The effects of R&D subsidies and network embeddedness on R&D output: evidence from the German biotech industry

Abstract

Although policy makers subsidize firms’ R&D projects to increase innovational output in an industry, region, or country, it remains unclear whether such subsidization triggers additional R&D efforts or crowds out private investments. To address this question, we assess the effectiveness of subsidization for individual and collaborative research in the German biotech industry while also examining the relevance of network embeddedness for patent output. Our results indicate that subsidies do lead to increased patent output, with additional benefits from involvement in more than one project. However, the amount of money is only significant under certain circumstances. Also, it is higher degree centrality in the firm network that significantly increases R&D success.

Keywords:

• R&D policy
• policy evaluation
• innovation networks
• biotech

JEL codes:

• O31
• O33
• O38

Notes

¹ Additionally, we argue that including only firms that have applied at least once for a patent, helps to reduce unobserved heterogeneity among firms. Because there are many common additional ways to protect innovations (e.g. secrecy) which are probably related to certain firm characteristics we assure that we have a comparatively homogenous sample by using only firms which have patented at least once in the observation period. For instance, if our data-set included firms which do not use patents as an instrument for the protection of IPR, we would expect to find a zero correlation between subsidies and patents for these firms (by assumption). This in turn would lead to a severe bias of the estimates (depending on the proportion of the firms with a different patenting strategy in the sample). We think that limiting our sample to firms which show at least a small patenting activity is the best way to avoid sample selection.

² This database is publicly accessible via the website at www.foerderkatalog.de.

³ SUM_SUB, SUM_JSUB, as well as SUM_NJSUB are measured in 1000 Dollars.

⁴ Overdispersion is also suggested by rejection of the null hypothesis that the alpha of the Poisson distribution equals zero. This implies that using negative binomial regression is more convenient compared to a Poisson model.

⁵ Using a 90% confidence interval.

⁶ Additionally, we tested a model incorporating the betweenness centrality as a measure for the global network embeddedness together with the degree centrality. Since the results yield no further insights, we stick to the basic model as described in the text.

⁷ For more information on FMMs, see Deb et al. (2011).

⁸ To rule out sample selection with regards to the heterogeneity between subsidized and unsubsidized firms we additionally estimate our model by including only firms which have been involved in at least one subsidization scheme during the observation period. Since the results between this estimation method and our basic model differ only slightly, we assume that selection into subsidization schemes seems to be a minor issue.

⁹ Our collaboration data begin in 1977, so a four-year lag takes us into 1981. Descriptive statistics for the different periods are presented in the Appendix 1.