R&D investment and innovation performance under vertical partner concentration

ABSTRACT

Vertical exchange partners, both customers and suppliers, are key sources of knowledge about new components, technologies, and markets, exerting a critical influence on firms’ innovation performance. While firms’ reliance on a few suppliers and/or a few customers has become a regularly occurring phenomenon, there is only an insufficient understanding of the effect of vertical partner concentration on the returns to firms’ R&D investments. Building on transaction cost and evolutionary economics theories, we conjecture that vertical partner concentration negatively moderates the relationship between R&D investments and innovation performance. Our analysis is built on a dataset of 768 R&D-active French manufacturing firms derived from two independent and temporally separated surveys, enabling the use of lagged values of the key independent variables. The results reveal that customer concentration, whether accompanied or not by supplier concentration, weakens the positive relationship between R&D investment and innovation performance, but this is not observed in dynamic industry settings.

KEYWORDS:

• Vertical exchanges
• partner concentration
• R&D investment
• innovation performance
• industry dynamism

JEL CLASSIFICATION:

• O32
• O30
• L20

1. Introduction

In a recent high-profile corporate feud, Qualcomm accused its leading customer Apple of stealing and passing over its technology to rival Intel, while Apple alleged that Qualcomm, one of its biggest suppliers, charged unfairly high royalties (Financial Times, 2018).¹ This and several such incidences highlight the constraints on firms’ capacity to generate returns on their Research and Development (R&D) investments when vertical exchanges are concentrated with a few partners (henceforth vertical partner concentration), either suppliers or customers (Dedrick, Kraemer, and Linden 2010; Elfenbein and Zenger 2014, 2017; Mawdsley and Somaya 2021; Wagner and Bode 2006).

There is a wide recognition that close interactions with vertical exchange partners are critical for fostering knowledge development (Aoki and Wilhelm 2017) and value creation activities associated with innovation (Adner and Kapoor 2010; Afuah 2000; Kang and Afuah 2010). This perspective, in which innovation is viewed as involving feedback loops rather than as a linear process (Fleming and Sorenson 2004; Kline 1985; Mowery and Rosenberg 1979), suggests potential advantages offered by vertical partner concentration. It may make it easier to create routines and processes of cooperation that smoothen the exchange of ideas and knowledge, especially of the tacit variety (Aoki and Wilhelm 2017; Zollo, Jeffrey, and Singh 2002). Despite these benefits, exchanges with a limited number of vertical partners may impose significant search barriers (Laursen 2011; Uzzi 1996), resulting in cognitive biases that can turn firms’ search more deeply and narrowly oriented towards familiar rather than new domains (Zhong et al. 2021). Ultimately, this can limit firms’ capacity to generate new products and reduce their competitiveness in a changing environment (Laursen 2011; Uzzi 1996). Indeed, Hsu et al. (2021) show that while downstream partners positively contribute to new product development in a firm, the effect is negative when there is downstream partner concentration.

While research underscores that exchanges with a few vertical exchange partners can exert significant pressures on firms’ innovation activities (Hsu et al. 2021; Laursen 2011; Zhong et al. 2021), there is still a lack of understanding of whether and how concentration of exchanges with value-chain partners may influence the expected innovation returns to R&D investment. These questions assume particular importance given that the close involvement of vertical partners is a major constituent of firms’ innovation processes (Helfat and Raubitschek 2000; Rerup and Feldman 2011). The limited research in this area has focused on specific dimensions of innovation process, revealing a reduction in growth opportunities associated with customer concentration (Mawdsley and Somaya 2021) and a reduction in the capacity for value capture associated with supplier concentration (Elfenbein and Zenger 2017). It is then the aim of this paper to shed light on the potential consequences of vertical partner concentration on the relationship between R&D investments and innovation performance.

