Content
Volume 48, January/February 2013, Number 1 · pp. 5966
Articles
Convergence in the Spanish and Portuguese NUTS 3 Regions: An Exploratory Spatial Approach
Miguel Viegas, Micaela Antunes
Since their accession to the European Union in 1986, both Portugal and Spain have benefited from strong financial support. Both countries have experienced considerable growth in income per capita, converging towards average European levels. However, several studies suggest a high degree of persistence of regional asymmetries within the countries. This paper empirically analyses convergence among NUTS 3 regions of the Iberian Peninsula between 1995 and 2008. The results reveal divergent national trends and indicate no evidence of catchingup effects among the poorest regions, confirming the existence of economic clusters.
Miguel Viegas,
University of Aveiro, Portugal.
Micaela Antunes,
University of Coimbra, Portugal.
Over the last several decades, economies have been experiencing mixed performances regarding economic growth, thus motivating economists to study convergence. While some economies have been improving their income levels, others have become progressively poorer on a worldwide scale, thus widening income gaps and deepening disparities.
Despite the large literature dealing with the topics of economic growth and convergence, we emphasise the neoclassical perspective following Solow's longrun growth model.1 According to that formulation, countries with an initially lower capital stock grow faster than others at an earlier stage before converging to grow at similar rates in the long term. The explanation lies in the decreasing marginal returns to capital, implying that the lower the stock of capital, the higher the corresponding marginal productivity.
Therefore, the further an economy is from a steady state, the faster the rate of growth. However, the rate declines when the economy moves from a low per capita income level to a higher one. At the end of the transitional dynamics, the initially poorer economy reaches the per capita income level of the richer economy (catching up). According to this view, divergence is a transitory shortterm phenomenon reflecting adjustments towards a longrun equilibrium level of per capita income. This basic kind of convergence to a common income level, showing an inverse relationship between the initial income level and the corresponding growth rate, is known as absolute (or unconditional) beta convergence, a term first applied in the early 1990s.2 Absolute convergence is a strong assumption, as it implies that economies possess the same structural parameters (saving rate, population growth, capital depreciation and technology level), differing only in terms of capital endowment. Therefore, it is more probable that such a condition is met for a group of homogenous economies with common institutional and legal features and similar economic parameters. As similarities in terms of common economic policies as well as the higher mobility of factors of production and technological diffusion are more common within national boundaries, it is expected that regions of a given country tend to converge to a certain steady state.3
Our focus is specifically on regions of the Iberian Peninsula. The first reason is geographical: Portugal and Spain share a common border. Moreover, these countries have been historically engaged, becoming increasingly integrated throughout the years and having entered the European Economic Community (EEC) in the same year. The countries differed in their structural indicators, but both managed to fulfil the nominal targets and be among the 11 initial countries to enter the third stage of the Economic and Monetary Union (EMU).
Since their accession to the EU, both Portugal and Spain have experienced considerable economic growth and convergence towards the EU average GDP per capita. In general terms, Spain has had better economic performance than Portugal since adhesion, both at the national and the regional level. Several improvements have been achieved in both countries, particularly in infrastructure, accompanied by increased public investment. However, regional disparities have persisted; indeed, regional gaps have increased in many cases.4 By the time of the adhesion to the EEC, most Portuguese and Spanish regions were Objective 1 regions, meaning that their per capita GDP was less than 75% of the Community average, and thus they were eligible to receive Structural Funds from the EU in order to catch up with the richer nations. As such, Portugal and Spain have benefited extensively from EU assistance funds. During the period between 1995 and 2008, Portugal and Spain together received about €140 billion (at current prices, including Structural and Cohesion Funds) through the Delors II Package (19941999) and the Agenda 2000 (20002006).
This monetary support has been at the core of a wide debate on the efficiency of Structural Funds, as it was mostly oriented towards infrastructure and thus did not result in convergence but rather in the concentration of technologically advanced activities in specific points in space.5 Qualification of the workforce and human capital improvement were not a real priority. Moreover, since the entry of Eastern European countries to the EU, it is interesting to observe how these two main receivers of funds deal with the challenge of smaller allocations of financial resources.
