Industrial Districts, Innovation and I-district Effect: Territory or Industrial Specialization?

Abstract

The I-district effect hypothesis establishes the existence of highly intense innovation in Marshallian industrial districts due to the presence of external localization economies. However, industrial districts are characterized by specific manufacturing specializations in such a way that this effect could be due to these dominant specializations. The objective of this research is to test whether the effect is explained by the conditions of the territory or by the industrial specialization and to provide additional evidence of the existence and causes of the highly intense innovation in industrial districts (I-district effect). The estimates for Spain of a fixed-effects model interacting territory and industry suggest that the high innovative performance of industrial districts is maintained across sectors, whereas the industrial specialization behaves differently depending on the type of the local production system in which it is placed. The I-district effect is related to the conditions of the territory more than to the industrial specialization. The territory is a key variable in explaining the processes of innovation and should be considered a basic dimension in the design of innovation and industrial policies.

Acknowledgements

The authors wish to thank Marco Bellandi and two anonymous referees for helpful comments. We also thank the Spanish Patent and Trade Mark Office (Ministry of Industry, Tourism and Trade), the Centre for the Development of Industrial Technology (CDTI) and the Spanish Institute of Statistics (INE) for providing most of the data used in this research.

Notes

Precise compared results are provided by Boix and Galletto (2008). Detailed and comparable maps for Italy, Spain and the UK can be found in A Handbook of Industrial Districts (Becattini et al., 2009).

The difference is noted between the “district effect” (productivity/efficiency, competitiveness and innovation) and other “characteristics of districts” such as the degree of vertical integration, smaller size of establishments or a premium on wages.

Innovations affect not only static efficiency reducing costs but also dynamic efficiency since they allow for changes and improvements in products and their introduction into markets.

Bagella and Becchetti (2000) propose a theoretical model based on a game where proximity reduces the appropriability of knowledge, positively affects the imitative capacity of firms and fosters knowledge spillovers from firms with R&D expenditures to other firms in the neighbourhood. As a result, the expenditure on R&D of individual firms and aggregated R&D effort is lower in industrial districts, although other forms of technological innovation take on the same role.

The complete patent database includes 70,000 documents from 1991 to 2006. Patent counts include “utility models”, a figure granted by the OEPM, which is similar to the patent, although legal requirements are less strict and protection covers only 10 years. Similar figures exist in Austria, Denmark, Finland, Germany, Greece, Italy, Japan, Poland and Portugal. Employment data come from the 2001 Census of the Spanish Institute of Statistics (INE). Patent data do not differentiate between incremental and radical innovation so that this aspect has been not addressed in the research.

Data treatment follows international standards: patents are located according to the first applicant with an address in Spain (inventor's address is not available for national patents); reference date is the oldest application data in any register because it is the closest to the invention date and does not introduce biases due to legal or procedural delays.

Boix and Galletto (2009) provide a detailed explanation of the procedure to divide the 806 local labour markets by typology of LPS and dominant industry. Boix and Galletto (2008) and Boix (2009) describe thoroughly the elaboration of the map of industrial districts in Spain, following the Sforzi-ISTAT (2006) procedure.

This was indeed expected because the procedure to divide the LPS by typology is based on the characteristics of the industry in the territory.

In a posterior step, Boix and Galletto (2009) relate the fixed effects to the existence of external economies: δ* = f(Z_j ).

Although the interactive form is preferable in this context, the two-ways fixed-effect model and two-ways fixed effects with interactive term can be tested. Anticipating the results of the next section with the data used in this research, the comparison statistics (Akaike and Schwartz) favours simple models (fixed effects by typology of LPS, fixed effects by dominant industry and interactive effect) against combined two fixed effects and two fixed effects plus interactive effects, which produce a large number of non-significant coefficients.

The reported estimates refer to business R&D expenditures from micro-data, which are considered to be more precise. Fixed effects reported using R&D assigned from regional data are very similar.

The subsequent study of the separated effects (Table 5, central part) shows the internal heterogeneity of this group, mainly composed by micro-LPS, and suggests a cautious interpretation of the averaged effect.

In leather and footwear, the territorial typology and specialization are basically the same because only two specialized LPSs are not industrial districts.