From a transaction cost perspective, exchanges with a limited number of partners may permit operational efficiencies, but it may increase the risk of opportunistic behaviour by partners and may engender knowledge spillovers to competitors (Arslan 2018; Elfenbein and Zenger 2014, 2017; Martínez-Noya and García-Canal 2018; Mawdsley and Somaya 2021). The evolutionary economics theory provides a complementary perspective through its conceptualisation of innovation as a cumulative, path-dependent, and recombinatory search process (Dosi 1982; Katila and Ahuja 2002; Nelson and Winter 1982). Viewed from this perspective, exchanges with a limited number of vertical partners can facilitate cost- and time-efficient searches in local domains, but they may also lead to learning traps and cognitive lock-ins (Ahuja and Lampert 2001; Katila 2002; Laursen 2011, 2012; Nelson 2008; Zhong et al. 2021).

We combine these insights to argue that vertical partner concentration may lower the positive effect of R&D investment on firms’ innovation performance (Christensen and Bower 1996; Leonard-Barton 1992). We further propose that the dynamism of a firm’s industry, affecting the conditions for appropriation of innovation rents (Chen, Coviello, and Ranaweera 2021; Fabrizio and Tsolmon 2014; Malerba and Orsenigo 1996; Visnjic, Ringov, and Arts 2019), moderates the influence of vertical partner concentration on the ability of firms to improve their innovation performance through R&D investments.

We argue that certain features of the dynamic industries can enhance firms’ capacity to appropriate the fruits of their R&D efforts. In these industries, the rapidly evolving technological and market conditions, periodic shakeouts, and fierce competition demand the development of new product designs, functionalities, and technologies (Agarwal, Sarkar, and Echambadi 2002, Christensen, Suarez, & Utterback, 1998; McGahan and Silverman 2001; Tatarynowicz, Sytch, and Gulati 2016). This increases the opportunities to commercialise outputs of R&D investments not only as final products but also in disembodied form from a firm’s previous innovations, and thereby enhances the appropriability capabilities of the firm (Cohen and Klepper 1996; Grimpe et al. 2017). Furthermore, greater thrust on product innovation in these industries, unlike process or product-service innovation, can render firms’ own contribution to the innovation process more discernible from that of their exchange partners (Chen, Coviello, and Ranaweera 2021; Cohen and Klepper 1996; Nelson and Winter 1982).

Results from an analysis of a sample of 768 R&D-active French manufacturing firms suggest that the positive relationship between R&D intensity and innovation performance is reduced in the presence of customer concentration, more so when there is also the simultaneous presence of supplier concentration. Furthermore, the limiting influence of vertical partner concentration on the relationship between R&D intensity and innovation performance is observed only in industries that are less dynamic.

Our study contributes to research on innovation and vertical linkages where the focus has been predominantly on the advantages of embedded relationships with exchange partners (Aoki and Wilhelm 2017; von Hippel 1998). We complement this literature by highlighting that vertical partner concentration can result in reduced learning opportunities that may outweigh the benefits of embeddedness. Specifically, we highlight certain conditions that allow for realising the benefits of embeddedness. Our research adds to the limited studies that focused on biases in search (Laursen 2011; Zhong et al. 2021) and cognitive lock-ins (Hsu et al. 2021) arising from customer concentration, and to the discussion around the paradox of embeddedness that suggested trade-offs associated with embedded relationships (Laursen 2011; Uzzi 1996).

This paper also contributes to previous research on the negative effects of vertical partner concentration by shifting the focus away from pricing (Elfenbein and Zenger 2017) and production volume (Mawdsley and Somaya 2021) to innovation. We further extend the partial focus in this literature on either supplier concentration (Elfenbein and Zenger 2017) or customer concentration (Hsu et al. 2021; Mawdsley and Somaya 2021; Zhong et al. 2021), by considering both these dimensions and studying their interdependent effects. The results of our analysis reveal an important nuance that firms should be particularly concerned about customer concentration. We discuss the implications of this and other findings for both research and practice.

