Productivity convergence in the European regions, 1980–2003: a sectoral and spatial approach

Abstract

This article analyses the evolution of the EU productivity between 1980 and 2003, both across regions and sectors. By making use of various techniques (cross-section, nonparametric and spatial approaches) it concludes that: (1) the regional and sectoral dispersion of productivity is quite high; (2) the gains experienced in aggregate productivity are due entirely to the sectors productivity growth effect; (3) there is a weak beta-convergence process at the aggregate and sectoral levels; (4) the accounting decomposition of the aggregate productivity convergence process reveals the sectoral productivity growth effect to be the only responsible factor for regional catching-up; (5) finally, there are clear signs of spatial dependence which, when properly addressed, increase the speed of convergence at the aggregate level.

Acknowledgements

The research was partly funded by the Spanish Institute for Fiscal Studies.

Notes

¹ NUTS means Nomenclature of Territorial Units for Statistics. It refers to the European regional classification established by Eurostat.

² Another common interpretation refers to total factor productivity (TFP).

³ A similar approach can be seen, among other recent articles, in (Laurini et al. 2005).

⁴ For a different but complementary approach to the analysis of convergence in productivity, see Tsionas (2000).

⁵ As is usual in the analysis of beta convergence, we have introduced the variable object of analysis – in this case, productivity – in relative terms. In fact, we have taken the value of each region with respect to the mean value for EU-15 without considering the value registered by a ‘leader’ region (as is done in other articles). Here the degree of disaggregation is so high that in our opinion this second option is not advisable.

⁶ We have also carried out a conditional beta-convergence analysis. The results are available upon request.

⁷ To calculate this we use the following expression: .

⁸ If this number of years is denoted h, it can be easily calculated according to the expression: .

¹¹ Also negative is the effect of ‘Paper and printing products’, but its value is practically zero.

⁹ When the initial productivity of region i exceeds that of the EU, the expression to apply is as follows: _UE −  i =  + .

¹⁰ We should point out that the results shown are weighted by the share of the GVA in each region with respect to the EU as a whole.

¹² For this, we used the programs ArcView GIS 3.2 and SpaceStat 1.90. Other recent studies that have used spatial econometric techniques are, for example, López-Bazo et al. (2002), Toral (2002), Villaverde and Maza (2003), Maza and Villaverde (2004) and Villaverde (2006).

¹³ Spatial autocorrelation can be of two types: substantive, which, through phenomena such as technological diffusion, externalities and factor mobility, links the behaviour of a particular variable in various different spaces; and noise, which comes from a poor specification of the model and refers to the residuals of the estimated regression.

¹⁴ As well as other possibilities, such as including spatial lags of the explanatory variables.

¹⁵ It should be pointed out that the traditional measure of fitness (R ²) is unreliable because of the inclusion of spatial errors. For the sake of comparison, the values of these coefficients in the estimation of the traditional beta convergence are as follows: LIK = 601.15, AIC = −1198.30 and SC = −1191.65.