Regional technological diversification and the global network of embodied R&D: evidence from the exposure of European regions

ABSTRACT

We investigate whether regions’ participation to the global network of embodied R&D (GNRD) facilitates their technological diversification. Filling a gap about the role of global research and development (R&D) networks, we maintain that by patenting in pivotal GNRD industries, regions become more exposed to global knowledge and increase their capacity to diversify also in technologies less cognitively related to pre-existing ones. Using novel GNRD data, we test this using a panel (2004–19) of NUTS-2 regions for the EU-13. GNRD regional exposure positively correlates with technological diversification, mainly at the intensive margin. A higher exposure makes technological diversification less related to existing technologies, though the relationship is non-linear.

KEYWORDS:

• relatedness
• global innovation networks
• diversification
• revealed technological advantage

JEL:

• O31
• O33
• R11
• R15

ACKNOWLEDGEMENTS

We thank the editor who handled the paper and the anonymous referees for their helpful suggestions. An earlier version of this paper was presented at the: XIX Annual Workshop of the Italian Association of Industrial Economics and Politics (SIEPI) (Virtual), 10-11 June 2021; 18th International Schumpeter Society (ISS) Conference, LUISS Rome (virtual), 8-10 July 2021; 6th Global Conference on Economic Geography, University College Dublin & Trinity College Dublin, 7-10 June 2022; and 6th Geography of Innovation Conference 2022, Universitá Bocconi, Milan, 4-7 July 2022. We are grateful to the participants of these events for their useful comments.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

Notes

1. As will also be shown in section 3, the cognitive proximity between technologies can be measured by looking at the co-occurrence of the relative classification codes within patent documents.

2. By looking at co-inventor and patent citations networks, the regional participation to patent-based global networks can in fact be easily mapped. Indeed, its relationship with regional diversification has been already investigated by recent studies (e.g., Miguelez & Moreno, 2018; Whittle et al., 2020; Balland & Boschma, 2021).

3. Because of the discussed shortage of sufficiently fine-grained data, the construction of an interregional version of the GNRD, whose nodes are region–industries, is to date technically unfeasible. This is due to a lack of required data at the regional level, which regional modellers have already faced in the literature, and for whose solution different approaches have been proposed: such as the development of compensation methods to estimate inter-industry and interregional trade effects and of methodologies to regionalise national input–output coefficients (e.g., Flegg et al., 1995; McCann & Dewhurst, 1998; Spoerri et al., 2007; Bonfiglio, 2009; Kowalewksi, 2015). Still, the application of these approaches to the GNRD is impeded by the lack of fine-grained regional R&D data at the industry level.

4. For extensive treatments of these standard, and other network analysis indicators, see Newman (2003), among others.

5. This precaution has been undertaken to attenuate the potential distortion introduced by inherent patent volatility.

6. Following the previous literature, to mitigate the sensitivity of RTA to sporadic changes in the number of patents, only regions in which the average number of patents over the sample is at least equal to 10 are maintained in the analysis (Santoalha et al., 2021).

7. The analytical definition of the hub and authority score indicators of a network node is reported in Appendix B in the supplemental data online.

8. The stock of FDI is calculated by applying the perpetual inventory method (PIM) to the yearly sum of inward and outward greenfield FDIs located in a given region, using a depreciation rate of 15%. Data on the greenfield cross-border investment projects are extracted from the fDI Markets database.

9. Formally, this amounts to defining the degree of proximity between each technology s and x at t as ${\phi }_{\mathrm{sxt}}=min\{P\left(\mathrm{RT}{A}_{\mathrm{st}}|\mathrm{RT}{A}_{\mathrm{xt}}\right),P\left(\mathrm{RT}{A}_{\mathrm{xt}}|\mathrm{RT}{A}_{\mathrm{st}}\right)\}$. Intuitively, the cognitive proximity between the two technologies is proxied by the frequency with which regions co-specialise in them, by reflecting their reliance on similar capabilities. Since such co-specialisation is not symmetrical, the indicator retains the minimum between: (1) the probability of a region being specialised in technology s conditional on being already specialised in technology x; (2) and the probability of a region being specialised in technology x conditional on being already specialised in technology s.

10. The use of fixed effects, while controlling for unobserved, time-invariant, effects correlated with the error term, allowing the alleviation of potential omitted variable bias, may come at the cost of increasing the measurement error. Though we are confident that this represents a second-order issue, we acknowledge that our estimates could be affected by an attenuation bias.

11. Given the dichotomic nature of the extensive margin diversification variable, in the specification with $\mathrm{Div}E{M}_{\mathrm{rt}}$ linear probability model (LPM) is employed.

12. As expected, and consistently with previous studies, our dependent variables are highly correlated with AvRD, confirming that the proximity between new and existing technologies measured by this variable accounts for an important part of technological diversification, as from the technological branching hypotheses (Montresor & Quatraro, 2017).

13. For further graphical evidence, see Figure C1 in Appendix C in the supplemental data online.