Income dynamics in an enlarged Europe: the role of capital regions

Ann Reg Sci (2012) 48:663–693DOI 10.1007/s00168-010-0400-x

ORIGINAL PAPER

Income dynamics in an enlarged Europe: the roleof capital regions

Sheila A. Chapman · Stefania Cosci ·Loredana Mirra

Received: 5 January 2010 / Accepted: 27 July 2010 / Published online: 14 August 2010© Springer-Verlag 2010

Abstract This article is aimed to study, by means of both a nonparametric and aparametric approach to convergence, whether after the recent enlargements of theEuropean Union the traditional twofold spatial regime of regional per capita incomegrowth, envisaging a north/south and/or a cohesion/non-cohesion countries divide,should be replaced by an alternative east/west spatial pattern. A second relevant issueis whether new member regions where capital cities are located are benefiting from thesame “network effect” that stimulated growth in old member capital regions duringthe 1980s and early 1990s. We find evidence, by means of spatial econometrics tools,of significant spatial effects in the enlarged Europe which seems to be formed by agroup of old member regions, slowly becoming more homogeneous, and a newcomersgroup which represents a separate “convergence club” but whose capital regions arerapidly integrating into the west. The European regional policy may play a crucial rolein this context.

JEL Classification C14 · C21 · O18 · O52 · R11 · R12

1 Introduction

The literature on integration and convergence is far from unanimous and the empiricalresearch on the subject has not yet reached a common answer to the question if, and

S. A. Chapman · S. CosciUniversity Lumsa of Rome, Rome, Italye-mail: [email protected]

S. Coscie-mail: [email protected]

L. Mirra (B)University of Rome Tor Vergata, Rome, Italye-mail: [email protected]

123

664 S. A. Chapman et al.

under which conditions, convergence actually takes place. A consistent strand of theliterature maintains that overall inequality among countries has increased over time,while convergence has occurred only within small groups of economies (Magrini2004). This last finding is consistent with “club convergence” theory1 predicting thateconomies converge to one another only if, on the basis of their initial conditions, theybelong to the same basin of attraction. One important implication of this theory is thatricher countries, or regions, will converge toward a middle-rich position and poorerones toward a middle poor-one (Quah 1996a,b).

Several studies find evidence of a permanent polarization in income distributionacross European Union regions, with rich regions clustering in the North and poorones in the South (López-Bazo et al. 1999; Le Gallo and Ertur 2003; Ertur and Koch2006). Studies in the New Economic Geography show that, for high enough tradecosts, integration is likely to cause agglomeration in sectors characterised by increas-ing returns to scale and strong inter-firm linkages (spillovers, market effects and so on.See Krugman 1991; Ottaviano and Puga 1998; Fujita et al. 1999). This may explainwhy growth is concentrated in certain areas, while others are left behind: a poor regionsurrounded by rich regions has a bigger probability of achieving growth than a poorregion surrounded by poor ones.2 To acknowledge the fact that regions are naturallyopen to interactions among one another calls for an explicit treatment of spatial effectsin regional convergence studies. To this aim, several works, using spatial economet-rics tools, analyse the regional convergence process in Europe taking into account thepossible presence of spatial dependence and/or of spatial heterogeneity. Ertur et al.(2006) prove both the presence of strongly significant spatial spillover effects amongEuropean (EU-15) regions and of significant differences in the convergence processbetween North and South regions over 1980–1995. Ramajo et al. (2008) find strongsupport for the existence of two different spatial regimes over 1981–1996 concerningcohesion and non-cohesion regions, with convergence being stronger for regions ofthe cohesion group than for non-cohesion ones.3 Ertur and Koch (2006), analysingregional income disparity in the EU-27 by exploratory spatial data analysis (ESDA),highlight a new north-west/east polarization pattern replacing the previous north/southdivide characterising the EU-15. Also Fischer and Stirböck (2006) test for the pres-ence of club convergence in EU-25 during 1995–2000 using a spatial econometricsframework and find evidence of a heterogeneous pattern in what they call the “pan-European” convergence process. Their conclusions are confirmed by Ezcurra et al.(2007) who analyse labour productivity in the EU-25 regions over 1991–2003 usinga nonparametric approach. This framework allows to overcome some of the meth-odological limitations of the standard parametric approach used, among the others,by Fischer and Stirböck (2006). Cross-section analysis, in fact, raises a number ofeconometric problems (Quah 1993b, 1996b,c, 1997). Moreover, it does not show

1 See Islam (2003) for an overview.2 In other terms “location matters”. See Quah (1996b, p. 952).3 Cohesion countries are identified as having national income below 90% of the EU average. At the time,they included Greece, Ireland, Portugal and Spain. Following the latest accessions all new members, plusGreece and Portugal, are Cohesion countries, Ireland having become ineligible in 2001 and Spain beingcurrently on a transitional regime.

123

Income dynamics in an enlarged Europe 665

intra-distribution mobility or the shape of the distribution in the long run, since itaccounts for a single “representative” economy model (Magrini 2004; Durlauf et al.2005).

Another relevant feature of regional growth dynamics in Europe is the signifi-cant “network effect” that stimulated growth in the EU-15 capital regions during the1980s and early 1990s. Buoyant growth in large urban areas, coupled with the oppo-site dynamics in old industrialised and peripheral lagging regions led to increasingregional disparities within countries (Meliciani 2005; Rodrìguez-Pose 1998). Somestudy suggests that this may concern newcomer members as well. In these coun-tries regional disparities rose significantly between 1990 and 2001 mainly because ofthe strong dynamism of the major urban centres, fostered by the presence of skilledlabour force, relatively developed infrastructure, advanced services and a markedincrease in foreign investment (Ezcurra et al. 2007; Resmini 2000). Frenken andHoekman (2006) find that network cities (that are mostly capital regions), operatingin global trade networks and relatively independent from their hinterland, convergemore quickly than other areas in the EU-25. They also test if the formerly dominantpolitical geography of the nation-state still affects growth opportunities of the regionswithin a country. This is done by introducing capital cites, border regions and otherregions in their cross-section regressions. They find that the core-periphery patternat a national level is still present. An interesting question is whether the more Euro-pean countries become integrated the more large urban centres become important asthey enter in a super-national network of administrative centres so that the impulseof the integration process to capital regions’ growth could well increase in the nearfuture.

The present article addresses the issue of regional per capita income dynamics in theenlarged Europe (EU-27). Our first aim is to test more accurately the results, respec-tively, of Fischer and Stirböck (2006) and of Ezcurra and Rapún (2007), both sug-gesting that the traditional north/south and/or cohesion/non-cohesion countries divideshould be replaced by an alternative north-west/east spatial pattern. In this respect,for the first time in the literature, we use the same data set to carry out both a non-parametric and a parametric analysis of convergence in the EU-27. In the parametricapproach spatial spillover effects are evaluated by means of spatial econometric tools.We test for the presence of a twofold old/new member spatial regime and also investi-gate spatial heterogeneity in the EU-15 group after the enlargement. In particular wetest for the presence of a Core/Objective 1 regime in the EU-15 to see whether thecohesion/non-cohesion divide, documented in the literature cited above, still survivesin old members countries.

Another aim of this paper is to deepen the analysis on the role of capital regions innewcomers. To do this, we use both nonparametric and parametric tools. In particularwe study capital regions in the context of a cross-section analysis where we test for thepresence of spatial dependence and spatial heterogeneity. This allows us to evaluatemore accurately the relative importance of the growth enhancing effect of capital citiesin new entrants and in old members.

The article considers three groups of regions over the period 1995–2004: thosebelonging to the enlarged Europe (EU-27), to old member states (EU-15) and to a

123


group named the European Core (formed by EU-15 without Objective 1 areas).4 Theanalysis relies on Eurostat data and chooses a territorial breakdown predominantly atthe NUTS 2 level.5

Section 2 analyses per capita income absolute convergence in the three groups ofregions by adopting a nonparametric (i.e. distributional) approach, based on both tran-sition matrices and stochastic kernels. The results arising from the analysis reportedin Sect. 2 are tested in Sect. 3 also by means of a parametric (i.e. regression) approachto conditional convergence that takes into account both spatial heterogeneity and spa-tial dependence by means of spatial econometrics tools. The conclusions in Sect. 4highlight some implications of the results for regional development and for cohesionpolicies.

2 The growth path of European regions in the enlarged EU: a nonparametricapproach analysis

Magrini (2004) argues that the “distributional approach to convergence” appears gen-erally more informative than convergence empirics based on the regression approach.This methodology, suggested by Quah (1996a,b,c, 1997), describes the law of motionof the cross-sectional distribution of regional per capita income, allowing to study theshape and the dynamics of regional per capita income for a given sample over a periodof time.

