Survival of entrepreneurial firms: the role of agglomeration externalities

Abstract

This paper analyzes the role of various types of agglomeration externalities on the survival rate of entrepreneurial firms. In particular, we trace the population cohort of newly-established and self-employed Swedish firms in the Knowledge-Intensive Business Service sector in 1997 up to 2012 and investigate the role of Marshallian and Jacobian externalities on the survival of these firms. We find that only Jacobian externalities (diversity) is positively associated with the survival of entrepreneurial firms. Not all Jacobian externalities matter though. Only the higher the ‘related variety’ of the region in which an entrepreneurial firm is founded, the higher will be the survival chance of the firm, while ‘unrelated variety’ barely has any significant correlation. The result is robust after controlling for extensive firm characteristics and individual characteristics of the founders. The main message here is: for a newly-established entrepreneurial firm, not only it matters who you are, but also where you are.

Keywords:

• Entrepreneurial firms
• region
• agglomeration externalities
• survival analysis
• related variety
• unrelated variety

Acknowledgment

We are grateful to the comments received by the editor, two anonymous referees, and participants of ACERE2015 conference in Adelaide, Australia. Sam Tavassoli is financed by the funding of the Swedish research council FORMAS (grant number: 2011-80). Viroj Jienwatcharamongkhol gratefully acknowledges financial support from Forte and Norface.

Notes

1. This highlights the role of geography for innovation of firms, in so-called ‘geography of innovation’ as a field of study (Feldman and Kogler 2010).

2. Neffke, Henning, and Boschma (2012) does not study the very small firms, although majority of firms in the economy are actually very small. We precisely focused on such sample size in this paper.

3. On the other hand, looking at the literature on geography (of innovation and entrepreneurship), the firm or entrepreneur as the unit of analysis is underdeveloped (Feldman 1994; Acs and Varga 2005). When the focus is on entrepreneurial individual or firm, usually the evidences are based on case study of a particular region and systematic evidence is still lacking (Fredin 2014).

4. Agglomeration externalities are costs or benefits that firms gain by being located in geographical proximity of other economic agents.

5. This is in line with routine regime (Winter 1984).

6. As Jane Jacob argued, many innovations are the result of ‘adding new work to old ones’.

7. This is in line with entrepreneurial regime (Winter 1984).

8. Similarly, Aldrich and Ruef (2006) made a distinction between reproducer and innovator start-ups and Samuelsson and Davidsson (2009) made a distinction between imitator and innovator start-ups. But again, it is not always clear who these innovator start-ups are.

9. KIBS is among the top innovative sectors in Sweden with over 60% of the firms engage in some forms of innovative activities (Statistics Sweden 2014). These innovating activities are: introducing product, process, marketing, and organization innovations as well as engaging in ongoing/abandoned product or process innovation.

10. Focusing on KIBS sector allows us to go beyond the ‘wide’ definition of entrepreneurship offered by Austrian school (Kirzner 1978) and focus on the Schumpeterian innovation-based definition of entrepreneurship (Landström 2007). The similar sectoral choice is preferred in previous Swedish survival analysis dealing with geography (Wennberg and Lindqvist 2010). Other recent Swedish studies also dealt with KIBS sector. Andersson and Hellerstedt (2009) used the total number of those individuals graduated in natural sciences, technology, and medicine over the period of 1994–2000. Delmar, Wennberg, and Hellerstedt (2011) used all the firms in the knowledge-intensive industries from 1995–2002.

11. Although there are other datasets to be used for identifying an entrepreneurial firm, such as Community Innovation Survey or patent data (Tavassoli 2015), none of these data-set offers a possibility of tracing newly established entrepreneurial firms over an extended period of time. Therefore we cannot use them in our survival analysis.

12. Using the full population of firms based on high-quality and registered data, allow us to dramatically reduce problems related to inferences and internal validity, since our estimates are not based on a sample of firms.

13. KIBS sectors are classified as two-digit European NACE codes 72–74. Moreover, 1997 is chosen since this is the earliest year that we can have the full set of control variables at the level of individual, firm, and region.

14. The longitudinal individual level data-set provides information on various characteristics of each individual in Sweden. Among other things, it also classifies the main occupation type of each individual to be either (i) a sailor (ii) a farmer (iii) an employed (in public or private sectors) or (iv) an entrepreneur (founder of a business).

15. There is a small portion (about 1%) of KIBS self-employed firms founded in 1997 that their founders moved out, while the firm continued to exist (i.e. ‘harvest sale’ exit route (Wennberg et al. 2010)). Excluding this particular type of firms from our analysis did not change our main result.

16. After accounting for missing values in variables, the actual number of firms used in the subsequent empirical analysis is 4617.

17. Those studies which used municipality as the geographic unit of analysis constructed their agglomeration externality indicators based on a number of geographical potentials (Neffke, Henning, and Boschma 2012).

18. Assume the survival time T follows a distribution with density function , such that the survival function , the commonly-used Weibull distribution implies that and the hazard function , where λ denotes model parameters. On the contrary, the Cox model does not require a specification of the distribution, so h(t) is simply , where h₀(t) is the baseline hazard rate that corresponds to when parameters equal zero.

19. Since the firm’s employment in our study is always 1 in the beginning of the period (1997), value is far bigger than all of the other variables. We then decide to use the logged value. We also used population density as an alternative measure of the size of the region. Main results are very similar.

20. Calculating Unrelated Variety which is limited to only KIBS sectors (i.e. same fashion as calculating Related Variety) did not change our main results.

21. An alternative strategy would be to estimate competing risks models, which would have allowed us to separate all possible exits, as separate events. Unfortunately, we cannot identify and distinguish all possible types of exits in our data-set in order to perform a proper competing risks models. For instance, we are not able to properly identify the death of a founder or when a founder migrate abroad. Since a founder is strictly attached to his/her firms in our empirical setting (except 1% of the founders that moved out of his/her firm and the firm still continued to exist), a death or migration of the founder means exit of his/her firm in our data-set. Therefore, we are not able to identify at least two exit events (at firm-level outcome). This issue indeed makes us hesitant to go for competing risks models.

22. Various studies find that male entrepreneurs lead firms that stay in business longer than female counterpart (Taylor 1999; among other). On the other hand, Watson (2012) finds that there is little difference in networking between male- and female-controlled Australian SMEs after controlling for size, education, industry and size.