6. EU-27 Regions: Absolute or Club Convergence?

Essentially, the NUTS system is a hierarchical classification established by EUROSTAT to provide comparable regional breakdowns of the EU Member States. The first version of the NUTS system was set up in the early 1970s. A legal basis was obtained in 2003 (Regulation of the European Parliament and Council 1059/2003). NUTS-1 corresponds to Government Office Regions in England; NUTS-2 to English Counties.

A list of the NUTS-2 regions used in this study is provided in Appendix IV.

Boldrin and Canova (2001), for instance, put forward three objections. First, the large size of the NUTS-2 regions, second commuting in several NUTS-2 regions, e.g. the metropolitan area (an agglomeration zone constituting of several urban centres or a very large city) of Hamburg, which is defined as NUTS-2 region, and yet half of the population of this area lives in the nearby regions of Schleswig-Holstein and Lower Saxony; a similar situation emerges for Ile de France, the Bassin Parisien and Madrid and Castillas and Flevoland in the Netherlands, and third factor endowments and population density are very heterogeneous across the NUTS-2 regions of the EU. Furthermore, GDP is abnormally inflated in most capital cities of Europe due to a large concentration of state governments and headquarters of large national companies. As a result, GDP is attributed to headquarters or central government offices, even when production is taking place elsewhere.

GVA is the net outcome of output at basic prices less intermediate consumption valued at consumers’ prices. The estimates are in accordance with the European System of Accounts 1995. It is possible to use Gross Regional Product at market prices or to measure GVA in purchasing power standards (PPS). However, at the regional level this raises several problems (Ertur et al. 2007). To be more precise, the conversion should be made using regional PPS. However, since such data are not available, the adjustments are made using national price levels. Moreover, the relative figures of regional GVA can change not only due to differences in the rate of GVA growth in real terms but also due to changes in the relative price level. Changes due to reductions in the relative price level might have a different implication than one resulting from a relative growth in real GVA.

Using total labour force may be misleading if the primary concern of the study is regional productivity. To be more specific, inclusion of unemployed labour force would distort the productivity measure when unemployment levels vary significantly across regions.

Using the coefficient of variation for GDP per-head over the period 1995–2005 Michelis and Monfort (2008) reach similar conclusions.

It is argued that within the EU income disparities have diminished between Member-States but increased between regions. Indeed, the richest regions are eight times richer than the poorest regions. Socio-economic inequalities within regions and countries constitute about 80 % of overall inequalities (Kanbur and Venables 2007) and for the majority of the Member-States were higher in 2007 than in 1980. For a more detailed description of regional inequalities and income polarization in the EU see Paci (1997), Magrini (1999), Puga (1999), Maza and Villaverde (2004), Martin (2005), Ezcurra (2009), Shucksmith et al. (2009) Bracalente and Perugini (2010), among others.

This is an exercise that a number of authors, including, Fingleton (1997), Quah (1996a), Puga (2002) have undertaken. Neven and Gouyette (1995), however, argue that this approach is mostly a descriptive analysis.

These regions are characterised as Objective-1 regions. In 2003, 84 regions were below the 75 % threshold with total population about 154 million inhabitants. The Objective-1 regions cover the entire area of the 10 Member States that joined the EU in 2004 (with the exception of Bratislava, Prague and Cyprus). A second criterion applies to the definition of Objective-1 regions: a low population density (less than eight inhabitants per square kilometre). This criterion covers a number of region in northern Finland and Sweden, the French overseas departments, the Canary Islands, the Azores and Madeira. The European Commission (1996) observes that ‘regions with more than 500 inhabitants per square kilometre account only for 4 % of the land area of the Union but for more than half the population. This implies that between two-thirds and three quarters of the EU’s total wealth creation occurs in urban areas’ (p. 12). Regions under economic and social restructure are classified as Objective-2 regions. In this case the following criteria are applied: changes in key sectors due to declining employment in industrial and services sectors, economic and social crisis in urban areas, decline of traditional activities and depopulation of rural areas. Regions in which efforts are made to reduce unemployment are characterised as Objective-3 regions. The disadvantages areas of the EU, Objective-1 and 2 regions correspond to almost half of the EU-25’s total population (about 225 million inhabitants).