2. Theoretical Framework and Hypotheses

2.1. Vertical exchange partners in R&D and innovation processes: A literature review

R&D is central to firms’ competitive advantage by supporting the development of new technologies and designs, improving efficiency, advancing new competences to respond to technological change, and improving the organisation of knowledge creation activities (Cohen and Levinthal 1994; Hall, Mairesse, and Mohnen 2010; Kogut and Zander 1992; Mansfield 1980; Pavitt 2005). While these characteristics of R&D make it strongly influence firms’ innovation performance (Leiponen and Helfat 2010; Mairesse and Mohnen 2004; Roy and Sarkar 2016), innovation returns on R&D investment are uncertain (Soete and Freeman 2012). This is because innovation is not a linear process, following a traditional production function, where changes in the amount of inputs will automatically lead to a change in the outcomes (Hall, Mairesse, and Mohnen 2010; Kline 1985, Mansfield, 1965; 1980). Innovation is instead a trial-and-error recombinatory search process characterised by path dependence and feedback loops (Fleming and Sorenson 2004; Mowery and Rosenberg 1979), in which customers² and suppliers³ occupy a central position (Clark 1989; Rerup and Feldman 2011; Thomke, von Hippel, and Franke 1998; von Hippel 1998). The interactive nature of this process suggests that the structure of a firm’s vertical exchange relationships can be critical for leveraging partners’ knowledge, while mitigating potential opportunistic behaviour, to create successful innovations (Helfat and Raubitschek 2000).

Despite the recognition of the importance of the structure of vertical exchange relationships for innovation, empirical research on this topic is rather scarce. The few studies that have analysed firm innovation under vertical partner concentration highlight a paradox. On the one hand, concentration of vertical exchanges with a limited number of partners may facilitate the development of efficient inter-organisational routines to communicate, monitor, and coordinate exchanges (Zollo, Jeffrey, and Singh 2002) and thus speed up innovation (Uzzi 1996). On the other hand, it can pose the risk that partners are incapable or are unwilling to respond to exogenous shocks such as new technologies or new markets, with the result that the focal firm may get locked-in in particular technologies and markets, and hence can miss out on promising new technological and market trajectories and advancements (Afuah 2000; Hsu et al. 2021; Uzzi 1996).

In particular, when a firm’s downstream exchanges are with a limited number of customers, it may end up channelling substantial resources to meet the specific needs of these few customers. This may restrict the flow of new ideas to the firm, making it likely to miss out on the changing trends in markets, technologies and product designs (Christensen and Bower 1996; Laursen 2011; Leonard-Barton 1992). In this respect, recent studies show that customer concentration limits firms’ search to local knowledge domains (Zhong et al. 2021) and reduce the prospects of positive technological spillovers (Hsu et al. 2021).

Contrary to the research on the effect of downstream concentration on innovation that on the effect of upstream concentration on firms’ innovation activities is almost non-existent. The limited range of knowledge, critical technologies and inputs possessed by the firm and its few suppliers is likely to reduce the variety in the firm’s knowledge pool and the recombinatory possibilities from its R&D investments (Ahuja and Katila 2001; Davis and Eisenhardt 2011; Forti, Sobrero, and Vezzulli 2020). This may create learning traps whereby firms have limited awareness of evolving technologies, inputs and market trends, leading them to focus on non-optimal technologies and designs with poor prospects for innovation (Afuah 2000; Davis and Eisenhardt 2011; Levitt and March 1988; Uzzi 1996).

Overall, while the literature points out that vertical partner concentration can have a significant influence on the effectiveness of firms’ innovation activities, there is a lack of sufficient understanding of how and to what extent vertical partner concentration influences firms’ ability to create and appropriate value from their innovation activities. Below, we develop a framework to conceptualise how vertical partner concentration may influence the relationship between R&D investment and innovation performance.