Several empirical studies on regional convergence have pointed to a significant regional convergence process in Spain until the late 1970s and in Portugal from the 1980s until the mid1990s.6 Unfortunately, empirical evidence is not conclusive concerning recent decades. Sanchez and Roura7 state that regional disparities have remained essentially constant, while Marelli8 finds a slowing, although positive, crossregional convergence process between 1990 and 2005. RodriguezPose,9 RodriguezPose and Fratesi,10 and Costa and Fonseca11 analyse the regional disparities since 1989 and find considerable growth in the standard deviation of both Portuguese and Spanish regions at the NUTS 2 level.12
Therefore, considering the spatial continuity of Spain and Portugal, together with their similar historical paths concerning EEC entry, we proceed with an empirical convergence exercise among the NUTS 3 regions of the Iberian Peninsula, using a spatial econometric approach as a means to embody the role of space and geography. Regional economies tend to be more open and more specialised than national ones. As spatial units become smaller, economic specialisation increases and spatial dependence becomes more relevant. The convergence literature also pays particular attention to the national effect, according to which each region is closely linked to the respective national economic performance.13 As such, we differentiate between Portuguese and Spanish regions in order to check for the presence of a national convergence club effect.
The analytical framework
The neoclassical growth model is based on Solow's14 approach and assumes that in the long run all economies converge to the same steadystate level of per capita income, as they grow more rapidly the further the economy is initially from the equilibrium level. Whenever a negative and statistically significant relation is found between the initial per capita GDP level and the corresponding growth rate, we can assume the presence of absolute beta convergence.15 Absolute beta convergence was investigated by Baumol16 through the following equation:
where y _{i,0} and y _{i,T} correspond to the per capita GDP of region i at the initial and final periods respectively and T is the time interval. The lefthand side represents the average annual log growth rate of per capita GDP of region i, α and β are the parameters to be estimated, and ε is the error term. From the estimation of β, we obtain the annual speed of convergence, and the halflife of convergence,.17
Another concept of convergence is that of sigmaconvergence, which analyses the evolution of income disparities across economies over time through measures of dispersion like the coefficient of variation (an indicator of relative dispersion given by the ratio of the standard deviation over the sample mean). A reduction in this indicator implies a decrease in dispersion and thus the existence of sigmaconvergence. Betaconvergence is a necessary but not sufficient condition for sigmaconvergence to occur.18
Many convergence studies use crosssection analyses. However, there are several criticisms of these models, mostly related to the existence of multicollinearity, endogeneity bias and the existence of specification errors. These problems may seriously affect the robustness of the convergence coefficient and produce misleading outcomes.19 Moreover, the introduction of the geographical dimension allows one not only to capture the spatial effect but also to improve the estimation and forecasting, since spatial dependence violates some of the GaussMarkov assumptions of the OLS estimation (crosssection observations are no longer independent), producing inefficient estimators.20
Several studies focusing on the importance of spatial location for growth argue that when spatial correlation is ignored, the results regarding economic growth may be biased. Two kinds of spatial effects are pointed out in the literature: (i) spatial autocorrelation, revealing that contiguous regions may influence each other's performance through spillover effects, and (ii) spatial heterogeneity, whenever the same functional form is erroneously considered for all regions.21 Spatial autocorrelation can be of two types: spatial autoregressive dependence in the variables due to interrelationships between economic variables of contiguous regions and spatial autocorrelation in the disturbance term, which can be due to omitted variables or deficient functional form.
