.97 4.04


The stationarity (i.e. high values of the diagonal) is less concentrated than with this definition of categories than with the preceding one. This simply reflects the fact that the former are narrower in the centre of the distribution and hence are record more movements than between categories at the tail of the distribution which much wider. However, the stationary distribution still indicates convergence of the poorest towards higher levels of GDP per head. Results are therefore not qualitatively different from those obtained with the categories previously chosen.

Given the particular evolution, we also conducted a Markov chain analysis of the regional distribution of GD P per head among the new Member States.

23




regions 0-75 75-100 100-125 125-150 150- 0-75 5% 91.7% 8.3% 0.0% 0.0% 0.0% 75-100 17% 21.7% 60.9% 13.0% 4.3% 0.0% 100-125 43% 0.0% 16.7% 83.3% 0.0% 0.0% 125-150 31% 0.0% 0.0% 40.0% 40.0% 20.0%

1995


0-75 75-100 100-125 125-150 150- Stationary distribution 55% 21% 22% 2% 1% Half-life 5.3 periods S 0.65 C0-75 0.44


0-75 1.83 12.00 33.48 113.00 332.00 75-100 9.98 4.78 21.48 101.00 320.00 100-125 15.98 6.00 4.58 107.00 326.00 125-150 18.90 8.92 2.92 52.76 219.00 150- 19.73 9.75 3.75 44.80 131.90


The dynamics of the distribution within the group of New Member States is radically different. Persistence is particularly high at the low end of the distribution. In addition, for poor regions the frequency of downward moves within the distribution is higher than that of upward moves. This reveals a tendency for the distribution to evolve toward concentration in low GDP per head categories as indicated by the stationary distribution according to which 55% of EU-12 regions could end up with a GDP per head below 75% of the EU-12 average.13 This is clearly revealing of a divergence process which could take place among EU-12 regions. This result is obviously strongly influenced by the commuting effect discussed above and holds under the rather strong and un-discussed assumption that the 1995-2005 trend can be projected in the next decades.

2.3. Growth, convergence and regional disparities

What is the relationship between growth, regional disparities and convergence? Have countries characterised by high growth rates recorded growing or decreasing disparities among their regions? To what extent is the (regional or global) growth process supportive or detrimental to regional convergence? And to what extent is convergence supportive or detrimental to (regional or global) growth? These are very complex questions for which we cannot expect to provide an answer. But we can aim at clarifying these issues by reviewing and discussing results in the literature.

13 For this group of regions, the categories of GDP per head had to be adapted so as to have sufficient

observations in each category.

24

2.3.1. Disparities, economic development and growth

In a seminal paper on income distribution, Kuznets (1955) examines the historical relationship between the level of development (measured by GDP per head) and inequalities in the distribution of personal income (measured by the Gini coefficient) for a sample of countries. The main result of his analysis is that the relationship between the two variables yields an inverted-U curve. At the first stage of development, inequalities would tend to grow with GDP per head while at later stage of development, they would tend to decrease when the level of GDP per head increases.

The issue tackled by Kuznets is a central one: whether development leads to rising or diminishing inequality among individuals. We’ll see in section 5 that the availability of better data and of the experience of five more decades show that the pattern assumed by Kuznet is not at work. But here a different issue must be addressed. The relationship envisaged by Kuznets has opened up a very different field of works where the inverted-U curve has been claimed to hold for disparities across regions (not across individuals) and for the stages of development of different regions in the same moment in time (not of the same region in different moments in time).

In particular, Williamson (1965) compares the level of development of the US States (measured again by GDP per head) with that of regional disparities internal to each State. The results are similar in nature to those obtained by Kuznets. For regions at lower stage of development, regional disparities seem to be the higher, the higher is the GDP per head while at a higher stage, the link between the two variables seems inverted, regional disparities being the lower, the lower is the level of GDP per head. Some authors (see for instance Barrios and Strobl, 2005 or Ezcurra and Pascual, 2007) have examined this relationship using more sophisticated statistical techniques and confirm the existence of an inverted-U curve linking GDP per head and regional disparities for the EU.

An inverted-U curve seems indeed lo link in 2005 the level of country GDP per head and the extent of internal regional disparities, as measured by the standard deviation of regional GDP per head (in logarithm). Based on similar observations, it has sometimes been argued that: (i) development will inevitably create regional disparities; and (ii) that regional disparities will "naturally" vanish as economies develop. However, it should be stressed that this graph corresponds to a cross-section analysis and must be interpreted as such. In particular, it says nothing concerning the relationship between growth and regional disparities, nor does it allow to infer that regional disparities will be “naturally” reduced by growth along the path towards development.

25

Figure 19: Inverted-U curve: GDP/head (EU-27=100), EU-27 NUTS-2 regions, 2005

Österreich

Belgique-België

Balgarija

?eská Republika

DanmarkEspaña

Ellada

Magyarország

Ireland

Italia

Nederland

Polska Portugal

Romania

Slovenija

Slovenská Republika

United Kingdom

Suomi / FinlandFrance

Sverige

0.1

0.2

0.3

0.4

0.5

0.6

-1.5 -1 -0.5 0 0.5 1Log GDP/head

St Dev of Log GDP/head


This caution is confirmed once we introduce time and dynamics in the above chart. The following figure also plots country GDP per head and the extent of internal regional disparities, but also adds a temporal dimension by showing the trajectory of each country on these two dimensions between 1995 and 2005. For every country, each point corresponds to a year and shows the level of GDP per head level and the extent of regional disparities. Different patterns seem to co-exist.

Let us first focus on the new Member States. In some countries, an increase in GDP per head is accompanied through time by an increase in the extent of regional disparities (Hongrie, Poland, …) (green arrows). However, there are countries (the Czech Republic and Romania) for which disparities increase, but no substantial increase in GDP per head takes place (orange arrows)14.

