![Sample our Economics, Finance,Business & Industry journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5931/TOC-Subject-Sample-2021-Economics-Finance-Business-Industry.jpg"})
Abstract
A stream of literature has demonstrated that venture capital generates externalities that enhance the knowledge base of a given region and accordingly assist firms in high-technology industries to improve their innovative performance. What has gone largely unexamined in this literature is the geographic extent of such externality impact. In this research we address the issue at hand. We do so by analyzing the association between the patenting rate of all life sciences firms (LSFs) that have won Small Business Innovation Research grants from 1983 to 2006 and the venture capital investments that have occurred at increasingly distant spatial units from those firms. Controlling for firm-specific and environmental factors as well as for endogeneity concerns, we document that LSFs tend to produce more patents whenever they are situated in very close proximity to where venture capital investments occur. Further, we find that improvements of the patenting rate of focal firms largely emanate from investments that reflect the expertise of venture capitalists on advancing existing prototypes closer to commercialization. We conclude the paper with a discussion on research and policy implications of our findings.
Acknowledgements
Research funding provided by the Ewing Marion Kauffman Foundation Strategic Grant #20050176 is gratefully acknowledged.
Notes
1. Using Germany as their case study, Fritsch and Schilder (Citation2008) demonstrate that country-specific features can undermine the significance of spatial proximity between venture capitalists and target firms.
2. To illustrate the geographic scope of an MSA, Stuart and Sorenson (Citation2003) report that in their study the average area of an MSA was 10,515 square miles.
3. Note that besides informing the literature on the geographical limits of venture capital, this research can also contribute to a better understanding of the geographical dimensions of entrepreneurship (e.g. Harrison, Hong, and Lee Citation2010; Plummer Citation2010; Zacharakis, Shepherd, and Coombs Citation2003) because venture capital is mostly associated with entrepreneurial ventures.
4. At an aggregate level, despite their wide use, patents have certain shortcomings as a measure of innovation. For instance, innovative firms may not patent for strategic reasons (Teece Citation1986) or maybe patents relate more to invention rather than to innovation (Moser Citation2005). Albeit a less than perfect proxy for innovation, patents are generally still a reliable measure of innovation (Acs, Anselin, and Varga Citation2002). Specifically for our sample, we expect patents to represent a strong proxy for innovation for a number of reasons. First and foremost, because we focus on life sciences we expect patents to measure innovations largely because the distance between invention and innovation in that industry is relatively short due to the immediate commercial applications of basic life sciences research (Argyres and Liebeskind Citation1998; Mcmillan and Narin Citation2000). Further, we analyze small firms which generally employ patents as a defensive protection measure to block competitors at a significantly lower rate than larger firms and other types of institutions (Giuri et al. Citation2007; Veer and Jell Citation2012). Relatedly, patents tend to be economically active for such firms mainly because they allow them to attract external finance primarily by signaling their innovative performance to interested parties (Audretsch, Bönte, and Mahagaonkar Citation2012; Baum and Silverman Citation2004). Altogether these observations support our choice of patents as a measure of innovation and allow us significant comfort in the robustness of our empirical estimates.
5. The SBIR program is the largest federal program in the US and it provides seed and early stage funding to promising small firms in cutting edge research areas such as life sciences, electronics, materials and energy conversion. Promoting innovation is among the stated goals of the program and it has generally been found successful in achieving that goal (Audretsch, Link, and Scott Citation2002).
6. It is important to note that mainly because of the complexity and scope of relationships that underlie the creation and maintenance of regional innovation systems, venture capital can be a consequence rather than an impetus of cluster development (see for intance the work of Harrison, Cooper, and Mason Citation2004 for the Ottawa cluster).
7. Note that the composition of those networks is not necessarily confined only to VCs but it can also include other industry professionals such as lawyers and accountants (Powell et al. Citation2002; Zook Citation2005).
8. This is not to say that positive externalities cannot arise from a single VC. But, because externalities typically emanate from the flow of knowledge, networks of VCs are expected to be stronger in that respect.
9. In this section we concentrate on the impact of spatial proximity on post-investment activities; proximity is relevant for pre-investment activities as well but such discussion is beyond our scope.
10. In related empirical studies, that do not focus on the spatial extent of VC relationships but estimate what drives the investment decision of VCs ex ante, CitationLutz et al. (in press), Cumming and Dai (Citation2010) and Sorenson and Stuart (Citation2001) also highlight how significant geographic proximity between VCs and target firms is in shaping that decision.
11. Alternatively we could use a dependent variable that reflects the number of patents per year. Nevertheless, we did not opt for this approach because differential and often unobserved lags in the dates of discovery, patent submission and patent issuance could make the allocation of innovative performance by year an exceedingly difficult task. Relatedly, we performed robustness checks for potential temporal lag effects between the timing of the venture capital investments and the strength of the externality impact. In these tests, the venture capital investments were limited to those that occurred only 1, 2, 3 or 4 years from the birth of the focal LSF. The tests yielded qualitatively similar results to those presented in Table .
12. We use the Negative Binomial maximum likelihood estimator. The Negative Binomial estimator was chosen to address observed overdispersion and to overcome the standard assumption of the Poisson model of equal conditional means and variances, which was not met for the dependent variable in our sample.