For our exploratory spatial analysis, we use per capita GDP at the NUTS 3 level between 1995 and 2008, as published by the Portuguese and Spanish national statistics offices, deflated by a national GDP deflator.22 We only collect information for regions of mainland Portugal and Spain. Regions like the Azores, Madeira and the Canary Islands are excluded from our analysis as they do not have spatial contiguity with other regions (see Table 1). Summing up, our database comprises 75 spatial units, 28 in Portugal and 47 in Spain. After assessing sigmaconvergence, we test the presence of spatial autocorrelation in average per capita GDP and in growth rates using the Moran's I autocorrelation coefficient. Finally, we estimate the betaconvergence process. We introduce a national dummy to test the presence of spatial heterogeneity. This specification allows us to estimate the possibility of two different convergence patterns in each country. All estimations are carried out in MATLAB using the general maximum likelihood method.23
Table 1 (back to the text)
Iberian Peninsula regions (NUTS 3)
CODE  NUTS 3  CODE  NUTS 3  CODE  NUTS 3  CODE  NUTS 3 
ES243  Zaragoza  ES412  Burgos  ES617  Málaga  PT161  Baixo Vouga 
ES242  Teruel  ES418  Valladolid  ES612  Cádiz  PT168  Beira Interior Norte 
ES241  Huesca  ES411  Ávila  ES615  Huelva  PT166  Pinhal Interior Sul 
ES230  La Rioja  ES419  Zamora  ES618  Sevilla  PT163  Pinhal Litoral 
ES220  Navarra  ES416  Segovia  ES614  Granada  PT16C  Médio Tejo 
ES212  Guipúzcoa  ES432  Cáceres  ES611  Almería  PT16B  Oeste 
ES213  Vizcaya  ES431  Badajoz  ES613  Córdoba  PT162  Baixo Mondego 
ES211  Álava  ES424  Guadalajara  ES616  Jaén  PT164  Pinhal Interior Norte 
ES300  Madrid  ES423  Cuenca  ES620  Murcia  PT16A  Cova da Beira 
ES120  Astúrias  ES422  Ciudad Real  PT184  Baixo Alentejo  PT169  Beira Interior Sul 
ES130  Cantabria  ES421  Albacete  PT182  Alto Alentejo  PT114  Grande Porto 
ES112  Lugo  ES425  Toledo  PT183  Alentejo Central  PT117  Douro 
ES114  Pontevedra  ES514  Tarragona  PT185  Lezíria do Tejo  PT111  MinhoLima 
ES111  A Coruña  ES513  Lleida  PT181  Alentejo Litoral  PT118  Alto TrásosMontes 
ES113  Ourense  ES512  Girona  PT150  Algarve  PT116  Entre Douro e Vouga 
ES413  León  ES511  Barcelona  PT171  Grande Lisboa  PT112  Cávado 
ES414  Palencia  ES522  Castellón/Castelló  PT172  Península de Setúbal  PT113  Ave 
ES417  Soria  ES523  Valencia/València  PT167  Serra da Estrela  PT115  Tâmega 
ES415  Salamanca  ES521  Alicante/Alacant  PT165  DãoLafões 
Source: Eurostat.
The exploratory spatial data analysis: results and discussion
Figure 1 illustrates the dispersion, measured by the coefficient of variation of the logarithm of per capita GDP during the 19952008 period, in the Portuguese and Spanish regions separately and in all 75 regions combined. In the combined result, the regional dispersion decreases until 2001 and thereafter increases, reaching a higher level of dispersion relative to the initial point. The dispersion across Portuguese regions shows a downward path over the whole period, while the Spanish coefficient of variation increases a little at first and decreases steadily from 1999 onwards to a point below the initial dispersion level. Moreover, the Portuguese regional dispersion is above the Spanish regional dispersion levels throughout the entire period. The apparent contradiction between the overall sigmadivergence process and the two national sigmaconvergence processes can be explained by a divergence process between the two countries. In fact, Portuguese per capita GDP represented 65% of the Spanish per capita GDP in 1995, rose to 68% in 1999 and decreased to 54% in 2008.
Figure 1 (back to the text)
Sigmaconvergence between 1995 and 2008
Sources: National statistics offices.
The spatial autocorrelation is widely measured by Moran's I statistic, which can be represented by the expression:
where w_{ij} represents the element of the spatial contiguity matrix, W, such that w_{ij} = 1 if municipalities i and j are neighbours and w_{ij} = 0 otherwise; x_{ij} represents the logarithm of per capita GDP (in deviation from the mean) of region i at time t; and n is the number of observations.
Moran's I estimates the linear dependence between a variable in a specific location and the mean of the same variable in the neighbourhood. Moran's I statistic and the respective marginal probability relative to the logarithm of per capita GDP are shown in Table 2, revealing a positive and significant spatial dependence in all years and in each scenario (in all regions of the Iberian Peninsula as well as when national regions are separated). This means that richer regions tend to be located near other rich regions, while poor regions tend to be located near other poor regions. Moran's I statistic for all regions (Portuguese and Spanish) shows a similar trend as that for the coefficient of variation, i.e. decreasing initially and increasing from 2000 onwards (the correlation between Moran's I statistic and the coefficient of variation is 0.95). This result points out that spatial dependence increases with spatial dispersion, which may be interpreted as a shadow effect of richer regions over poor ones, leading to a more unequal distribution of economic activity. Regarding Portugal and Spain separately, however, we observe decreasing trends for the whole period, rather similar to the respective coefficients of variation (the correlations are 0.78 and 0.53 respectively). Spain exhibits a stronger pattern of spatial autocorrelation.24
Table 2 (back to the text)
Moran's I statistic
Portugal and Spain  Portugal  Spain  