2.1 Data

This study uses data obtained from the Regional statistics database released by Euro-stat (the Statistical Office of the European Communities) with reference to the period1995–2004. The main sample includes 184 NUTS (Nomenclature of Territorial Unitfor Statistics) regions at a different aggregation level (NUTS-level 0, 1, and 2) becausefor some countries the level mostly used, i.e. NUTS-2, does not allow to distinguishhomogeneous regional areas possibly giving rise to a form of the modifiable areal unitproblem (MAUP) that can affect spatial analysis since, due to a scale and/or a zoneeffect, results can change under different boundary definitions.

The 184 regions of the sample belong to 27 European countries (EU-27): Austria(9 regions), Belgium (3 regions), Bulgaria (6 regions), Cyprus (1 region), CzechRepublic (8 regions), Denmark (1 region), Estonia (1 region), Finland (5 regions),France (22 regions), Germany (16 regions), Greece (4 regions), Hungary (7 regions),Ireland (2 regions), Italy (21 regions), Latvia (1 region), Lithuania (1 region), Luxem-bourg (1 region), Malta (1 region), The Netherlands (4 regions), Poland (16 regions),Portugal (5 regions), Romania (8 regions), the Slovak Republic (4 regions), Slove-nia (1 region) Spain (16 regions), Sweden (8 regions) and the United Kingdom(12 regions). The sample includes all the regions of the 27 members of the EU save

4 For EU-27 we consider 184 regions, for EU15 129 regions and for the Core 95 regions. The breakdownis in “Appendix”.5 Save in some cases. For further details, see Sect. 2.1.

123


Ciudad Autónoma de Ceuta, Ciudad Autónoma de Melilla and Canarias in Spain,French overseas departments in France, Regiao Autonoma dos Acores and RegiaoAutonoma de Madeira in Portugal. As in Ramajo et al. (2008), these latter regions havebeen excluded due to their geographical remoteness that could bias the subsequent spa-tial econometrics analysis. Per capita GDP figures are measured in Purchasing PowerStandards (PPS).

2.2 The transition probability matrix

A first idea of regional per capita income dynamics in a group of regions is gained bylooking at the transition probability matrix. This tool shows the probability of a regionfalling in a particular income class in the base-year to move to another class withinthe final year of observations. The higher the number of regions moving towards thecentral column of the matrix, the higher is per capita income absolute convergence. Onthe contrary, the higher the number of regions remaining on the matrix’s main diago-nal, the higher is per capita income persistence. The transition probability matrix forthe regions of the enlarged Europe is shown in Table 1. Each cell of the matrix reports,together with the transition probability, the number of regions belonging to each coun-try. Objective 1 regions are in bold characters;6 newcomer regions are underlined whilecapital regions are indicated by a star.

6 The following NUTS2 are considered as Objective 1 regions:

1) the whole territory of Bulgaria;2) the whole territory of the Czech Republic (save region cz01-Praha);3) de4-Brandenburg; de8-Mecklenburg-Vorpommern; ded-Sachsen; dee-Sachsen-Anhalt;

deg-Thüringen in Germany;4) the whole territory of Estonia;5) ie01-Border, Midlands and Western in Ireland;6) the whole territory of Greece;7) es11-Galicia; es12-Principado de Asturias; es41-Castilla y León; es42-Castilla-la Mancha;

es43-Extremadura; es52-Comunidad Valenciana; es61 Andalucia; es62 Murcia in Spain;8) itf3-Campania; itf4-Puglia; itf5-Basilicata; itf6-Calabria; itg1-Sicilia; itg2-Sardegna in Italy;9) the whole territory of Latvia;

10) the whole territory of Lithuania;11) the whole territory of Hungary;12) the whole territory of Malta;13) at11 Burgenland in Austria;14) the whole territory of Poland;15) the whole territory of Portugal (save region pt17-Lisboa);16) the whole territory of Romania;17) the whole territory of Slovenia;18) the whole territory of Slovakia (save region sk01-Bratislavsky kray);19) fi13-Itä-Suomi; fi1a-Pohjois-Suomi in Finland;20) ukl-Wales in Great Britain.

Sometimes Objective 1 areas cover territories that are smaller than NUTS-2 regions. In these cases, thecorresponding region is (is not) included in the sample according to whether the majority of its territoryfalls (does not fall) under Objective 1. Thus, for instance, Wales is considered an Objective 1 region, whileNorra Mellansverige, in Sweden, is not. Regions phasing out until 2006 are not considered Objective 1areas.

123


Table 1 Regional per capita income dynamics over 1995–2004 in the EU-27

1995/2004 [0,00–0,46) [0,46–0,83) [0,83–1,06) [1,06–1,22) [1,22–∞)

[0,00–0,46) n. obs. (38) 74% 26%

(5) bg (l) bg*

(5) hu (1) lt*

(1) lv* (4) pl

(9) pl (1) ro*

(7) ro (2) sk;

(1) sk (1) ee*

[0,46–0,83) n. obs. (35) 83% 14% 3%

(7) cz (1) hu* (1) gr*

(2) de (1) sl*

(2) hu (2) es

(1) mt* (1) ie

(2) pl*

(3) gr

(4) es

(4) it

(4) pl

[0,83–1,06) n. obs. (37) 13% 70% 13% 3%

(2) de (l) it (2) es (l) sk*

(2+1) it (1) cy* (3) uk

(l) de

(l) at

(2+1) es

(12) fr

(2+1) fi

(2+1) uk

(1) be

[1,06–1,22) n. obs. (37) 38% 46% 16%

(2) at (2) at (3) es*

(2) de (l) de (l) ie*

(7) fr (2) es (l) nl*

(1) it (l) fr (1) uk

(2) se (2) nl

(1) pt∗

(4) se

(4) uk

[1,22–∞) n. obs. (37) 3% 24% 73%

(1) de* (1) at (4) at*

(l) de (2) be*

(l) fr (5) de

(5) it (l) dk*

(l) se (l) fr*

(2) fi*

(7) it*

(l) nl

(l) lu*

(l) se*

(l) uk*

(1) cz*

Percentages in bold are transition probabilities. In parentheses, the number of regions belonging to each country. Objective 1 regions are in

bold characters; newcomers’ regions are underlined; capital regions are indicated by a star. For Malta and Romania, the base-year is 1998

123


Following Quah (1993a,b), we choose the five income classes in order to ensure thepresence in the base-year of a homogeneous number of regions in each class. How-ever, this kind of discretization, that is the partition of per capita income into classescorresponding to Markov chain states, as pointed out by Quah himself, has somedrawbacks. In particular, different discretizations give different results in distributiondynamics studies because the Markov property—namely that future occurrence duringany transition period t to t + 1 does not depend on anything else but its own startingvalue at t—can vanish. As a result, the estimated probabilities cannot be interpreted asMarkov transition probabilities. Nor do the methods suggested by Magrini (1999) andBulli (2001) aimed at providing objective criteria for defining optimal choice seem tosolve the problem. We therefore deepen the nonparametric analysis by means of thestochastic kernel analysis presented in the following section.

Table 1 shows that for the sample as a whole persistence prevails: the only incomeclass to register mobility of some entity is that of the regions falling in the next-to-the-highest class. For all the other classes of income the percentage of regions remainingby 2004 in the same income class where they were in 1995 is never below 70%.7

Furthermore,

• all regions in the lowest income class belong to newcomer countries. Among them,only about one quarter of the total (26%, equal to 10 regions) move towards thefollowing income class. Of these, 4 (out of 10) are regions where capitals arelocated;8

• as already mentioned, mobility prevails only for the next-to-richest income class(the fourth row in Table 1) where no Objective 1 region is present. For the groupof the richest regions (fifth row), where obviously there is no Objective 1 regionas well, persistence is instead very high;

• almost all capital regions tend to move towards, or to persist in, the last columnof the Table. Without counting the ten capital regions that persist in the highestincome class throughout the period (all belong to the older members save Praha),more than two thirds of the remaining ones (and most of those belonging to newmembers) move.

Considering a second matrix (not shown in the paper)9 with the same income classesbut covering sub-period 1995–1999 the elements in the main diagonal are consistentlyhigher. Starting from the lowest income class we have, respectively, 87, 94, 87, 89 and84% for the highest class. This means that the moves reported in Table 1 were mainlyconcentrated after the year 2000. For the lowest income class (that includes only East-ern country regions) this is possibly due to the tightening of association links betweenprospective members and the EU. In December 1999, Eastern future members eachsigned an Association Agreement with the EU, moving from looser pre-existing Part-nerships to closer coordination with the Commission. This allowed them to accede

7 With respect to 1989–1996, Cosci and Sabato (2007) find roughly similar results, even if on a differentsample. Therefore results are strictly not comparable.8 All new entrants regions are Objective 1 save three small capital areas (Berlin, Praha and Bratislavia).9 The matrix results are available upon request from the authors.

123


to specific funds directed at financing convergence processes (alongside with Phare,ISPA and Sapard).

