Does Related Variety Affect Regional Resilience? New Evidence from Italy PDF

The Annals of Regional Science (2019) 62:657–680 https://doi.org/10.1007/s00168-019-00911-4 ORIGINAL PAPER Does related variety affect regional resilience? New evidence from Italy Giulio Cainelli1 · Roberto Ganau1,2 · Marco Modica3 Received: 3 October 2017 / Accepted: 25 April 2019 / Publish...

The Annals of Regional Science (2019) 62:657–680 https://doi.org/10.1007/s00168-019-00911-4 ORIGINAL PAPER Does related variety affect regional resilience? New evidence from Italy Giulio Cainelli1 · Roberto Ganau1,2 · Marco Modica3 Received: 3 October 2017 / Accepted: 25 April 2019 / Published online: 6 May 2019 © Springer-Verlag GmbH Germany, part of Springer Nature 2019 Abstract Although several contributions have studied the effect of related variety on the eco- nomic performance of firms and regions, its influence on regional resilience—that is, regions’ capacity to adapt to external shocks—has received little attention. This paper contributes to this debate by analysing empirically the relationship between related variety and regional resilience at the local labour market (LLM) level in Italy. The analysis uses a standard definition of regional resilience and employs spa- tial econometric techniques to analyse the role played by related variety as a short- run shock absorber with respect to the 2008 Great Recession. The results obtained from the estimation of Spatial Durbin Error Models suggest that LLMs character- ised by a higher level of related variety have shown a higher capacity to adapt to the Great Recession with respect to the 3-year period 2010–2013. On the contrary, there is evidence of a negligible role played by related variety as a shock absorber with respect to the 1-year resilience period 2012–2013. JEL Classification B52 · C21 · R11 * Roberto Ganau [email protected]; [email protected] Giulio Cainelli [email protected] Marco Modica [email protected] 1 Department of Economics and Management “Marco Fanno”, University of Padova, Via del Santo 33, 35123 Padua, Italy 2 Department of Geography and Environment, London School of Economics and Political Science, Houghton Street, London WC2A 2AE, UK 3 Gran Sasso Science Institute, Via Iacobucci 2, 67100 L’Aquila, Italy 13 Vol.:(0123456789) 658 G. Cainelli et al. 1 Introduction The 2008 Great Recession—which affected all European regions, although with different intensity—prompted interest in the concept of regional resilience, which refers, from an evolutionary perspective, to a region’s capacity to react positively to a short-term external shock or disruption (Simmie and Martin 2010; Martin 2012). With different European regions reacting differently to the same external shock, several attempts have been made to identify empirically the main determi- nants of this regional resilience heterogeneity (e.g. Fingleton et al. 2012; Martin 2012; Modica and Reggiani 2014; Bristow and Healy 2018; Faggian et al. 2018; Fratesi and Perucca 2018; Cainelli et al. 2019). Among the drivers identified, one of the most important appears to be the region’s current industrial structure and, in particular, its level of related diversification (Boschma 2015). As it is well known, the concept of related variety, which has been widely discussed and inves- tigated during the last 10 years (Frenken et al. 2007), stresses that what matters it is not the level of productive diversification per se, but the presence of diversified domains allowing complementarities to be exploited across different industries. These complementarities arise from the existence of shared competencies and depend on the cognitive proximity among local actors (Nooteboom 2000). In spite of the fact that the concept of related variety, and the ways through which it can be measured, have been criticised (e.g. Ejermo 2005; Bishop and Gripaios 2010; Brachert et al. 2011; Desrochers and Leppälä 2011; Wixe and Andersson 2017), it is widely employed in many empirical studies. These works generally identify a positive effect of related variety on the economic performance of firms and regions, regardless of the country analysed or the econometric meth- odology adopted (e.g. Quatraro 2000; Cainelli and Iacobucci 2012; Cainelli et al. 2016; Lazzeretti et al. 2017). Despite the relevance of the notions of related variety and industrial relatedness for economic geography and regional economics, the analysis of their impact on regional resilience is scarce. This is surprising given that, according to the evolution- ary approach to regional resilience (Boschma 2015), industrial relatedness seems to play a relatively important role with respect to the short-run ability of regions to absorb an external shock. The few studies that focus on this issue find a positive effect of related variety and industrial relatedness on regional resilience, that is, they confirm the “shock absorbing” ability of related diversification. In fact, these char- acteristics of an industrial structure seem to exert a positive effect on local resilience since skills, capabilities and technologies can be reallocated more rapidly across industries, thus improving the ability of a local system to adjust to an external shock (Cainelli et al. 2019). The contribution by Sedita et al. (2017) is of particular inter- est for the Italian case. They employ an Ordinary Least Squares (OLS) economet- ric strategy and confirm that related variety has a positive effect on local resilience measured as the growth rate of the