Geographical distance puzzle in patent citations: intensive versus extensive margins

ABSTRACT

This paper explores the effects of geographical distance on knowledge spillovers through patent citations across 270 European regions. Despite decreasing transport and communication costs, geographical distance effects are strong and not decreasing. To address this distance puzzle, we distinguish between the extensive margin (the number of cited technologies) and the intensive margin (the average number of citations per technology) of patent citation flows. We confirm an increasing distance effect on knowledge flows at the extensive margin. We show it is compatible with decreasing transport and communication costs.

KEYWORDS:

• Knowledge flows
• distance puzzle
• extensive margins
• intensive margins
• patent citations

JEL CLASSIFICATION:

• O30
• R12
• R15

ACKNOWLEDGMENT

We thank Lorenzo Zirulia for his helpful comments and suggestions.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ Following Jaffe, Trajtenberg, and Henderson (1993), the analyses of the diffusion of knowledge in geographical space mainly use patent citations data. In parallel to this substantial literature, there is a methodological debate on the measurement of geographic localization in patent citations (see, e.g.: Thompson and Fox-Kean 2005). With this regard, several studies suggest that patent citations are valid but noisy indicator of knowledge flows (Jaffe and de Rassenfosse 2017; Corsino, Mariani, and Torrisi 2019).

² http://www.wipo.int/classifications/ipc/en/.

³ For patents with multiple inventors residing in the same region (citing or cited), citations are counted only once. Similarly, for patents with multiple technologies in the same IPC 4-digit class, citations are counted only once. Self-citations within firms and between inventors, as a tradition in the literature, are excluded.

⁴ $Distanc{e}_{is}$ is calculated using the great circle distance method based on the geographical coordinates of the centre point of the regions (see, e.g. Cappelli and Montobbio 2016).

⁵ Santos Silva and Tenreyro (2011) show that the PPML estimator performs well also when the sample has a large proportion of zeros. Moreover, as a robustness check, we perform zero-inflated negative binomial regressions (ZINB) on the base-line gravity model, i.e. a model with the inventive mass of the citing and cited regions ($\mathrm{P}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{n}{\mathrm{t}}_{\mathrm{i}\mathrm{t}}$ and $\mathrm{P}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{n}{\mathrm{t}}_{\mathrm{s}\mathrm{t}}$) and the geographical distance between the citing and cited regions ($\mathrm{D}\mathrm{i}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{c}{\mathrm{e}}_{\mathrm{i}\mathrm{s}}$). The results, not reported here, but available from the authors upon request, are similar.

⁶ According to these temporal windows, the variables $Paten{t}_{st}$ and Technology_ist are computed referring to the patents developed by region s during the period from (t – 5) to t.

⁷ Bacchiocchi and Montobbio (2010) show that the average modal citation lag for the EPO patents is 2.6 years. As a robustness check, we perform analysis using a fixed 10-years time window between the citing and cited patents. The results, not reported here, but available from the authors upon request, are very similar..