2.3 The stochastic kernel analysis

The results obtained through the transition probability matrix are analysed by meansof another nonparametric tool used to study convergence, i.e. the stochastic kerneldensity analysis. This kind of nonparametric analysis of convergence was developedby Quah (1996a,b) and represents the evolution over time (from time t to time t + k)of per capita income distribution. It is a sort of transition probability matrix with con-tinuous classes of income.10 The stochastic kernel corresponds to the estimate of theconditional distribution of per capita income yt+k at time t + k given its value at timet(yt ); this can be obtained by the formula:

f (yt+k/yt ) = f (yt , yt+k)

f (t)(1)

The marginal distribution of yt can be calculated via a numerical integration ofthe joint distribution f (yt , yt+k) with respect to yt+k whereas f (yt , yt+k) can beestimated by the following bivariate kernel density estimator:

f (yt , yt+k) = 1

N

N∑

i=1

1

h1

1

h2K

(yt − yti

h1

)K

(yt+k − yt+ki

h2

)(2)

K (·) is the kernel function that must integrate to 1, h1 and h2 are the so-called smooth-ing parameters or bandwidths. We use Gaussian kernel functions and the normal opti-mal smoothing parameters suggested by Silverman (1986).

The stochastic kernel is represented using a three-dimensional graph or by the“contour plot”, i.e. a graph in which contours of the “mountain” are drawn such aslevel curves in an orographic map. In other words the three-dimensional curve hasbeen cut at different heights (probability values) to allow an easier interpretation ofthe graph, especially with reference to the task of capturing multimodality. In fact,the presence of disjoint borders is a clear sign of more than one mode in the distribu-tion. With reference to the analysis of income convergence, stochastic kernel densityshows persistence of income if the bulk of the distribution lies along the main diagonal,convergence if the distribution appears to collapse around a single value at time t + k.

The evolution of per capita income dynamics in the EU-27 (i.e. in all the 184regions in the sample) is contrasted with that in the pre-enlargement EU-15 members(i.e. a sub-sample of 129 regions obtained by eliminating newcomer regions). Thetwo stochastic kernels (see Figs. 1, 2) are both significantly distributed along the maindiagonal, indicating strong persistence.11 Lower income regions are situated at a lowerlevel in the EU-27 sample than in the EU-15. Moreover, lower income regions in the

10 Durlauf and Quah (1999) give a detailed definition and a description of some of the main properties ofstochastic kernels in the study of distribution dynamics.11 The stochastic kernels are calculated by means of a matlab routine developed by Magrini.

123


Fig. 1 Stochastic kernel densities of regional per capita income in EU-27 (184 regions), transition1995–2004, surface (a) and contour plot (b)

Fig. 2 Stochastic kernel densities of regional per capita income in EU-15 (129 regions), transition1995–2004, surface (a) and contour plot (b)

Fig. 3 Stochastic kernel densities of regional per capita income in the Core (98 regions), transition1995–2004, surface (a) and contour plot (b)

EU-27 sample show a peak near the origin that is absent in the EU-15 sub-sample(compare Figs. 1b, 2b). The peak is due to poor regions tending to become relativelypoorer at the end of the period.

Figure 3 shows per capita income dynamics in a more homogeneous group ofregions, i.e. in the so-called Core Europe (a sub-sample of 95 regions obtained byeliminating Objective 1 areas).12 This kernel presents three peaks and higher mobility

12 The full list of the regions of each sample is reported in “Appendix”. There is some overlapping betweenthe last two sub-samples given that all new members’ regions are Objective 1 save three small capital areas(Berlin, Praha and Bratislavia).

123


than the other two kernels, also if it does not show a clear absolute convergenceprocess.

As a first conclusion it appears evident that mobility essentially characterises regionsin the income classes around or above the average, given that in all samples mobilityfor these classes—in both directions—is higher than for other classes. Newcomers,as well as other Objective 1 areas, are scarcely mobile. Moreover, newcomers gen-erally show a tendency to polarize towards the lowest level of per capita income.These results obviously rule out the possibility of absolute convergence. However,they leave open the question if, instead of absolute convergence, there is an ongoingprocess of “conditional convergence” in Europe and/or groups of regions are con-verging towards different steady states because they belong to different “convergenceclubs”. The regression analysis reported in Sect. 3 will allow us to test if the evolutionof regional per capita income described by the kernel analysis is the result of hiddenprocesses of “conditional” or “club convergence”.

2.4 The role of capital regions

Table 1 shows that regions where capitals are located tend to move towards higherincome classes or to persist in the highest one. Urban geographers argue that qual-itative differences between cities have to be taken into account in order to under-stand the convergence or divergence between regional income levels. Frenken andHoekman (2006) suggest that European policymaking should take into account theleading role of network cities and the problems of other cities and rural areas. Net-work cities are expected to profit from policies enhancing science and technology(which is concentrated in network cities) as well as investments in transnational infra-structure networks (which connect network cities). In contrast, rural areas profit, ifall, from income subsidies for farmers and structural funds. Consequently, cities thatare not part of the global urban network and neither benefit from income subsidiesnor structural funds, are ‘stuck in the middle’, and deserve special attention, be itfrom national or European governments. In particular, one expects different rates ofgrowth for cities that are central nodes in global networks of trade, finance and knowl-edge (“network cities”) compared to cities that serve only their respective regionalhinterlands (“christallerian cities”).13 On the basis of these considerations, after theenlargement, Eastern Europe capital cities should have benefited from a significantnetwork effect and this might justify the relevant increase of within-country regionaldisparity which characterised new member countries during the last 10 years. Duringthe 1980s and early 1990s convergence patterns of Cohesion countries were char-acterised by income catching-up on the EU average but also by increasing regionaldisparity within each country (Dall’erba and Hewings 2003). Between 1995 and 2004disparities grew for newcomer countries as well while they fell or remained constant

13 See Glaeser et al. (1992, 1995) and Van Oort (2004), who stress the growth-enhancing effects of agglom-eration economies in cities and Hohenberg and Lees (1996) and Castells (1996) who stress that networkcities have a higher growth potential than ‘christallerian’ cities.

123


Per capita GDP 1995 (nat. adj.)

0 50 100 150 200 250

160

140

120

100

80

60

40

20

0

Industrial restructuring

Intermediate

Capital andUrban Regions

Peripheral

Per

cap

ita

GD

P g

row

th 1

995-

2004

(n

at a

dj.)

Fig. 4 Regional groups according to nationally weighted growth rates, 1995–2004

for most of the older members (the only exceptions being Ireland, Greece, Swedenand the UK).

Some first indication may be drawn by considering Fig. 4 where the averagegrowth rate of European regions during 1995–2004 is plotted against GDP level in thebase-year 1995. For both indicators regional GDP is weighted against the nationalaverage, in order to wipe out the national component of growth from regional perfor-mance. Following Rodrìguez-Pose (1998), regional growth rates (Yw) are definedas follows: Yw = (Yr/YN )YE where Yr is the average growth rate of per-capitaregional income; YN is the national average growth rate and YE is the average growthrate of European per-capita income. GDP is always calculated in purchasing powerparities. According to the prevailing literature Fig. 4 identifies four different groups,each selected on the basis of their performance patterns and sharing common socio-economic features.14 Thus, Fig. 4 shows the performance of the following groups:capitals regions, regions affected by industrial decline and restructuring, peripheralregions and intermediate regions. Capital regions are almost entirely concentrated inthe north-west quadrant. Furthermore, Table 2 shows that among this group15 newmember capital regions record growth rates significantly higher than those of oldmember regions. In other terms, a strong “capital effect” appears to be at work innewcomer countries, while it appears to have somewhat weakened over time in oldermembers with respect to previous performance. Rodrìguez-Pose (1998) shows thatduring the 1980s EU-12 capital regions recorded buoyant above-the-average growth

14 See the subdivisions identified by Rodrìguez-Pose (1998) for EU-12 regions and by Chapman (2008)for the remaining new entrants’ ones.15 Only 19 capitals (plus two urban areas in Germany, according to Rodríguez-Pose’s classification,1998) figure in Table 2. This occurs because adjusted GDP, being defined as regional performance againstnational data, leaves out one-region countries (in the 1995–2004 data-set: the three Baltic countries, Cyprus,Denmark, Luxembourg, Malta and Slovenia).

