Is there a life cycle in all industries? First evidence from industry size dynamics in West Germany

ABSTRACT

We propose a novel nonlinear regression approach to test whether the size of industries develops over time along the stylized pattern assumed by the industry life cycle theory. We apply our model on data covering the full spectrum of 205 NACE industries including services in West Germany between 1976 and 2009 and four indicators describing industry size (employment, establishments, entries and exits). The results of our large scale analysis show that in most industries indeed size develops along a cyclical path, albeit this development is not universal. Furthermore, we provide first empirical evidence on service industries where we show that the number of establishments and employees frequently develop in line with what is found for most (product) industries.

KEYWORDS:

• Industry life cycle
• industry size dynamics
• nonlinear modelling

JEL CLASSIFICATION:

• C23
• L16
• O33

Acknowledgements

We thank participants at the 2015 European Meeting on Applied Evolutionary Economics (EMAEE) in Maastricht and at the Geography of Innovation Conference 2016 in Toulouse as well as EVO-seminar participants in Marburg for helpful comments and suggestions. Comments received from an anonymous referee are gratefully acknowledged. Matthias Dorner acknowledges funding from the Graduate Programme (GradAB) of the Institute for Employment Research (IAB). The usual disclaimer applies.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ An alternative approach to explain the evolution of industry populations is provided in the organizational ecology literature (Hannan and Freeman 1977, 1989).

² Note that although we study the full spectrum of industries, we do neither hypothesize nor expect that all industries follow the ILC pattern (Malerba and Orsenigo 1996). Natural exceptions should be noncommercial industries or industries where, for example, technological change is irrelevant for the size dynamics.

³ The administrative data that we use record 5-digits industry codes in the NACE system since 1998 and earlier at the 3-digits level of a national industry classification. Due to the changes in the classification systems, the generation of a time consistent classification back until 1976 comes at the cost of aggregation to the 3-digits level. Nevertheless, we argue that our approach has also significant advantages (see Section V). An extensive discussion of potential aggregation bias in industry data from administrative sources for ILC analyses is provided by Dinlersoz and MacDonald (2009).

⁴ Standard R routines did not work in most cases. Therefore, a modified Levenberg–Marquardt algorithm with varying starting values was programmed.

⁵ The total number of 3076 regressions results from the combination of 205 industries × four indicators of industry size × four model approaches. For data confidentiality reasons, not all four indicators were available in all 205 industries (see Table 1).

⁶ The regression approach assumes normally distributed residuals which are tested using the Shapiro test. We find deviations from the normal distribution (significance level: 0.05) in some cyclical industries. In 31% of the cases residuals are not normally distributed. In these cases there are some additional trends or dynamics which represent an interesting topic for further studies, but are ignored here.

⁷ A detailed classification list of the industries can be requested from the authors.

⁸ The combination of industry data is usually problematic. Several concordance tables have been proposed in the literature (Schmoch et al. 2003; Lybbert and Zolas 2014). For our study plan to use a novel concordance table at the 3-digit level which is based on linked inventor–establishment data generated by Dorner, Harhoff and Hoisl (2016).

Additional information

Funding

This work was supported by the Institute for Employment Research (IAB).