14 In the first group, growth in the other regions, although not as high, is still substantial. For instance, in

Slovakia, the growth rate over the ten years period in the capital city regions of Bratislavský kraj is 45% but the average growth rate in the other regions of the country is 22%. In the second group, growth in the peripherical regions is much lower, which eventually leads to poor growth performance at the national level. For instance, in Romania, the capital city regions of Bucureşti-Ilfov has grown by 77% but on average the other regions have only grown by 5%, GDP per head in Sud-Vest Oltenia even decreasing by 5%.

26

Figure 20: Inverted-U curve: GDP/head (EU-27=100), EU-27 NUTS-2 regions, 1995-2005

Österreich

Belgique-België

Balgarija

Česká Republika

España

Ellada

Magyarország Ireland

Italia

Nederland

Polska Portugal

Romania

Sverige

Slovenija

Slovenská Republika

United Kingdom

0.1

0.2

0.3

0.4

0.5

0.6

-1.5 -1 -0.5 0 0.5 1Log GDP/head

St Dev of Log GDP/head


As for the EU-15, the pattern predicted by the inverted-U curve, whereby an increase in relative GDP per head is accompanied by a decrease in regional disparities, holds only in Ireland (only partly) and Spain (with very small changes). For a number of other countries, disparities have reduced but at the same time relative GDP per head has decreased. In fact, cyclical effects seems to be dominating by which periods of high (low) growth are generally associated to periods of increasing (decreasing) regional disparities.

2.3.2. Beta-convergence

These observations underline the complex relationship between economic growth and regional disparities. For some countries, episodes of high growth seem to be associated with growing regional disparities. For others, regional disparities are increasing even in the absence of growth.

The relationship between regional disparities and growth in economic theory

An attempt of discovering regularities in the relation between disparities and growth has been made by the literature on beta-convergence.

Beta-convergence refers to a process in which poor regions grow faster than rich ones and therefore catch-up on them. The concept of beta-convergence is directly related to neo-classical growth theory (Solow, 1956) whose one of the key assumptions is that factors of production, in particular capital, are subject to diminishing returns. Accordingly, the growth process should lead economies to a long run steady-state characterised by a rate of growth which depends only on the (exogenous) rates of technological progress and the growth of labour force. Diminishing returns also implie that the growth rate

http://en.wikipedia.org/wiki/Technological_progress

27

of poor economies should be higher and their income and /or GDP per head levels should therefore catch up with those of rich economies.

When all economies are assumed to converge towards the same steady-state (in terms of GDP per head), beta-convergence is said to be absolute. However, one can envisage that the steady-state depends on features specific to each economy, in which case convergence will still take place but not necessarily to the same long-run levels. This will be the case when GDP per head is supposed to depend on a series of determinants, such as for instance factor endowment or institutions, which can vary from one economy to an other even in the long-run. Beta-convergence is then said to be conditional. In any case, in the context of a beta-convergence process, growth contributes to convergence as poor regions grow faster than rich ones.

However, some schools of thought inspired by the work of Myrdal (1957) argue that growth is in fact a spatially cumulative process, which is likely to increase disparities. Different strands of research, like theories of development (Rosenstein and Rodan 1943, Fleming 1955, Hirschman 1958), urban economics (Segal 1976, Henderson 1988), new economic geography (Krugman 1991, 1993, Fujita, Krugman and Venables 1999, Thisse 2000), and endogenous growth (Romer 1986), have in common the idea that there are local economies of scale and increasing rates of return. In such context, economic growth will tend to concentrate in a few places in a self-sustained, spatially selective and cumulative process15. Under these assumptions, economic growth fosters regional divergence rather than convergence. But several of these theories also argue that, beyond a certain level, due to congestion or social effects, agglomeration can reduce efficiency, i.e. increasing return come to an end. Ultimately the theory does not offer strong general predictions, but rather offers tools to explore every specific case.

At the same time, the extent of regional disparities could also impact on the global efficiency of an economy and therefore be a determinant of economic growth. A rich literature focuses on social/income inequalities and their impact on growth but it does not lead to univocal conclusions, some contributions suggesting a positive impact of inequalities on growth, others a negative one (see Aghion et al., 1999 for a extended survey). In addition, the extent to which these results can be used to highlight the role of regional disparities on global growth is limited. In particular, the link between social/income inequalities and regional disparities is not unidirectional and important social/income inequalities can co-exist with low regional disparities (see section 4 below).

Testing the relationship between regional disparities and growth

If economic theory does not offer firm general conclusions on the relationship between regional disparities and economic growth, clear-cut results are not offered by empirical analysis either.

15 See also Perroux (1955) concept of growth poles.

28

The seminal papers by Barro and Sala-i-Martin (1992) and Mankiw et al. (1992) have launched a huge empirical literature attempting to detect and measure the extent of beta-convergence in various contexts. In particular, number of beta-convergence analysis have been conducted on the EU regions.

Some are based on absolute convergence frameworks. For instance, Cuadrado-Roura (2001) tests the hypothesis that regions with an initial level of GDP per head below the EU average have an above-average growth rate. Data cover EU-12 old Member States over the period 1977 to 1994. The main finding is that there is a convergence process but it is rather weak and very slow. The estimated convergence rate is less than 2 %, with a half-life of about 35 years. This type of result is also obtained by other analyses like for instance López-Baso (2003) on data covering the EU-12 old Member States over the period 1975 to 1996.

A number of contributions have also examined the possibility that the convergence rate is not necessarily constant in time. For instance, Cuadrado-Roura (2001) or Martin (2001) find that the convergence speed decreases in time. However, other analyses find that the convergence rate decreases until the mid 1980’s and increases afterwards (see for instance Yin et al., 2003 or Geppert et al. 2005).