employment after the 2008 Great Recession. The present paper develops this kind of analysis in two directions. First, regional resilience is defined following the approach proposed by Lagravinese 13 Does related variety affect regional resilience? New evidence… 659 (2015), who proposes an index which captures the ability of a region to “resist” to an adverse shock with respect to the rest of the country. Thus, resilience is not identified simply with local employment growth, but rather with the ability of the local system to react to an exogenous shock—in this case, the 2008 Great Recession. A value of this resilience indicator lower than zero means that the local system is highly sensitive to an exogenous shock, while a value greater than zero indicates a highly resilient local system. Second, the empirical analysis, per- formed at the Italian local labour market (LLM) level, employs spatial econo- metric techniques in order to provide a more accurate picture of the related vari- ety–resilience relationship. The empirical results, obtained through the estimation of Spatial Durbin Error Models (SDEM), suggest that LLMs characterised by a higher level of related vari- ety have shown a higher capacity to adapt to the Great Recession over the 3-year resilience period 2010–2013, while not over the 1-year resilience period 2012–2013. In other words, the analysis partly identifies related variety as being a shock absorber. These findings are strongly linked to the recent debate on the Smart Spe- cialisation Strategy (S3). In fact, related diversification at the regional level is one of the key concepts underlying this European regional strategy, which assumes that new productive and technological specialisations can arise only through a dynamic process that moves from a set of related industries. In this sense, the results pre- sented in this paper—which emphasise the role played by related variety in reinforc- ing the capacity of a region to resist to an external shock—confirm the relevance of industrial relatedness in fostering new regional development patterns by exploiting vertical and complementary relationships with the pre-existing activities of a region. The rest of the paper is organised as follows. The second Section discusses the related literature; the third Section describes the dataset and the econometric meth- odology adopted; the fourth Section presents and discusses the empirical findings; the fifth Section concludes the work. 2 Related literature 2.1 The concept of regional resilience The concept of resilience, whose popularity has increased in recent years, is multi- faceted and, if not properly defined and contextualised, can result in confusion. The literature considers different dimensions of resilience, including its definition, and the ways it can be operationalised and measured empirically.1 In relation to its definition, the literature proposes three different interpretations (Angulo et al. 2018). The so-called “ecological approach” defines regional resil- ience as the region’s capacity to move from one possible steady-state path to another without changes to its structure, identity and function (Holling 1973; Reggiani 1 Note that, frequently, both definition and measurement depend on the analytical context (e.g. economic vs. natural shocks). For details, see Modica and Reggiani (2015) and Faggian et al. (2018). 13 660 G. Cainelli et al. et al. 2002). The so-called “engineering approach” defines regional resilience as the region’s capacity to return to a persistent steady-state equilibrium following a shock (Pimm 1984; Rose 2004; Fingleton et al. 2012). Finally, the so-called “evolutionary approach” defines regional resilience as the ability, following a shock, to adapt in the short run or to develop new growth paths in the long run (Martin 2012; Boschma 2015). All these definitions share a common feature: the presence of a certain threat- ening event, such as a natural disaster (e.g. the Northern Italy earthquake), a terror- istic attack (e.g. the September 11 attacks) or a financial crisis (e.g. the 2008 Great Recession). Similarly, operationalisation and measurement issues are important. For instance, typically, resilience to natural disasters is analysed through indices, while the analy- sis of economic shocks is based mainly on econometric models. Both simple and composite indicators can be used to assess the resilience of a given territory (Mod- ica and Reggiani 2015). According to Martin (2012) and Martin and Sunley (2015), the ratio between the drop in regional employment or output and the corresponding drop in the country as a whole is an appropriate simple indicator to evaluate the regional resistance to recessions. In the case of composite indicators, the selection procedure of variables ranges from the identification through the study of previ- ous literature (Cardona et al. 2008; Cutter et al. 2008; Briguglio et al. 2009; Foster 2011) to statistical methods based on factor analysis (Graziano 2013). In the con- text of econometric analyses, most of them are based on time series approaches. For instance, Fingleton et al. (2012) and Cellini and Torrisi (2014) test for differences in regional resilience through Seemingly Unrelated Regression Equations (SURE) models. Sensier et al. (2016) operationalise regional economic resilience by adopt- ing a business cycles approach, which allows for the measurement of comparability in a cross-country analysis. Finally, Di Caro (2017) analyses both engineering and ecological resilience using nonlinear smooth transition auto-regressive models. 