This scatterplot is known as the ‘convergence picture’ (Romer 1987b).

Labour productivity is expressed in logarithmic terms given that the empirical literature has concentrated mainly on logarithms instead of levels.

This slow process of regional convergence can, possibly, be explained by the low degree of labour mobility that characterises the European regions, due to linguistic and cultural barriers. As Boldrin and Canova (2001, p. 243) state ‘while capital is moving around Europe, labour is definitely not’. Obstfeld and Peri (1998) report that labour mobility in Germany, Italy and the UK over the period 1970–1995 was only about one-third of the US level.

Recall that the notion of absolute convergence derives from the standard neoclassical model, which treats all economies (countries or regions) as similar. If this was true, then the economies will display absolute β-convergence as well as σ-convergence (Sala-i-Martin 1996a). While there is evidence for σ-convergence, the slow process of β-convergence implies that structural characteristics and overall conditions differ markedly across the EU-27.

This is computed as \( F = \frac{{ESS/k - 1}}{{RSS/n - k}} \), where \( ESS \) is the explained sum of squares, \( RSS \) is the residuals sums of squares (the total sum of squares is \( TSS = ESS + RSS \)), k is the number of parameters including the constant and n is the number of observations. The null hypothesis associated with this test is that all the regression coefficients are equal to zero. A low probability value implies that at least some of the regression parameters are nonzero and that the regression equation does have some validity in fitting the data (i.e., the independent variables are not purely random with respect to the dependent variable).

As a matter of fact, Chatterji’s specification was tested using these four regions as leaders. In any case, the results clearly indicate that the NUTS-2 regions of the EU-27 diverge from the leading regions. Nevertheless, using LU as the leading region produces more robust results from econometric point of view, compare to the four other regions, and therefore was chosen.

Neven and Gouyette (1995) exclude the region of Groningen from their empirical analysis. They argue that output recorded in that region is somewhat of artificial nature, which includes all production of gas from the North Sea in the Netherlands. This region was the most affluent in the EC in the early 1980s, but declined markedly relatively to others as energy prices fell throughout the second part of the 1980s. The authors argue that the inclusion of Groningen in the sample would bias the estimates in favour of finding convergence. On similar lines, Maurseth (2001) takes account of the fact that Groningen was hard hit by the ‘Dutch disease’ in the 1980s and excludes this particular region from the empirical analysis.

The presence of ‘outlying’ or ‘leading’ regions might influence the process of overall regional convergence (Button and Pentecost 1995).

Equation (6.1.1) is used extensively in the empirical literature on spatial econometrics (e.g. Richardson 1974; Cliff and Ord 1981; Attfield et al. 2000; Ravallion and Jalan 1996; Fingleton 2000; Frizado et al. 2009). Nevertheless, the spatial matrix can be constructed also using the inverse distances to the square as denominator. Results using this kind of spatial matrix were very similar.

Choosing cities according to their level of GVA per-worker may cause problems of endogeneity in estimating spatial econometric models. However, such a criterion is used extensively in the relevant empirical literature. Nevertheless, one way to overcome this endogeneity problem is to choose the most populated cities. Such a choice does not serve any purpose since cities with high GVA per-worker are normally associated with high population and most economic activities are concentrated here also.

The measures of σ-convergence are not reliable in the presence of spatial autocorrelation (Rey and Dev 2006). Nevertheless, these measures are used only indicative.

Baumont et al. (2003) put forward the argument that convergence clubs can be detected using a Moran’s scatterplot. They support this argument using data for the EU regions. Similar results are reported by Maza and Villaverde (2004) and Mora (2004, 2005).