Year  Moran I  Mg. Prob.  Moran I  Mg. Prob.  Moran I  Mg. Prob. 
1995  0.7277  0.0000  0.3709  0.0002  0.6886  0.0000 
1996  0.7319  0.0000  0.3554  0.0003  0.6789  0.0000 
1997  0.7089  0.0000  0.3881  0.0001  0.6690  0.0000 
1998  0.6914  0.0000  0.3786  0.0001  0.6495  0.0000 
1999  0.6731  0.0000  0.3423  0.0004  0.6280  0.0000 
2000  0.6705  0.0000  0.3251  0.0008  0.6239  0.0000 
2001  0.6890  0.0000  0.3131  0.0011  0.6399  0.0000 
2002  0.7029  0.0000  0.3198  0.0009  0.6396  0.0000 
2003  0.7140  0.0000  0.3144  0.0011  0.6355  0.0000 
2004  0.7290  0.0000  0.2993  0.0018  0.6292  0.0000 
2005  0.7293  0.0000  0.2774  0.0035  0.6102  0.0000 
2006  0.7483  0.0000  0.3322  0.0006  0.6129  0.0000 
2007  0.7581  0.0000  0.3399  0.0005  0.6322  0.0000 
2008  0.7469  0.0000  0.3025  0.0016  0.6319  0.0000 
Source: Own calculations based on data from national statistics offices.
Figure 2 presents the distribution of average per capita GDP for the 19952008 period across all Iberian regions. The map shows a concentration of rich regions in the northeast of Spain (the Portuguese Grande Lisboa rich region is the exception); a broad area composed of mediumsize income regions in the centre, south and north of Spain with some central and southern Portuguese regions; and a poor area composed of the northern and inland Portuguese regions.
Figure 2 (back to the text)
Average per capita GDP 19952008, Spanish and Portuguese regions (NUTS 3)
Source: National statistics offices.
Figure 3 (back to the text)
Average per capita GDP 19952008, Spanish and Portuguese regions (NUTS 3).
LISA cluster map (lefthand side) and Moran scatter plot (righthand side) (Moran's I = 0.7277)
Source: Own calculations based on data from national statistics offices.
Figure 3 presents the LISA cluster map and the Moran scatter plot of the same variable. The Moran scatter plot depicts the variable on the horizontal axis with the average values of the neighbouring regions for the same variable on the vertical axis. The four quadrants in the scatter plot show (i) the regions with high per capita GDP associated with neighbouring regions with high per capita GDP (topright), (ii) the regions with low per capita GDP associated with neighbouring regions with low per capita GDP (bottomleft), (iii) the regions with low per capita GDP associated with neighbouring regions with high per capita GDP (topleft) and (iv) the regions with high per capita GDP associated with neighbouring regions with low per capita GDP (bottomright). The first and second quadrants (highhigh and lowlow) highlight the existence of positive autocorrelation, while the third and fourth ones show negative autocorrelation. Therefore, the presence of a large number of 61 regions (81%) in the first and second quadrants is a clear sign of positive spatial autocorrelation. The remaining 14 regions occupy the atypical locations of quadrants three and four. The LISA cluster map depicts these types of spatial association. The spatial autocorrelation pattern can be identified by the wide stretches of areas with the same colour, indicating that similar regions tend to aggregate geographically. As such, we can see a group of rich regions in the north and east of Spain which contrasts with the south and west of the Iberian Peninsula, including almost all the Portuguese regions. The map also shows some atypical cases of rich regions without spatial dependence with their neighbouring regions (La Coruna, Pontevedra, Salamanca and Grande Lisboa, among others).
The distribution of the growth rate of per capita GDP between 1995 and 2008 (Figure 4 and 5) is more heterogeneous across the territory. The highest growth rates belong to Spanish regions, namely Badajoz, Huelva, Cadiz and Almeria in the south, and Vizcaya, Alava, Guipuzcoa, Pontevedra, Asturias, Cantabria and Zamora in the north. The lowest growth rates are observed in five Portuguese regions: BaixoVouga, Ave, Grande Porto, Lezíria do Tejo and Península de Setúbal. The Moran scatter plot shows a concentration of regions in the first and second quadrants (56 regions, equivalent to 75% of all regions), confirmed by the Moran's I indicator of 0.44 with a high level of significance.
Figure 4 (back to the text)
Per capita GDP growth rate 19952008, Spanish and Portuguese regions (NUTS 3)
Source: National statistics offices.
Figure 5 (back to the text)
Per capita GDP growth rate 19952008, Spanish and Portuguese regions (NUTS 3).
LISA cluster map (lefthand side) and Moran scatter plot (righthand side) (Moran's I= 0.4412)
Source: Own calculations based on data from national statistics offices.
Finally, we use a spatial econometric methodology to estimate a model of absolute betaconvergence for the Iberian NUTS 3 regions for the 19952008 period. First, we estimate the simple model of betaconvergence according to equation (1) with and without a national dummy. The next step is to detect the presence and type of spatial effects in order to evaluate whether the spatial lag model or the spatial error model is most appropriate to describe the data. We follow the robust LM tests described in Elhorst,25 which test the type of spatial dependence based on the residuals of the nonspatial models.26
Table 3 (back to the text)
Estimations results and spatial tests. Betaconvergence, 19952008
^{(pvalues in parentheses)}
Models  OLS  SEM  OLS  SEM  OLS  OLS  SEM 

