Abstract
This article aims to demonstrate how the industry characteristics of manufacturing sectors affect the patterns of their spatial agglomeration. It also addresses several intricate issues concerning the measurement of localization economies and estimation of their main determinants in manufacturing industries. The original empirical analysis employs annual industrial data from the Hellenic Statistical Authority (ELSTAT) during the period 1993–2006 in Greece at the prefecture level, i.e. for 51 prefectures. The data processing reveals three important findings. The first is the temporal persistence of localization economies in the Greek manufacturing. The second refers to the high level of agglomeration associated with the high-technology industries as well as the resource- and scale-intensive industries. Lastly, there are significant effects of industry characteristics related to knowledge externalities, labour skills and productivity, scale economies and own-transport expenditure on spatial agglomeration, as resulted from the use of alternative geographic concentration indices and panel data models. Results obtained have implications for policy-makers, who can enhance the regional manufacturing activity by affecting these industry-specific factors. Amongst others, planning measures and policies which aim at promoting the local development and regional convergence should focus on reducing transport costs for firms or sectors, by improving the infrastructure capacity, interconnectivity and quality of services.
Notes
The NUTS (Nomenclature of Territorial Units for Statistics) classification is used in the Community legislation for the sub-national division of regions into three levels: Development Regions at NUTS I, Regions at NUTS II and Prefectures at NUTS III.
The use of labour productivity index as proxy for the LM was not found to significantly change the results of the other coefficients and the overall model, compared to the use of the sectoral wage per employee.
One industry (ISIC 30) is dropped from the analysis due to insufficient data.
It is noted that the results obtained from the use of output data lead to quite similar conclusions, although the parameter estimates bear somewhat less statistical significance, compared to those obtained from the use of employment data.
In all the cases, the Hausman test was found to reject the hypothesis of non-systematic difference between the fixed-effects (FE) and random effects (RE) models, based on the asymptotic chi-square (χ2) distribution, thus leading to the adoption of FE panel regression models. For instance, in the case of the LT model with the LGC, that hypothesis was rejected with a χFootnote2(6) = 13.34 and P value = 0.0379.
The panel truncated regression model was found to have superior performance (in terms of the model fit and overall statistical significance) and better convergence behaviour, in comparison to the panel tobit regression model, although their coefficient estimates are quite similar to each other.