Others frameworks allow for conditional convergence, notably by taking into account the different economic conditions of countries or groups of countries. Results differ from one analysis to the other. For instance, Fagerberg and Verspagen (1996), Cappelen et al. (2003) or Geppert et al. (2005) point to low or even to absence of a convergence process. On the contrary, Neven and Gouyette (1995) consider two different regimes for Northern and Southern European regions and find a significant convergence rate. Identically, Basile et al. (2005) find evidence of a significant convergence process. Martin (2001) distinguishes various groups of regions among which Objective 1 regions and different sub-periods. Convergence is detected for all groups and periods, with a convergence rate that decreases in time for all groups except for Objective 1 regions among which convergence rate increases.

More recent contributions also introduce a spatial dimension in the formulation of the problem (see for instance Baumont et al., 2003 or Dall’erba and Le Gallo, 2006). There are indeed reasons to believe that the omission of a space from the analysis of regional beta-convergence is likely to produce biased results. In particular, working with regional data imposes to address the specific issue of spatial dependence. The proximity and numerous linkages between (more or less) neighbouring regions imply that regional economic variables are likely to be interdependent which conflicts with the assumptions under which convergence equations can be validly estimated. Solution consists in introducing so-called spatial lags or cross-regressive models (accounting for the fact that the growth rate of one region also depends on either the growth rate or the level of income of surrounding regions) or considering spatial error models (accounting for possible systematic measurement errors due to the spatial correlation of the variables

29

included in the model and which makes the assumption of spatial independence of the error terms too restrictive).

In general, the inclusion of spatial effects is found highly relevant (presence of significant local spatial autocorrelation and of spatial heterogeneity) and tends to reduce the estimated speed of the global convergence process while highlighting that the speed of convergence is higher for the poorest regions of Europe.

Other empirical analyses are based on new growth theory and/or economic geography, trying to detect the possible positive link between economic growth and regional disparities. For instance, works by Quah (1996, 1997) emphasise that it is likely that countries and/or periods characterised by high growth rates also feature rises in the extent of regional inequalities. Analysing the evolution of income disparities in European regions and transition countries in the period 1977-2002, Arbia et al. (2005) note that if there is an EU-wide tendency for regional disparities to decrease, intra-country regional disparities increase. However, their estimates from cross-country data show a positive relationship between the extent of regional disparities (measured by various disparity indices) and growth. Petrakos et al. (2005) find that short-term divergence and long-term convergence processes may coexist. According to their results, regional inequalities could follow a pro-cyclical pattern (dynamic and developed regions growing faster in periods of expansion and slower in periods of recession) while at the same time, significant dispersion effects partly offset the cumulative impact of growth on space.

Overall, current empirical analysis does not lead to a clear-cut conclusion concerning the relationship between growth and regional disparities. Results strongly depend on the specification adopted (absolute or conditional convergence, set of control variables, incorporation of spatial effects) and on the observations (period and regions considered, dataset used) and it is therefore difficult to draw a single general conclusion from the vast panel of existing studies (see for instance the survey by Eckey and Türk, 2006). One common finding however is that a convergence process is taking place among EU regions (at least when considering EU-15 or EU-27) but that the process is rather slow.

3. ANALYSING THE MACRO-EFFECTS OF COHESION POLICY

The frameworks mentioned in the preceding section, as well as others, have also been used to analyse the effect of cohesion policy on convergence. Three approaches have generally been adopted to assess the impact of cohesion: econometric analysis of convergence models; macroeconomic simulation models; qualitative evaluation studies. Since very little counterfactual impact evaluation of projects has been conducted in the realm of cohesion policy, no attempt has instead been made to conduct meso-analysis of such studies.

30

The first is econometric analysis such as those described in the preceding section. It mostly focuses on the impact of cohesion policy on macroeconomic variables like GDP per head, employment or productivity. In a majority of cases, the approach amounts to estimate a model (in a reduced or structured form) which borrows from growth theory so that the analysis also provides information on the extent and pace of beta-convergence.

For instance, de la Fuente and Vives (1995) analyses the contribution of different policy instruments to convergence among 17 regions of Spain between 1980 and 1991. They find that the impact of public investment on regional convergence is rather small (accounting for around 1% of the inequality reduction) but that promoting education has significantly contributed to the reduction of regional disparities in Spain.

Cappelen et al. (2003) analyse regional convergence and the impact of cohesion policy for 95 regions of the EU-9 between 1980 and 1997. They present evidence that the 1989 reform of cohesion policy has increased its effectiveness in generating growth in poorer regions and promoting smaller disparities in productivity and income in Europe. However, they also stress that growth in poorer regions is hampered by an unfavourable industrial structure (such as an important agricultural sector) or lack of R&D and point to the need to accompany the support provided by cohesion policy with policies that facilitate structural change and increase R&D capabilities in poorer regions.

Rodriguez-Pose and Fratesi (2004) examine how Structural funds support is allocated among different development axes in Objective 1 regions for the period 1989 to1999. The categories of expenditure they consider are infrastructure, education-human capital, business support and support to agriculture. They find no significant impact of funds devoted to infrastructure or to business support. Only investment in education and human capital has medium-term positive effects, while support for agriculture has short-term positive effects on growth.

Ederveen et al. (2006) attempt to assess the effectiveness of Structural Funds and whether this is conditioned by the quality of regional "institutions" proxied by quantitative measures of corruption, inflation or openness to trade. Their approach is in fact following the one Burnside and Dollar (2000) applied to assess the effectiveness of aid on growth in developing countries, i.e. the estimation of a beta-convergence specification where measures of institutional quality and the amount of Structural Funds are introduced as additional regressors. Their findings points to the absence of a global significant impact of Structural Funds on regional growth but that support allocated to regions with high quality of institutions are effective, leading to the conclusion that EU Structural Funds are conditionally effective16.