123


Table 2 Nationally adjusted growth rates, 1995–2004 average, and nationally adjusted GDP per inhabitantin 1995: capital and Urban Regions

NUTS code NUTS region Adjusted growth rate (EU27=100) GDP in 1995

CZ01 Praha 146,8 169,9GR3 Attiki 146,2 70,6RO32 Bucuresti-Ilfov 138,9 164,0ITE4 Lazio 133,8 112,2PL12 Mazowieckie 131,4 127,4BG41 Yugozapaden 126,7 135,5SE01 Stockholm 116,8 129,0HU1 Közép-Magyarország 116,0 143,8UKI London 111,2 145,2SK01 Bratislavsky kraj 110,2 213,3DE5 Bremen 106,7 131,7FR1 Île de France 101,2 153,9FI18 Etelä-Suomi 101,1 114,4DE6 Hamburg 101,0 167,2PT17 Lisboa 100,9 140,3NL3 West-Nederland 99,9 109,4IE02 Southern and Eastern 99,8 109,9ES3 Comunidad de Madrid 99,6 130,7BE1 Région de Bruxelles-Capitale 94,8 202,6AT13 Wien 90,9 143,3DE3 Berlin 49,9 101,3

All calculations refer to Chapman (2008). For Romania, (Bucuresti-Ilfov) the base-year is 1998. Italicsindicates new member regions

rates, in a substantially similar way to that occurring to new member states during1995–2004.

3 The growth path of European regions in the enlarged EU: a Barro-regressionanalysis introducing space

3.1 Beta-convergence analysis with spatial effects

The nonparametric analysis of Sect. 2 is now integrated by means of cross-section,or β-convergence, analysis widely exploited in the literature notwithstanding somedrawbacks that will be dealt with later on.

Galor (1996) distinguishes three kinds of long run per capita income convergence:absolute convergence, conditional convergence and club convergence. While abso-lute convergence16 models consider countries characterized by the same structure ofpreferences and technology, according to conditional convergence theories (which arebased on endogenous growth models)17 convergence in income levels occurs onlyif countries’ structural characteristics (such as preferences, technologies, populationgrowth, government policies, etc.) are identical. Finally, club convergence arises not

16 See Ramsey (1928); Solow (1956); Cass (1965) and Koopmans and Tjalling (1965).17 See Lucas (1988) and Romer (1986).

123


only if the structural characteristics are identical but also if initial economic conditionsare similar. The majority of empirical works18 on absolute and conditional conver-gence adopt a cross-section approach and estimate conditional convergence on thebasis of the following model:

log yit − log yi0

t= α − β log yi0 + γ j x j,i0 + εi for j = 1, . . . , k, (3)

where yi0 and yit represent per capita income of region i respectively at time 0(the initial year) and at time t (the final year), x j,i0 are k control variables thatkeep the steady state of each economy constant and γ j are k parameters to be esti-mated with α and β. A negative value for β may indicate convergence and hasoften been interpreted as such, even if it could be simply due to a general phenom-enon of reversion to the mean. In this case interpreting a negative value for β asa sign of convergence leads to “Galton’s fallacy” (Quah 1993b). In second place,the cross-section approach does not allow to identify the dynamics of income dis-tribution; this is the reason why this paper adopts both nonparametric and para-metric methods to study convergence. A third limit of this kind of approach (thisdrawback however concerns nonparametric methods as well) is that it completelyneglects the role played by space and geographical location in regional convergenceprocesses. In other terms, this approach does not account for the role of spatialinteractions among regions and countries (geographical spillovers) or for that ofproximity in enhancing spatial interactions. These elements have come to be con-sidered extremely important de facto in shaping regions’ economic activity (see Quah1996b).

However, in recent years, an emerging new “spatial thinking” perspective is com-pelling a growing numbers of scholars to analyse growth and convergence by means ofspatial econometrics tools that take into account both spatial heterogeneity and spatialdependence.

Abreu et al. (2005) define spatial heterogeneity as a concept of absolute location,depending on a subject being located at a certain position in space. Spatial heteroge-neity therefore takes place when parameters in growth models differ across regionsor countries conditional on their geographical position and growth processes differaccording to location. Therefore the idea of spatial heterogeneity is strictly linked tothe hypothesis of club convergence determined by geographical location. A numberof studies find evidence of club convergence in Europe through the detection of spa-tial regimes (see, among others, Dall’erba and Le Gallo 2008; Ramajo et al. 2008;Dall’erba et al. 2008).

Spatial dependence is a relative location concept, i.e. it changes if a country or aregion is located more or less close to another one. In particular there is spatial depen-dence if the values of some variables of a certain country or region are affected by thevalues of the same variables in other countries or regions at a certain location (e.g.neighbour behaviour influences local behaviour). Spatial dependence leads to spatial

18 Among others it is worth mentioning Baumol (1986); Barro and Sala-i-Martin (1991, 1992) and Mankiwet al. (1992).

123


autocorrelation (Anselin 1988), which is the correlation among values of a single var-iable strictly due to their relatively close geographical positions. This departs fromthe traditional assumption of statistics concerning the independence of observationsand therefore violates the main hypothesis of OLS estimation, i.e. the independenceof the error terms. Possible model misspecification due to the non-inclusion of spatialautocorrelation is considered in several works that study the path of European regionalconvergence (see among others Baumont et al. 2003; Ertur et al. 2006; Dall’erba and LeGallo 2008). These studies, incorporating spatial autocorrelation into β-convergencemodels, obtain more reliable estimates since, in this case, the hypothesis of indepen-dence of error terms is not violated while measuring spatial spillover effects betweenregions.

In our regression analysis we take into account both spatial dependence and spatialheterogeneity19 trying to detect the presence of spatial regimes in Europe over theperiod 1995–2004 corresponding to different convergence clubs. These are selectedon the basis of an a priori criterion dividing regions with reference to political, eco-nomic and historical reasons. In particular, we first test for the presence of two differentspatial regimes between old member and new member regions (like in Fischer andStirböck 2006) in the enlarged Europe (EU-27). We then perform the same type ofexercise in order to verify if in EU-15 we still find evidence of the existence of twoseparate spatial regimes between the group of Core regions and Objective 1 regions.20

A brief presentation of some results of exploratory spatial analysis have been used tocorroborate this choice.

3.2 Data and spatial weight matrix

The analysis uses the database described in Sect. 2. In order to mitigate the effects ofpossible outliers per capita GDP is expressed in logarithms.

We test for conditional convergence using three conditioning variables.21 In partic-ular we employ a dummy variable for capital and urban regions (as suggested by theresults of Sect. 2.3), the share of employment in agriculture (as a proxy for structuralcharacteristics) and the share of population with a university degree (as a proxy forthe quality of human capital).22

In order to obtain information concerning the spatial configuration of regional datawe adopt a distance-based contiguity approach. We assign a weight to the spatial

19 As highlighted by Anselin (2001) spatial heterogeneity often takes place together with spatial autocor-relation; in a cross-section they are “observationally equivalent”.20 A description of these subgroups of regions is in Sect. 2. For the complete list see the “Appendix”.21 The European regional data available for all the EU-27 regions in the time period taken into considerationunfortunately are not enough to represent all the economic, political and social characteristics of Europeanregions at the beginning of the period.22 The share of employment in agriculture is calculated as the ratio between employment in agricultureand total employment in 1995; the share of university degree holders is given by the ratio between thepopulation with a university degree and total population aged 15 and over (for lack of data we use datarelative to 1999).

123


effects between regions based on geographical distance due to its unambiguous exo-geneity (see Anselin and Bera 1998; Anselin 1996). Due to the presence of islands wedo not use simple contiguity matrices to exclude the presence in the weight matrix Wof rows and columns with only zeros.

In particular, we calculate Great circle distances between the couples of centroidsof the 184 regions. Geometrically the centroid is the centre of an area or of a polygon.Regional areas usually are irregularly shaped polygons; thus, the centroid is derivedmathematically and is weighted to approximate a sort of “centre of gravity”. In Geo-graphical Information System (GIS) these discrete locations expressed in geographicalcoordinates are often used to index or reference the polygon (hence the regions) withinwhich they are located. We assume that centroids are regions’ “centres of gravity” froman economic point of view.

To exploit the information on distances we use a n × n symmetric spatial weightmatrix W , where element wi j specifies the intensity of the effects between two regionsi and j ; i.e. how region i is spatially connected to region j (Anselin and Bera1998). Furthermore, the spatial weight matrix is usually row standardized so thateach row’s elements sum up to one in order to normalize the effects of all regionson their neighbours. Each element of the row standardized weight matrix thereforemeasures the regional share of the global spatial effects on a particular observa-tional unit (Niebuhr 2001). The spatial weighting matrix is based on an “inversedistance” function that downweights observations that are geographically moredistant (Badinger et al. 2004). We use the inverse of the squared distance toreflect a “gravity model” function. The elements of the matrix are therefore thefollowing:

⎧⎪⎨

⎪⎩

wi j = 0 if i = j

wi j = 1/

d2i j if di j ≤ Me and wi j = wi j

/ ∑j wi j

wi j = 0 if di j > Me

(4)

where Me is the median of the great circle distance distribution and di j is the distancebetween regions i and j . We obtain similar results using other weighting matricescalculated substituting the median (the critical cut-off parameter) respectively withthe first and the third quartile of the geographical distance distribution.23

3.3 Exploratory spatial data analysis

Figure 5 gives an idea of the spatial distribution of regional per capita GDP (expressedin logarithms) in the EU-27 in base-year 1995. As expected, it shows poorer regions

23 In addition, the robustness of results is checked using spatial weight matrices based on the k-nearestneighbours with k = 10, 15, 20, 25 neighbours (see Ertur et al. 2006). A k-nearest neighbours weight matrixis computed from the distance between the regions’ centroids and implies that each region is connected tothe same number k of neighbours. Complete results on robustness analysis are available from the authorsupon request.