13. In fact, Parhankangas (Citation2007) specified $100 K as the lower bound of that stage. In order to include venture capital disbursement between $25 K and $100 K in the analysis we included such amounts in the start-up stage. Robustness checks in which these amounts were either included in the seed stage or were excluded from the analysis yielded qualitatively similar results to the results presented here.
14. The equivalence between the size of the investment and the firm growth stage it corresponds can potentially mask cases under which investments of different size but of the same stage are allocated to a given firm in tranches. To address that concern we leverage the feature of our data source of venture capital to report the firm growth stage of each venture capital investment (i.e. early stage, later stage, expansion, etc.). The classification of stages in our data source is largely comparable to the classification of Parhankangas (Citation2007). In a few cases, the classification in Parhankangas (Citation2007) is slightly more aggregate where for instance ‘later stage’ in Parhankangas (Citation2007) includes what our data source reports as ‘later stage’ and ‘expansion’. Importantly, the threshold levels for each stage used in Parhankangas (Citation2007) were, for the vast majority of investments, in line with the information in our data source. For example, similar to Parhankangas (Citation2007) in our data source all investments below $25,000 were reported as seed investments and all investments categorized as startup received more than $500,000. In sum, the overlapping of the relevant information in Parhankangas (Citation2007) and of our data source ameliorates the potential impact of the concern identified above.
15. Note that besides research intensity, we expect the NIH variable to also be capturing the underlying quality and subsequent reputation of the local universities because NIH funds are awarded on a very competitive basis. As we explain in detail in Section 4, this observation is significant for the robustness of the instrumental variable we use in the empirical analysis.
16. Among others, contributions from Breschi and Lissoni (Citation2001) and Tappeiner, Hauser, and Walde (Citation2008) take a critical stand towards positive spatial externalities.
17. Examples of private organizations that offer consulting services on securing SBIR grants include Foresight S&T in Rhode Island and the Larta Institute in California and the District of Columbia.
18. Because our sample covers a lengthy period (23 years) in which relevant state characteristics are expected to change, we do not include associated independent variables, whose historical availability is also limited, in the empirical specification. Further, some of these state characteristics may be difficult to observe (e.g. business climate), and hence to approximate with associated variables which adds to our methodological approach.
19. We, however, followed the codification scheme described in Table because the number of employees is typically reported by firms in discrete categories.
20. There is also a Phase 3 but federal agencies do not provide funds during it.
21. Notice that while direct investments from university endowments to young firms are rare, the local density of young innovative firms may be correlated with university endowments through an alternative process. Typically, the most reputable academic institutions realize the largest endowments. At the same time, these kinds of institutions tend to be research-intensive, which often prompts young innovative firms to locate close to them mainly in order to reap spatial externalities and other sorts of proximity effects. By extension, university endowments maybe correlated with the innovative character of regions that attract VCs. We account for this potential relationship by including the NIH variable in the analysis, which can capture the research intensity and underlying quality/reputation of the local universities as NIH funds are awarded on a very competitive basis.
22. See the correlation table in the Appendix for more details.
23. In the first stage of the instrumental variable approach presented in Table we construct variables that employ the number of patents that each of the proximate firms that eventually received venture capital investments was granted before such investments took place. To collect the number of patents per firm we searched in the online patent search engine maintained the United States Patent and Trademark Office for patents issued before the first venture capital investment where the focal firm was listed as the applicant/assignee. To ensure that our search was not prone to different name recordings of the same firm, we run the searches with different versions of the name of each firm (e.g. instead of ‘inc.’ we tried ‘inc’).
24. To sort out life-sciences grants from the total population of grants from the National Institutes of Health we consulted with life-sciences researchers employed at the authors' institution. The list was composed of more than 400 terms, including the following: enzyme, peptide, antigen, mutation, clone, immunoassay, coli, hormone, neuron, PCR, cytokines, gene, collagen, bioreactor, ELISA, nucleotide, plasmid, biomass, bacillus, bioassay, embryo, and genetic.
25. The first year in which the Chronicles of Higher Education report data on university endowments is 1999. Therefore, the average endowment of each state university is calculated with the corresponding value for 1999 as the starting point.
26. The use of narrow units in the analysis offers as an additional methodological advantage in that we escape estimation issues that relate to required spatial corrections in the data when the unit of analysis is administrative units such as states (Arauzo-Carod and Manjón-Antolín Citation2009).
27. The estimation with standard errors clustered at the state level is carried out with generalized estimating equations which is a method to estimate the standard errors which first estimates the variability within the defined cluster and then sums across all clusters (Zorn Citation2006).
28. The marginal effects for dummy variables were computed as the change in the expected number of counts (patents) when the value of the variable goes from 0 to 1 and keeping the remaining variables at their mean value.
29. McFadden's R 2 is analogous to the OLS R 2 where the log likelihood value for the null model replaces the total sum of squares and the log likelihood value for the unrestricted model replaces the residual sum of squares. An increase in the McFadden statistic indicates better model fit (Long Citation1997).
30. 0.1738 × 12.876 (the maximum value of the First_01 variable) = 2.24.
31. Note that in one model with standard errors clustered at the state level, the SBIR_05 estimate is negative and statistically significant. However, the statistical significance is marginal (p-value 0.095). Accordingly, this estimate should be interpreted with caution.