Fagerberg and Verspagen (1996) analyse regional growth in the EU in the postwar period and examine the levels and growth of per capita GDP for a sample of 70 regions, covering six of the EU Member States. They find that during most of the post-war period, regional disparities have steadily declined but that since the early 1990's, there is

16 Note that Bradley and Untiedt (2008) reproduce the estimates of Ederveen et al. using the same set of

data and demonstrate that their results are in fact not robust to the inclusion of Greece in the sample of countries considered. More fundamentally, this type approach has been severely criticised (see for instance Rodrick and Rodríguez, 2001) for its incapacity to assess policy effectiveness. The argument is that policy is not a variable independent of the other determinants of growth, which is particularly sensitive for Cohesion policy as support provided through Structural Funds explicitly depends on the level of GDP per head.

31

a reversal in this trend. Moreover, differences in levels of productivity and income across European regions have remained substantial. According to their findings this would mainly be due to variables, notably R&D effort, investment support from the EU, the structure of GDP and differences in unemployment that have a diverging impact on regional economic performance. They also find some support for the idea of different 'growth clubs' characterized by different dynamics, productivity and unemployment levels.

Other contributions develop arguments borrowed from the Economic Geography. Martin (1999) discusses the role of public infrastructures in a two-region endogenous growth model and analyses the contribution of different types of public policies on growth, economic geography and spatial income distribution. Its main conclusion is that public policies that reduce the cost of innovation can attain the objectives of higher growth and more even spatial distribution of both income and economic activities. On the contrary, public policies targeting transport infrastructure face a trade-off between growth and the reduction regional disparities.

Puga (2002) discusses the role of regional policies, especially transport infrastructure improvements, in the EU context where Member States have developed different production structures and have witnessed a increase in the polarisation of regional unemployment rates. In particular, the paper stresses that the impact the reduction of transport costs between regions may not foster convergence and can in fact harm the industrialisation prospects of less developed areas. Moreover, the framework also shows how the impact of lower transport costs on less developed regions depends on certain aspects of the economic environment (such as mobility and wage rigidities) and on characteristics of the projects. In particular, while TransEuropean Transport Network give better access to the main activity centres, it is also likely to increase the gap in relative accessibility between core and peripheral areas, therefore reinforcing the position of core regions as transport hubs.

Relying on spatial econometrics to include spatial effects in the estimation of a conditional Beta-convergence model, Dall’Erba and Le Gallo (2007) assess the impact of structural funds on convergence among 145 European regions over the period 1989 to 1999. They analyse separately each of the five objectives of regional support. The results indicate either insignificant impact or very small and even negative in some cases. However, some of the figures obtained should be considered with caution. In particular, support under Objective 1 is found to have a positive impact in the core regions but an insignificant one in the periphery regions which shed some doubts on the capacity of such specification to capture and measure the determinants of the regional growth process.

The second approach is based on macroeconomic simulation models.

32

The use of macroeconomic models for analysing cohesion policy is hampered by two very relevant factors. First, most impact of cohesion policy is supposed to be transmitted through supply side channels, a context in which the Lucas' critique is particularly relevant. The policy indeed partly aims at changing the behaviour of economic agents (e.g. in terms of education or research and development). Its implementation should therefore lead, if it is successful, to a break in the parameters estimed to account for such behaviour. Second, cohesion policy principally targets key engines of growth such as the stocks of physical and human capital and of knowledge, variables whose level and effect on growth are extremely difficult to measure. Results should then be considered as simulations useful in order to make policy assumptions accountable, rather than estimations.

Having clarified this caveats, existing analysis present different results, often suggesting positive impact of cohesion policy.

De la Fuente (2002) assesses the impact of EU Cohesion policy on growth and convergence in the Spanish regions using a supply oriented model estimated with regional panel data covering a period of 30 years. He finds that the contribution of the 1994-2000 Community Support Framework (CSF) to the growth of output and employment in the poorer Spanish regions is substantial. The model also shows that the growth effects of the CSF vary significantly across territories, reflecting differences in both the volume of investment and in its rate of return, which in turn positively depends on whether or not regions have reached a a saturation point in terms of infrastructure.

Bradley et al. (2007) bases their analysis on a review of Structural funds impact assessment carried out using the HERMIN model. The model highlights the central role played by supply side effect of Structural funds in order to generate long-lasting impact of the policy. The magnitude of such effects is likely to be affected by the design and/or implementation of the programmes and the model suggests a sensitivity of the impact to the quality of the programmes. In addition, the analysis emphasises that the real, long-term benefits of the Structural funds are more likely to be associated with the way in which each of the lagging economies responds to opportunities arising in the rest of the EU and the world rather than with the Structural funds in isolation. They also stress that structural effects are typically smaller than the demand-side effects of the Structural funds, albeit of different magnitudes from one Member State to another.

Honohan et al. (1997) conducted a model-based analysis of the impact of Cohesion funds on the Irish economy. They find that, depending on the assumptions embodied in the model, on average one percentage point of the Irish economy growth rate in the 1990's could be attributed to support provided under Cohesion funds. Using the HERMIN framework, Sosvilla-Rivero et al. (2006) find that support provided under the Structural funds raised the growth rate of Castilla la Mancha by 0.64 percentage points during the period 1988 to 1999.

Finally, Arcalean et al. (2007) develop a two-regions endogenous growth model with public investment in infrastructure and education. They calibrate the model to Portugal and find that the Structural funds can enhance growth in the lagging regions and reduce regional disparities without necessarily producing convergence, the impact being not always sufficient to counterbalance agglomeration economies benefiting the advanced urban regions.

33

The third approach followed in the literature is based on qualitative evaluation studies. They examine the effects of cohesion policy not only on macroeconomic variables but also on other aspects, like the quality of the institutions, or on variable pertaining to the meso and microeconomic levels relevant for the interventions. The results obtained generally point to substantial beneficial effects of cohesion policy.

