1 Introduction

Economists view technical change as a driving force of industrial transformation (Freeman and Louçã, 2001). As industrial transformations do not happen overnight, and the outcomes of technical change are not only uncertain and enduring but co-depend on the regional context, the choice of future technical trajectories in regions justifies policy intervention.

Two distinct ways are open to policymakers aiming to stimulate economic growth through new technology, possibly resulting in industrial transformation. Both see spillovers as drivers of the industrial transformation of regions (Caniëls and Romijn, 2005). First, following the 3% objective in the EU, policymakers may attempt to stimulate internal research and development (R&D) activities to increase the productivity of domestic firms (Hall et al., 2010; Acharya, 2015). R&D spillovers are externalities resulting from an inability of R&D active firms to appropriate all benefits from their in-house R&D activities (Ugur et al., 2020) and thereby influence productivity levels of other firms and industries (Gonçalves et al., 2017). This early definition has been complemented by including all voluntary transfer of R&D activities (Gonçalves et al., 2017). R&D spillovers by firms thus induce industrial transformation from within the region as new knowledge and technologies created in local firms spill over to other industries and regions (Li & Bosworth, 2020).

Second, policymakers may want to attract foreign direct investment (FDI) using instruments such as providing infrastructure, tax measures and regulatory advantages, expecting this imported new technology to boost the productivity of domestic firms (Howell, 2020; Liang, 2017). This paper considers FDI as a source of non-regional knowledge with the potential to create new development paths (Padilla-Pérez & Gaudin, 2014; Trippl et al., 2018). FDI spillovers, generated by foreign-owned firms, may impact domestic firms’ productivity through the effect of non-regional knowledge on technology transfer, integration with global value chains and access to new markets (Alfaro et al., 2004) thus inspiring industrial transformation (Crespo et al., 2009; Barge-Gil et al., 2020).

R&D and FDI spillovers are considered important drivers of future industrial activities resulting in new path creation and future technological trajectories. Stimulating R&D and attracting FDI demand a continuous effort because these investments have substantial gestation periods, and their results co-depend on the type of linkages and the regional context. But, as we will demonstrate, the type of spillovers and their contribution to industrial transformation differ according to the existing regional development path as characterised by organisational thickness and specialisation of the regional innovation system.

The future path development of regions depends on different factors (Asheim, 2019; Hassink et al., 2019; Sotarauta et al., 2021). So far, the role played by different sources of spillovers on the productivity of other firms received little attention, although the debate has been launched recently (Barge-Gil et al., 2020; Ben Hassine et al., 2017). This paper contributes to this debate by looking at the effects of two sources of spillovers: R&D activities by indigenous firms creating new knowledge and technology, and the presence of foreign-owned firms bringing in advanced knowledge and technologies from abroad. Originally, research on spillovers focused at the national level, but an awareness of the importance of spillovers at regional levels has been increasing (OECD, 2013). From this perspective, regions have to invest in their factor endowments to transform their industry to avoid lock-in and path dependency (Martin & Sunley, 2006). FDI spillovers differ across regions as the capacity of domestic firms to absorb these spillovers may vary between regional innovation systems (Kallio et al., 2010), and increasing physical distances between firms can lead to decreasing spillover effects (Merlevede & Purice, 2016).

The paper brings together two unrelated sets of literature. The first one discusses types of spillovers within and between industries, reflecting technological relatedness. Second, these spillovers occur in a regional context described by three related concepts: regional innovation systems, path dependency, and smart specialisation.

The first research question is whether R&D or FDI spillovers translate in productivity effects on domestic non-R&D active firms. Additionally, we study the combination of R&D and FDI, which has only recently been of interest (Liang, 2017, Ben Hassine et al., 2017; Li & Bosworth, 2020). Conflicting views exist on the relation between FDI and R&D inputs. R&D inputs can increase when foreign-owned firms expand their R&D activities to realise scale and scope economies. But they can decrease if foreign-owned firms downsize their R&D activities or transfer them to other locations abroad.

The second research question investigates different types of linkages (Barge Gil et al., 2020; Merlevede, 2020). Linkages are especially relevant as these provide insights for regions as to what extent their future path development hinges on the creation of internal technology and knowledge, or the attraction of external technology and knowledge. The paper investigates if the productivity effects for domestic non-R&D firms vary according to the type of regional linkages of the firm.

The paper is structured as follows. Section two reviews the relevant literature on spillover types, industry linkages and technological relatedness in a regional context. The regions in Belgium, introduced in the third section, serve as a framework for studying the role of spillovers and industry linkages as they cover different regional development paths. Section four presents the data, variables, and empirical strategy. Section five discusses the analyses and interprets the mechanisms behind the main findings. Section six summarises the findings and looks at policy implications.

2 R&D and FDI spillovers: industry linkages and regional context

Spillover literature focuses on the existent linkages between industries. These linkages can be of an intra-industry or inter-industry nature. The assumption of intertwined industries builds on a perception of the economy as an interdependent system where interaction is a decisive factor in economic growth. Empirical verification of economic interaction in terms of flows of physical goods and services starts with input–output linkages. The spillover literature often draws on input–output tables (e.g. Gonçalves et al., 2017; Merlevede et al., 2014), which are also applied at the regional level (e.g. Cerqua & Pellegrini, 2020).

Based on the literature, industry-level linkages play a crucial role in future technological development and regional economic performance (Asheim & Isaksen, 2002; Boschma & Iammarino, 2009). The literature on technological relatedness and related variety in regional economics can be combined to look at spillovers, and the first subsection discusses both briefly. The second subsection aims to add context to these industrial linkages by discussing regional innovation systems, which are often relatively static descriptions necessitating the concept of path dependency to determine the possible outcomes. The discussion on smart specialisation introduces the way policymakers could intervene to stimulate firm-level productivity.

2.1 Spillovers, industry linkages and technological relatedness

In performing R&D, enterprises are constantly looking for future benefits by developing new technologies, introducing new products and implementing new processes. However, their R&D efforts create the possibility of spilling over to other firms (Hall et al., 2010). R&D active firms add to the existing R&D stock, which facilitates intra-industry R&D spillovers because of the potential to lower the cost of inputs to firms within the industry (Li & Bosworth, 2020). Two potential effects result from this (Hall et al., 2010; Lucking et al., 2019; Ugur et al., 2020). First, a positive effect because spillovers increase the productivity of domestic non-R&D active firms using related technology. Second, a negative effect because spillovers reduce the value of firms as their competitors outperform them. Lychagin et al. (2016) demonstrate that these negative effects are not always relevant and, at the same time, stress the relevance of the geographical nature of spillovers because proximity tends to increase the spillover effects. However, being close to one another does not guarantee R&D spillovers. A large body of literature on the existence of technologically related activities versus unrelated ones states that spillovers may ultimately transform a region’s industrial structure as new industries emerge from different but related ones (Frenken et al., 2007).

