![Sample our Geography journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5937/TOC-Subject-Sample-2021-Geography.jpg"})
Abstract
This article adds to the empirical evidence on the impact of agglomeration externalities on regional growth along three main dimensions. On the basis of data on 259 Europe NUTS2 (Nomenclature of Territorial Units for Statistics) regions and 15 NACE (Nomenclature statistique des Activités économiques dans la Communauté Européenne) 1.1 2-digit industries for the period 1990–2007, we show that agglomeration externalities are stronger in technology-intensive industries, also after controlling for sorting; that specialization externalities are stronger for low density regions, while diversity matters more for denser urban areas; and, finally, that Jacobs externalities comprise a pure diversification effect (related variety) and a portfolio effect (unrelated variety), although evidence of positive effects on regional growth is only found for the latter. An additional contribution of this article is to extend the analysis on the basis of a full geographical coverage of European NUTS2 regions, with the aim to generalize the empirical identification of the impacts of specialization and diversification externalities with respect to the existing literature. Our results are robust to a rich set of consistency checks, including the use of spatial autoregressive models with autoregressive disturbances, used to assess to what extent the effects of agglomeration externalities are localized.
Acknowledgments
The views expressed are purely those of the author and may not in any circumstances be regarded as stating any official position of the European Commission. Laura de Dominicis acknowledges financial support from the European Union’s Sixth Framework Programme for Research and Technological Development. We are grateful to the reviewers and handling editor of this journal for highly useful and constructive comments on earlier versions of this article. The usual disclaimer applies.
Notes
1 A recent research avenue on the growth-enhancing effects due to related and unrelated variety is represented by Cainelli and Iacobucci (Citation2012), where the focus is on interindustrial vertical integration and the effects on vertical integration of unrelated and vertically related variety at the local level.
2 Our study complements the two in covering the full set of EU27 regions (unlike Dogaru, Van Oort and Thissen Citation2011), and explaining labor growth (unlike Marrocu et al. Citation2013).
3 ‘The NUTS classification (Nomenclature of territorial units for statistics) is a hierarchical system for dividing up the economic territory of the EU’ (EUROSTAT Citation2015).
4 NACE is the Statistical Classification of Economic Activities in the European Community. It can be compared to the SIC (Standard Industrial Classification) and NAICS (North American Industry Classification System) systems, in use in the US and UK.
5 This process works also the other way around: workers get together in search of better access to diversified job posts, and firms tend to concentrate where a large pool of human resources is available.
6 In the absence of EU27-wide data on five-digit industries, data at the most detailed level of industrial disaggregation are available for two-digit industries.
7 In particular, industries are ranked in decreasing order of employment share, and the sum of the employment shares of the second through the sixth industry are calculated. Finally, the complement to one of these sums is calculated.
8 A full list of sources for the raw data is shown in Table S1 in Appendix S1.
9 Data on the number of local units active in the regions industries have been collected, with several missing values, within EUROSTAT’s structural business statistics. Such data are used in order to run a consistency check testing whether the results found with our chosen indicator of Porter externalities also hold when a different set of raw data is used (see Appendix S1 for further details).
10 A possible limit to this approach is that our data do not offer a detailed industrial classification. A breakdown of employment data by 15 NACE rev. 1.1 two-digit industries is employed here. This issue may represent a further improvement of our work.
11 The estimated parameters are robust to the choice of different indicators for all three measures of externalities. A detailed account of the different consistency checks performed to support this statement is provided in Supplemental data.
12 An increasing body of literature has emphasized structural differences across industries in the extent to which agglomeration externalities exert their effects. We thank an anonymous referee for pointing out this issue, and we believe this to be a very promising future research avenue.
13 R&D, engineering and computing are typical science and technology–based service sectors, being major propagators and diffusers of technological knowledge both within the service sector and in the manufacturing industry. Transport and communications and the financial industry represented in the 1990s and early 2000s an engine for growth in advanced countries. Firms in this industry devote large financial resources to innovation, and a large share of them is used to carry out R&D and design activities. The innovation process in this sector is also characterized by the presence of close interactions with both final customers, consultancy firms, as well as with private research institutes. This sector provides customized answers to a variety of technical needs and requirements of clients, exploiting the technologies available in the market and in the broader science and technology system.
14 Strictly speaking, our estimates do not control for heterogeneity at the individual level, which is a focal point in the recent sorting literature.
15 For instance, Pohjois-Suomi in Finland and Övre Norrland in Sweden, taken together, are roughly equal to the size of Italy.
16 Using data on employment dynamics in a number of Local Labor Systems (LLS) in Italy over the period 1991–2001, Paci and Usai (Citation2008) find evidence of remarkable differences in the influence of local specific externalities between manufacturing and services.
17 The remaining sectors, including agriculture and fishing and forestry, and the construction and utilities sectors, do not enter this last analysis.
18 By convention, the diagonal elements of a weight matrix are set equal to zero.
19 When the spatial interaction is determined by factors linked to economic variables, authors have proposed the use of weights with a more direct relation to the particular phenomenon under study (i.e., travel time, social or economic distances). It is important to keep in mind that the standard estimation and testing approaches assume the weight matrix to be exogenous. Therefore, indicators for the socioeconomic weights should be chosen with great care to ensure the exogeneity, unless their endogeneity is considered explicitly in the model specification (Anselin and Bera Citation1998).
20 In Europe, as pointed out in Ertur and Le Gallo (Citation2003), regions have on average five to six contiguous neighbors. A choice of k = 7 yields a ring around each region of approximately the first and second order contiguity and moreover connects the United Kingdom as well as some islands, such as Sicily, Sardinia, and Baleares, to continental Europe, so that the block-diagonal structure of the simple contiguity matrix is avoided. In order to normalize the outside influence upon each country, the weight matrix is row-standardized. An alternative spatial weight matrix with k = 5 has been used as a robustness check. Where results are not reported in the article, they are available from the authors upon request.
21 The authors would like to thank Raymond Florax and Vanessa Daniel for providing the ©R code for the estimation of the SARAR (1,1) model.
22 As a robustness check, Tables S5 and S6 report the results for an alternative spatial weight matrix based on k-nearest neighbors, with k = 5. The complete set of estimations (OLS, SARAR) for the k-nearest neighbors spatial weight matrix with k = 5 is available from the authors upon request.
23 The decomposition of the impacts (direct, indirect, and total) for the remaining explanatory variables are available from the authors upon request.
24 Interesting to note – and consistent with the findings in this article – is that in an update of their earlier meta-analysis, De Groot et al. (Citation2015) find evidence that specialization is more important in lower-density places and that more recent studies are less supportive of the importance of diversity externalities.