2.2. Theoretical Framework: Vertical partner concentration, R&D investment, and firms’ innovation performance

This study aims to understand whether and how the association between R&D investment and firms’ innovation performance changes in the presence of vertical partner concentration. From a transaction cost perspective, market exchanges are governed by incomplete contracts that can potentially create the risk of partner opportunism, manifesting in, for example, non-compliance and unfair bargaining (Langlois 2006; Williamson 1975). Although vertical partner concentration can help develop coordination efficiencies, which is another dimension of transaction cost, it can also elevate the risk of getting locked-in with a few partners that may encourage opportunistic behaviour (Elfenbein and Zenger 2014, 2017), potentially weakening firms’ capacity to create and capture value from their R&D investments.

The evolutionary economics theory, emphasising that innovation results from a recombinatory search process (Katila and Ahuja 2002; Laursen 2012; Nelson and Winter 1982), complements these insights related to coordination and contracting from transaction cost theory. In this perspective, firms’ search patterns influence whether they identify high-performing paths or get locked-in to less interesting paths of technologies and markets (Helfat 2018; Malerba and Orsenigo 1996, 1997; Nelson 2008; Winter, Cattani, and Dorsch 2007). Viewed thus, interacting with a limited number of vertical exchange partners can mean that firms’ search primarily revolves around the technologies of their suppliers or the market applications demanded by their customers (Hsu et al. 2021; Laursen 2011).

Our analytical framework integrates these views and conceptualises vertical partner concentration as reducing firms’ potential to create and appropriate value from their innovation activities. As value-chain partners are a strategic knowledge source (Helfat and Raubitschek 2000; Rerup and Feldman 2011; von Hippel 1998), vertical partner concentration can restrict firms’ knowledge search to the technological advances of their suppliers and/or the market applications demanded by their customers (Laursen 2011). Firms may thus be unable to pursue search paths beyond the knowledge domains of their vertical exchange partners (Laursen and Salter 2006; Zhong et al. 2021). They may thus be unable to explore new product designs to address changing market developments or to integrate promising new inputs and technologies into their product development activities, weakening capacity to create and capture value from their R&D investments (Hsu et al. 2021; Laursen 2011).

We further conceptualise that this attenuating influence of vertical partner concentration on the relationship between R&D investments and innovation performance is contingent on the appropriability conditions in firms’ environment (Grimpe et al. 2017; Malerba and Orsenigo 1997; Nelson and Winter 1982; Visnjic, Ringov, and Arts 2019). In highly dynamic environments, where cognitive frames are fluid, innovation activities are mainly driven by the idiosyncratic characteristics of firms, and are typically directed towards new product designs, functionalities, and technologies rather than improved processes, product-service offerings or improved quality or reliability of existing products (Kaplan and Tripsas 2008; McGahan and Silverman 2001). These features of the dynamic environment can enhance firms’ capacity to appropriate value from their innovation activities (Malerba and Orsenigo 1997; Nelson and Winter 1982). In addition, as firms and their vertical exchange partners are simultaneously exposed to changing market and technological conditions, they have a common need to seek new and advanced knowledge to better cope with evolving competitive pressures (Chen, Coviello, and Ranaweera 2021; Christensen, Suarez, and Utterback 1998). Changing market and technological conditions and the associated need for new complementary capabilities also mean lower opportunity costs associated with changing vertical partners, reducing the cost of potential partner opportunism (Elfenbein and Zenger 2014, 2017; Langlois 2006). Overall, thus, a dynamic environment can mitigate the deleterious effect of vertical partner concentration on firms’ ability to generate innovation returns to their R&D investments.

Next, we develop our hypotheses specific to manufacturing industries, in which the mechanisms we outline are more applicable than in services. Even though the boundaries between manufacturing and services are increasingly blurred (Djellal, Gallouj, and Miles 2013; Miles 2007) and services consist of a large and heterogeneous set of activities (Castellacci 2008; Miles 2007; Miozzo and Soete 2001), innovation in services differs greatly from that in manufacturing. While formal R&D investments in services have increased in recent years, service innovation activities tends to be more informal and ad hoc than manufacturing innovation (Miles 2007; Yacoub, Storey, and Haefliger 2020). This may have implications for the relationship between R&D investment and innovation performance. Additionally, service innovation has a shorter life cycle than manufacturing innovation (Castellacci 2008; Leiponen 2012; Miozzo and Soete 2001). This has significance for the characteristics of firms’ exchanges with their value-chain partners as well as for the potential for cognitive lock-ins that these exchanges may exert during the innovation process.