ECOTEC (2003) conducts a comprehensive evaluation of the Objective 1 programmes and finds substantial gains in terms of jobs created, business supported and economic activities that have been enabled by cohesion policy interventions. However, the study emphasises that the wide variety of quantitative and qualitative targets of the programmes makes it difficult to evaluate the overall effectiveness and impact of the policy.

In 2003, the Centre for Strategy and Evaluation Services conducted an ex-post evaluation of Objective 2 programmes 1994-99 and found that the programmes contributed to the development of regions were they were implemented, although the sharp fall in the unemployment rates that it observed could well be attributed to national or EU-wide policies. The evaluation anyway points again to the difficulty of assessing the overall impact of the programmes due to lack of appropriate data and methodologies.

4. THE RELATIONSHIP BETWEEN REGIONAL DISPARITIES AND INTERPERSONAL INEQUALITY

If regions converge and close the gap in terms of development level, should we expect to observe a reduction in the interpersonal inequalities in both income and other dimensions of well being? In general, what is the relationship between growth, convergence and social inequalities? And what is the regional dimension of such relationship?

On the whole, we now know that the simple inverted-U relation between stages of development and levels of interpersonal inequalities does not hold. The relevant rebounce in income inequalities in the Anglo-saxon countries and in some Nord-European countries in the last twenty years and the stationary level of inequalities in the majority of the other countries shows, as Atkinson (2007) puts in, that “If there was an inverse-U, the pattern has now become a U”, or more precisely, that “it is misleading to talk of «trends» when describing the evaluation of income inequality… it is better to think in terms of «episodes» when inequality rose or fell”.

The evolution of personal income inequalities is captured by Figure 21. Personal income inequalities have recently increased in some countries while they have decreased in others. However, they have increased in 75% of the new Member States against 46% of the EU-15 countries, with substantial surges in Latvia, Lithuania and Hungary. These observations are in line with the findings of Förster et al. (2002) who analyse the evolution of personal income inequalities during the 1990's in the Czech Republic, Hungary, Poland and Russia.

34

Figure 21: Change in the personal income inequalities, EU-27*

EU-25

EU-15Belgium

Czech Republic

Denmark

Estonia Spain

FranceItaly

Cyprus

Latvia

Lithuania

Luxembourg

Hungary

MaltaAustria

Poland

Portugal

Romania

SloveniaSlovakia

Sweden

Netherlands

U-KFinland

Ireland Greece

Germany

Bulgaria

-1

-0.5

0

0.5

1

1.5

2

2.5

3

* Measured as changes in the income quintile share ratio (S80/S20), which is the ratio of total equivalent disposable income received by the top richest 20% of the population to that received by the bottom 20% of the population. Changes are computed for the period 1995-2005 for EU-15 countries and 2000-2005 for the new Member States, except for Cyprus (2002-2005), Latvia (1999-2005), Malta (1999-2005) and Slovakia (2004-2005). Source: EUROSTAT database.

Unfortunately, the issue of the linkage between regional disparities and income inequality has received very little attention. This is part of the overlapping of efficiency and equity considerations when discussing convergence objective, as if reducing regional disparities amounted to reduce income inequality.

Unbalanced growth can generate regional disparities (if it is not evenly distributed in space) and /or social inequalities (if it is not evenly distributed in society). For both processes to coincide, growth should be located in rich regions, affect rich people and rich people should live in rich regions, a configuration which is likely but not systematic. As emphasised by Amos (1983), coincidence between the two processes may indeed not remain along the development path. In particular, as rich individuals often live in the growth areas in the early stages of the development process, regional disparities and personal inequalities first rise together. However, when the economy develops, rich individuals leave the growth spots (e.g. moving from cities to the suburbs), leading regional disparities to decrease but leaving personal income inequalities unchanged.

The relationship between regional disparities and income inequalities is therefore mostly an empirical question. Amos (1988) addresses the issue for the United States, investigating if States with greater social convergence are also those with greater regional convergence. For this analysis, he principally relies on contingency tables and simple linear regressions. His main finding is that on average, States in which regions (counties) have converged also experienced convergence in personal income. There is therefore a positive correlation between the two variables. However, this correlation and its sign may not be constant in time.

The evidence available for Europe suggests that income inequality is indeed highly concentrated within countries. Jesuit et al (2002) find a very high variation of poverty rates of different regions within each country. Disparities in poverty rates are particularly important in some countries, such as Italy or the United-Kingdom. Moreover, as reported by Förster et al. (2002), the majority of inequality in each of the countries is due to intra-regional rather than interregional disparities. Some regions are indeed characterised by relatively high rates of poverty computed using a local poverty line. For instance, more than 17% of the population in Greater London have an income (after tax and transfer) below 50% of the area media equivalent income (1995 estimates). In Calabria and Sicily,

35

this rate reaches 16 and 18% respectively. Achieving convergence does not seem to be enough to significantly reduce inequality. There is a genuine regional dimension in the determination of inequalities that needs to be addressed region by region.

But is there at least a positive correlation in Europe, as there seems to be in the US, between the convergence of regions within a country and the degree of income inequality in the same country? Obviously, with average per capita GDP converging all personal income would also converge unless a strong counterbalancing effect is at work by which: a) the benefit of the improvement are strongly concentrated, in catching-up regions, on a minority of the population; b) a significant reduction takes place in the net income public transfers to the less wealthy.

The issue can be very preliminary explored with the available data. EUROSTAT provides statistics concerning the extent of regional disparities (measured by the coefficient of variations) and the extent of personal income inequalities (measured by the Gini coefficient). The figures for the EU Members States are displayed in the following table17.