The literature referring to spillovers induced by input–output linkages proves relevant. Upstream spillovers refer to upstream inter-industry linkages from R&D active or foreign-owned firms interacting with input-producing suppliers in other industries. Downstream inter-industry spillovers result from linkages where R&D active or foreign-owned firms sell their products or services to customers active in other industries. Upstream and downstream spillovers thus occur across different industries. Javorcik (2004) studied the productivity effects from FDI from upstream and downstream linkages. Spillover effects of R&D activities by domestic firms and FDI associated with the presence of foreign-owned firms, multinational enterprises or their affiliates are the result of intra-industry or horizontal linkages, upstream inter-industry linkages, and downstream inter-industry linkages (Li & Bosworth, 2020; Li & Tanna, 2018; Merlevede et al., 2014). Firms are classified in industries with a particular place in the supply chain and stand relative to other industries. The input–output idea captures this interconnectedness. Intra-industry spillovers occur between firms or competitors in the same or a closely related sector (Li & Bosworth, 2020). The mechanism behind it is that R&D active or foreign-owned firms interact with domestic firms as they transfer resources or share market information, technical equipment, and product specifications (Li & Tanna, 2018).

FDI spillovers cover embodied spillovers and differ from R&D spillovers in being more likely to affect labour productivity in upstream industries (Li & Bosworth, 2020). Lorentzen and Barnes (2004) point out that FDI spillovers prevent unnecessary duplication of research. Li and Tanna (2018) find, for Chinese manufacturing firms, that short-run spillovers from FDI exert a negative impact. In contrast, they become positive in the long run, irrespective of their stemming from intra- or inter-industry linkages. This ambiguity makes it difficult to estimate the expected effects on productivity. Liu (2008) corroborates this absence of positive spillover effects on productivity for intra-industry linkages but finds positive FDI spillover effects for upstream inter-industry linkages. Intra-industry linkages in studies on FDI found no spillover effects because foreign-owned firms are reluctant to share their knowledge and technology with domestic competitors (Liang, 2017). Others find a negative effect of FDI on the productivity of domestic firms (Lu et al., 2017) or an ambiguous effect (Atallah, 2002).

Rodríguez-Pose and Crescenzi (2008) acknowledge that knowledge spillovers have internal and external origins. This raises the question of optimising the effects of internal R&D spillovers and external FDI spillovers on productivity, and if there is a possibility that regional development can thrive on FDI spillovers alone. Both R&D and FDI spillovers are sources of productivity growth as new technological insights can result from internal R&D activities deployed within firms or bring in externally through embedded technology in foreign-owned firms. These internal or external sources of technological progress may not be unconnected (Ben Hassine et al., 2017; Liang, 2017). The productivity effects of FDI have been extensively studied in the recent decade (Ugur et al. 2020), as are the productivity effects of R&D activities (Hall et al., 2010).

In addition to this discussion on the importance of R&D and FDI spillovers, it is helpful to look at the empirical findings of these spillovers on productivity because spillovers do not happen automatically and involve a costly learning process. Inter-industry R&D spillovers from foreign-owned firms occur because of the imitation effect. Ben Hassine et al. (2017) find that upstream spillovers are more likely than downstream spillovers. Javorcik (2004) reports evidence for positive productivity spillovers from foreign-owned firms to their local suppliers in upstream industries in Lithuania.

Next, the ideas on technological relatedness and related variety are also relevant in the study of spillovers because of their connection with regional new path development and economic renewal (Asheim et al., 2011). Related variety refers to a diversity of existing industries that are technologically and cognitively related (Frenken et al., 2007). This concept captures the extent to which different industries in a region are sufficiently alike to facilitate knowledge exchange and spillovers. A common way to measure relatedness uses entropy measures to estimate the concentration of industries with similar activities. Another way to capture the related- or unrelatedness between industries is through input–output linkages. The method used will inevitably influence the estimations of the effects of the relation between linkages and productivity. Frenken et al. (2007) argue that spillovers primarily happen between related industries and, to a lesser extent, between unrelated industries. Inter-industry spillovers occur more frequently between industries using related knowledge and technologies. Technological and cognitive proximity facilitate these spillovers. On the other hand, unrelated variety points to diversified industrial linkages, which reduce regional dependence on a few industries and safeguards from industry-specific shocks that may cause unemployment and even a regional downturn (Boschma & Iammarino, 2009). Related variety in regions with a diversified industrial structure positively impacts regional economic development (Boschma & Iammarino, 2009; Frenken et al., 2007; Neffke et al., 2011).

Empirical studies on unrelated variety provide mixed results (Content & Frenken, 2016). Boschma and Frenken (2011) highlight the importance of the regional context in which firms operate and where the industrial structure is an essential factor, and firms’ ability to capture technological information affect the regional path development. This regional context invites a thorough discussion of the economic history of a region, necessitating quantitative and qualitative information. Longitudinal research is needed to examine how technological competencies accumulate over time and how the economic structure’s diversification is impacted (Boschma and Gianelle, 2014).

The literature on related variety tackles the optimal regional mix of industries that open possibilities for R&D and FDI spillovers to allow for a promising future path development. This question invites a policy response such as offered by smart specialisation. If related variety can be used to evaluate the industrial structure, smart specialisation policy can encourage regional diversification by identifying new economic activities leading to future development paths (Boschma and Gianelle, 2014). The geographical aspects of knowledge linkages between related industries point to the possibility of integrating foreign knowledge with regional connections to transform the industrial structure (Fløysand et al., 2017).

2.2 Insights on the regional context

The issue of future regional path development invites insights from three related streams of literature: regional innovation systems, path dependency, and smart specialisation.