(ES+PT)  (ES+PT)  (ES+PT)  (ES+PT)  (ES)  (PT)  (ES)  
1  2  3  4  5  6  7  
Estimation  
Obs  75  75  75  75  47  28  47 
R2  0.1017  0.3875  0.5710  0.5978  0.0682  0.1698  0.2333 
Constant  0.0025 (0.9283)  0.0678 (0.0786)  0.1558 (0.0000)  0.1621 (0.0000)  0.1556 (0.0000)  0.1727 (0.0003)  0.1636 (0.0005) 
β  0.0091 (0.0053)  0.0019 (0.6587)  0.0092 (0.0034)  0.0099 (0.0035)  0.0072 (0.0763)  0.0111 (0.0293)  0.0081 (0.1141) 
δ      0.0176 (0.0000)  0.0182 (0.0000)       
λ    0.6740 (0.0000)    0.3180 (0.0549)      0.4930 (0.0046) 
Autocorrelation tests  
Moran I  55.444 (0.0000)    24.788 (0.0132)    36.993 (0.0002)  0.3302 (0.7413)   
LM lag (robust)  0.0196 (0.8890)    0.4294 (0.5120)    0.4459 (0.5040)  0.0112 (0.9160)   
LM error (robust)  191.410 (0.0000)    43.358 (0.0370)    89.440 (0.0030)  0.0027 (0.9590)   
Source: Own calculations based on data from national statistics offices.
The results for all the regions of the Iberian Peninsula are presented in the first column of Table 3. They indicate that the specification of the spatial error model in which only disturbances exhibit spatial dependency, given by Equation (3), is adequate for the convergence process (λ represents the spatial autoregressive parameter in the error term). The results of the three spatial autocorrelation tests can be seen at the bottom of the table. The LM robust error test27 indicates the presence of spatial correlation in the residuals of the regression model.
With the spatial error dependence model (column 2), the slight betadivergence process estimated by the OLS model ceases to be significant. This excludes the presence of a catchingup effect among the poorest regions, as would be predicted according to the sigmadivergence detected above. Therefore, the results, namely the presence of a strong spatial effect, confirm the existence of a polarisation of economic activity at the Iberian Peninsula scale.
Columns (3) and (4) of Table 3 present the results of estimating the same equation with the inclusion of a dummy variable for Spain. They indicate the spatial error model given in Equation (4) as the more appropriate one. As expected, the estimation shows a highly significant dummy coefficient and a slow betaconvergence process (with a velocity of convergence of 1% per year and a halflife of 69 years), compatible with individual sigmaconvergence processes in each country, as well as a significant spatial dependence on the error term.28
After confirming the spatial heterogeneity in the form of a national effect, we estimate separately the convergence process in the two countries. The results provide little evidence for a betaconvergence process in Spain (OLS in column 5 and Spatial Error Model in column 7) and are insignificant in the Spatial Error Model (with a pvalue just above 10%). The strong spatial dependence in the error term confirms the effect of nonobservable variables that may have contributed to the development of contiguous areas, improving (slightly) the income distribution. As for the Portuguese regions (column 6), the OLS estimation reveals a statistically significant betaconvergence process, although one that is rather slow and without spatial dependence. The velocity of convergence is 1.2% per year with a halflife of 58 years.
