Technological novelty and key enabling technologies: evidence from European regions

ABSTRACT

This paper deals with the determinants of technological novelty at the local level and, in particular, with the impact of the local endowment of key enabling technologies (KETs). Looking at local innovations as recombinations of pre-existing knowledge, we argue that the local endowment of KETs facilitates regions introducing inventions with novel technological origins, either in absolute or in relative terms. We test for this argument by focusing on a sample of 1,255 EU (NUTS3) regions over the 2000–2014 period and propose an original instrumental variable strategy that allows us to maintain the local endowment of KETs as exogenous. The results confirm our main hypotheses. In particular, a 1% increase in KETs increases by ∼1.8% the number of novel patents generated at the local level. KETs therefore appear to be ‘enabling’ of technological innovations that are unique in recombination across the board. However, KETs promote ‘new-to-the-region’ innovations more than ‘new-to-the-world’ innovations, representing a policy leverage to which regions could resort in targeting the local replication of technological advancements already present at the global frontier for some time.

KEYWORDS:

• Technological novelty
• knowledge recombination
• regional novelty
• key enabling technologies (KETs)

JEL codes:

• R11
• R58
• O31
• O33

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 See https://knowledge4policy.ec.europa.eu/publication/key-enabling-technologies-kets-observatory_en.

2 For a recent survey on the use of patent citation data in social science research, see Jaffe and de Rassenfosse (2017). For a more critical view of the implications of patent citations, see also Kuhn, Younge, and Marco (2020).

3 More precisely, patents marked by novelty in terms of technological origins are identified in the following way: “We construct ‘backward citation pairs’ of IPC codes, i.e., combinations between distinct IPC codes from, on the one hand, all patents cited by the focal patent and, on the other hand, all distinct IPC codes the focal patent belongs to. We compare each of the focal patents’ ‘backward citation pairs’ to all citation pairs previously used to assess whether a certain pair is new (has never occurred before)” (Verhoeven, Bakker, and Veugelers 2016, 711).

4 As Rigby (2015) recognizes, the relatedness that the co-occurrence of technological classes in local patents reveals could be due to “unspecified economic relationships that display positive spatial autocorrelation”; similarly, the absence of these economic relationships could mask the absence of the relatedness they entail with actually spurious radical inventions. This is in line with what Balland (2016) also recognizes in observing that the “co-production of knowledge can capture much more than knowledge relatedness understood as a reflection of cognitive proximity between organisations. … [and] reflect[s] the need for similar institutions, infrastructure, physical factors, technology or a combination of these factors. So, using such an outcome-based measure of relatedness for knowledge domains will not necessarily capture scientific or technological relatedness, but probably much more factors that lead to the co-production of knowledge domains” (p. 132).

5 We exploit the International Patent Classification (IPC) and consider 4-digit IPC classes.

6 It is worth noting that a novel combination may appear in more than one patent, as well as in more than one region in the same year. We take the patent priority year as a time reference to assign patents to local areas.

7 We check the robustness of our estimates to the inclusion of KETs patents when measuring novel recombination. Results, available upon request, fully confirms the main findings of our analysis.

8 In spite of the debate surrounding the pros and cons of choosing the inventor rather than the applicant address (see, for example, Santoalha 2019), this is widely considered as a good approximation of local innovative outcomes.

9 For example, if a patent $p$ is filed by three inventors resident in two distinct NUTS3 regions (let’s say one inventor resides in region $A$ and two inventors reside in region $B$), $p$ weighs one third for region $A$ and two thirds for region $B$. This information is provided by the Regpat database.

10 As anticipated, in our analysis we consider KETs as a whole, by pooling the endowment of their six constitutive technologies at the regional level. Considering them separately would be for sure interesting, but goes beyond the scope of this paper. Accordingly, investigating KETs heterogeneity in driving technological novelty at the regional level represents a promising line of future research.

11 See Marco et al. (2015) and Graham, Marco, and Myers (2018) for a precise description of the data on USPTO patent transfers.

12 US buyers are companies with a registered address is in the US, as per the information reported in PAD. Similarly, US-invented patents are patents whose inventors reside in the US, again as per the information reported in PAD and in PatentsView.

13 Our definition of innovative company is broad. We define as innovative companies that filed at least one patent before acquiring a KET patent. We exclude companies that had already filed a KET patent before acquiring a KET patent to minimize cases of merely strategic patent acquisitions within the KETs domain.

14 This further restriction serves the goal of minimizing the risk of capturing confounding factors that drive technological acquisitions taking place between technologies possibly not related to KETs.

15 We assign sectors to patent applicants using the concordance tables proposed by Lybbert and Zola (2014). Specifically, we merge 4-digit IPC classes contained in US patents with 2-digit NAICS sectors. Each patent applicant in year t is assigned to the NAICS 2-digit sector that includes the largest number of IPCs contained in its patents filed in that year (or in the last year in which it patented).

16 We run several robustness checks, moving the starting year of exposure back to 1990 and 1985. Results are fully consistent with those reported in the text and are available upon request.

17 We apply the perpetual inventory method to calculate $SNOKETS$, with a decay rate of 15%. Formally, $SNOKET{S}_{r,t}=NOKET{S}_{r,t}+\left(1-d\right)\times SNOKET{S}_{r,t-1}$, where $d$ is the decay rate and $NOKET{S}_{r,t}$is the number of non-KET patents filed by inventors resident in region $r$ at time $t$.

18 The index measures the degree of disorder (or randomness) of the regional knowledge base looking at the probability of co-occurrence of patent technological classes contained in local patents. Technological class j is linked to class m whenever they co-occur within the same patent. The higher the number of patents showing co-occurrence of j and m, the stronger the link.

19 In order to deal with nil values of KETs and of other non-binary variables in the log transformation, we have opted for the inverse hyperbolic sine transformation, which allows us to not lose any zeros in the variables. With respect to a generic variable y, this transformation is defined by the following formula: $\mathrm{i}\mathrm{n}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{s}\mathrm{e}\mathrm{\_}y=log\left[y_i+{\left(y_i^2+1\right)}^{1/2}\right]$. Except for very small values of y, the inverse sine can be interpreted as a standard logarithmic variable. However, unlike a logarithmic variable the inverse hyperbolic sine is defined at zero as well (Burbidge, Magee, and Robb 1988; MacKinnon and Magee 1990).

20 More precisely, we apply Poisson fixed effects and the fixed effects Poisson quasiMLE proposed by Lin and Wooldridge (2019), which need to be properly adapted to accommodate our IV strategy.

21 The coefficients of the control variables are not reported in panel A but are available from the authors upon request.

22 A drawback of the data is that the federal recording of a change in ownership (entire or partial) is not mandatory. However, both patent statute and federal regulations provide some incentive for recording this. For a discussion of the requirements for assignment recording, see Marco et al. (2015).

23 We do not consider mergers and acquisitions. Moreover, we exclude all first assignments that refer to inventor-applicant formal assignment cases. All further assignments referring to “change of name”, “change of address”, etc. have been excluded as well.

Additional information

Funding

This work was supported by Ministero dell'Istruzione, dell'Università e della Ricerca: [Grant Number PRIN 20177J2LS9].