Merging macroeconomic and territorial determinants of regional growth: the MASST4 model

The MASST model bears important similarities w.r.t. the GMR model by Varga and coauthors, in that both strive to integrate a macroeconomic and a regional sub-model. The main difference between these two approaches lies in the forecasting nature of the MASST model, against the GMR’s aim to assess policy impacts. See, e.g., Varga et al. (2020) for a recent discussion of the state-of-the-art of the GMR model.

NUTS (Nomenclature of Territorial Units for Statistics) is the European Union’s classification of regional units, comprising a hierarchical structure form Countries (NUTS0) to the rough equivalent of US counties (NUTS3) (EUROSTAT 2016a).

For a comprehensive list of endogenous and exogenous variables, see Technical “Appendix 1.1”.

The regional sectoral input–output table has been made available by the JRC group responsible for the Rhomolo model in Seville. We would like to thank the colleagues in Seville for sharing their input–output matrix.

The spatial connectivity definition adopted is based on simple geodesic distance between centroids.

The alphabetical order with which industries are listed also (imperfectly) matches the increasing technological complexity of the involved manufacturing activities, as also testified by the high-tech and medium–high-tech reclassifications of manufacturing activities discussed in EUROSTAT (2018) and regularly updated on the EUROSTAT data base.

A detailed definition of how agglomerated, urban, and rural regions are defined according to the ESPON 1.1.1 project is provided in the Technical “Appendix 1”.

The estimates here presented are based on pooled OLS. Still, given the panel structure of our regional data set, the GMM would be preferable (Belotti et al. 2017). However, GMM estimates require a perfectly balanced panel data set, a condition that is unfortunately not met by our data. As a robustness check, GMM estimates have also be obtained with the same spatial weights matrix adopted in the pooled OLS regression, and this alternative specification required us to restrict the data set to the 128 regions for which we have a full-fledged panel data set. Results, available upon request, qualitatively confirm the main messages shown in Table 5, although some parameters become no longer significant, due (probably) to the major loss of information from the missing data. The main difference lies in the estimated spatial spillover effect for Western regions, that now becomes positive significant. Still, this comes at a major loss of information from previous period observations. Also, the missing observations prevent us from using this specification in the simulation stage of the MASST model that needs full data availability for all NUTS2 regions in the simulation data base, from which lever variables are moved to their simulation targets.

Results are not reported; they are available upon request from the authors.

In Capello and Lenzi (2017), the unit of observation is the NUTS2 region in the 2010 version.

In Capello and Lenzi (2017), this is shown in Column 5 in Table 2, p. 7.