Regional Convergence in Central and Eastern European Countries: A Multidimensional Approach

Abstract

The paper examines the dynamics of regional income at the NUTS3 level of the new EU Member States from Central and Eastern Europe in the years 1998–2005. The authors apply a wide range of methods and tools including classical beta and sigma convergence analysis supplemented by transition matrices, kernel density estimations and spatial autocorrelation statistics. Results of such a multi-dimensional empirical study reveal some previously unrecognized patterns of regional growth in Central and Eastern European Countries (CEECs). Well-acknowledged metropolization and marginalization processes that cause regional divergence at the national scale are accompanied by the following processes. Firstly, at the macroregional scale, regional convergence has been observed as a result of differences in growth rates between individual countries. Secondly, at the national scale, petrification of existing regional structures has been prevailing in majority of the countries. Furthermore, weak convergence of clubs has been observed separately among the richest metropolitan regions and between the group of the poorest regions. In general, the polycentric spatial structure of the macroregion has reduced the impact of rapid growth of rich capital city-regions on convergence processes. Simultaneously, diffusion of development processes had a rather limited range and polarization in larger metropolitan regions have been a characteristic feature of CEECs.

Notes

The detailed description of the methodology can be found in Quah (1996) or Wójcik (2004).

In case of linear regression for spatial data, the assumption of independence of error terms might not be satisfied, which requires a different approach. Here, regression residuals are spatially correlated for both 179 and 169 regions’ analyses (significant global Moran's I statistic). However, that finding does not change the conclusions—on 5% significance level conclusions concerning convergence (sign and significance of beta parameter) are identical when Spatial Lag Model or Spatial Error Model is used—detailed results available upon request.

Romania was the only country with spatially autocorrelated error terms from beta convergence regression. Using Spatial Lag Model or Spatial Error Model did not change the conclusions about sign and significance of convergence parameters for that country.

We also estimated transition matrices for group borders based on the quintiles of the initial distribution. The results were very much alike.

The results are quite similar and robust regardless of investigated spatial weight matrix. The main difference is the higher value of Moran's I statistics by ca. 0.12 pp in case of GDP dynamics (Models A and B) for matrix elaborated on the basis of first-order contiguity that indicates higher significance of positive spatial autocorrelation. Detailed results are available upon request.