Is there really an inverted U-shaped relation between competition and R&D?

Abstract

We test whether predictions of the Aghion et al. (Aghion, P., Bloom, N., Blundell, R., Griffith, R. and Howitt, P. (2004) Competition and Innovation: An Inverted U Relationship. NBER Working Paper series, No. 9269.) model are supported by firm-level data. In particular, we analyze if there is an inverted U-shaped relation between competition and R&D. Results show that the inverted U-shaped relation is supported by the Herfindahl index but not by the price cost margin. Using the Herfindahl index, results suggest that breaking up monopolies increases R&D, whereas further increases in competition most likely lead to reduced R&D. Comparing different estimators, we find that time series-based estimators typically result in less clear-cut results, probably driven by a lack of time series variation in measures of competition.

Keywords:

• R&D
• Competition
• Firm size
• Spillovers

Acknowledgements

We are grateful to Lars Lundberg, Juhana Vartiainen and seminar participants at FIEF for valuable comments. Financial support from Jan Wallander and Tom Hedelius Research Foundation is gratefully acknowledged by both authors.

Notes

¹Schumpeter made a distinction between actual and anticipated monopoly power. For a discussion of this issue, see Geroski (1990).

²The main goal of the Crépon study is not to study competition and R&D but rather to link R&D, innovation and productivity.

³Blundell et al. find that dominant firms innovate more than non-dominant firms while industry concentration dampens innovative activity.

⁴The main purpose of the Crépon study is not to analyze the impact of competition on R&D but rather to link R&D, innovation and productivity.

⁵Since we build on a well-known model, the description is kept brief. See Aghion et al. (2004) for details.

⁶This approach is also taken by Aghion et al. (2004).

⁷See e.g. Klette and Griliches (1996) and Klette (1999).

⁸R&D can be seen as an input in the innovation process where innovations and patents are the output. Increased input, R&D, is therefore expected to give rise to increased output, that is, innovations.

⁹For example, the US competition authorities use the Herfindahl index as a guideline for making decisions on approving mergers and acquisitions, see Federal Trade Commission (1992).

¹⁰For a survey of the theory of oligopoly, competition and price, see e.g. Weiss (1989).

¹¹Herfindahl index, σ_be=2556, σ _w =1756. PCM; σ_be=0.92, σ _w =0.35, see Table A4.

¹²We instrument the PCM with industry import penetration, capital intensity, Herfindahl index, TFP, fixed industry effects and period dummies.

¹³Wolfgang Keller, in an array of papers (see e.g. Keller 1997, 2000, 2002a, 2002b), has studied both national and international technology spillovers. In line with Griliches (1992), he finds strong support for the existence of technology spillovers.

¹⁴For a similar approach, see e.g. Coe and Helpman (1995). For details on variable construction, see Appendix A.

¹⁵For a similar approach, see e.g. Van Reenen and co-workers (2000) and Aghion et al. (2004). For details on variable construction, see Appendix A.

¹⁶Relating competition to firm size may incur a spurious relationship. We adjust for endogeneity using a set of lagged values of firm size and industry and time dummies as instruments.

¹⁷See e.g. Grossman and Helpman (1991), Romer (1990) and Jones (1995).

¹⁸In Sweden, approximately 21% of workers with post-secondary education within the manufacturing industry are involved in R&D-related work (Statistics Sweden, 2001).

¹⁹Similarly, we lack patent statistics and therefore treat the appropriability conditions as an unobservable characteristic captured by industry/fixed effects.

²⁰An alternative to the FS R&D data is the bi-annually collected Research Statistics (RS), based on all firms within the FS with at least 200 employees and on a sample of firms with 50–200 employees, given that these firms report R&D expenditures of at least 200,000 SEK to the FS. In the context of statistical reliability, the bi-annually collected Research Statistics is of higher quality but has less coverage. The RS and FS data yield similar results but the RS reduces the sample size by more than 50%.

²¹The instrumented variables are {PCM, Size and wH}.

²²This result also holds for the Herfindahl index.

²³For parameter estimates, see Table A3.

²⁴Aggregating the PCM to the industry level yields similar curvature as for the firm level and a drop in significance.

²⁵Test of coefficients of H and H ² between neck-and-neckness specifications and full models indicates insignificant differences at the 5% level.

²⁶See Table A2.

²⁷0.45 when using the Herfindahl index.

²⁸See e.g. Aghion and Howitt (1999).

²⁹The 2SLS estimated elasticity spans the interval 1.03–1.31.

³⁰R&D is an activity which takes place on a systematic basis to increase the body of knowledge, including the knowledge of people, culture and society as well as the application of this knowledge to new areas and to develop or improve products, systems and methods (definition by Statistics of Sweden).

³¹See Gunnarsson and Mellander (1999).

³²The share of highly skilled labor (with post-secondary education) in R&D-related activities equals 21% in 1999.