2.3. Hypothesis development

2.3.1. Vertical partner concentration and the R&D-innovation performance nexus in manufacturing industries

Vertical partner concentration most often means the adoption of narrow local search and problem-solving routines, which are likely to decrease firms’ ability to generate novel and valuable innovations from their R&D investments (Kogut and Zander 1992; Mawdsley and Somaya 2021; Rerup and Feldman 2011). Concentrating downstream exchanges with a few customers can make firms’ search centre around the needs of these core customers, resulting in the creation of technologies and products with narrow applications and functionalities (Ahuja and Lampert 2001; Christensen and Bower 1996; Katila 2002; Laursen 2011; Zhong et al. 2021). This can make firms blind to paths and solutions that could otherwise permit developing alternative product designs, as well as new applications and functionalities with broad commercialisation potentials (Christensen and Bower 1996; Kogut and Zander 1992; Thomke, von Hippel, and Franke 1998).

Likewise, supplier concentration may result in preferences for knowledge and inputs provided by firms’ limited number of suppliers, leading to the shrinking of their search maps. Dependence on a few suppliers may be especially damaging during unexpected market shocks when suppliers fail to adapt their technologies to match focal firms’ new technologies and product designs (Adner 2006; Adner and Kapoor 2010; Afuah 2000; Pisano 1990).

Thus, vertical partner concentration is likely to orient firms’ search towards local domains (Laursen 2012; Zhong et al. 2021), limiting the likelihood of identifying high performing solutions that could otherwise lead to valuable innovations (Nelson 2008; Winter, Cattani, and Dorsch 2007). It can also lead the firm into learning traps and cognitive lock-ins, thus reinforcing search for advances in technologies and business models for which the firm and its value-chain partners may have strong capabilities, but which may have outlived their usefulness in the marketplace (Christensen and Bower 1996; Forti, Sobrero, and Vezzulli 2020; Laursen 2012; Leonard-Barton 1992). Vertical partner concentration may, then, increase the risk of firms’ innovation development process focusing on local search paths and ignoring critical knowledge inputs, market signals, and novel knowledge combinations, thus weakening the value creation potential of their R&D efforts (Christensen and Bower 1996; Levitt and March 1988; Mawdsley and Somaya 2021; Rerup and Feldman 2011; Winter, Cattani, and Dorsch 2007).

The focus on local search paths may be reinforced if vertical partner concentration encourages value-chain partners to be opportunistic. Opportunistic behaviour of vertical partners, when they are limited in number, may be especially challenging for a firm to address given the incomplete nature of contracts (Oxley 1997). It may deepen a firm’s lock-in into its local search path, thereby further limiting its ability to identify high performing solutions and to develop valuable products from their R&D investments. Opportunism arising from supplier concentration may manifest in hold-up problems where the suppliers do not offer components or inputs with the best quality and the prices (Wagner and Bode 2006). Suppliers may view their relationship with the focal firms as generating cashflows from their older products and hence may be unwilling to provide their cutting edge products and may also be unresponsive to firms’ adaptation needs (Martínez-Noya and García-Canal 2018; Mitręga and Zolkiewski 2012). These dangers may be realised only too late when significant sunk investments have already been made, making it time-consuming and costly to switch to new suppliers (Afuah 2000; Langlois 2006). In response to the lost time and opportunities, the firm may attempt to speed up the innovation development process. Selecting the paths early to bring R&D outputs to the market, the firm may choose the paths most reliant on its existing knowledge and assets (Laursen 2011). This focus on short-term solutions in response to opportunistic behaviour by suppliers can reinforce the firm’s reliance on local search paths, which may paradoxically, cement the bargaining power of its suppliers and exacerbate the hold-up problem.

