Too fast to live? Effects of growth on survival across the growth distribution

ABSTRACT

Do moderate-growth new firms have higher survival rates than fast-growing new firms? To address this question the customer bank records of 6578 new ventures are tracked over their first 10 years, and survival is measured either in terms of continued use of the bank account, or by entry into financial default. Simple bar charts show that it is the 7^th or 8^th decile of the growth distribution that has the highest survival chances. Although growth enhances survival on average, nevertheless the highest decile of the growth distribution never has the highest survival rates.

KEYWORDS:

• Firm growth
• survival
• failure
• Penrose effects
• high-growth firms
• post-entry growth
• scale-up

Acknowledgments

We are grateful for many helpful comments from Carl-Magnus Bjuggren, Ross Brown, Roberto Camerani, Sven-Olov Daunfeldt, Frederic Delmar, Matthias Deschryvere, Niklas Elert, Ken Guy, Magnus Henrekson, Jolanda Hessels, Werner Holzl, Yuji Honjo, Barbara Larraneta, Per Larsson, Jonathan Levie, Assar Lindbeck, Andy Lockett, Mariana Mazzucato, Kevin Mole, Nicos Nicolaou, Gabriele Pellegrino, Niklas Rudholm, Josh Siepel, Erik Stam, Noni Symeonidou, Karl Wennberg, Haibo Zhou, and seminar participants at the European Commission’s 6th IRIMA workshop (Brussels), Universidad Pablo de Olavide (Seville), IFN Stockholm (Sweden), JRC-IPTS (Seville), Université de Caen, Henley Business School (Reading), HUI’s Malargarden workshop (Sweden), SPRU (University of Sussex), Tokyo Workshop on Entrepreneurship and Innovation (March 2017) and Warwick Business School. Furthermore, many helpful comments and suggestions from the Editor and three anonymous reviewers are gratefully acknowledged. The usual caveat applies.

Notes

¹ Nevertheless there could be doubts about the interpretation of these results because of the regression specifications. Gjerløv-Juel and Guenther (2012) look at the effect of early growth on survival, while holding constant initial size and also final size. Controlling for final size is a “bad control”, because final size lies on the causal path from growth to survival. Delmar et al. (2013) look at the effect of sales growth on survival, holding constant Return On Assets (ROA). Controlling for ROA could also be a “bad control”, because ROA lies on the causal path from sales growth to survival. As an example of the “bad control” problem, one would probably find no significant effects of drinking beer on driving performance if you ceteris paribus control for blood alcohol levels. The bad control problem is discussed in Angrist and Pischke (2009) and Pearl (2009).

² Storey (2011) shows that only one third of new firms surviving until the end of year 3 grew their sales in year 2 and year 3. The link between volatility and survival is more formally examined by Lundmark et al. (2019).

³ See also Zhou and van der Zwan (2019).

⁴ Coad and Guenther (2013) investigate the effects of growth through diversification (i.e., entry into new product submarkets) on survival and observe that though diversification enhances survival, these benefits are somewhat offset by an opposite-signed quadratic effect, although the relationship between diversification and survival only occurs if a firm more than doubles its product lines in a given year.

⁵ Bruno and Leidecker (1988), for example, made the case that “incompetent management was responsible for nearly 90% of these failures.” However more recently there have been efforts to “rehabilitate” business failure on three grounds: first that some exits are successful, second that many owners view the closed business as a success and third that exit is a learning experience (Josefy, Harrison, Sirmon, & Carnes, 2017).

⁶ The metric captures real sales and not financial account transfers.

⁷ In practice, there are no IPOs in our sample.

⁸ Firms are coded as “switchers” if there is an “explicit” closure of the account and a record of exit to another provider. A (small) number of other firms may have moved “below the radar”, however, which means that it is possible that not all “switchers” may have been identified.

⁹ There is of course the likelihood that some financial transactions take place with proceeds being placed “under the mattress.”

¹⁰ Zhou et al. (2012, p. 6) write that: “The number of control variables in our dataset is very limited; we can only control for size, industry and age.” Pe’er et al. (2016, p. 36) explain that, “A third limitation of our study stems from the lack of information on the characteristics of the founders.”

¹¹ The threshold for VAT registration was £58,000 annual turnover for the year starting 1 April 2004, and £73,000 annual turnover from April 2011 onwards. Hence, many of the firms in our dataset are below this threshold and would not appear in standard administrative data sets.

¹² Because the cohort starts in 2004, and year 2 corresponds to 2006, then a focus on growth deciles in year 2 means that the analysis focuses on the years before the Great Recession. To investigate whether the relationship between growth and survival is a regularity that holds irrespective of the business cycle, we repeat the analysis in Figure 2 by measuring growth in year 5 (which corresponds to 2009), and the results (not shown here) confirm the nonmonotone relationship between growth and survival. We conclude that the financial crisis did not lead to a change in the patterns observed between growth rates and survival.

¹³ See, e.g., Dunne and Hughes (1994), Headd and Kirchhoff (2009), and Delmar et al. (2013).

¹⁴ Further bar charts for different years and for different survival horizons present similar results.

¹⁵ Our polynomial regression specification has the advantages of being continuous and parsimonious; it permits the pooling together of observations across years and is fairly comparable with other studies on the same topic. However, an alternative to the polynomial regression specification involves the use of dummy variables for each growth decile. In further analysis, we fix the omitted baseline case as the 10th decile, and observe that the deciles corresponding to moderate growth have significantly higher survival rates than the dummy corresponding to the fastest growth decile.

¹⁶ Inspection of the Akaike Information Criteria (AIC) and the Bayesian Information Criteria (BIC), as well as an inspection of how model fit statistics (Cox-Snell and Nagelkerke R² statistics as well as the percentage of cases correctly classified) improve with the inclusion of higher-order polynomials, show that the specification improves steadily when up to the 10^th-order polynomial is included. For simplicity, however, our preferred specification is a polynomial model that includes the quintic term (5th power).

¹⁷ To further verify that business exit is due to poor performance (rather than being cases of successful “entrepreneurial exit” such as IPO or acquisition, as in, e.g., Wennberg et al., 2010), we ran 10 logistic regressions for each of the 10 deciles of growth in year 2, regarding survival into year 3. Indeed, exit events in the fastest-growth decile appear to be associated with poor performance (e.g., high revenue volatility, unauthorized overdraft activity, sole trader legal form). These extra results are available from the corresponding author upon request.

¹⁸ Bank account activity variables – use of (unauthorized) overdraft and bank account volatility – might be endogenous to failure (i.e., these variables may lie on the causal path between rapid growth and survival, making them “bad controls”; Angrist & Pischke, 2009). However, our results for rapid-growth firms did not change noticeably whether or not these variables were included here.

Additional information

Funding

This work was supported by the National Research Foundation of Korea Grant funded by the Korean Government (NRF-2018S1A3A2075175).