2.2 Related variety and regional resilience The industrial structure is generally considered a key determinant of local resilience. Starting from this insight, a new stream of research in economic geography and regional economics has enabled a deeper analysis through a focus on a specific fea- ture of the regional industrial structure: the level of related diversification or indus- trial relatedness. This literature strand, which originated from the debate on related variety (Frenken et al. 2007), investigates the role played by related variety in terms of regional resilience from two different time perspectives. In fact, according to the evolutionary approach (Boschma 2015), industrial relatedness may have a positive effect not only on the ability of a region to absorb an external shock in the short run, but also on its ability to develop new long-run growth paths. Looking at the short-run effects of industrial relatedness, Balland et al. (2015) investigate the technological resilience of US cities over the period 1975–2002. They find that cities with knowledge bases with high levels of relatedness with respect to the set of technologies in which they do not (yet) possess a comparative advantage have a higher tendency to avoid crises, or to limit the intensity and duration of a 13 Does related variety affect regional resilience? New evidence… 661 crisis event. Diodato and Weterings (2015) use Dutch data on 12 regions and 59 sectors to investigate how embeddedness of input–output linkages, skills relatedness and connectivity contribute jointly to the resilience of regional labour markets to economic shocks. They find that labour markets in centrally located and services- oriented regions recover more quickly irrespective of the type of shock hitting the economy. Sedita et al. (2017) measure regional resilience simply as the growth rate of the employment following the Great Recession and, through an OLS approach, find that related variety has a positive effect on the resilience of Italian LLMs. They also investigate the role played by the local differentiated knowledge base (synthetic, analytical, symbolic) and find some interesting counter-intuitive results: symbolic and synthetic knowledge bases have a positive effect on regional resilience, while the role of the analytical knowledge base is negligible. Cainelli et al. (2019) analyse the relationship between industrial—technological vs. vertical, i.e. market-based— relatedness and economic resilience across European Union regions over the crisis period 2008–2012 and find that technological relatedness has a positive effect on the resilience probability in the very short run (i.e. over the period 2008–2009), while vertical relatedness has a negative effect that seems to persist for longer. The long-run evolutionary approach to regional resilience was developed by Xiao et al. (2018). The main idea is that industrial relatedness can be a determinant of both long-run economic development and long-run regional resilience. Xiao et al. (2018) investigate the ability of 173 European regions to develop new industrial speciali- sations after the 2008 Great Recession, assuming industrial relatedness as a major determinant. They propose four measures of industrial proximity: unrelated variety, related variety, industrial relatedness—measured as the average proximity among the industries of specialisation with respect to the other industries located within a region—and technological relatedness—measured using a Los (2000) index. Their main finding is that industrial relatedness has a positive effect on regional resilience following the crisis only in the case of knowledge intensive sectors. The present paper focuses on the short-term, adaptive dimension of local resil- ience and on the role played by related variety as a shock absorber. The concept of related variety assumes that it is not the level of productive diversification per se which matters, but the presence of diversified domains that allows the exploitation of complementarities across different sectors (Frenken et al. 2007). Complementa- rities arise from existing shared competencies, and their diffusion depends on the level of cognitive proximity among local actors (Nooteboom 2000). The diversified productive structure of a local system can improve the opportunity to interact, copy, modify and recombine ideas, practices and technologies across industries. These processes can lead to the development of new products and services. Also, they can favour the transfer of skills, capabilities and technologies among the industries in the same local system. For these reasons, related variety is expected to have a positive effect on regional resilience since skills, capabilities and technologies can be rapidly reallocated across different local industries sharing the same knowledge base, thus improving the capacity of a local system to respond to an external shock. From this perspective, related variety can be considered as a short-run shock absorber. It is worth noting that the present paper adopts a measure of related variety which is based on a standard statistical classification of industries. As it is known, this measure 13 662 G. Cainelli et al. has been criticised from different perspectives in the literature on related variety. Some authors (e.g. Boschma et al. 2012) underline that standard statistical classifications use some priors to establish industrial relatedness, such