123


Fig. 5 Map of regional per capita income distribution in EU-27 (year 1995)

Table 3 Moran’s I statistics for per capita GDP in logarithm in 1995 and 2004 for EU-27

Year Moran’s I Mean Standard deviation Z value p value

1995 0.6141048 −0.005 0.030 20.900 0.000

2004 0.5774348 −0.005 0.030 19.692 0.000

mostly clustered at the East (future newcomer regions) and richer areas grouped in theCore zone.

In order to detect and represent the main spatial features in European regional percapita income we use the global spatial autocorrelation Moran’s I statistic (Cliff andOrd 1981) both for the initial and the final year of our observations (respectively, 1995and 2004).

Moran’s I statistics, that represents one of the main tools of ESDA, can be writtenfor each year t as follows:

It = n

S0

z′t W zt

z′t zt

(5)

where n represents the number of observations (regions), zt is the logarithm of percapita GDP at time t calculated in deviation from the sample mean, W is the row stan-dardized spatial weight matrix described in the previous section and S0 is a scalingfactor given by the sum of all the elements of W .

Results in Table 3 show a positive and statistically significant spatial autocorrela-tion for 1995 and 2004 (both with zero p values), even if Moran’s I is slightly lowerin 2004 (I = 0.5774) than in 1995 (I = 0.6141).

123


Fig. 6 Moran scatterplot of per capita GDP in logarithms in 1995

The presence of spatial autocorrelation is better seen by means of the Moranscatterplot24 and Moran scatterplot map.25 These are shown for the EU-27 regionsin 1995 respectively in Figs. 6 and 7. Results for 2004 are very similar.

What appears at first sight from Fig. 7 is the presence of a fairly neat divisionbetween a large number of old member regions and new entrant ones. In a certainway this recalls the north-west/east regime described by Ertur and Koch (2006) orby Ezcurra and Rapún (2007). This adds up to the a priori choice of testing for thepresence of spatial regimes between old member regions and newcomers. Alongside,however, deeper examination of Fig. 7 suggests the presence of a spatial “break”between the club of the Core regions and that of the Objective 1 regions. Actually, thevery definition of Objective 1 regions implies the presence of a break, but it should beinvestigated if this break has some geographical connotation as well.

24 The Moran scatterplot represents the regression of a spatially lagged variable (multiplied for the spatialweight matrix W ) on the original variable’s standardized values. The slope of the regression line correspondsto Moran’s I value. Spatially lagged variables are a weighted sum of neighbours’ values. Each quadrant ofthe scatterplot contains information on the value of the observation relatively to its neighbour. In our casewe have:

– first quadrant (upper right): regions with high per capita income bordered by high income regions (HH,positive spatial autocorrelation);

– third quadrant (lower left): poor regions surrounded by regions with low per capita income (LL, positivespatial autocorrelation);

– second quadrant (lower right): rich regions with poor neighbours (HL, negative spatial autocorrelation);– fourth quadrant (upper left): poor regions surrounded by rich ones (LH, negative spatial autocorrelation).

In the map of Fig. 7 the results of the Moran scatterplot are mapped by assigning different colours to thevalues of the four quadrants.25 See Anselin et al. (1996).

123


Fig. 7 Moran scatterplot map of per capita GDP in logarithms in 1995

3.4 Econometric specification

As a first step we estimate the conditional β-convergence (described in Eq. 3 above)using standard OLS, without incorporating spatial effects but performing some testsin order to detect the actual existence of spatial heterogeneity and/or of spatial depen-dence. In other terms, we have two main goals: (1) to test for the presence of spatialregimes and (2) to check for the presence of spatial autocorrelation. If this were tooccur, then the most suitable way of incorporating it into the model must be identified,i.e. whether by the substantive form (spatial lag) or the nuisance form (spatial error).Once the presence of regimes is detected and the decision concerning the form ofspatial dependence is taken, the chosen model will be a regime switching one withcoefficients allowed to differ across regimes.

The spatial error model (SEM) and the spatial lag model (SLM) described byAnselin (1988) are both estimated by maximum likelihood estimation (MLE). In thespatial error model (SEM) spatial dependence is limited to the error term. In our model(without taking into account the presence of regimes) we have:

log yit − log yi0

t= α − β log yi0 + γ j x j,i0 + εi for j = 1, . . . , k

with εi = (λW )i + ui (6)

where, in addition to the model in (1) we have a spatially autoregressive error termwhere λ is the spatial autocorrelation coefficient, (λW )i is the spatially weighted sumof errors for region i while εi and ui are independent, normally distributed error terms.The SEM, therefore, allows for spatial autocorrelation in errors across regions.

The spatial lag model (SLM) is more appropriate when there is a direct geograph-ical spillover effect and income growth rate in a region is directly influenced by thegrowth rate of neighbouring regions. The specification of the SLM in β-convergence

123


analysis with a spatially lagged dependent variable is:

log yit − log yi0

t= α + ρ

(W

log yt − log y0

t

)

i− β log yi0 + γ j x j,i0 + εi

for j = 1, . . . , k (7)

where ρ is the spatial autocorrelation coefficient and W the spatial weight matrix usedto calculate the income growth rate spatial lag.

The first useful test to detect spatial autocorrelation is the application of Moran’s Itest on regression residuals (Cliff and Ord 1981). To discriminate between the twoforms of spatial autocorrelation we use four Lagrange Multiplier tests: LMERR,LMLAG, and their robust form, R-LMERR and R-LMLAG (Anselin et al. 1996).Anselin and Florax (1995) provide the following rule: the spatial lag model is cho-sen as a more suitable specification if LMLAG is more significant than LMERR andR-LMLAG is significant while R-LMERR is not. The opposite situation would implythe choice of the spatial error model. In this case, a classic “specific to general”approach is suggested to find the true data generating process. Even if this approachis not exempt from important drawbacks, Florax et al. (2003) show, via some MonteCarlo simulations, that this approach offers a better performance than Hendry (1979)“general to specific” one.

3.5 Estimation results and discussion

Table 4 shows the first results obtained by using the methodology just described. Inthis case we test for the presence of spatial regimes between the group of old memberregions (EU-15) and the group of newcomer regions.26

From OLS estimation in column 1 we have the first preliminary results that confirmconditional β-convergence in the sample with a negative and significant β coefficient(−0.019), a convergence speed of 2.08% and a half-life of about 36 years.27 Theconditioning variables coefficients estimates are all significant.

Examining the diagnostics relative to this model, the Jarque–Bera test rejects thenull hypothesis of normality of the errors28 whereas the Koenker–Basset test does not

26 All calculations in this section and in the preceding one are performed using Spacestat 1.91, the spatialeconometrics software elaborated by Anselin (1999) and its extension in Arcview GIS 3.2.27 The convergence speed measures the speed at which an economy reaches its own steady-state and iscalculated as b = −ln(1 + β)/T . The half-life, defined as τ = −ln(2)/ ln(1 + β), is the time necessaryfor an economy to reduce the gap between per capita income and its steady state value by one half.28 As already underlined by Fischer and Stirböck (2006), in large samples, Jarque and Bera (1987) testis very powerful. Therefore, the presence of even a small number of outliers can lead the test to identifysignificant departures from normality in the residuals. In order to understand if this is the case, we havetried to introduce some dummies to control for outliers. In this case the Jarque and Bera test does not rejectthe null hypothesis of normality of the errors both in the regression without regimes (p value=0.218) andin the regression with the two regimes (p value=0.141). This leads to a certain confidence on the reliabilityof the subsequent tests that depend on the normality assumption, such as the various Lagrange Multipliertests used to choose the suitable spatial model. Moreover, introducing the dummies to control for outlierswe obtain results analogous to those shown in Table 4. In particular, the Chow–Wald test rejects the nullhypothesis of equality of coefficients across the two groups. This result will be commented upon later.

123


Table 4 Estimation results for the spatial regimes old/newcomers, conditional β-convergence, OLS andML-ERR

There are N = 184 observations. p values are in parenthesesO L S ordinary least squares estimation, ML-ERR maximum likelihood estimation for spatial error model(see Eq. 9), L I K the value of the maximum likelihood function, L M E R R the Lagrange Multiplier test forresidual spatial autocorrelation and R-LMERR is its robust form, L M L AG the Lagrange Multiplier testfor spatially lagged dependent variable and R-LMLAG is its robust form, L R likelihood ratio, L M standsfor Lagrange Multiplier

reject homoskedasticity. Moran’s I test reject the null hypothesis of absence of spatialdependence in the residuals thus confirming spatial misspecification.

