Competition and Persistence of R&D

Abstract

This paper investigates the R&D persistence of R&D active firms in different markets with different intensities of competition, based on firm-level panel data for the period 1996–2008. In a dynamic setting of the empirical model, it turns out that persistence is strongly related to market competition (measured by the number of principal competitors). Persistence of R&D expenditures is more likely to be observed in markets with few principal competitors (between six and 10) and is very unlikely to be observed in polypolistic market types (more than 50 competitors).

Keywords:

• R&D expenditures
• Persistence
• Innovation
• Competition
• Panel data
• Empirical investigation

JEL Code:

• O30

Acknowledgements

I am grateful to insightful comments from Douglas R. Nelson, Peter Egger, and Spyros Arvanitis, from the participants of the ‘Jena Economic Research Seminar’ held in Jena, December 2010, as well as from the participants of the ETH Zurich D-MTEC Seminar (October 2011). I am also grateful for the valuable input from the participants of the IIOC (International Industrial Organization Conference) 2012 in Arlington (VA) and the comments and suggestions from the participants of the Workshop on Innovation Persistence held in Turin, March 2013, especially from C. Antonelli and C. Les Bas. Mistakes are author's alone.

Notes

1. Latham and Le Bas (2006) looked also at patent data and extended the investigation of persistence to the level of individuals.

2. Studies that are based on the CIS (Community Innovation Survey) apply the following definition: An innovation is the introduction of a new or significantly improved product, process, organizational method, or marketing method by an enterprise. An innovation must have characteristics or intended uses that are new or which provide a significant improvement over what was previously used or sold by an enterprise. However, an innovation can fail or take time to prove itself. An innovation need only be new or significantly improved by an enterprise. It can have been originally developed or used by other enterprises.

3. Measured in the ordinal scale.

4. Antonelli et al. (2013) investigated the effects of external factors on long-term persistence.

5. They found significant results only when they used patents as a proxy for innovation activities.

6. ‘In general, the greater the number of competitors the greater is the incentive to act earlier rather than wait’ (Wernerfelt and Karnani 1987, p. 190).

7. It is clear that most of the papers referred to in the theoretical notions are based on static models but persistence is a dynamic concept. However, the investigation at hand does not test those theories; the latter simply provides the background for formulating a dynamic hypothesis.

8. According to Boone (2001, 2008), the intensity of competition is not necessarily related to the number of competitors, the price–cost margin of firms, or the Herfindahl index. However, in our survey we asked the firms to indicate the number of principal competitors in their main sales markets worldwide. Hence, we can assume that all firms within the respective markets have very similar marginal costs. It is therefore rather safe to assume that an increase in the number of principal competitors does indeed increase competition and lowers the total industry profits. If firms had very different marginal cost levels, the rise in competition would be likely to reallocate output from low price–cost margin firms or inefficient firms to more efficient firms (see also Boone 2001, 2008).

9. Questionnaires are available in German, Italian, and French.

10. For the econometric estimations we could only use firms that answered in 1996 and in two consecutive surveys, i.e. 1999 and 2002, 2002 and 2005, or 2005 and 2008, since the estimations also include the lagged dependent variable.

11. Frequently used measures of competition is the Herfindahl index. However, this type of measure has the disadvantage that it implies identifying the relevant market referring to the industry classification. For a small, open economy like Switzerland with many exporting and technologically advanced firms, the industry in the home country is clearly not the relevant market. Furthermore, survey data are subject to non-response. Thus, if, for example, very large firms in an industry did not answer the questionnaire in one cross-section, a Herfindahl type of measure would react very sensitive to such a non-response.

12. Since this information is available only for one-cross section, it is impossible to use these competition measures in a dynamic model.

13. Clearly, industry dummies are a very broad control for industry-level information like the nature of technology; especially, if we consider the findings of Levin et al. 1985; Breschi, Malerba, and Orsenigo 2000; Sutton 1998, or Peneder 2010. However, the available data lack more detailed information in terms of technological orientation of firms or industries.

14. The categories (number of principal competitors) are given in the questionnaires and cannot be changed.

15. An instrumental variable estimation (IV) and a generalized method of moments (GMM) estimator are two alternative procedures for dynamic panel estimations. However, these procedures were not applied for the following reasons:

• (a) IV is consistent only if N (number of observations) or T (time) or both tend to infinity (see Hsiao 2003, 83). Additionally, Alvarez and Arellano (2003, 1122) stated that a large T would be desirable for applying a two-stage least square estimator (IV). We have a comparable small T (5 periods) and a large N.

• (b) A consistent application of a GMM estimator would require the use of lagged differences of the dependent variables as instruments (see Blundell and Bond 1998). However, firm-level R&D expenditures are only available as a three-year average in our data set. Consequently, we would need to use the difference of two 3-year averages of R&D expenditures in order to have valid instruments. Three years is a long period to build reasonable differences in terms of R&D expenditures. Moreover, the number of observations (given our unbalanced panel) would decrease, and consequently the risk of measurement errors would increase. Hence, given the data, such an approach is not feasible.

16. In a theoretical sense, persistence would require a marginal effect of 1. However, in the empirical test a positive marginal effect indicates persistence and a negative marginal effect indicates no persistence.