Does the position in the inter-sectoral knowledge space affect the international competitiveness of industries?

ABSTRACT

This paper empirically investigates how the inter-sectoral knowledge flows affect the international competitiveness of industries, once controlling for both cost and other technological factors. Using patent data on 14 manufacturing industries in 16 OECD countries over the period 1995–2009, we apply a network-based approach to capture the effect of industries' position in the flows of technical knowledge across industries, which we label inter-sectoral knowledge space. We find that (i) centrality and local clustering in the inter-sectoral knowledge space positively affect the export market shares of an industry, (ii) such two effects are rather redundant and (iii) national-level knowledge flows' impacts on international competitiveness are way stronger than international ones. Network measures of position in the knowledge space are found to be more relevant than standard technological indicators such as patent counts. Our results point to the importance of industries being well located in the stream of knowledge flows, rather than being innovative per se, and offer a novel yet robust proxy to measure technological factors affecting trade performances. In addition, we find evidence of geographical boundaries of knowledge flows.

KEYWORDS:

• International trade
• industry competitiveness
• knowledge flows
• patent data

JEL CLASSIFICATION:

• D85
• F14
• L60
• O3

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 Network methods have been employed to quantitatively measure the impact of relatendeness on diversification/specialization patterns of countries and regions. Recently, Alshamsi, Pinheiro, and Hidalgo (2018) and Petralia, Balland, and Morrison (2017) provided evidence that the probability of diversification in terms of products, research areas and technologies increases with the number of related activities.

2 The method is the backbone of the so-called ‘Yale-matrix’ that relies on the Canadian Patent Office data.

3 In the EPO data supplementary classes may contain invention information (claimable) and additional information (unclaimable).

4 Formally, Jaffe (1986) employs the so-called cosine index to capture such distance.

5 See Griliches (1998) and Jaffe and Trajtenberg (2002) for a complete treatment of the topic.

6 We will discuss the home market effect in greater detail later in the paper.

7 Moreover, Fagerberg (1997) examined the effect of domestic and foreign R&D on export performance.

8 Along these lines, the interested reader may want to look also at the literature on the role of embeddedness in boosting performances at different levels (e.g. Ahuja 2000; Andersen 2013).

9 PatStat (i.e. EPO Worldwide PATent STATistical Database) is a single patent statistics raw database, held by the European Patent Office (EPO) and developed in cooperation with the World Intellectual Property Organisation (WIPO), the OECD and Eurostat.

10 The timespan for which we collected and analyzed the data stops in 2009. Such choice is driven by the occurrence of the Great Recession, that severely affected all the OECD countries in our dataset.

11 For compatibility reasons our classification is based on ISIC3 codes. The initial NACE2 classification has been converted into ISIC3 by means of standard conversion tables.

12 From patent data we match technology classes (IPC) with industry classes (ISIC3). In particular, we rely on the information on the NACE code associated to patents from the PatStat database (see Van Looy, Vereyen, and Schmoch 2015 for the conversion table IPC-NACE2) and then use the EUROSTAT RAMON conversion tables to move from NACE to the desired ISIC classification employed by the OECD STAN database.

13 Investigations of citation patterns in our dataset show a clear tendency of a country-specific dimension. See Figure A2 in the Appendix.

14 The distinction between national and international measures is not a matter of differences among countries/industries, it rather concerns the nature of co-occurrence and citation data. Using co-occurrences, we are not able to disentangle, and thus to count in a meaningful way, every IPC-country link. We overcome such difficulty by relying on citations, which include, in a way, an additional layer of information to map within-country industry relationships as well as across-country industry linkages.

15 $c\in \{AT,\dots ,US\}\equiv \mathcal{C}$ and $t\in \{1995,\dots ,2009\}\equiv \mathcal{T}$

16 For completeness, in the Appendix we include the unweighted degree centrality and the eigenvector centrality.

17 Figure A1 in the Appendix shows how our measures (including the ones derived from national citation networks) correlate with each other.