Conclusion
Using a spatial econometric framework, this paper empirically analyses convergence among NUTS 3 regions of the Iberian Peninsula between 1995 and 2008. The reduction of disparities in the levels of development of the various regions and of the backwardness of the least favoured regions represents one of the main objectives of the EU. Since joining the EU, Portugal and Spain have recorded impressive economic growth, converging towards the EU average. However, there remain concerns about persistent regional asymmetries.
At the Iberian Peninsula scale, our results point to a sigmadivergence process between 1995 and 2008, while at the national level, both countries have followed a sigmaconvergence process during the same period. This apparent contradiction reveals a worrying divergence between Spain and Portugal (from 2000 onwards) as well as a strong national effect that has apparently precluded some Portuguese border regions from benefiting from the impressive economic growth of some Spanish border regions like Huelva, Badajoz, Zamora or Pontevedra.
The results also point to some qualitative differences in the convergence pattern between Spanish and Portuguese regions. In the former, we found limited and insignificant betaconvergence with strong spatial dependence in the error term, while in the latter, the estimation reveals a slow but significant betaconvergence process, without spatial dependence. This subtle difference means that in the Spanish case, the spatial effects are crucial to the decrease of regional dispersion, while in the Portuguese case, in which spatial effects were not detected, a catchingup process of depressed regions seems to be at the core of the improvement of income distribution. As shown above, Spanish regions with high growth rates, e.g. Badajoz, Vizcaya, Pontevedra or Almeria, are always surrounded by regions with equally high growth. Conversely, the Portuguese regions with the highest growth rates (Alentejo Litoral and Serra da Estrela, two regions that have received significant public investments) did not seem to have any positive effects on the respective contiguous regions. These results raise multiple issues about the application of Structural Funds and the types of growth they generate, leading the way for further investigation.
In 1990 the European Commission integrated a special initiative for border regions into EU cohesion policy instruments known as INTERREG in order to promote crossborder cooperation (INTERREGA). Since then, two other INTERREGA generations were completed (19941999 and 20002006), and another one is currently included in the Territorial Cooperation objective (20072013). We did not formally test the presence of spatial effects across the border. However, our results confirm a strong national club effect and the incapacity of the Portuguese border regions to capture positive crossborder effects from prosperous Spanish regions. The inclusion of physical and human capital, population, and additional explaining factors is another reasonable line of investigation to further explore the behaviour of the NUTS 3 regions of the Iberian Peninsula with regard to growth and convergence.