Similarly, having only a few customers accounting for the bulk of a firm’s downstream exchanges might also elevate the risk that they behave opportunistically. They may resort to unfair bargaining for a larger share of the value from the firm’s innovations or restrict the focal firm from supplying parts and components for their older products (Langlois 2006). Being locked-in with opportunistic customers, the firms may, likewise, become too focused on generating short-term returns on its innovation investments. It may search for low-cost improvements to its existing products, technologies, and designs, whose value is likely to be recognised by its customers (and/or match the contractual definition of improvements) and may end up further customising its innovation activities for their customers rather than opting for search paths that integrate high value-generating recombinatory knowledge that may help it cater to newer market needs, beyond existing customers (Laursen 2012; Shin, Kraemer, and Dedrick 2009, 2017). Thus, customer concentration can make it increasingly difficult for the firm to change its status quo and pursue knowledge search away from the current needs of its few customers.

In sum, by turning the firm’s attention to the capabilities, resources, inputs, and the needs of a reduced number of partners, vertical partner concentration is likely to engender search maps that favour local learning (Zhong et al. 2021) and application of knowledge advances to a narrow set of products (Hsu et al. 2021), limiting the recombination processes in the firm’s R&D. This suggests a situation of eventual knowledge lock-in, and consequently, of low market innovation returns from R&D investments (Leonard-Barton 1992; Levitt and March 1988; Mawdsley and Somaya 2021; Thomke, von Hippel, and Franke 1998). In other words, vertical partner concentration may significantly reduce the positive effect of R&D investment on the innovation performance.

On the contrary, firms with more distributed vertical exchange relationships are exposed to incentives to try a larger number of recombination possibilities, often suggested or influenced by their vertical exchange partners, for improving product functionalities, design, and quality. Less concentrated supplier exchanges oblige the firm to interact with many suppliers and identify relevance of their offered solutions, for its innovation process based on the trial and comparison of alternatives from various suppliers (Afuah 2000; Langlois 2006). Similarly, more distributed exchanges with customers are likely to create access to more varied knowledge elements that can help generate products or suites of products with multiple functionalities to satisfy the demands of their wide customer base (Hsu et al. 2021; Zhong et al. 2021). This suggests an increased ‘multiplier effect’ of a firm’s R&D investments as the firm is able to serve several market segments and stay more competitive (Davis and Eisenhardt 2011; Rerup and Feldman 2011; Shin, Kraemer, and Dedrick 2017), while reducing the likelihood of being locked-in with less relevant vertical partners (Afuah 2000; Langlois 2006

These arguments lead us to propose that:

Hypothesis 1a(H1a):

Supplier concentration weakens the positive association between R&D investment and innovation performance for manufacturing firms.

Hypothesis 1b(H1b):

Customer concentration weakens the positive association between R&D investment and innovation performance for manufacturing firms.

2.3.2. Industry setting and R&D-innovation performance nexus under vertical partner concentration

Manufacturing industries differ in the rate at which technologies and markets evolve (Suarez and Lanzolla 2007) and these differences influence firms’ innovation processes and their capacity to create and capture value (Malerba and Orsenigo 1997; McGahan and Silverman 2001). In dynamic industries, technologies, products, product mixes and business models are characterised by high levels of variety and rapid transformations (Agarwal, Sarkar, and Echambadi 2002; Cohen and Malerba 2001; Sarkar et al. 2006). In these industries, new knowledge and technologies are the main driver of innovation and product development, as well as the lever for value capture (Kim and Lee 2016; Nelson and Winter 1982; Pavitt, Robson, and Townsend 1989). In industries where technologies and markets evolve more slowly, collective cognitive frames are rigid and solidified (Kaplan and Tripsas 2008; McGahan and Silverman 2001; Pavitt, Robson, and Townsend 1989). Although innovation may still be crucial for competitiveness in this setting, the direction of a firm’s innovation activities is shaped largely by the complementary assets provided by its vertical exchange partners (Pavitt 1984; Suarez and Lanzolla 2005) and the technological advances available in its value chain (Cohen and Levinthal 1990, 1994; McGahan and Silverman 2001).