Table 6: Dispersion of GDP/head and personal income in 2005

2005 Regional disparities Personal income

inequalities Netherlands 11.6 26 Ireland 14.7 32 Finland 15.7 26 Denmark 16.3 24 Sweden 16.3 24 Germany (including ex-GDR from 1991) 17.4 26 Austria 17.4 25 Spain 18.2 32 Slovenia 18.9 24 Poland 19.4 33 France 19.9 28 United Kingdom 20.4 32 Portugal 23.2 38 Italy 23.8 33 Czech Republic 25.1 26 Romania 25.4 33 Belgium 25.8 28 Bulgaria 26.2 25 Greece 26.5 33 Slovakia 32.3 28 Hungary 35.7 28


In 2005, for the Members States for which regional disparities can be computed, the correlation between the two variables is 0.21. The next figure plots the data and the linear regression between the two variables, revealing a positive relationship between them.

17 Cyprus, Estonia, Latvia, Lithuania, Luxemburg and Malta have only one region coinciding with the

country. These countries are therefore not included in the analysis.

36

Figure 22: Relationship between regional disparities and personal income inequalities: Cross-section, 2005

y = 0.1424x + 25.71R2 = 0.0454

20

22

24

26

28

30

32

34

36

38

40

10 15 20 25 30 35Regional disparities (coefficient of variation)

Personal income inequalities (Gini coefficient)


The correlation between regional disparities and personal income inequalities can also be examined in time. The next table shows the correlation between the two variables computed for each Member States over the period 1995-2005. The sign of the correlation is not systematically positive or negative. However, there is a majority of countries (11 over 17 or 65%) for which it is positive.

Table 7: Correlation between regional disparities and personal income inequalities, 1995-2005

Correlation 1995-2005 Poland 0.89 Portugal 0.61 United Kingdom 0.69 Slovenia 0.40 Hungary 0.21 Finland 0.15 Austria 0.27 Belgium 0.22 Spain 0.19 Greece 0.07 Bulgaria 0.01 Romania -0.11 Netherlands -0.29 Italy -0.15 Ireland -0.68 Germany (including ex-GDR from 1991) -0.57 France -0.43 Source: EUROSTAT database. DG REGIO own calculation.

Finally, one can also analyse the contingency of changes in the two dimensions. Figures are reported in the following table.

37

Table 8: Contingency of changes between regional disparities and personal income inequalities, 1995-2005

Change in regional disparities* Change in personal

income inequalities* Poland 2 6 Portugal 0.5 2 United Kingdom 1 2 Romania 1.5 2 Slovenia 1.6 2 Hungary 3.1 2 Netherlands -0.3 -2 Italy -0.9 4 Finland -1.9 2 Austria -0.5 2 Ireland -0.1 2 Germany (including ex-GDR from 1991) -0.2 1 Belgium 0.5 -2 Spain -2.4 0 France -1 0 Greece 5.8 0 Bulgaria 8.6 0 * Changes are measured in percentage points. Source: EUROSTAT database. DG REGIO own calculation.

Three groups of countries can be distinguished. For the first one, accounting for 41% of the sample, there is a positive contingency as regional disparities and personal income inequalities have moved in the same direction. For the second group, representing 35% of the countries included in the analysis, the contingency is negative. Finally, for 24% of the countries, the variation in the index of personal income inequalities is not sufficient to establish the direction of the contingency.

Considering only the first two groups of countries, the following figure plots the changes in regional disparities against changes in personal inequalities between 2000 and 2005.

38

Figure 23: Changes in regional disparities and personal income inequalities, 1995-2005

y = 0.4447x + 1.4474R2 = 0.1059

-3

-2

-1

0

1

2

3

4

5

6

7

-3 -2 -1 0 1 2 3 4Change in regional disparities

Change in personal income inequalities


The linear regression reveals a positive relationship, meaning that on average Member States where regional disparities have grown fast from 1995 to 2005 are also those where personal income inequalities have increased the most substantially.

5. CONCLUSIONS

This paper has examined various aspects of the relationship between growth, regional disparities and interpersonal-inequalities. It first analysed the recent evolution of convergence among EU regions and Member States using different methods and instruments. It then discussed the relationship between economic growth and convergence and extended the analysis to personal income inequalities.

Serious assessments of convergence cannot be based on a single measure but rather on a panel of instruments and a sound interpretation of their results exploiting their complementarities. In particular, simple or too aggregate measures may fail to capture important aspects of the convergence process. The analysis of regional convergence detects a convergence process at the level of the EU-27 but also at within the group of EU-15 regions. However, there is evidence that regional divergence may occur within the new Member States and within the group of EU-12 regions taken as a whole. Several indicators indeed point to growing disparities within a number of new Member States. This trend is mainly driven by the fact that very rapid growth is taken place in a limited number of regions (mostly urban areas and capital city regions) while poorest regions are often left out of this development process.

However, even if the analysis of regional disparities is conducted thoroughly, it says little concerning the effectiveness of the EU cohesion policy. Keeping track of regional disparities and monitoring their evolution is definitely of key importance for the design and conduct of cohesion policy. One must keep in mind that the analysis of disparities, whether pointing to the presence or absence of convergence, can generally not be used to infer firm conclusions concerning the success or failure of the policy.

The examination of the data and the review of the literature pointed to the complexity of the relationship between regional inequalities, convergence and economic growth.

39

Regional disparities show a tendency to increase in periods of rapid economic expansion and decreasing in periods of slow growth but in general, the theoretical and empirical literature remains inconclusive concerning the nature of this relationship. This conclusion extends to contributions attempting to assess the impact of cohesion policy.

As for the relation between regional convergence and inter-personal inequality, the very limited evidence shows that intra-regional inequality largely outweighs inter-regional inequality. Furthermore, on the whole (but not in all countries) regional disparities are positively correlated with personal income inequality.