First, the vast literature on regional innovation systems states that innovation results from complex interactions between different actors in the economy (Doloreux & Parto, 2005). Regional innovation systems consist of a network of nodes and linkages between different actors, agencies and organisations beneficial to initiate innovative activities (Cooke et al., 1997). A regional innovation system offers a framework that shape firm-level activities by describing inter-firm linkages, support infrastructures, and institutional configurations (Asheim & Isaksen, 2002). Human and social capital are a precondition for linkages to take place. Knowledge spillovers are pivotal for the functioning of regional innovation systems and should be promoted (Rodríguez-Pose and Crescenzi, 2008). Exchanging knowledge between co-located firms in a region is facilitated by sharing a homogeneous institutional context with similar norms, values, language, and laws (Agrawal & Cockburn, 2003). Co-location also helps to establish a shared interpretative framework facilitating inter-organisational learning (Berchicci et al., 2016).

Regional innovation systems refer to spatial entities possessing autonomous political power (Asheim & Isaksen, 2002; Cooke et al., 1997; Uyarra & Flanagan, 2010). The resurgence of the concept owes to the need of EU and regional policymakers to promote innovation through smart specialisation policy. Asheim (2019) argues that regional innovation systems offer a strategy for regional economic development. It describes a process of economic diversification generating structural change and offers a good starting point for future policy design directed to regional economic development and highlights the relevant components. Moreover, the concept marries top-down and bottom-up policy approaches. Firms’ linkages are instrumental in the bottom-up approach because of the entrepreneurial discovery process that requires input from regional stakeholders.

As Rodríguez-Pose and Crescenzi (2008) posit, innovation is spatially embedded and dependent on social and institutional conditions present in every space. The concept of regional innovation systems highlights these institutional contexts (Cooke et al., 1997). The concept of regional innovation systems acknowledges the uneven spatial economic development due to region-specific capabilities in innovation and competitiveness.

Can each autonomous region characterise as a regional innovation system? Doloreux and Parto (2005) review the literature and find that regions can be classified as regional innovation system: advanced regions, old industrial regions, peripheral regions, rural regions and regions in transition, although they raise doubt that all regions can be characterised in this way. Cooke et al. (1997) consider the regions in Austria, Belgium, and Germany to be examples of strong regional innovation systems. Isaksen and Karlsen (2013) demonstrate that even small regions, such as the Brussels-Capital Region in Belgium, can exert regional advantages.

Trippl et al. (2018) add the dimension of organisational thickness to the regional innovation systems. Three types of regional innovation systems are distinguished: organisationally thick and diversified, organisationally thick and specialized, and organisationally thin. The difference between thick and thin organisational structures refers to the extent of the endowment of more or less innovative firms and other innovative organisations such as universities, research centres, etc. (Trippl et al., 2018). This paper draws on the conceptual insights concerning the organisational thickness of regions and posits that the role played by R&D and FDI spillovers in a region will vary according to these types. We explore empirically if the three types of regional innovation systems differ in their reliance on R&D or FDI spillovers by estimating our model by region. Spillovers depend on the capacity of regions to create (R&D) and attract (FDI) technology.

New path development is possible in organisational thin regions without a critical number of R&D active firms, provided that they use exogenous knowledge sources or if the location of a dominant technological firm spurs local innovation and development (Isaksen & Trippl, 2017). Therefore, organisationally thin regional innovation systems rely more on FDI spillovers than on R&D spillovers. Tapping into extra-regional technology and knowledge flows occurs in two forms: either by the presence of foreign-owned firms, MNEs and foreign subsidiaries or by setting up permanent or temporary knowledge pipelines to sources abroad (Bathelt et al., 2004). The interaction between R&D spillovers and FDI spillovers emphasise the need for absorptive capacity—often proxied by R&D—to identify, assimilate and transform external knowledge (Cohen and Levinthal, 1990). The absence of absorptive capacity hinders the firms in organisationally thin regions to absorb this foreign-owned technology and knowledge.

Second, the literature on path dependency bears relevance for its insights into the evolution of regional development. Early approaches to path dependence emphasise the persistence of regional industrial specialisations leading to lock-in problems (Martin & Sunley, 2006). The recent literature focuses on economic restructuring and new regional path development (Trippl et al., 2018). When regions host existing firms predominantly focused on developing incremental innovations and are specialised in low tech industries, this may lead to low productivity levels and a stagnating or declining economy by path extension, which may even result in path exhaustion (Hassink, 2010). When regional industries are populated by firms that can switch to different but related industries, they may follow path renewal (Boschma, 2015). When firms look for innovations in unrelated industries to forge more radical innovations, the regional economy may embark on a new path creation (Martin & Sunley, 2006; Tödtling & Trippl, 2013). To spur new path creation, firms must be at the forefront of R&D activities and the regional authority provides a policy framework and supportive institutional structures (Boschma & Capone, 2015). Martin and Sunley (2006) explicitly recognize that path dependency allows technology sources outside the region to incite new path creation. Consequently, the advent of dominant firms that embed in the regional economy may drive new path development. FDI spillovers from foreign-owned firms are one way of initiating a new development path (O’Malley & O’Gorman, 2001). MacKinnon (2012) posits that the link between FDI and regional economic effects is intricate, path-dependent, and varies with the type of region. The assumption is that the link between FDI and regional development is positive in advanced regions, while negative in vulnerable regions because of the danger that these might become locked-in and dependent on the whims and exigencies of foreign-owned firms and multinationals.

Third, the popular policy concept of smart specialisation was born out of productivity differences between the US and the EU (Foray et al., 2009). Technological linkages and spillovers between industries and regions have been heralded as one of the explanations of productivity differences and hence also one of the drivers of smart specialisation (McCann & Ortega-Argilés, 2015).

Smart specialisation is an evidence-based policy aimed at identifying the engines of regional development conditional on the available resources and technological domains present in industries. By studying the related industries that may improve the economic structure, smart specialisation prioritises industries that draw on the available resources (Asheim, 2019). This policy draws on the region’s existing strengths and reveals concealed opportunities with potential high value-added activities (Balland et al., 2019). Three aspects underpin smart specialisation: specialised diversification; identification of technology domains; and specific roles of private stakeholders, entrepreneurs and regions.

The smart specialisation policy aims to select and prioritise the most promising technology domains (instead of envisaging industries) that complement the region’s existing economic structure (McCann & Ortega-Argilés, 2015). This selection is made in a bottom-up way to spur specialised diversification, enhancing the region’s competitiveness and expand or create promising developments paths to bring about structural change generating economic growth (OECD, 2013). However, smart specialisation explicitly acknowledges the importance of inherited economic structures, capabilities, resource endowments, and therefore their previous growth paths (McCann & Ortega-Argilés, 2015).