These differences in innovation processes associated with different industrial environments (Kaplan and Tripsas 2008; Nelson and Winter 1982; Pavitt, Robson, and Townsend 1989) also have important implications for the type of innovation outcomes (Agarwal, Sarkar, and Echambadi 2002; Christensen, Suarez, and Utterback 1998; Cucculelli and Peruzzi 2020; McGahan and Silverman 2001; Sarkar et al. 2006; Visnjic, Ringov, and Arts 2019) that may impact the returns to firms’ R&D investments.

In dynamic industries with substantial variety in products and technologies in the marketplace, firms’ R&D and innovation activities are highly idiosyncratic in nature (Kaplan and Tripsas 2008; McGahan and Silverman 2001). Innovation activities in these industries tend to focus on product innovations, which involve new or improved designs and features and novel or breakthrough technologies, rather than on advancing cost-efficient processes (Ahuja and Lampert 2001; Bos, Economidou, and Sanders 2013; Grimpe et al. 2017; Visnjic, Ringov, and Arts 2019). These features of the industry can help augment the ability of a firm to create and appropriate value from its R&D investments (Malerba and Orsenigo 1997; Nelson and Winter 1982). Product innovations (as opposed to process or product-service innovation) offer the opportunity to commercialise a firm’s innovation outcomes in disembodied form, enhancing the firm’s ability to appropriate, and thereby, generating discontinuous growth (Cohen and Klepper 1996). In addition, innovation outputs in the form of distinct new products allow for easier recognition of the focal firm’s value addition by end customers, enabling the firm to establish a justified price hike more easily than with process or product-service innovation.

Moreover, innovation activities in dynamic industries are relatively less related to firms’ current product portfolios, less aimed at satisfying the needs of the current customers, and less based on the know-how and quality of current suppliers’ inputs (Robertson and Patel 2007; Sarkar et al. 2006). This may act as a catalyst for introducing change to the portfolio of a firm’s vertical partners, as the firm may need to (re-)align its network of partners to match its innovation efforts in exploring and unlearning (Chen, Coviello, and Ranaweera 2021). In fact, switching to a new set of partners is a more normal practice in this setting, for example, when a firm launches a new product that requires new types of inputs and components or when it targets markets different from its existing ones (Chen, Coviello, and Ranaweera 2021; Gesing et al. 2015; Mina et al. 2007). As the distribution of value along the value-chain varies with the direction of technical change (Dosi 1982), in these industries, the bargaining power of different players, including key ones, is then in a state of flux (Agarwal, Sarkar, and Echambadi 2002; Christensen, Suarez, and Utterback 1998; Sarkar et al. 2006). As a result, vertical partner concentration may pose fewer constraints on firms’ ability to appropriate returns from their R&D investments.

Furthermore, as both the firm and its vertical exchange partners are simultaneously exposed to similar kinds of changes and uncertainties in the technological, regulatory and demand environments, vertical exchange partners, for their own competitiveness, may have to (re-)align their capabilities to match the changes in the firm’s innovation processes (Chen, Coviello, and Ranaweera 2021; Suarez and Lanzolla 2005; Teece 1986). Additionally, since innovation exerts a significant commercial impact in these industries (Malerba and Orsenigo 1997; Nelson and Winter 1982), sourcing-contracts are likely to include clauses related to knowledge leakages and allocation of propriety rights (Argyres, Bercovitz, and Mayer 2007; Pavitt 1984; Pavitt, Robson, and Townsend 1989). This further increases the prospects of a firm appropriating value from its R&D investment and reducing the likelihood of opportunistic behaviour by partners. Overall, in a dynamic industrial setting, the relationship between R&D investment and innovation performance may depend little on the structure of exchanges with vertical partners.