Before applying the most suitable model incorporating spatial effects we have per-formed a simple OLS switching regression (results are presented in columns 2 and 3of Table 4) of the following form:

log yit − log yi0

t= αA D A

i + αB DBi − (β A D A

i + βB DBi ) log yi0

+ (γ Aj D A

i + γ Bj DB

i )x j,i0 + εi

for j = 1, . . . , k

εi ∼ N

(0,

[σ 2

A IA 00 σ 2

B IB

])(8)

where D Ai and DB

i are dummy variables assuming value one if a region belongs togroup A (EU-15 regions) or to group B (newcomer regions) and zero otherwise.

123


Results displayed in columns 2 and 3 of Table 4 show preliminary evidence ofthe presence of heterogeneity in per capita income patterns between the two clubsof regions. In fact, the Chow–Wald test rejects the null hypothesis of equality ofcoefficients across the two groups. Even if also this model is affected by spatial mis-specification, as the OLS without regime switching, it is interesting to observe that β

coefficient estimates differ across the two groups (both are negative and significanteven if β is lower for newcomer regions). The dummy for capital and urban regionsis positive and significant for both groups and so is the share of the population witha university degree. The share of occupation in agriculture is positive and significantonly for the newcomer regions’ group.

The rule concerning the choice of the most suitable form of spatial correlationbased on the Lagrange multiplier tests points to the spatial error model as the mostappropriate one, instead of the spatial lag model. In fact, both LMERR and LMLAGare significant whereas only R-LMERR is significant.

The spatial error model with spatial regimes is the following:

log yit − log yi0

t= αA D A

i + αB DBi − (β A D A

i + βB DBi ) log yi0

+ (γ Aj D A

i + γ Bj DB

i )x j,i0 + εi

for j = 1, . . . , k

with εi = (λW )i + ui and ui ∼ N

(0,

[σ 2

A IA 00 σ 2

B IB

])(9)

The maximum likelihood estimates of this model (ML-ERR) are in the last two col-umns of Table 4. As expected, the SEM specification is characterized by a higher loglikelihood relative to OLS. A lower value of the Akaike Information Criterion (AIC)and of the Schwarz Information Criterion (SC) gives evidence of a better fit of thisspatial specification with respect to OLS.29 The coefficient λ of the spatial error termis 0.863 and is significant, confirming the presence of positive spatial autocorrelation.As for the OLS estimates β coefficient estimates are negative and significant for bothEU-15 and newcomer regions but now β is greater (in absolute value) for newcomerregions (β = −0.021) than for EU-15 regions (β = −0.015). Accordingly, the con-vergence speed is over 2% for new entrants (with a half-life of 33 years), whereas it isonly 1.59% for the EU-15 group (with a half-life of almost 47 years). It is interestingto notice that when we take into account spatial effects, the convergence rate estimateis augmented for regions belonging to the club of newcomer regions whereas it islowered for EU-15 regions. A possible interpretation of this result can be the fact thatgeographical spillovers contribute negatively to convergence of newcomer regions,which are mostly poor regions bordered by poor regions. On the other hand spatialspillovers have a positive, convergence enhancing effect, in the case of the richerregions belonging to EU-15 group that are predominantly rich regions surrounded byother rich neighbours.

29 We use AIC instead of R2 as a measure of model fit because in maximum likelihood estimation thecoefficient of determination is only a pseudo-measure.

123


The dummy for capital and urban regions is positive and significant for both groupsand the coefficient is again greater for newcomer regions (moreover, the differencebetween the two groups’ coefficients is bigger than in the OLS regressions reported incolumns 2 and 3). While the share of employment in agriculture is not significant forneither of the two clubs, the share of population with a university degree is positiveand significant only for new entrants.

Again, the Chow–Wald test rejects the null hypothesis of equality of coefficientsacross the two groups confirming the presence of the two different spatial regimes.Through the spatial Breusch–Pagan test the SEM specification reveals the presence ofremaining heterogeneity, probably due to the presence of more than two convergenceclubs. This suggests that it would be interesting to further test for the presence of otherconvergence clubs especially within the group of newcomers. In fact there is evidencethat foreign capital flows directed to these countries are mainly concentrated aroundmajor metropolitan areas and/or around the western regions closer to the EU-15 bor-ders (Ezcurra et al. 2007). On the contrary, the more eastern areas remain largelyoriented toward the former Soviet-bloc markets (Boeri et al. 2000).

Finally, the Likelihood Ratio test on spatial error dependence, the Lagrange Multi-plier test on spatial lag dependence and the Wald-test on the common factor hypothesisare all significant, possibly indicating that the SEM specification is preferable.

We also test for the possible presence of spatial regimes between Core and Objec-tive 1 regions in EU-15 (129 cases) over the same period. In this case the Chow–Waldstructural instability test does not reject the null hypothesis of equality of coefficientsacross the two groups (the test value is equal to 1.366 with a p value of 0.242), i.e.we do not find evidence of the presence of two separate spatial regimes. This resultsuggests that EU-15 is becoming progressively more homogeneous with respect tothe previous decade (Ertur et al. 2006; Ramajo et al. 2008) and that the old Objec-tive 1 group of regions does not represent a separate convergence club. This supportsour previous suggestion that further investigation testing for the presence of moreconvergence clubs within the group of the new member regions is needed.

Given that these results could be affected by the choice of the conditioning variables,we perform the same analysis excluding each variable at a time. We find the presenceof the spatial break in all the regressions.30 Moreover, we carry out an absolute conver-gence analysis by using the same sequence of regressions as in Table 4—OLS over theentire sample and OLS and SEM with regime switching. Again the results, reportedin Table 5, show the presence of the old/new members spatial regime. By looking atTable 5 we also notice that there is no absolute convergence in the group of newcom-ers, a result consistent with the conclusions arising from the kernel analysis reportedin Sect. 3. The Chow–Wald structural instability test rejects the null hypothesis ofequality of coefficients across the two groups both in the OLS switching regressions(columns 2 and 3) and in the SEM specification with regimes (columns 4 and 5).Also in this case the spatial Breusch–Pagan test reveals the presence of remainingheterogeneity, probably due to the presence of more than two convergence clubs.

30 Results available upon request.

123


Table 5 Estimation results for the spatial regimes old/newcomers, absolute β-convergence, OLS and ML-ERR

There are N = 184 observations. p values are in parenthesesOLS indicates ordinary least squares estimation, ML-ERR maximum likelihood estimation for spatial errormodel (see Eq. 9). LIK the value of the maximum likelihood function, LMERR the Lagrange Multiplier testfor residual spatial autocorrelation and R-LMERR is its robust form, LMLAG the Lagrange Multiplier testfor spatially lagged dependent variable and R-LMLAG is its robust form. LR likelihood ratio, LM LagrangeMultiplier

4 Conclusions

In the EU-27 sample we find evidence of the presence of a twofold old/new memberspatial regime. Furthermore we do not find evidence of the presence of two separatespatial regimes in the EU-15 for Core and Objective 1 regions. On the basis of thislatter result it seems that, consistently with what emerges from the nonparametricanalysis, Objective 1 regions are no more a separate “convergence club”. This is aninteresting result; it suggests that for the majority of old member regions integration isfinally taking place. We also find that, when spatial effects are taken into account, new-comer regions’ convergence speed is higher than that of the EU-15 group. Moreover,consistently with the results of previous studies, our analysis shows that the growth-enhancing effect of capital cities is particularly strong in the group of new entrants.Newcomer capitals are converging rather quickly towards the EU-15 network cities.Old member capitals, while still characterised by relatively high growth rates, growless than during the 1980s, and generally less than newcomer capitals. We find thatthis is true also when we take into account spatial effects. The enlarged Europe seemstherefore to be formed by a group of old member regions slowly becoming morehomogeneous and a newcomers group which still represents a separate “convergenceclub” but whose capital regions are rapidly integrating into the west.

123


These results have important implications for the future design of EU regionalpolicies. During the last forty years the convergence process in the EU-15 has beenheavily fostered by EU policies (even if resources were not always used efficiently).At present it cannot be expected that newcomers may follow a similar path withoutsimilar support on behalf of the EU. However, the EU is now called to promote con-vergence in a situation in which resources are far more scarce than in the past. Indeed,there is an increasing disproportion between the amount of disposable resources andthe amount of those needed by EU lagging regions. In the future, EU regional policiesmay well achieve their goals with less resources but, in this context, it is essential todeepen the knowledge of the special features of the different lagging regions. In fact,intervention should be appropriately tailored in order to face problems that may beradically different from one region to another even within a given country.