Smart specialisation owes its popularity in science policy milieus to being a condition for European funding to increase synergies and avoid duplication, fragmentation and imitation (OECD, 2013). By selecting existing or new technology domains for innovation-led growth, smart specialisation policy aims to concentrate the limited public funds in knowledge-related investments. Smart specialisation policy seeks to stimulate innovation through careful use of public spending. Regional policymakers have to acknowledge that innovation will not be equally successful in all industries. They will have to choose and prioritise industries in which they want to invest public money conditional on the available resources in their region. Regions are considered the appropriate spatial level to spur innovation and productivity as smart specialisation policy complements the top-down directives from the European Union with bottom-up insights emerging from region-level considerations (Asheim et al., 2016). Smart specialisation thus investigates the potential of regional innovation systems to develop new economic activities and change the development path using distinctive and diversified areas of specialisation instead of duplicating other regions (Foray, 2015). Many regions are susceptible to such changes (OECD, 2013; Rissola & Sörvik, 2018).

3 The heterogeneous regional development paths in Belgium

We exemplify the research questions by focussing on (the regions of) Belgium. Belgium bears a slight resemblance to the European Union as each federated authority is, according to its constitution, responsible for the science, technology, and innovation policy on its territory (Belspo, 2010). From the 1970s, the process of decentralisation of the Belgian government leads to increased differentiation of regulations, institutions, and instruments dealing with each separate region’s economic and social needs. In Belgium, there are three regions (at major socio-economic or NUTS1 level) with very different development paths. Anecdotally Belgium consists of the ‘administrative’ Brussels Capital Region, the ‘high-tech’ Flemish Region, and the ‘restructuring’ Walloon Region (Capron, 2000). And the divergence is not limited to the economic dimension but also visible at the political, social and religious level (Billiet et al., 2006). In terms of science policy, the regions face different challenges and each targets specific instruments (Kelchtermans & Robledo-Böttcher, 2018).

Depending on the objective of peer group comparisons, many regional categorisations are possible (Marsan & Maguire, 2011). It comes as no surprise that official indicator-based sources arrive at diversified classifications. The Regional Innovation Scoreboard (2019) data uses 18 indicators at the NUTS1 level classified into four categories: innovation activities, framework conditions, impacts, and investments. According to these indicators, the Scoreboard considers the Brussels Capital Region an innovation leader, the Flemish Region posits as a strong + innovator, and the Walloon Region figures as a strong innovator (Hollanders et al., 2019).

The Regional Competitiveness Index uses 11 separate dimensions in three pillars to classify regions, defined at NUST2 level, i.e. the primary regions relevant to European regional policies. In this view, Brussels ranks as a highly competitive region outperforming the Flemish Region because of its score on the innovation pillar. Relative to the other two regions in Belgium, the Walloon Region is the least competitive (Annoni & Dijkstra, 2019).

The regional classification offered by the European Quality of Government Index captures the opinions of respondents on the quality of public services, impartiality and corruption (Charron et al., 2019). This is a relevant indicator for innovation as a growing consensus develops in academic and policy circles that the quality of institutions and governments matters for economic development (Rodríguez‐Pose and Garcilazo, 2015). According to this ranking, the Flemish Region has the most qualitative government, followed by the Walloon Region and the Brussels Capital Region.

The OECD uses cluster analysis on input and output indicators on the socio-economic and industrial structure to categorise NUTS1 regions (Maguire and Martins, 2011). It classifies Brussels as a knowledge hub because of its high GDP per capita linked to its small size and dense population, a substantial patent and R&D activity, a high share of knowledge-intensive service sectors and an educated labour force. The Flemish Region and Walloon Region as medium-tech manufacturing and service providers because of their outspoken medium–high and medium–low base in manufacturing linked with relatively high absorptive capacities due to a large percentage of the labour force with tertiary education.

Apart from their different characterisation in official statistics and rankings, the three regions in Belgium also differ according to their regional innovation systems. Regional innovation systems are chiefly relevant in regions with a well-endowed organisational and institutional support structure (Asheim, 2019). Trippl et al. (2018) offer a theoretical conceptualisation of regional differences and categorise regions according to their organisational thickness and specialisation. Organisational thickness refers to many firms in various industries, a critical mass of higher education institutes and research centres, and public support structures that stimulate innovative activities in promising technological domains (Binz et al., 2016; Boschma, 2015). In this respect, the Brussels Capital Region and the Flemish Region can be catalogued as organisationally thick. At the same time, the old industrial economic structure of the Walloon Region can be classified as organisationally thin. Trippl et al. (2018) further distinguish between specialised and diversified regions, whereby specialised regions show a presence of firms with similar economic and innovation activities. When organisational thickness characterises regional innovation systems, they provide a context promoting trust relations, reducing the danger of opportunism and uncertainty.

The Brussels Capital Region is an international administrative centre, hosts many knowledge-intensive business services, has a critical mass of knowledge organisations with many higher education institutions, and is home to many headquarters of MNEs. As such, it has a more diversified economic structure than the Flemish Region. Following the typology developed by Trippl et al. (2018), the Brussels Capital Region classifies as diversified and organisationally thick; the Flemish Region as specialized and thick; and the Walloon Region as specialized and organisationally thin.

4 Empirical strategy, data and variables

4.1 Empirical strategy

Although the empirical strategy closely follows the FDI spillover literature, this framework has recently also been applied to R&D spillovers (Li and Bosworth (2020); Barge‐Gil et al., 2020). We use an extended production function framework to identify spillover effects. Indicators on the foreign presence and R&D activity explain the total factor productivity (TFP) of domestic firms that do not perform R&D. As such, estimations capture pure spillover effects. We employ the typical two-step procedure in the literature where first an estimate of TFP is obtained, and then TFP is related to indicators of foreign presence and R&D activity. Endogeneity problems occur when estimating TFP because firms observe their productivity (shocks) and adjust their input choices accordingly. Several semi-parametric techniques account for this problem, among which those of Olley and Pakes (1996) and Levinsohn and Petrin (2003). We use the estimator introduced by Wooldridge (2009), which combines the benefits of the mentioned authors, whilst applying a joint GMM estimation that both enhances efficiency and accounts for serial correlation and heteroskedasticity while dealing with the points raised by Ackerberg et al. (2015).