as similarities in product charac- teristics or in production technologies. Moreover, the use of statistical classifications of industries does not allow to take into account “similarities in regulatory framework, complementarities in their use, the intensive use of a certain type of infrastructure, the use of advertisement to build trade marks, etc.” (Boschma et al. 2012, p. 242). Other authors suggest that related variety, and, consequently, its measures, should capture localised knowledge spillovers in terms of individual skills (Desrochers and Leppälä 2011; Wixe and Andersson 2017). The main idea behind these studies is that much of the learning and knowledge spillovers in bounded geographic areas, such as regions or local systems, takes place at the level of individuals and not at the level of indus- tries (Wixe and Andersson 2017). This is the reason why some authors adopt different measures of related variety. For example, Boschma et al. (2012) define related variety on the basis of two indicators, namely the geographic correlation of employment across traded industries and the products’ proximity index which is based on the probability that a region develops a comparative advantage in two products (Hidalgo et al. 2007). On the contrary, Wixe and Andersson (2017) develop measures of relatedness in edu- cation and occupation to capture knowledge spillovers among individuals. We acknowledge the relevance of these criticisms. However, the present paper focuses explicitly on the role of specific characteristics of the industrial structure of a region as key determinants of local resilience. What really matters for the present anal- ysis is the role played by industry-based related variety as a short-run shock absorber. This is the main motivation for the adopted measurement approach. 3 Empirical framework 3.1 Measuring local resilience and related variety The empirical analysis focuses on the relationship between industrial relatedness and local economic resilience. The spatial unit of analysis is the Italian LLM, which is a functional area encompassing the municipality, and identified on the basis of workers’ commuting flows. Therefore, LLMs are defined according to economic—rather than administrative—criteria. The literature has proposed different indices to capture resilience at the local/ regional level. This paper employs the index developed by Lagravinese (2015), who modified that proposed by Martin (2012) to account better for asymmetric regional behaviours and longer time periods. Specifically, economic resilience is evaluated over two post-crisis periods, i.e. the 1-year period 2012–2013 and the 3-year period 2010–2013, according to the following equation: ET −Et ∑686 ∑686 l=1 ElT − l=1 Elt l Elt l − ∑686 t l=1 El ResilienceT−t = (1) l ∑686 ∑ ElT − 686 t l=1 El l=1 ∑ 686 t l=1 El 13 Does related variety affect regional resilience? New evidence… 663 Fig. 1  Italian GDP dynamics where El denotes employment in LLM l = 1, … , 686, t = 2010, 2012 and T = 2013. According to Eq. (1), a LLM can show a resilient behaviour with respect to the country if ResilienceT−t l > 0, a non-resilient behaviour if ResilienceT−t l < 0 or a neu- tral pattern if Resiliencel = 0.2 T−t The resilience index is constructed for two periods: the 3-year period 2010–2013 and the 1-year period 2012–2013. As Fig. 1 shows, the Italian GDP recorded an increase following the 2008 Great Recession in the year 2009, although it reached its pre-crisis level only in 2010, when it also recorded a positive annual growth rate. However, a new recession affected the Italian economy over the 2011–2012 period, which was characterised by a negative annual GDP growth. Then, the GDP started to increase again from the year 2012. Therefore, the analysis considers these two short-run post-crisis periods in order to evaluate the effects of industrial relatedness over time periods of pure recovery. 2 According to Martin (2012), a simple indicator to evaluate regional resilience is the ratio of the change in regional employment (or output) to the respective change for the country as a whole. Specifically, the index of resilience is defined as follows: ∑686 ElT l=1 El T Martin_ResilienceT−t l = −1 ∑686 t −1 Elt l=1 El where all terms are defined as for Eq. (1). Hence, LLMs characterised by a resilience index value greater than one are highly sensitive to exogenous shocks, that is, they show a low level of resilience, while LLMs characterised by a resilience index value lower than one show a high level of resilience to exog- enous shocks. However, such an indicator works only if the country (as a whole) and all its sub-national units show a negative performance during the period considered. With particular reference to the Italian case, the country recorded a negative employment growth rate of about − 2% during both periods con- sidered in the present analysis. However, about the 16% of Italian LLMs recorded a positive employment growth rate during the 1-year period 2012–2013, and about the 19% recorded a positive employment growth rate during the 3-year period 2010–2013. In addition, two out of 686 LLMs experienced a zero- growth rate in employment during the 3-year period 2010–2013. Given this evidence, the index proposed by Martin (2012) could not behave properly to capture regional resilience in the present analysis. 13 664 G. Cainelli et al. Table 1  Structure of resilient Resilience period LLMs ResilienceT−t regions l No. % 2013–2012 >0 257 37.46 =0 0 0.00 0 238 34.69 =0 0 0.00