On the contrary, in industries characterised by low technological dynamism, innovation opportunities are scarcer and imitators may obtain higher returns than innovators, especially if they possess complementary assets (Nelson and Winter 1982; Pavitt, Robson, and Townsend 1989; Visnjic, Ringov, and Arts 2019). Given the rigidity of collective cognitive frames, innovation activities tend to focus on incremental improvements around reliability, costs of processes, products, and technologies, most often through the adoption of externally developed technologies and practices (Bos, Economidou, and Sanders 2013; Kaplan and Tripsas 2008; McGahan and Silverman 2001; Pavitt, Robson, and Townsend 1989). Most of these improvements are process-related, embodied in existing products and product-service offerings, making it challenging for firms to appropriate the benefits of their innovations (Cohen and Klepper 1996; Visnjic, Ringov, and Arts 2019).

A limited number of vertical exchange partners in this setting can increase the risk of a propinquity search trap, i.e. search in close proximity to existing solutions, resulting in innovations that offer limited additional value (Ahuja and Lampert 2001). Being tethered to the technologies and inputs of a few suppliers, firms in these industries may lose control over the technological direction of their innovation processes, while dependence on a few consumers could mean premature saturation to the applications of firms’ technologies and products (Suarez and Lanzolla 2005; Teece 1996). Vertical partner concentration in this setting is likely to exacerbate asymmetric bargaining power in the value-chain, increasing the risk of opportunistic behaviour by partners (Pavitt, Robson, and Townsend 1989; Suarez and Lanzolla 2005). Consequently, in this context, exchanges with a limited number of vertical partners may weaken the positive relationship between R&D investment and innovation performance.

Based on the above discussion, we propose that:

Hypothesis 2(H2):

The dampening effect of vertical partner concentration on the positive association between R&D investment and manufacturing firms’ innovation performance is weaker in manufacturing industries with high dynamism than those with low dynamism.

3. Methodology

We use self-reported data of French manufacturing firms collected via two secondary surveys – Changements Organisationnels & Technologies de l’information et de la Communication (COI) and the Community Innovation Survey (CIS). Both surveys were administered by the French statistics service (INSEE, French national institute for statistical and economic studies), hence following the same standards of data collection. The surveys were administered among a stratified sample of firms with more than 10 employees, representative of the population of about 25,000 firms with more than 10 employees on the statistics public register (INSEE 2023). Samples were stratified for each sector of activity (defined by the first two digits of the nace code) across five classes of firm size defined by number of employee (10–19, 20–49, 50–249, more than 250) (INSEE 2023). The surveys yielded response rates of over 80%. For CIS, this response rate is one of the highest in Europe (EUSTAT 2023).

The COI was administered in 2006 and focused on firms’ strategic organisation, relationships with suppliers and customers, products, strategic intent, resource availabilities, resources management and information technology tools. The CIS is a biannual survey coordinated by EUROSTAT and focuses on firms’ innovation activities. We used the 2008 survey, which investigates firms’ innovation activities from 2006 to 2008. Both surveys comply with the Frascati and Oslo manuals’ harmonised data collection methodologies for research on R&D and innovation (OECD 2005) and had response rates of over 80%. Moreover, most of the questions in the survey were objective in nature, which minimises any subjective interpretations. The two surveys were most likely completed by different individuals, reducing the common method bias (Podsakoff et al. 2003).

From the COI, we extracted information on the structure of firms’ vertical exchanges and from CIS we derived all the remaining variables used. The temporal gap between COI (2006) and CIS (2008) allows for lagged information on firms’ vertical partner structure (derived from COI 2006 data set) with respect to R&D investments and innovation performance (derived from CIS 2008 data set).⁴

Merging the two datasets resulted in a sample of 885 firms with complete information for all variables used in the study. Since we are interested in the innovation returns to R&D investments, we excluded 117 firms with no R&D investments, yielding a final sample of 768 firms. Because being in both CIS and COI datasets is more likely for firms in larger size classes, the merging of these two independent datasets slightly biased our data towards larger firms.⁵

Therefore, our data build on data sources from the end of the 2000s, when the prevalence of value-chains in the organisation of production, markets and innovation activities was already high (Dedrick, