Family firms in European regions: the role of regional institutions

ABSTRACT

Our study investigates whether the quality of regional institutions influences firms’ likelihood of being a family firm. We explore our conjecture using the EU-EFIGE/Bruegel-UniCredit dataset, which provides comparable cross-country data on manufacturing firms in seven European countries. We use a multilevel framework to analyse how firm- and regional-level variables influence firms’ likelihood of being a family firm. We find evidence that location matters in explaining firms’ probability of being a family firm but that differences between countries are more relevant than are differences between regions. Our results show that the lower the quality of regional institutions, the higher the likelihood of a firm being a family firm. Our results are robust to alternative regional-level control variables and persist after several robustness checks.

KEYWORDS:

• Family firm
• manufacturing firms
• regional familiness
• regional institution; European regions; multilevel model; context

Disclosure of potential conflicts of interest

No potential conflict of interest was reported by the author(s).

Notes

1. NACE-CLIO refers to the General Industrial Classification of Economic Activities in the European Communities (Nomenclature générale des Activités économiques dans les Communautés Européennes), used for the input-output tables. NUTS refers to ‘Nomenclature of territorial unit for statistics’, which are EU statistical regions. For more information, see http://ec.europa.eu/eurostat/web/nuts/overview.

2. Given their relevance in national economies, large firms were oversampled.

3. Given the representativeness of the EFIGE database, we are confident that the distribution of family firms in our dataset reflects the distribution in each country.

4. The regional quality of government index has been employed to analyse the impact of institutions on regional performance (Ketterer and Rodríguez-Pose 2018; Rodríguez-Pose and Di Cataldo 2014) and firm performance (Ganau and Rodríguez-Pose 2019; Ricotta 2019).

5. The territorial aggregation is NUTS 1 for Austria, Germany, Hungary, and the United Kingdom and NUTS 2 for France, Italy, and Spain.

6. For Austria, the data for the QoG indicators available for NUTS 2 were aggregated at NUTS 1 using the simple mean. This was also done for the Italian region of Trentino Alto Adige, for which two provinces (NUTS 3) are presented separately in Charron at al. (2014).

7. This continuous variable is centred on its sample mean to make the interpretation of parameter estimates easier and to facilitate convergence.

8. Size is also measured by the logarithm of firm employees.

9. We test the difference in the human capital mean between family firms and non-family firms. The t-stat = 11.4476 (p value = 0). We conclude that the difference is different from 0.

10. The ability to employ contextual factors to explain variability in random components is the main difference between multilevel models and random coefficient regressions. A particular advantage of multilevel modelling is its ability to explore the effects of Level 2 predictors while simultaneously including random effects to allow for the effects of unobserved group-level variables (Steele 2009).

11. The models for dichotomous responses can be viewed as latent-response models. In this case, $\rho$ represents the estimated intra-class correlation of the latent responses $\rho =\frac{{\sigma }_u}{{\sigma }_u+\frac{{\pi }^2}{3}}$, where the latter term in the denominator represents the variance of the underlying latent continuous response if it is assumed to be distributed as a standard logistic distribution (Rabe-Hesketh and Skrondal 2008).

12. For Italy, Fazio and Piacentino (Fazio and Piacentino 2010) showed that only 5% of the variability in firms’ labour productivity was due to provincial (NUTS 3 aggregation) spatial variations. Aiello, Pupo, and Ricotta (2014) found that Italian regions (NUTS 2) explained 4.6% of the variance in total factor productivity. For the Netherlands, van Oort et at. (2012) found that between-region variance was approximately 3% for firms’ survival and employment growth.

13. Aiello and Ricotta (2016) found a similar outcome using the EFIGE dataset and multilevel approach when explaining firm productivity.

14. The literature proposes various cut-off thresholds. Some authors suggest a value of 10, while others suggest 5 and even 4 (Micheal and Abiodun 2014).

15. The result of the specification in which size is measured as the logarithm of firm employees (available upon request) is equivalent to Model 6 of Table 3.

16. Using the Pavitt taxonomy leads to the loss of up to 600 firms because their Pavitt classification is missing in the database.

17. The data for these control variables are collected from EUROSTAT.

18. The calculation of this index for European regions by the Directorate-General for Regional Policy (DG REGIO) is based on a slight variation of the methodology developed by the United Nations Development Programme (Bubbico and Dijkstra 2012).

19. The index varies between 0 and 100, where 0 represents low levels of human development and 100 represents high levels.

20. This variable is taken from Tabellini (2010). It is available in the QoG database, which is freely accessible at http://www.qog.pol.gu.se/data/. Social trust is measured as the percentage of respondents who answer ‘Most people can be trusted’ to the question ‘Generally speaking, would you say that most people can be trusted or that you can’t be too careful in dealing with people?’ in the World Value Surveys .