1 R. Solow: A contribution to the theory of economic growth, in: Quarterly Journal of Economics, Vol. 70, No. 1, 1956, pp. 6594.

2 R. Barro, X. SalaiMartin: Economic growth, Cambridge 2004, The MIT Press.

3 X. SalaiMartin: Regional cohesion: evidence and theories of regional growth and convergence, in: European Economic Review, Vol. 40, No. 6, pp. 13251352.

4 E. Costa, M. Fonseca: Convergência económica e coesão social e territorial da Península Ibérica na União Europeia, in: X Colóquio Ibérico de Geografia – A geografia ibérica no contexto europeu, Universidade de Évora 2005.

5 For an interesting survey about structural funds and Objective 1 regions see: F. Torres, M.L. de Freitas, F. Pereira: Convergence among EU Regions, 19902001 – Quality of National Institutions and "Objective 1" Status, in: Intereconomics, Vol. 38, No. 5, 2003, pp. 270275.

6 A. Sanchez, T. Roura: Regional convergence in productivity and productive structure. Application to European Southern countries, in: Institute of Social and Economic Analysis, Working paper, No. 11, 2008.

7 Ibid.

8 E. Marelli: Specialisation and convergence of European regions, in: The European Journal of Comparative Economics, Vol. 4, No. 2, 2007, pp. 149178.

9 A. RodriguezPose: Economic convergence and regional development strategies in Spain: The case of Galicia and Navarre, in: European Investment Bank Papers, Vol. 5, No. 1, 2000, pp. 89115.

10 A. RodriguezPose, U. Fratesi: Between development and social policies: the impact of European Structural Funds in Objective 1 regions, in: European Economy Group – Working Paper, No. 28, 2003.

11 E. Costa, M. Fonseca, op. cit.

12 Most of the references about empirical studies use the NUTS 2 regional level.

13 See J. LopezRodriguez, A. Faina: Regional policy and convergence in Europe: the case of backward regions, in: Economics Bulletin, Vol. 29, No. 2, 2009, pp. 10461053; and A. Sanchez, T. Roura, op. cit.

14 R. Solow, op. cit.

15 R. Barro, X. SalaiMartin, op. cit.

16 W. Baumol: Productivity growth, convergence and welfare: what the longrun data show?, in: American Economic Review, Vol. 76, No. 5, 1986, pp. 10721085.

17 For details, see N. Islam: Growth empirics: a panel data approach, in: The Quarterly Journal of Economics, Vol. 110, No. 4, 1995, pp. 11271170; S. Silva, M. Silva: Crescimento económico nas regiões europeias: Uma avaliação sobre a persistência das disparidades regionais no período 198095, in: FEP Working Paper, No. 96, 2000.

18 M. Chatterji: Convergence clubs and endogenous growth, in: Oxford Review of Economic Policy, Vol. 8, No. 4, 1992, pp. 5769.

19 See D. Quah: Galton's fallacy and tests of the convergence hypothesis, in: The Scandinavian Journal of Economics, Vol. 95, No. 4, 1993, pp. 427443; P. Evans: Using crosscountry variances to evaluate growth theories, in: Journal of Economic Dynamics and Control, Vol. 20, No. 67, 1996, pp. 10271049; and F. Caselli, G. Esquivel, F. Lefort: Reopening the convergence debate: a new look at crosscountry growth empirics, in: Journal of Economic Growth, Vol. 1, No. 3, 1996, pp. 363389.

20 See, among others, L. Anselin: Spatial econometrics: methods and models, Dordrecht 1988, Kluwer Academic Publishers; J. LeSage, R. Pace: Introduction to spatial econometrics, USA 2009, CRC Press, Taylor and Francis Group.

21 For comprehensive references on spatial econometrics see for instance L. Anselin, op. cit.; J. LeSage, R. Pace, op. cit; J. LeGallo: Econometrie spatiale: l'autocorrelation spatial dans les modèles de regression lineaire, in: Economie et Provision, Vol. 155, No. 4, 2002, pp.139158.

22 AMECO database.

23 We use the LeSage Spatial Econometrics Toolbox functions available at http://www.econ.utoledo.edu.

24 The positive correlation between the Moran's I statistic and the coefficient of variation can be found in other empirical studies (see, for instance, J. LeGallo et al., op.cit.).

25 J. Elhorst: Spatial Panel Data Models, in: M.M. Fischer, A. Getis (eds.): Handbook of Applied Spatial Analysis, Berlin, Heidelberg, New York 2009, Springer, pp. 377408.

26 MATLAB routines available at www.regroningen.nl/elhorst.

27 J. Elhorst, op. cit.

28 We also introduced a national dummy variable multiplied by the initial income level in order to detect differences in the convergence rate, but the estimation does not detect any statistical significance.