Related variety and firm heterogeneity. What really matters for short-run firm growth?

ABSTRACT

In recent years, two concepts have become key elements in economic geography: related variety and firm heterogeneity. The first one predicts that knowledge spillovers within a region/local system occur among firms operating in ‘different but related’ sectors. The second one assumes that knowledge spillovers can occur among ‘different’ firms belonging to the same localised sector/industrial cluster. Using a sample of 27,817 Italian manufacturing firms observed during the period 2010–2013, this paper analyses the role played by related variety and within-sector firm heterogeneity on short-run employment growth. The results suggest that both related variety and within-sector firm heterogeneity have a positive effect, although the latter has a higher impact than the former. These results confirm the role played by related variety, but identify firm heterogeneity as a potential additional source of local knowledge spillovers.

KEYWORDS:

• Related variety
• firm heterogeneity
• short-run firm growth
• regional productive structure
• Italy

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental data

Supplemental data for this article can be accessed here.

Notes

1. Maskell suggests that since firms located within a cluster have ‘different perceptive powers, divergent insights and dissimilar attitudes . . . they develop a variety of solutions . . . to similar problems’, which leads to a ‘parallel process of experimentation and testing’ of a variety of solutions (2001, 928–929). By observing, discussing and comparing these different and dissimilar solutions, firms engage in a process of interactive learning and local knowledge creation.

2. The Ateco 2007 classification of economic activities is the Italian version of the European nomenclature NACE Rev. 2 adopted with Regulation (EC) No.1893/2006 of the European Parliament and of the Council of 20 December 2006 (Istat 2009).

3. Table A1 in Appendix A, reported in the Online Supplemental Material, presents the two-digit classification for the manufacturing sectors, and their corresponding technological levels (low, medium-low, medium-high, high) according to the classification adopted by the Statistical Office of the European Communities (Eurostat). Table A2 provides a sector-based comparison between the sample of firms and the population of manufacturing plants in 2010 (Istat). Tables A3 to A7 report the distribution of the sample firms according to two-digit sector, technological regime, firm size and geographic level 1 of the Nomenclature des Unités Territoriales Statistiques (NUTS) adopted by the European Union.

4. Related variety is captured by an inverse HHI, rather than an entropy measure as proposed by Frenken, van Oort, and Verburg (2007), in order to proxy for sub-sector diversification within a sector and firm diversity using a similar measure. In fact, the entropy measure has more restrictive mathematical properties than the HHI in terms of sub-units (e.g. five-digit sectors) belonging to a unique unit (e.g. the two-digit sector), and this could be an issue since size classes (as sub-units) are not exclusively identified within a two-digit sector (as the unit).

5. Cainelli, Ganau, and Giunta (2018) exploit a similar identification strategy, and construct similar instrumental variables, in the context of spatial agglomeration economies.

6. In particular, the year 2009 was characterised by very negative macroeconomic performances. The Gross Domestic Product (GDP) decreased in real terms by 5%, the number of employed persons declined by 380,000, and the unemployment rate rose to 7.8%. The situation improved a little bit during the year 2010, but not significantly.

7. The results of the estimated first-step selection equations, and the results of the first-stage regressions of the TSLS and GMM specifications are not reported here, but are available from the authors upon request.