Equation (1) shows a stylized version of the specification to test for productivity spillover effects. It relates TFP of firm i in industry j at time t to spillover proxies defined at the industry level (SPILLOVER), firm and time fixed effects, and a vector of other firm and industry controls (X). Spillover proxies are lagged one year reflecting that they are not instantaneous, but do not take long to materialise. The size, sign, and significance of the coefficients with the spillover variables then provide evidence of the existence, direction, and size of productivity spillover effects.

$${TFP}_{ijt}={\alpha }_{i}+{\alpha }_{t}+\Theta {SPILLOVER}_{jt-1}+\Gamma {X}_{ijt-1}+{\varepsilon }_{i}$$
(1)

Recent efforts in TFP estimation use firm-level quantity data (TFPQ) rather than deflated revenue data (TFPR). Katayama et al. (2003) emphasise that the use of TFPR will often confound higher productivity with higher mark-ups. As data on quantities are not available to us, results should be interpreted bearing this caveat in mind. To capture the evolution of mark-ups as much as possible we include the change in a firm’s market share and within industry competition defined by a Herfindahl index. We use the WIOD industry classification to determine industries. We further include initial market share and initial productivity levels, but these are subsumed in the firm fixed effects. The latter will help in capturing any TFP-contaminating factors that are stable over time (when mark-ups are relatively stable compared to technological change fixed effects will also help in better isolating technology). Finally, we include firm age and firm size, measured as lagged real total assets, as further control variables.

For the construction of spillover variables, Caves (1974) proposed to capture intra-industry FDI spillover potential as the share of output produced by foreign-owned firms in each industry. This measure is still the preferred proxy in most of the literature today. Typically, a firm is foreign-owned (F = 1) when a single foreign shareholder participates at least 10%. The intra-industry or horizontal spillover variable (HR) is then a measure of the degree of foreign presence in sector j at time t and is constructed as:

$${HR}_{jt}^{FDI}=\frac{{\sum }_{i\in j}{F}_{it}{Y}_{it}}{{\sum }_{i\in j}{Y}_{it}}$$
(2)

with Yit output of firm i in industry j at time t. \({HR}_{jt}^{FDI}\) is industry j’s share of output produced by foreign-owned firms. In the same spirit, we define \({HR}_{jt}^{R\&D}\), with R&Dit a dummy variable indicating whether firm i performs R&D in year t, as:

$${HR}_{jt}^{R\&D}=\frac{{\sum }_{i\in j}{R\&D}_{it}{Y}_{it}}{{\sum }_{i\in j}{Y}_{it}}$$
(3)

\({HR}_{jt}^{R\&D}\) is then the share of industry j’s output produced by firms doing R&D.

Javorcik (2004) introduced supply chain spillovers and defined the standard measure to capture them. Ideally, one would like to know the share of firm output sold to foreign or R&D firms to measure the potential for upstream inter-regional or backward spillover effects (BK). There are at least two problems associated with such a choice. First, information at this level of detail is rarely available for a large panel of firms. Second, even when available, the share of firm output sold to foreign-owned firms is likely to cause endogeneity problems if the latter prefer to buy their inputs from more productive domestic firms. Following Javorcik (2004), the literature therefore resorts to the industry-level proxy variable, \({BK}_{jt}^{FDI}\). It is defined as:

$${BK}_{jt}^{FDI} ={\sum }_{k if k\ne j}{\gamma }_{jkt}{HR}_{kt}^{FDI}$$
(4)

where \({\gamma }_{jkt}\) is the proportion of industry j’s output supplied to sourcing industry k at time t. These \({\gamma }_{jkt}\) s are calculated from input–output (IO) tables for domestic intermediate consumption. As firms cannot easily switch between industries for their inputs, one avoids a particular source of potential endogeneity by using the share of industry output sold to downstream domestic industries k with a given level of foreign presence \({HR}_{jt}^{FDI}\). Following Javorcik (2004), inputs sold within the same industry are explicitly excluded from the calculation of \({BK}_{jt}^{FDI}\) because this is captured by \({HR}_{jt}^{FDI}\). In the same spirit, the downstream inter-industry or forward spillover variable \({FW}_{jt}^{FDI}\) is defined as:

$${FW}_{jt}^{FDI}={\sum }_{l if l\ne j}{\delta }_{jlt}{HR}_{lt}^{FDI}.$$
(5)

where the IO-tables reveal the proportion \({\delta }_{jlt}\) of industry j’s inputs purchased from upstream industries l. Inputs purchased within the industry are again excluded. Equations (4) and (5) can be straightforwardly extended to their R&D counterparts using (3) rather than (2) as a basis for the calculation.

4.2 Data and variables

The analysis combines two primary data sources for Belgium: the R&D survey by Belspo and firm-level data from the Amadeus database by Bureau Van Dijk Electronic Publishing. Datasets are matched through firms’ VAT-numbers.

European Union member states are obliged to collect data on the R&D activities of firms located on their territory. In Belgium, these are biannually organised at the regional level according to and comply with the Frascati Manual’s guidelines (OECD, 2015). This R&D survey, organised since 1992, targets the population of firms known as R&D active in Belgium. Due to methodological breaks, this paper uses the period from 2000 to 2017. The responsible regional authority updates this list of R&D active firms regularly. The unit of analysis is the firm, i.e. the smallest legal entity. This R&D dataset contains 60,768 observations from 11,011 firms. Some observations have estimated results to comply with Eurostat regulations to calculate the nations’ R&D expenditure. Deleting these estimates yields 44,506 observations which contain annual information on the internal R&D expenditure, the R&D personnel in full-time equivalents (FTE) and sector information.

Amadeus is a widely used pan-European database of financial information on public and private companies. We focus on firms that file unconsolidated accounts. Every month Bureau van Dijk issues a new version of the database with updated information. However, a single version only contains the latest information on ownership, and firms that go out of business are quickly dropped from the database. Furthermore, because Bureau van Dijk updates individual ownership links between legal entities rather than the full ownership structure of a given firm, the ownership information on a specific issue of the database often consists of ownership links with different dates, referring to the last verification of a specific link. To construct a dataset with entry and exit, we employ a series of database issues, yielding consistent data for all firm-level variables in the model for 2000–2017. Firms are foreign-owned if a single foreign shareholder owns at least 10% of shares. Averaged over industry-year combinations, the dataset covers 46% of firms (with at least one employee), 79% of employment, 72% of output in the NACE revision 2 Structural Business Statistics (SBS) provided by Eurostat.

