Intra-industry spillovers and innovation: An econometric analysis at the firm level

Abstract

This paper studies spillover effects of innovation at the firm level and the comparability of generalized method of moments (GMM) estimators with maximum likelihood estimators of the earlier studies. Two sources of spillovers are identified, i.e. intra-industry R&D expenditure and intra-industry innovation output. This paper estimates a negative R&D spillover effect and a positive output spillover effect. Because of the substitution effect of intra-industry R&D spillovers, the elasticity of patent with respect to firm's own R&D expenditure is greater than those estimated in the earlier studies. With GMM, individual effects are incorporated into the models either by developing proxies for them or attempting to eliminate them.

Keywords:

• Spillovers
• Innovation
• General methods of moments
• R&D expenditure

Acknowledgements

The author wishes to thank Bronwyn Hall, Adam Jaffe, and the National Bureau of Economic Research for providing the data used in this paper. The author is also grateful to Jim Jozefowicz, Kajal Lahiri, Jerry Marschke, and two anonymous referees as well as the editors of EINT for their extremely helpful comments on earlier versions of this work.

Notes

¹There are two empirical studies of spillover effects. Crepon and Duguet (1997b) estimated a negative effect of rival R&D on a firm's own patenting, whereas Cincera (1997) estimated a positive effect of intra-sector R&D spillovers. Although Cincera explains that his negative sign is due to that diffusion spillovers are more important than competitive ones, he actually only studied competitive spillovers. The technological spillovers are discussed in Griliches (1992) and Jaffe et al. (1993).

²HHG (1984) justify the introduction of a heterogeneity term captured by the mean of the Poisson distribution to cite differences among firms in the patent decision. For instance, if the number of patents, given the number of innovations, is distributed as a binomial variate with probability p and the number of innovation is Poisson distributed with parameter exp(X _itβ), the number of patents is Poisson distributed with parameter exp(X _itβ+log p). log p indicates the source and measure of heterogeneity among firms. Presample information used in this paper is an alternative.

³The full documentation and key issues of the new data set are described and addressed in Hall et al. (2001).

⁴Cincera (1997) considers the annual flow of R&D investment, the annual flow of spillovers, the technological time-invariant dummies, and geographical time-invariant dummies in the explanatory variables.

⁵Since only a minority of firms is engaged in R&D activities, firms' decisions on inventive inputs are also subject to such bias. Crepon et al. (1998) adapted a similar Heckman two-stage model to account for self-selection issue in R&D investment of French manufacturing firms. Piga and Vivarelli (2004), based on a survey dataset on Italian firms, consider explicitly the selectivity issue in firms' decisions on whether to engage in R&D activities and among the innovative firms, whether to engage in external R&D. Both studies found the selectivity bias significant.

⁶Adjustment: 0.5 is automatically added to 0-counts for a log form (Cameron and Trivedi, 1998).

⁷They develop a composite index of patent quality using four different variables: the number of claims specified in the patent, the number of subsequent citations to the patent, the number of citations the patent makes to previous patents and the number of countries in which the patent is applied for. The index also reveals that average patent quality has increased over time and that this accounts for a significant part of the apparent decline over time in research productivity.

⁸Kim and Marschke (2004) studied whether if the recent surge in patenting is contributed by more productive R&D input or increased patent propensity without any increase in inventive activities. They found that in two high-tech industries, increased R&D spending explains 70% of the patent increase.

⁹Crepon and Duguet (1997b) interpret their negative spillover effects as the dominance of competitive spillovers over diffusion ones. They actually did not include diffusion spillovers in the model.

Additional information

Notes on contributors

Vincent Wenxiong Yao

Tel.: [phone]; Fax: [phone]; E-mail: [email protected]