8. As stated in the main text, the estimated effects of the variables for related variety and within-sector firm heterogeneity are in line with previous contributions. Looking at firm-level studies, and underlining the lack of analyses on employment growth, a comparison can be made with previous findings concerning firm-level productivity and innovation. For the Italian case, Cainelli, Ganau, and Iacobucci (2016) consider manufacturing firms’ total factor productivity and a measure of vertical related variety defined using input-output relationships across two-digit sectors; their spatial unit of analysis in the Italian NUTS-3 region, and the time horizon is the period 1999–2007; their fixed effects results suggest an elasticity of the vertical related variety variable ranging from 0.04 to 0.36, while their IV results suggest an elasticity ranging from 0.04 to 0.37. Also Wixe (2015) uses a panel data approach, and analyses the average labour productivity of Swedish manufacturing plants; differently from our related variety variable − and that used by Cainelli, Ganau, and Iacobucci (2016) − which is specific to the sector and the region, Wixe (2015) uses a region-specific variable (i.e. all two-digit sectors are summed over the region), and finds a negligible fixed effects elasticity of 0.06, while a negative random effects elasticity of −0.04. Aarstad, Kvitastein, and Jakobsen (2016) use a region-specific variable for related variety defined at the Norwegian municipality level, and find a marginal effect ranging from 0.025 to 0.027 on enterprises’ labour productivity, while a margin effect ranging from 0.21 to 0.29 on enterprises’ product innovation. Eriksson (2011) uses a region-specific related variety variable defined at the municipality level in Sweden, and finds a marginal effect of 7.31 on plant-level labour productivity growth over the period 2001–2003. To the best of our knowledge, the only work which is similar to ours is Wixe (2018), where two different dimensions of relatedness are compared, i.e. a region-specific variable for industrial related variety vs. a region-specific variable for educational related variety aimed at capturing the level of relatedness in terms of employment’s educational background; the estimated odds ratio of industry related variety on firms’ product innovation ranges from 0.76 to 1.27 in the case of all firms, from 0.88 to 1.33 in the case of industrial firms, while from 0.71 to 1.23 in the case of services firms; the estimated odds ratio of the education-based related variety variable ranges from 0.86 to 1.54 in the case of all firms, from 0.7 to 1.36 in the case of industrial firms, while from 1.01 to 4.13 in the case of services firms. The second branch of selected studies we would focus on for a comparative analysis looks at regional employment growth. Frenken, van Oort, and Verburg (2007) focus on NUTS-3 regions in the Netherlands, and finds a marginal effect of the related variety variable on regional employment growth over the 1996–2002 period ranging from 0.46 to 0.64. Hartog, Boschma, and Sotarauta (2012), looking at Finnish NUTS-4 regions, find a marginal effect of related variety on employment growth equal to 0.01, even though the estimated effect increases up to 0.28 when related variety is defined on high-tech industries only. Wixe and Andersson (2017) analyse the effect of related variety defined in terms of industry, educational level and occupations; they look at (manufacturing) employment growth in Sweden over the periods 2002–2007 and 2002–2011, and find an estimated marginal effect of related variety which, overall, ranges from 0.06 to 0.29. Firgo and Mayerhafer (2018) consider Austrian LLMs, and find an elasticity of the related variety variable on employment growth equal to 0.06, even though the estimated elasticity increases up to 0.22 in the case of urban LLMs. On the Italian case, Boschma and Iammarino (2009) consider a related variety measure defined at NUTS-3 geographic level, and estimate a marginal effect on employment growth over the 1995–2003 period which ranges from 0.60 to 4.28; even larger effects are estimated using as dependent variable the growth in value added − the marginal effect of related variety ranges from 7.19 to 8.99 − and the growth in labour productivity − the marginal effect of related variety ranges from 2.42 to 7.38. Sedita, De Noni, and Pilotti (2017) analyse the effect of related variety on the employment growth of Italian LLMs during the recent period 2009–2013, and estimate a coefficient ranging from 0.05 to 0.12. Overall, this short overview of results concerning the concept of relatedness suggests two conclusions. First, the results are different across different countries, and the results on the Italian case tend to be similar across studies. Second, the estimated marginal effects and elasticities range over large intervals and, sometimes, the values reported in previous studies are much larger than those presented in our paper − i.e. an elasticity of 0.133 of related variety, and an elasticity of 0.251 of within-sector firm heterogeneity according to the results reported in column (3) of Table 3.

9. Specifically, the highest marginal effect for related variety is associated with the sector ‘15 – Leather and related products’, while the highest marginal effect for within-sector firm heterogeneity is associated with the sector ‘16 – Wood, wood and cork products (except furniture), articles of straw and plaiting materials’. On the contrary, the lowest marginal effect for related variety is associated with the sector ‘28 – Machinery and equipment N.E.C.’, while the lowest marginal effect for within-sector firm heterogeneity is associated with the sector ‘29 – Motor vehicles, trailers and semi-trailers’.

10. A series of further robustness exercises has been performed to check the results. First, two alternative variables defined at the LLM level, namely population and surface, have been used to proxy for urbanisation externalities and size effects. Second, the entropy measure has been used to construct the variety variables. Appendix B in the Online Supplemental Material provides a description of the robustness analyses, and reports the empirical results.