To calculate industry-level spillover measures defined in Eqs. (2) to (5), there are about 3 million firm-year observations on turnover in all sectors over the sample period. For the weights in constructing supply chain spillovers, we use the WIOD November 2016 Release (henceforth WIOD), which provides a time series of ‘World IO Tables’ for Belgium for 2000–2014Footnote 1 (Timmer et al., 2016). WIOD uses the Statistical classification of products by activity (CPA) which contains 56 industries, 19 of which are in the manufacturing sector (see Table 10 in “Appendix” for correspondence with the NACE Rev.2 2-digit industry classification). An advantage over other databases is that the WIOD varies over time and that information on goods imports does not rely on the standard proportionality assumption. Instead, we follow a more flexible approach whereby import proportions vary over end-use categories. This approach increases variability over time and intermediate input types. Table 1 shows summary statistics for the different spillover variables.

Table 1 Industry level summary statistics of main explanatory variables (1008 observations)
Full size table

Foreign-owned firms, on average, account for about 30% of industry output, but the 10th and 90th percentile reveal quite some heterogeneity. R&D active firms account for 29% of industry output on average and show a similar heterogeneity. Combining foreign-owned and R&D active firms yields three additional categories of firms. Foreign-owned R&D active firms account, on average, for 13% of industry output, suggesting that domestic and foreign-owned firms that do not perform R&D produce a non-trivial part of industry output. This raises the question of whether all three categories of firms carry the potential to generate spillover effects for domestic non-R&D active firms and be a catalyst for regional development. Table 6 in the “Appendix” shows that the average annual share in industry output of domestic R&D active firms and foreign non-R&D active firms mildly increases from 2000 to 2017, whereas the average annual share in industry output of foreign R&D performing firms doubles over the same period.

The estimation sample requires information on firms’ turnover, the number of employees, tangible fixed assets, and material inputs to estimate TFP. Given reporting requirements, especially material inputs are not available for many firms (see Table 2).

Table 2 Firm-level summary statistics of dependent and explanatory variables
Full size table

Nominal data are deflated with industry price data, taken from Eurostat, at the NACE 2-digit level. Real output is constructed by deflating operating revenues with industry-level producer price indices. Capital is calculated as tangible fixed assets deflated by the average of the following five industry deflators: machinery and equipment, office machinery and computing, electrical machinery and apparatus, motor vehicles, trailers and semi-trailers and other transport equipment. Real material inputs are calculated by deflating material inputs with a weighted intermediate input deflator. We construct weights from the input–output tables obtained from WIOD. We retain more than 10,000 firms per year (198,172 firm-year observations) in the TFP sample. The average firm in the TFP sample is larger than in the full sample but we retain smaller firms as well (see Table 2).

Table 3 presents a closer look at the TFP sample. At the Belgian level, 9.3% are domestic R&D active firms, 3.9% of observations consists of foreign-owned R&D active firms, and 20.1% are foreign-owned firms that do not perform R&D activities. The remaining two-thirds of the sample are domestic firms that do not perform R&D.

Table 3 Total factor productivity (TFP) summary statistics over firm categories and regions
Full size table

The Flemish Region hosts about two-thirds of all firms, 19% are in the Walloon Region, and 13% in the Brussels Capital Region. The distribution in the subgroups of firm categories is similar across the regions. However, foreign-owned firms are slightly more concentrated in the Brussels Capital Region and R&D active firms in Flemish Region.

Table 3 shows a clear ranking of firms in terms of TFP. Foreign-owned R&D active firms are the most productive firms, followed by foreign-owned firms that do not perform R&D, and domestic firms performing R&D. Domestic firms without R&D are least productive. We focus on the latter set of firms in the estimations as pure recipients of spillovers. Because of the use of lags and other control variables estimations use about 100,000 observations. Table 3 further shows that the Brussels Capital Region, irrespective of firm category, is the most productive region, followed by the Flemish Region and the Walloon Region.

5 Results and analysis

5.1 Foreign ownership versus R&D activity

Table 4 presents the basic results of the estimation of Eq. (1). Firms in the estimation sample are only domestic non-R&D active firms in Belgium.

Table 4 Productivity spillovers to domestic non-R&D active firms of R&D active and foreign-owned firms
Full size table

The first two models consider spillover effects from R&D active (Model 1) and foreign-owned (Model 2) firms separately. Model 1 shows the existence of upstream inter-industry spillovers of R&D active firms. Domestic non-R&D firms are thus more productive in industries that supply more inputs to industries with a higher presence of R&D active firms. In Model 2 there are significant inter-industry spillover effects from upstream foreign-owned firms. Model 3 confirms these findings only partially when we include both types of spillover variables simultaneously in the specification because only the existence of upstream inter-industry spillovers of R&D active firms plays a role. Based on two metastudies, Havránek and Irsova (2011) and Irsova and Havránek (2013) conclude that the FDI spillover literature consistently finds upstream inter-industry spillovers to be positive and significant (as in Model 2). In contrast, intra-industry and downstream inter-industry spillovers vary considerably across studies and average zero. A restrictive focus on foreign ownership may explain these findings of their metastudy. The result in Model 3 suggests that accounting for R&D activity might be essential and foreign ownership and R&D activity should be treated simultaneously.

To shed more light on this finding, Model 4 uses spillover variables for three separate categories of firms that potentially generate productivity spillover effects: domestic R&D active firms, foreign-owned firms without R&D activities, and foreign-owned firms with R&D activities. The first may be a critical spillover source as the rationale for FDI spillovers is foreign firms’ productivity bonus over local firms. From Table 3 we know that domestic R&D active firms outperform domestic non-R&D active firms in terms of productivity, making them a potential source of productivity spillovers as domestic non-R&D active firms can draw upon their technology stock. For intra-industry spillovers we find significant and positive spillover effects of domestic R&D active firms. A one standard deviation increase in the spillover variable is associated with a 3.6% higher productivity level. These firms could be either competitors or collaborators. The analysis suggests that the positive effect dominates. We do not find such an effect for intra-industry spillover effects for foreign-owned firms, irrespective if they engage in R&D activities or not.