40

6. REFERENCES

Arcalean, C., Glomm, G. and Schiopu, I. (2007), "Growth Effects of Spatial Redistribution Policies," Caepr Working Papers 2007-002, Center for Applied Economics and Policy Research, Economics Department, Indiana University Bloomington.

Aghion, P., Caroli, E. and Garcia-Peñalosa, C. (1999), "Inequality and Economic Growth: The Perspective of the New Growth Theories, Journal of Economic Literature, 37, pp. 1615-1660.

Amos, O. (1983), "The Relationship Between Regional Income Inequality, Personal Income Inequality, and Development," Regional Science Perspectives, Vol. 13, No. 2, 1983, pp. 3-14.

Arbia, G., De Dominicis, L., and Piras, G. (2005), "The relationship between Regional Growth and Regional Inequality in EU and transition countries - a Spatial Econometric Approach," ERSA conference papers, European Regional Science Association.

Atkinson, A.B., “Distribution and growth in Europe – the empirical picture: a long-run view of the distribution of income”, in Eu 2007 Conference.

Barrios, S. and Strobl, E. (2005), "The Dynamics of Regional Inequalities", Economic Papers of the European Commission, 229, Brussels.

Barro, R. and Sala-i-Martin, X. (1992), "Convergence", Journal of Political Economy, 100, 223-251.

Basile, R., Nardis, S. and Girardi, A. (2005), "Regional Inequalities and Cohesion Policies in the European Union", http://www.camecon.com/services/europe/Downloadable%20files/isae%20paper.PDF (2006/01/06).

Baumont, B., Ertur, C. and Le Gallo, J. (2003), "Spatial Convergence Clubs and the European Regional Growth Process, 1980-1995", in Fingleton, B. (ed.) European Regional Growth, Berlin et al., Springer, 131-158.

Boldrin, L. and Canova, F. (2001), "Inequality and Convergence in Europe's Regions. Reconsidering European Regional Policies", Economic Policy, 16, 207-253.

Bradley, J., Untiedt, G. and Mitze T. (2007), "Analysis of the Impact of Cohesion Policy: A Note Explaining the HERMIN-based Simulations", http://ec.europa.eu/regional_policy/sources/docgener/evaluation/pdf/hermin07.pdf.

Bradley, J. and G. Untiedt (2008), "EU cohesion policy and "conditional" effectiveness: What do cross-section regressions tell us?", GEFRA Working Paper 4/2008.

Burnside, C. and D. Dollar (2000), "Aid Policies and Growth", World Bank Working Paper N° 1777, Washington D.C.

Cappelen, A., Castellacci, F., Fagerberg, J. and Verspagen, B. (2003), "The Impact of EU Regional Support on Growth and Convergence in the European Union", Journal of Common Market Studies, 41, 621-644.

Centre for Strategy and Evaluation Services (2003), Ex-Post Evaluation of 1994-99 Objective 2 Programmes, Report for the DG REGIO off the European Commission.

Cuadrado-Roura, J. (2001), "Regional Convergence in the European Union. From Hypothesis to the Actual Trends", Annals of Regional Science, 35, 333-356.

http://ideas.repec.org/s/inu/caeprp.html

http://ideas.repec.org/s/wiw/wiwrsa.html

http://www.camecon.com/services/europe/Downloadable files/isae paper.PDF

http://ec.europa.eu/regional_policy/sources/docgener/evaluation/pdf/hermin07.pdf

41

Dall'erba, S. and Le Gallo, J. (2006), "Evaluating the Temporal and the Spatial Heterogeneity for the European Convergence Process, 1980-1999", Journal of Regional Science, 46, 269-288.

Dall'erba, S. and Le Gallo, J. (2007), "The impact of EU regional support on growth and employment", Czech Journal of Economics and Finance - Finance a Uver, vol. 57, n° 7-8, 325-340.

De la Fuente, A. and Vives, X. (1995), "Infrastructure and Education as Instruments of Regional Policy: Evidence from Spain", Economic Policy, 10(1), 11–54.

De la Fuente, A. (2002), "The effect of Structural Fund spending on the Spanish regions: an assessment of the 1994-99 Objective 1 CSF", Mimeo, Instituto de Análisis Económico, Barcelona.

Eckey H.-F and Türck, M. (2006), «Convergence of EU-Regions: A Literature Review», Discussion Paper at the Economic Department of the University of Kassel, 86/06, Kassel.

ECOTEC (2003), Ex-Post Evaluation of 1994-99 Objective 1 Programmes, Report for the DG REGIO off the European Commission.

Ertur, C., Le Gallo, J. and Baumont, C. (2006), "The European Regional Convergence Process, 1980-1995. Do Spatial Regimes and Spatial Dependence Matter?", International Regional Science Review, 29, 3-34.

Ezcurra, R. and Pascual, P. (2007), "Regional Polarisation and National Development in the European Union", Urban Studies, Vol. 44, No. 1, 99–122.

Escurra, R. (2007), "Is Income Inequality Harmful for Regional Growth? Evidence from the European Union", Urban Studies, Vol. 44, No. 10, 1953–1971.

Ederveen, S., Groot H.and R. Nahuis (2006), "Fertile Soil for Structural Funds?A Panel Data Analysis of the Conditional Effectiveness of European Cohesion Policy," Kyklos, Blackwell Publishing, vol. 59(1), 17-42.

Fagerberg, J. and Verspagen, B. (1996), "Heading for Divergence? Regional Growth in Europe Reconsidered", Journal of Common Market Studies, 34 (3), 431-448.

Fleming, J. (1955) "External economies and the doctrine of balanced growth", Economic Journal, 65, 241-56.

Fingleton, B. (1997), «Specification and Testing of Markov Chain Models. An Application to Convergence in the European Union», Oxford Bulletin of Economics and Statistics, 59, 385-403.

