Intra- and inter-regional knowledge spillovers: Evidence from European regions

Abstract

Is the knowledge creation process linear or characterized by feedback relations among actors involved in the regional innovation system? How can the innovation process of ‘lagging’ regions be strengthened? What is the role and extent of inter-regional knowledge spillovers? The paper aims at providing satisfactory answers in investigating a knowledge production function framework adapted to the specific questions and which is tested on an extended sample of European regions. On the basis of the results, concrete policy measures are derived aiming at upgrading the knowledge creation capacity of European regions.

Acknowledgements

The author wishes to thank Henri Capron and anonymous referees for their useful comments and suggestions.

Notes

1. The foundation for modern theories of innovation go back to Schumpeter (1928, 1935, 1939, 1954) who conceived economic development as the creation of new combinations of productive means which give rise to new products, new processes but also to new markets and organizational forms.

2. The extent of intra-regional knowledge spillovers or externalities also depends on the organization or the composition of the region's underlying productive system in terms of specialization and/or diversity (Feldman & Audretsch, 1999; Paci & Usai, 1999; Greunz, 2003, Greunz, 2004b). This aspect is not considered here.

3. In the related literature spillovers and externalities are characterized as pecuniary/non-pecuniary, pure/impure, static/dynamic, etc. Pecuniary externalities come close to the concept of rent spillovers (Griliches, 1979) while pure externalities strongly reflect the concept of knowledge spillovers. This paper concentrates on knowledge spillovers such as defined by Griliches (1979). For reasons of limitations in terms of data availability a dynamic approach in the spirit of Glaeser et al. (1992) could not be adopted.

4. However, it should be noted that patent applications as a proxy for innovations have some shortcomings. For several reasons, not all innovations are subject to patent applications. A strategy of secrecy in order to avoid disclosure of new knowledge may be preferred. The high cost associated with patent applications may discourage especially small firms to apply. Moreover the propensity to patent differs across industrial activities and so does the economic value of patents (for a review of patent literature see Griliches, 1990). Despite these shortcomings there is a strong link between patents and innovations (Acs et al., 2002). In any case, patent applications are the only available harmonized innovation measure at the European regional level.

5. For an overview of the effects of different government policy instruments on private R&D see David et al. (2000).

6. Varga (1998) provides a detailed overview of the literature with regard to university research and regional innovation.

7. The authors also investigate the system for 43 US states in using different measures for the ‘geographic coincidence index’ that aims at correcting for the relative inappropriateness of states for this kind of analysis.

8. Despite the fact that technological progress in recent years is increasingly generated by the service sector, industrial patenting activity is still predominant.

9. The system of equations has been estimated by full information maximum likelihood, two stage least squares and three stage least squares. The estimated coefficients obtained by the three estimation methods are statistically similar. The model has also been estimated allowing for different time lags. It appears that patent applications (equation A) are best explained by contemporaneous business R&D expenditures together with university R&D realized 3 years previous to the application. As far as business R&D is concerned (equation B), contemporaneous university R&D and government R&D with a time lag of 5 years is the best fitting specification. Finally for the university R&D (equation C), the best fitting formulation is the one which considers as explanatory variables contemporaneous business R&D and tow years lagged government R&D expenditures. For more details see Greunz (2002, 2003b).

10. See the appendix for a detailed description of the regional sample.

11. Formally the weight attributed to the R&D expenditure of region j when investigating patenting activity of region i is given by where d is the distance that separates region i and j. The choice of the applied distance decay parameter is somewhat subjective. A larger decay parameter would lead to a sharper distance decay and possibly to a more restricted set of neighbours with appreciable weight. However, similar empirical outcomes were obtained for different decay parameters.

12. The measure is defined in the following way: where p _ir is the measure of technological closeness and f _ik is the share of a particular patent class k in the total of patents of region i. If the technological profiles of region i and r are similar, p _ir is close to one. Conversely, the more the regions are technologically different, the closer p _ir is to zero.

13. Econometric estimations are performed by means of a panel data estimation method that allows for random effects. Given that that we wish to control for the qualification level of the working age population for which information is only available since 1997 the fixed effect panel data method is not appropriate. In order to determine whether the common or the random effect model should be applied, the Lagrange multiplier test derived by Breusch and Pagan (1980) has been performed which is in favour of the random effect model.

14. 1st-order geographical neighbours are the set of regions with which the ‘home’ region shares a common border. The median distance separating 1st-order geographical neighbours from the ‘home’ region is about 147 km.

15. 2nd-order geographical neighbours are the set of regions which are direct neighbours of the ‘home’ region's 1st-order geographical neighbours. The median distance separating 2nd-order geographical neighbours from the ‘home’ region is about 284 km.

16. 3rd–order geographical neighbours are the set of regions which are direct neighbours of the ‘home’ region's 2nd-order geographical neighbours. The median distance separating 3rd-order geographical neighbours from the ‘home’ region is about 400 km.

17. A pertinent question that naturally arises when considering that knowledge spillovers occur within a distance of 400 km concerns the underlying mechanisms. Despite the fact that this distance seems relatively important, it is likely that the mechanism is still face-to-face interaction. Indeed, in the European core where innovative activity is most developed, the transport system is sufficiently well developed to manage a 400 km round-trip within 1 day. This assumption can be formally tested using travel time distances between regions instead of geographical ones and will be tackled in future research.

18. This reflection is strongly confirmed when computing for each individual region the geographical distance with its 1st- and 2nd-order technological neighbours. While the average geographical distance for each pair of regions contained in the sample is 1242 km, the average distance that separates 1st- and 2nd-order technological neighbours from the ‘home’ economy is respectively 895 and 817 km.

19. MAR externalities arise when firms active in the same industry cluster geographically. In opposite to Jacobs (1969) externalities, it is assumed that only firms active in the same industry are able to internalize these externalities. However, the measure in this study of technological proximity does not allow inferring on this issue since a given technology can be exploited by quite different industrial sectors. Theoretically it is even possible that two regions exhibit a similar technological profile but are specialized in very different industries.

20. Compared to the previously estimated elasticities relative to total R&D for the geographical space (Figure 2), the effects obtained when splitting up total R&D may appear to be relatively limited. As shown, the innovation process is neither linear nor additive but characterized by feedback relations among institutional sectors. Since, feedback relations between ‘home’ and neighbourhood's university, business and government R&D efforts are not explicitly modelled by means of a simultaneous equations framework, it is quite logical that the estimated elasticities of institutional sector's inter-regional knowledge spillovers turn out to be lower.

21. When the model is estimated, assuming that R&D efforts instantaneously turn into innovations, the results are remarkably similar to the ones obtained by Anselin et al. (1997) for their ‘full’ model (p. 436, Table 2). University R&D expenditures realized in the 1st-order geographical neighbourhood significantly contribute to the ‘home’ knowledge creation but business R&D expenditures realized by the 1st- and the 2nd-order geographical neighbourhoods do not influence ‘home’ patenting activity.

22. The reduced form parameters are not reported.

23. This two-step procedure enables the following problems identified by the European Commissionto be addressed: “Too often border programmes were developed in parallel and in accordance with a national perspective and were then presented to the European Commission jointly with a neighbouring country. Quite often, therefore, these are only national projects related to border problems which were jointly adopted in the Monitoring Committees. Moreover, national border projects are bundled together in a sort of ‘package’, ‘added up’ and declared to be a cross-border measure” (European Commission, 2000, p. 3).