Concerning upstream inter-industry spillover effects there are indications that all three firm categories generate positive spillover effects. Domestic R&D active firms generate the most substantial effect. A one standard deviation increase augments TFP by 11.1%. A one standard deviation increase in foreign R&D active and non-R&D active firms in client industries generates a positive productivity effect of 5.4% and 2.5%, respectively. The impact of domestic R&D active firms on their suppliers seems to be counteracted by a sizeable negative impact on their clients. A one standard deviation increase in domestic R&D active firms in their supplier sectors is associated with a productivity level of domestic non-R&D active firms that is on average 5.3% lower. Tables 7 and 8 in “Appendix” show that these results, especially those for spillovers from domestic R&D active firms, are not driven by correlations among the spillover variables. Our results may capture such indirect effects if spillovers between domestic R&D active firms, foreign R&D active firms, and foreign non-R&D active firms exist. Although we are unable to disentangle this effect fully, Table 9 suggests that the spillover effects resulting from domestic R&D active firms are the only ones that carry through to other domestic R&D active firms and foreign-owned firms with and without R&D. This confirms and strengthens domestic R&D active firms as the main driver of spillovers as part of the positive spillovers from both types of foreign firms may originate from higher productivity through spillovers from domestic R&D active firms.

5.2 Regional heterogeneity setting the stage for divergent development paths

Table 5 investigates whether Belgium’s heterogeneous regional composition influences spillover effects. If these effects manifest, this gives empirical evidence of divergent (future) development paths in regional innovation systems (Binz et al., 2016; Isaksen & Trippl, 2016; Neffke et al., 2011; Tödtling & Trippl, 2013).

Table 5 Productivity spillovers to domestic non-R&D active firms of R&D active and foreign-owned firms by region
Full size table

Model 5 examines whether firms experience similar spillover effects of R&D active or foreign-owned firms in the same region or in the other regions. We do so by splitting the variables in Eqs. (2) to (5) in a part driven by R&D active or foreign-owned firms located in the same region as the domestic non-R&D active firms under consideration, and a part driven by R&D active or foreign-owned firms located in the other regions. One reason to do so is to emphasise physical distance (Audretsch & Feldman, 1996; Wang & Wu, 2016), but a more prominent reason is to stress the difference in institutional settings (Cooke et al., 1997), organisational thickness (Trippl et al., 2018), development stages between regions (Annoni & Dijkstra, 2019; Hollanders et al., 2019), and potential language barrier effects (Billiet et al., 2006).

Model 5 finds that, for all regions taken together, domestic R&D active firms have productivity effects on domestic non-R&D active firms. A positive productivity effect occurs in the case of intra-industry linkages. We also find positive upstream inter-industry spillover effects originating from domestic R&D active firms and foreign-owned firms without R&D when located in the same region. Here ‘pure’ R&D and FDI spillover effects are at play, because they do not exert an effect when these are coinciding. When located in a different region, domestic and foreign R&D active firms generate positive productivity effects on their domestic non-R&D active suppliers. Concerning downstream inter-industry spillover effects, in line with Table 4, domestic R&D active firms are associated with negative spillover effects on their clients, irrespective of whether these firms belong to the same or different regions.

Models 6 to 8 repeat the estimation in Model 5 for each separate region. This allows all spillover effects to vary across regions which is justifiable because indicator-based classifications of the three regions in Belgium, as highlighted in Sect. 3, emerge as different types, and we may expect regional differences in the patterns of spillover effects. These differences are what we observe in Table 5.

In Brussels (Model 6), the smaller capital region, we find no significant spillover effects that originate within the region itself. Organisational thick and diversified regional innovation systems, such as the Brussels Capital Region, are characterised by framework conditions conducive to new path creation. The significant positive spillover effects come from domestic and foreign R&D active firms in other regions when firms belong to the same industries. The empirical results point to upstream inter-industry FDI spillovers originating from other regions, irrespective if they perform R&D or not. The Brussels Capital Region is thus capable of using extra-regional foreign knowledge to impact the productivity of its domestic (Brussels) non-R&D active firms positively. In addition, the Brussels Capital Region is a hub for many non-R&D active MNEs that act as leading firms in global networks and are capable of absorbing knowledge from extra-regional sources. The region itself, with a diverse set of economic activities and a highly skilled labour force coming from inside and outside the region, the presence of higher education institutions, international organisations and well-established innovation support structures, has a high absorptive capacity which caters for future path developments (Miguélez & Moreno, 2015). The rich endowment in the Brussels Capital Regions of knowledge-intensive business services (KIBS) is less R&D intensive because they rely on consulting activities that are not affected in their productivity in terms of TFP. This may serve as an explanation for why there are no downstream inter-industry effects on productivity.

Overall, the pattern for the Flemish Region in Model 7 is in line with the overall pattern revealed in Model 5, which is not unexpected given that the Flemish Region accounts for two-thirds of the estimation sample. But there are differences as well: domestic R&D active firms in the Flemish Region, if in the same industry, have no productivity effect on domestic non-R&D active Flemish firms. Domestic (Flemish) R&D active and foreign-owned firms without R&D are associated with positive upstream inter-industry spillovers to domestic (Flemish) non-R&D active firms, while this is non-existent in the other regions, pointing to a sufficient regional knowledge base. Downstream inter-industry spillovers are negative when originating from domestic R&D active firms in the Flemish Region.

The Flemish Region is an organisationally thick and specialised regional innovation system (see Sect. 3) because the region is populated by many firms, of which some have substantial critical mass in particular industrial clusters (Trippl et al., 2018). This relative specialisation in several medium-tech manufacturing industries (e.g. chemicals) and emerging high-tech industries (e.g. biotech) renders the regional economy dynamic and susceptible to spillovers. In addition, the regions’ budget on innovation is large and growing (Belspo, 2019), as the Flemish Region nears the European goal of spending 3% of its GDP on R&D. As a result, the region possesses many supporting organisations and instruments that are critically for new path development (Asheim et al., 2011; Boschma & Frenken, 2011).

Model 8 presents results for the Walloon Region. Upstream inter-industry spillovers are the primary vehicle for positive productivity effects. Non-R&D active firms in the Walloon Region are more productive when there are more foreign-owned firms in the Walloon Region, especially when non-R&D active and when there are more domestic and foreign-owned firms located in other regions, provided that they are R&D active. New path development in this type of region is facilitated even when it lacks a critical number of R&D active firms as long as they can use external knowledge sources or a dominant technological firm spurs local innovation and development (Isaksen & Trippl, 2017). Downstream linkages with non-R&D active foreign-owned firms in the Walloon Region or domestic R&D active firms in other regions are associated with lower productivity levels of domestic non-R&D active firms in the Walloon Region. Here, the argument of having to cope with higher costs seems to be relevant in the case of these downstream linkages.