Förster, M., Jesuit, D., Rainwater, L and Smeeding, T. (2002), Regional Poverty and Income Inequality in Central and Eastern Europe: Evidence from the Luxemburg Income Study, Luxembourg Income Study Working Paper Series, Working Paper No. 324.

Fujita, M., Krugman, P. and Venables, A. (1999) The Spatial Economy, Cambridge Massachusetts: The MIT Press.

Geppert, K., Happich, M. and Stephan, A. (2005), «Regional Disparities in the European Union: Convergence and Agglomeration», DIW Working Paper, 525, Berlin.

Henderson, V. (1988) Urban Development, Oxford University Press. Hirschman, A. (1958) The strategy of economic development, New Haven, Conn.: Yale

University Press.

http://ideas.repec.org/a/bla/kyklos/v59y2006i1p17-42.html

http://ideas.repec.org/a/bla/kyklos/v59y2006i1p17-42.html

http://ideas.repec.org/s/bla/kyklos.html

42

Honohan, P. (ed.) (1997), EU Structural funds in Ireland: a mid-term evaluation of the CSF 1994-99, Policy Research Series Paper N° 31, Dublin, ESRI.

Jesuit, D., Rainwater, L. and Smeeding, T. (2002), Regional Poverty within the Rich Countries, Luxembourg Income Study Working Paper Series, Working Paper No. 318.

Krugman, P. (1991), "Increasing Returns and Economic Geography", Journal of Political Economy, 99, 183-99.

Krugman, P. (1993), "On the number and location of cities", European Economic Review, 37, 293-298.

Leonardi, R. (2005), Cohesion Policy in the European Union: The Building of Europe Basingstoke: Palgrave.

López-Bazo, E. (2003), «Growth and Convergence Across Economies. The Experience of the European Regions», in Fingleton, B., Eraydin, A. and Paci, R. (eds.) Regional Economic Growth, SMEs and the Wider Europe, Aldershot et al., Ashgate, 49-74.

Magrini, S. (2004), «Regional (Di)Convergence», in Henderson, J.V. and Thisse, J.-F. (eds.) Handbook of Regional and Urban Economics, Amsterdam et al., Elsevier, 2741-2796.

Mankiw, N., Romer, D. and Weil, D. (1992), «A Contribution to the Empirics of Economic Growth», Quarterly Journal of Economics, 107, 407-437.

Martin, Ph. (1999), "Public Policies, Regional Inequalities and Growth", Journal of Public Economics, (73)1 (1999), 85-105.

Martin, R. (2001), "EMU Versus the Regions? Regional Convergence and Divergence in Euroland", Journal of Economic Geography, 1, 51-80. Myrdal, G. (1957), Economic theory and underdeveloped regions, Hutchinson. Neven D. and Gouyette, C. (1995), «Regional Convergence in the European Union»,

Journal of Common Market Studies, 33, 47-65. Pellegrini, G. (2002), «Proximity, Polarization, and Local Labor Market Performances»,

Networks and Spatial Economics, 2, 151-173. Perroux, F. (1970). "A note on the concept of growth poles", in McKee, D., Dean, R. and

Leahy, W. (eds) Regional economics theory and practice, Free Press, 93-103. Petrakos, G., Rodriguez-Pose, A. and Rovolis, A. (2005) "Growth, Integration and

Regional Inequalities in Europe", Environment and Planning A, 37, 1837-1855. Puga, D. (2002), "European regional policy in light of recent location theories", Journal

of Economic Geography 2(4), 373-406. Quah, D. (1996), «Empirics for Economic Growth and Convergence», European

Economic Review, 40, 1353-1375. Quah, D. (1997), «Empirics for Growth and Distribution. Stratification, Polarization and

Convergence Clubs», Journal of Economic Growth, 2, 27-59. Rodrik, D. and F. Rodríguez (2001), "Trade Policy and Economic Growth: A Skeptic's

Guide to the Cross-National Evidence,", Macroeconomics Annual 2000, eds. Ben Bernanke and Kenneth S. Rogoff, MIT Press for NBER, Cambridge, MA.

http://team.univ-paris1.fr/teamperso/martinp/jpube.pdf

43

Rodriguez-Pose, A. and Fratesi, U. (2004), "Between Development and Social Policies: The Impact of European Structural Funds in Objective 1 Regions," Regional Studies, vol. 38(1), 97-113.

Romer, P.(1986), "Increasing returns and long-run growth", Journal of Political Economy, Vol. 94, 5, 1002-1037.

Rosenstein-Rodan, P. (1943), "Problems of industrialization of Eastern and Southeastern Europe", Economic Journal, 53, 202-11.

Segal, D. (1976), "Are there returns of scale in city size?", Review of Economics and Statistics, 89, 393-413.

Solow, R. (1956), «A Contribution to the Theory of Economic Growth», Quarterly Journal of Economics, 70, 1, 65-94.

Sosvilla-Rivero, S., Bajo-Rubio, O. and Diaz-Roldán, C. (2006), "Assessing the effectiveness of the EU's regional policies on real convergence: An analysis based on the HERMAN model", European Planning Stidies, 14, 383-96.

Thisse, J. (2000), "Agglomeration and regional imbalance: Why? And is it bad?", EIB Papers 5 2, 47-68.

Williamson, J. (1965), "Regional inequality and the process of national development: a description of patterns", Economic Development and Cultural Change, 13, 3–45.

Yin, L., Zestos, G. and Michelis, L. (2003), «Economic Convergence in the European Union», Journal of Economic Integration, 18, 188-213.

Documents

Regional Convergence, Growth and Interpersonal Inequalities …ec.europa.eu/regional_policy/archive/policy/future/pdf/9_monfort_final... · Regional Convergence, Growth and Interpersonal