On the one hand, policies should consider the leading role of capital regions inthe new EU members. This means, for instance, promoting the development and thediffusion of infrastructures, of networks and of research centres, but also regulatingmigration flows, facing transport and housing problems, environmental loads and soon. At the same time, other regions, both agricultural and industrial, are lagging behind.However, they are far from being a homogeneous group. While the ones that are closerto the EU-15 borders attract foreign investment, eastern ones remain cut off from thespillovers coming from the west, and possibly still rely on traditional links with theirformer Soviet partners. These regions, which are both agricultural and industrial, areoften the “losers” of transition.

This raises one last, important, question. The results of our spatial analysis, coupledwith the fact that we do not find evidence of a Core/Objective 1 regime in the EU-15,are consistent with the presence of more than two separate groups among newcomerregions. This point is left for further investigation.

Appendix

See Table 6.

Table 6 Regions included in the sample (EU-27) and in the different groupings (EU-15 and the Core)

NUTS Code Country NUTS Region EU-27 EU-15 Core

AT11 Austria Burgenland • •AT12 Niederösterreich • • •AT13 Wien • • •AT21 Kärnten • • •AT22 Steiermark • • •AT31 Oberösterreich • • •AT32 Salzburg • • •AT33 Tirol • • •AT34 Vorarlberg • • •BE1 Belgium Région de Bruxelles-Capitale • • •

123


Table 6 continued


BE2 Vlaams Gewest • • •BE3 Région Wallonne • • •BG31 Bulgaria Severozapaden •BG32 Severen tsentralen •BG33 Severoiztochen •BG34 Yugoiztochen •BG41 Yugozapaden •BG42 Yuzhen tsentralen •CY Cyprus Cyprus •CZ01 Czech Republic Praha •CZ02 Strední Cechy •CZ03 Jihozápad •CZ04 Severozápad •CZ05 Severovýchod •CZ06 Jihovýchod •CZ07 Strední Morava •CZ08 Moravskoslezsko •DE1 Baden-Württemberg • • •DE2 Bayern • • •DE3 Berlin • • •DE4 Brandenburg • •DE5 Bremen • • •DE6 Hamburg • • •DE7 Hessen • • •DE8 Mecklenburg-Vorpommern • •DE9 Niedersachsen • • •DEA Nordrhein-Westfalen • • •DEB Rheinland-Pfalz • • •DEC Saarland • • •DED Sachsen • •DEE Sachsen-Anhalt • •DEF Schleswig-Holstein • • •DEG Thüringen • •DK Denmark Denmark • • •EE Estonia Estonia •ES11 Spain Galicia • •ES12 Principado de Asturias • •ES13 Cantabria • • •ES21 Pais Vasco • • •ES22 Comunidad Foral de Navarra • • •

123


Table 6 continued


ES23 La Rioja • • •ES24 Aragón • • •ES30 Comunidad de Madrid • • •ES41 Castilla y León • •ES42 Castilla-la Mancha • •ES43 Extremadura • •ES51 Cataluña • • •ES52 Comunidad Valenciana • •ES53 Illes Balears • • •ES61 Andalucia • •ES62 Región de Murcia • •FI13 Finland Itä-Suomi • •FI18 Etelä-Suomi • • •FI19 Länsi-Suomi • • •FI1A Pohjois-Suomi • •FI20 Åland • • •FR10 France Île de France • • •FR21 Champagne-Ardenne • • •FR22 Picardie • • •FR23 Haute-Normandie • • •FR24 Centre • • •FR25 Basse-Normandie • • •FR26 Bourgogne • • •FR30 Nord-Pas-de-Calais • • •FR41 Lorraine • • •FR42 Alsace • • •FR43 Franche-Comté • • •FR51 Pays de la Loire • • •FR52 Bretagne • • •FR53 Poitou-Charentes • • •FR61 Aquitaine • • •FR62 Midi-Pyrénées • • •FR63 Limousin • • •FR71 Rhône-Alpes • • •FR72 Auvergne • • •FR81 Languedoc-Roussillon • • •FR82 Provence-Alpes-Côte d’Azur • • •FR83 Corse • • •GR1 Greece Voreia Ellada • •GR2 Kentriki Ellada • •

123


Table 6 continued


GR3 Attiki • •GR4 Nisia Aigaiou, Kriti • •HU10 Hungary Közép-Magyarország •HU21 Közép-Dunántúl •HU22 Nyugat-Dunántúl •HU23 Dél-Dunántúl •HU31 Észak-Magyarország •HU32 Észak-Alföld •HU33 Dél-Alföld •IE01 Ireland Border, Midlands and Western • •IE02 Southern and Eastern • • •ITC1 Italy Piemonte • • •ITC2 Valle d’Aosta/Vallée d’Aoste • • •ITC3 Liguria • • •ITC4 Lombardia • • •ITD1 Provincia Autonoma Bolzano-Bozen • • •ITD2 Provincia Autonoma Trento • • •ITD3 Veneto • • •ITD4 Friuli-Venezia Giulia • • •ITD5 Emilia-Romagna • • •ITE1 Toscana • • •ITE2 Umbria • • •ITE3 Marche • • •ITE4 Lazio • • •ITF1 Abruzzo • • •ITF2 Molise • • •ITF3 Campania • •ITF4 Puglia • •ITF5 Basilicata • •ITF6 Calabria • •ITG1 Sicilia • •ITG2 Sardegna • •LT Lithuania Lithuania •LU Luxemburg Luxembourg (Grand-Duché) • • •LV Latvia Latvia •MT Malta Malta •NL1 The Netherlands Noord-Nederland • • •NL2 Oost-Nederland • • •NL3 West-Nederland • • •NL4 Zuid-Nederland • • •

123


Table 6 continued


PL11 Poland Lódzkie •PL12 Mazowieckie •PL21 Malopolskie •PL22 Slaskie •PL31 Lubelskie •PL32 Podkarpackie •PL33 Swietokrzyskie •PL34 Podlaskie •PL41 Wielkopolskie •PL42 Zachodniopomorskie •PL43 Lubuskie •PL51 Dolnoslaskie •PL52 Opolskie •PL61 Kujawsko-Pomorskie •PL62 Warminsko-Mazurskie •PL63 Pomorskie •PT11 Portugal Norte • •PT15 Algarve • •PT16 Centro (PT) • •PT17 Lisboa • • •PT18 Alentejo • •RO11 Romania Nord-Vest •RO12 Centru •RO21 Nord-Est •RO22 Sud-Est •RO31 Sud-Muntenia •RO32 Bucuresti-Ilfov •RO41 Sud-Vest Oltenia •RO42 Vest •SE11 Sweden Stockholm • • •SE12 Östra Mellansverige • • •SE21 Sydsverige • •SE22 Norra Mellansverige • • •SE23 Mellersta Norrland • • •SE31 Övre Norrland • •SE32 Småland med öarna • • •SE33 Västsverige • • •SI Slovenia Slovenia •SK01 Slovak Republic Bratislavský kraj •SK02 Západné Slovensko •

123


Table 6 continued


SK03 Stredné Slovensko •SK04 Východné Slovensko •UKC United Kingdom North East • • •UKD North-West • • •UKE Yorkshire and The Humber • • •UKF East Midlands • • •UKG West Midlands • • •UKH Eastern • • •UKI London • • •UKJ South-East • • •UKK South-West • • •UKL Wales • •UKM Scotland • • •UKN Northern Ireland • • •

References

Abreu M, de Groot HLF, Florax RJGM (2005) Space and growth: a survey of empirical evidence andmethods, Région et Développement 21:13–44

Anselin L (1988) Spatial econometrics: methods and models. Kluwer, DordrechtAnselin L, Florax RJGM (1995) Small sample properties of tests for spatial dependence in regression mod-

els: Some further results. In: Anselin L, Florax RJGM (eds) New directions in spatial econometrics.In: Methodology tools and applications. Springer, Berlin, pp 21–74

Anselin L (1996) The Moran scatterplot as an ESDA tool to assess local instability in spatial association.In: Fisher M, Scholten HJ, Unwin D (eds) Spatial analytical perspectives on GIS. Taylor & Francis,London

Anselin L, Bera A, Florax R, Yoon MJ (1996) Simple diagnostic tests for spatial dependence. Reg SciUrban Econ 26:77–104

Anselin L, Bera A (1998) Spatial dependence in linear regression models with an introduction to spatialeconometrics. In: Ullah A, Giles DEA (eds) Handbook of applied economic statistics. Marcel Dekker,New York, pp 237–289

Anselin L (1999) SpaceStat, a software package for the analysis of spatial data, Version 1.90. Ann Arbor,BioMedware

Anselin L (2001) Spatial econometrics. In: Baltagi B (ed) Companion to theoretical econometrics. BasilBlackwell, Oxford, pp 310–330