For many years the economic development of the Walloon Region was in a lock-in situation as its structures, configurations and practices established in earlier phases of its industrial heyday remained untouched for decades. A stagnant development path characterises its regional development (Martin & Sunley, 2006). Recently, new specialisations such as life sciences and health replace the traditional industries, and the level of labour productivity is on the rise, although structural unemployment remains high (Rissola & Sörvik, 2018). This path renewal is the result of smart specialisation policies since 2000, which, among other measures, targeted business cluster creation and competitive poles. The Walloon Region issued several so-called Marshall Plans using hybrid top-down and bottom-up approaches.

6 Conclusions and policy implications

Even within the confines of a small open economy such as Belgium, the different organisational thickness of regional innovation systems, different development paths, and different types of linkages and spillover effects, justifies the need for a diversified regional innovation policy as offered by smart specialisation policies.

Policymakers anticipate R&D active or foreign-owned firms to exert positive effects on the productivity of their domestic (regional) non-R&D active firms as these provide the latest knowledge and technology which is either developed internally (R&D) or externally (FDI) or both. Therefore, regional policymakers seek to stimulate R&D expenditures and attract foreign-owned firms, and facilitate interactions through intra- and extra-regional linkages between firms in their regional innovation systems (Asheim, 2019; Boschma & Iammarino, 2009).

The quite diverse findings on the productivity effects of R&D or FDI spillovers suggests that these differ across regions with a specific organisational thickness of their regional innovation system. The Brussels Capital Region typifies, despite its smallness (Dotti et al., 2018; Isaksen & Karlsen, 2013), as a strong diversified regional innovation system (Trippl et al., 2018). However, the extent of industrial diversification seems so outspoken that the analytical results show no intra-regional spillover effect, irrespective if the source is R&D or FDI, to Brussels non-R&D active firms’ productivity. The effects for the Brussels Capital Region appear to lie in inter-regional spillovers as there are positive spillover effects (R&D and FDI/R&D) from intra-industry linkages and upstream inter-industry linkages (FDI and FDI/R&D) from other regions. This finding corroborates the ideas of Trippl et al. (2018) on thick diversified regional innovation system. A dedicated smart specialisation policy may identify the relatedness of the industries concerned. This way, regional policymakers can prioritise relevant industries, which is especially urgent given the limited (R&D) budget of the region and the many typical problems and challenges facing urbanised capital cities such as the lack of physical space, traffic congestion, etc. (Dotti et al., 2018; Martin & Simmie, 2008).

Although not an innovation leader, according to the Regional Innovation Scoreboard (2019), the Flemish Region seems to have firms affected by intra-regional linkages in their value chain. Especially the R&D spillovers exert an effect on productivity, which is positive in the case of upstream inter-industry linkages and negative for downstream inter-industry linkages. This asymmetric finding may point to the structure of the Flemish economy, which is very export-oriented and focused on medium–high and medium–low technology (Marsan and Maguire, 2011). Although the Flemish Region has a high level of regional absorptive capacity, the R&D spillovers through downstream inter-industry linkages exert a negative impact on the productivity of non-R&D active firms in the Flemish Region. A possible reason may be a mismatch in the quality of highbrow R&D results, which translate into high costs for industries when they implement the new technology. Policymakers could address this by stimulating the intra-regional R&D uptake.

The productivity of non-R&D active firms in the Flemish Region also benefits from upstream inter-industry FDI spillovers. Many foreign-owned firms, often active in the manufacturing sector, locate in the Flemish Region because of its workforce’s relatively high productivity and qualifications. Some of these foreign-owned firms are dominant multinationals. However, the results offer no evidence of foreign R&D spillovers, so it might be advisable to capitalise on these foreign-owned firms by adapting the smart specialisation policy to forge and strengthen the linkages with this industry to embed it in the regional innovation system.

Since the 1960s, the Walloon Region remained, economically, the weakest region in Belgium. This weakness is evidenced by their small public budget for R&D (Belspo, 2019; Rissola & Sörvik, 2018), making it a thin specialised regional innovation system (Trippl et al., 2018). The Walloon Region is a relatively slow-growing economy for many years (Capron, 2000). FDI spillovers from foreign-owned firms in the Walloon region, irrespective of whether they perform R&D activities, exert positive productivity effects on regional firms. These foreign-owned firms act as regional economic drivers of productivity (Agrawal & Cockburn, 2003). R&D spillovers generated by R&D active or foreign-owned firms in other regions positively impact the productivity of non-R&D active firms in upstream inter-industry linkages in the Walloon Region, presumably because these specialised suppliers may not be present in the Walloon Region itself. The supplied knowledge and technology is needed to boost the productivity of the non-R&D active firms in the Walloon Region. Therefore, policymakers in this region have the task of reinforcing the internal economic fabric of their region to endogenize the productivity effects of R&D spillovers. R&D spillovers generated by downstream inter-industry linkages with other regions negatively impact the productivity of Wallonian non-R&D active firms, as these may experience higher costs or firms must adapt their production processes because of these linkages.

In short, different types of spillovers exerting different effects show that regional innovation systems are highly path dependent. Although the three regions in Belgium are mature advanced economies, their smart specialisation policy will differ across them as they each face different challenges because their regional innovation systems differ (Trippl et al., 2018). Understanding the direction and extent of intra- and inter-industry linkages’ impact helps authorities to promote an efficient positioning in the global value chains with the potentials to develop new products, explore new markets, or seize technology opportunities. Pursuing and implementing adapted policy goals, identified by smart specialisation strategies, will transform their industry and set the regions to future development paths.

An obvious limitation of this research is its exclusive focus on linkages in the business sector. However, interactions between many different actors are an integral part of regional innovation systems. Universities and public research centres take on a substantial role in forming qualified labour, producing new knowledge and technologies, and providing troubleshooting solutions that can be extremely useful to firms.

The analysis points to at least one avenue for future research because it suggests that research on the empirical efforts on FDI spillovers should take the R&D activities of the foreign-owned firms into consideration. As such, this paper complements the research efforts recently started by Liang (2017), Ben Hassine et al. (2017), Li and Bosworth (2020) and Barge-Gil et al. (2020). A further research effort should complement the existing research focused on R&D or FDI by stressing the regional specificities in terms of organisational and institutional aspects, emphasising the dynamic path dependent nature of spillover effects on productivity, and linking the research to policy recommendations.