Innovation and skill premium

ABSTRACT

The relationship between innovation and skill premium is analysed on a panel of 12 countries for a 16-year span (2000–2015). According to a Schumpeterian view, a non-linear relationship between innovation and skill premium is found showing a threshold effect that reverses the relationship for relatively high levels of innovative activity. Moreover, the relationships change from convex to concave when variables representing different types of innovative activity are considered. In fact, with R&D a positive relationship with skill premium reverses once a threshold is exceeded, while the opposite holds for patents, for which the relationship is initially negative and then becomes positive. We argue that this is due to the different degrees of appropriability of the knowledge produced by innovators with these activities. We then show how to exploit these different patterns to provide a truly innovation-based analysis of the patterns of skill premium for the United States, France, Germany and Great Britain.

KEYWORDS:

• Innovation
• skill premium

JEL CLASSIFICATION:

• O33
• D63
• J24

Acknowledgments

The authors are grateful to the participants for their useful remarks and suggestions. The authors are very grateful to the Editor and two anonymous referees for their very insightful comments and suggestions that contributed to significantly improve the quality of the paper. The usual caveats apply. One of the authors (RL) started to work on this project when he was Senior Fellow and Marie Curie Fellow of the EU at the Freiburg Institute of Advanced Studies.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 ‘This is perhaps 5% empirical information and 95% speculation, some of it possibly tainted by wishful thinking’ (Kuznets 1955, 26).

2 It is also worth stressing that task offshoring can have similar restructuring effects to those stemming from the extensive introduction of computers (Grossman and Rossi-Hansberg 2008).

3 See, among the others, Aghion, Caroli, and Garcìa-Penãlosa (1999), Nolan, Richiardi, and Valenzuela (2019) and Chusseau, Dumont, and Hellier (2008).

4 This is confirmed by the recent explosion of the strategic use of patents (Hall 2004), which are put in place for reasons that are far from the protection of knowledge as a public good. See, for instance, Cohen, Gurun, and Kominers (2016) about the so-called patent trolls.

5 The release used is the following: EU KLEMS Growth and Productivity Accounts: Statistical Module, ESA 2010 and ISIC Rev. 4 industry classification – September 2017 release, Revised July 2018.

6 Data concerning patents are retrieved, as stated, from the WIPO Databank and follow standard imputation procedures outlined by, among the others, the World Bank Open Databank. According to these procedures, patent grants are allocated across time based on application date and to countries following standard fractionalised counting imputation techniques. Patent data used in this analysis cover worldwide patent applications filed through the Patent Cooperation Treaty procedure or with a national/regional patent office.

7 These data have been exported from OECD's SDBS Structural Business Statistics online data-warehouse.

8 GDP, national account figures and other monetary measures included in our model are retrieved either from the OECD and/or the World Bank databank and are expressed in millions of dollars PPP constant prices, the base year is 2010.

9 Data on trade are from the World Development Indicators, available on the World Bank Open Databank.

10 The data on the skill premium used in this paragraph differ from those used in the previous one. Indeed, on the one side, for France, Germany and Great Britain we used the same source as before (i.e. the EU KLEMS Growth and Productivity Accounts), but we exploited the availability of a longer series, which is taken from the March 2008 Release of that same dataset. On the other side, for the US we used the data taken from the publicly available Acemoglu dataset. This obtains two results: (i) we have a longer series; and (ii) we can make a comparison with the Acemoglu results.