Azariadis C, Drazen A (1990) Threshold externalities in economic development. Q J Econ 105(2):501–526Badinger H, Müller W, Tondl G (2004) Regional convergence in the European Union, 1985–1999: a spatial

dynamic panel analysis. Reg Stud 38(3):241–253Barro RJ, Sala-i-Martin X (1991) Convergence across States and Regions. Brookings Pap Econ Act 1:

107–182Barro RJ, Sala-i-Martin X (1992) Convergence. J Polit Econ 100(2):223–251Baumol WJ (1986) Productivity growth, convergence, and welfare: What the long-run data show. Am Econ

Rev 76(5):1072–1085Baumont C, Ertur C, Le Gallo J (2003) Spatial convergence clubs and the European regional growth process,

1980–1995. In: Fingleton B (ed) European regional growth. Springer, Berlin, pp 131–158Boeri T, Brücker et al (2000) The impact of eastern enlargement on employment and labour markets

in the EU member states. Report for European Commission’s Employment and Social Affairs

123


Directorate, European Integration Consortium (DIW, CEPR, FIEF, IAS, IGIER), Berlin and Milan(Brussels: European Commission)

Bulli S (2001) Distribution dynamics and cross-country convergence: a new approach. Scot J Polit Econ48(2):226–243

Buse A (1973) Goodness-of-fit in generalized least squares estimation. Am Stat 27:106–108Cass D (1965) Optimum growth in an aggregative model of capital accumulation. Rev Econ Stud 32:

233–240Castells M (1996) The rise of the network society. Blackwell, CambridgeChapman SA (2008) Regional growth patterns in the enlarged EU: are new members conforming to old

development schemes. Economia, Impresa e Mercati Finanziari 6(2):7–33Chatterji M, Dewhurst JHL (1996) Convergence clubs and relative economic performance in Great Britain:

1977–1991. Reg Stud 30(1):31–40Cliff AD, Ord JK (1981) Spatial processes: models and applications. Pion, LondonCosci S, Sabato V (2007) Income and employment dynamics in Europe. Reg Stud 41(3):295–309Dall’erba S, Hewings GJD (2003) European regional development policies: the trade off between effi-

ciency–equity revisited. Discussion paper 03, Federal Reserve Bank of Chicago, USADall’erba S, Le Gallo G (2008) Regional convergence and the impact of European structural funds over

1989–1999: a spatial econometric analysis. Pap Reg Sci 87(2)Dall’erba S, Percoco M, Piras G (2008) The European regional growth process revisited. Spat Econ Anal

3(1):7–25Dewhurst Mutis-Gaitan JHL (1995) Varying speeds of regional GDP per capita convergence in the Euro-

pean Union, 1981–91. In: Armstrong HW, Vickerman RW (eds) Convergence and divergence amongEuropean regions. Pion, London, pp 22–39

Durlauf SN, Quah DT (1999) The new empirics of economic growth. In: Taylor JB, Woodford M (eds)Handbook of macroeconomics. North-Holland, Amsterdam, pp 231–304

Durlauf SN, Johnson PA, Temple JRW (2005) Growth econometrics. In: Aghion P, Durlauf SN (eds)Handbook of economic growth. North-Holland, Amsterdam, pp 555–677

Ertur C, Le Gallo J, Baumont C (2006) The European regional convergence process, 1980–1995: do spatialregimes and spatial dependence matter? Int Reg Sci Rev 29(1):3–34

Ertur C, Koch W (2006) Regional disparities in the European Union and the enlargement process: anexploratory spatial data analysis, 1995–2000. Ann Reg Sci 40:723–765

Ezcurra R, Pascual P, Rapún M (2007) The dynamics of regional disparities in Central and Eastern Europeduring transition. Eur Plan Stud 15:1397–1421

Ezcurra R, Rapún M (2007) Regional Dynamics and convergence profiles in the enlarged European Union:a non-parametric approach. Tijdschr Econ Soc Geogr 98(5):564–584

Fagerberg J, Verspagen B (1996) Heading for divergence? Regional growth in Europe reconsidered. J Com-mon Mark Stud 34(3):431–448

Fischer M, Stirböck C (2006) Pan-European regional income growth and club-convergence. Ann Reg Sci40:693–721

Florax RJGM, Folmer H, Rey S (2003) Specification searches in spatial econometrics: the relevance ofHendry’s methodology. Reg Sci Urban Econ 33(5):557–579

Frenken K, Hoekman J (2006) Convergence in an enlarged Europe: the role of network cities. TijdschrEcon Soc Geogr 97:321–326

Fujita M, Krugman P, Venables AJ (1999) The spatial economy. Cities, regions, and international trade.MIT Press, Cambridge

Galor O (1996) Convergence? Inferences from theoretical models. Econ J 106:1056–1069Glaeser EL, Kallal HD, Scheinkman JA, Shleifer A (1992) Growth in cities. J Polit Econ 100(6):1126–1152Glaeser EL, Scheinkman JA, Shleifer A (1995) Economic growth in a cross-section of cities. J Mon Econ

36(1):117–143Hendry DF (1979) Predictive failure and econometric modelling in macroeconomics: the transactions

demand for money. In: Ormerod P (ed) Economic modelling. Heinemann, LondonHohenberg P, Lees L (1996) The making of urban Europe, 1000–1994. Harvard University Press, HarvardIslam N (2003) What have we learnt from the convergence debate? J Econ Surv 17(3):309–362Jarque CM, Bera AK (1987) A test for normality of observations and regression residuals. Int Stat Rev

55(2):163–172Koopmans, Tjalling C (1965) On the concept of optimal economic growth. In: The econometric approach

to development planning. North-Holland, Amsterdam

123


Krugman P (1991) Geography and trade. The MIT Press, CambridgeLe Gallo J, Ertur C (2003) Exploratory spatial data analysis of the distribution of regional per capita GDP

in Europe, 1980–1995. Pap Reg Sci 82:175–201Le Gallo J, Dall’erba S (2005) Evaluating the temporal and spatial heterogeneity of the European conver-

gence process 1980–1999. J Reg Sci 46(2):269–288López-Bazo E, Vayá E, Mora A, Suriñach J (1999) Regional economic dynamics and convergence in the

European Union. Ann Reg Sci 33(3):343–370Lucas RE (1988) The mechanics of economic development. J Mon Econ 22:3–42Mankiw NG, Romer D, Weil DN (1992) A contribution to the empirics of economic growth. Q J Econ

107:407–437Magrini S (1999) The evolution of income disparities among the regions of the European union. Reg Sci

Urban Econ 29:257–281Magrini S (2004) Regional (di)convergence. In: Henderson J, Thisse J-F (eds) Handbook of regional and

urban economics. Elsevier, Amsterdam, pp 2741–2796Meliciani V (2005) Income and employment disparities across European Regions. The role of national and

spatial factors. Reg Stud 40:75–91Mora T, Vaja E, Surinach J (2005) Specialisation and growth: the detection of European regional conever-

gence clubs. Econ Lett 86:181–185Niebuhr A (2001) Convergence and the effects of spatial interaction. Discussion Paper Series 26351,

Hamburg Institute of International EconomicsOttaviano G, Puga D (1998) Agglomeration in the global economy: a survey of the ‘New Economic Geog-

raphy’. World Econ 21:707–731Quah DT (1993a) Empirical cross-section dynamics in economic growth. Eur Econ Rev 37:426–434Quah DT (1993b) Galton’s fallacy and test of the convergence hypothesis. Scand J Econ 95:427–475Quah DT (1996a) Convergence empirics across economies with (some) capital mobility. J Econ Growth

1(1):95–124Quah DT (1996b) Regional convergence clusters across Europe. Eur Econ Rev 40:951–958Quah DT (1996c) Empirics for economic growth and convergence. Eur Econ Rev 40:1353–1375Quah D (1997) Empirics for growth and distribution stratification, polarization and convergence clubs.

J Econ Growth 2:27–59Ramajo J, Marquez MA, Hewings GJD, Salinas MM (2008) Spatial heterogeneity and interregional spill-

overs in the European Union: Do cohesion policies encourage convergence across regions? Eur EconRev 52:551–567

Ramsey FP (1928) A mathematical theory of saving. Econ J 38:543–559Resmini L (2000) The determinants of foreign direct investment in the CEECs: New evidence from sectoral

patterns. Econ Transit 8:665–689Rodrìguez-Pose A (1998) The dynamics of regional growth in Europe. Social and political factors.

Clarendon Press, OxfordRomer PM (1986) Increasing returns and long-run growth. J Polit Econ 94:1002–1037Silverman BW (1986) Density estimation for statistic and data analysis. Chapman & Hall, New YorkSolow RM (1956) A contribution to the theory of economic growth. Q J Econ 70:65–94Van Oort FG (2004) Urban growth and innovation. Spatially bounded externalities in the Netherlands.

Ashgate, Aldershot

123

Income dynamics in an enlarged Europe: the role of capital regions

Documents

Transcript of Income dynamics in an enlarged Europe: the role of capital regions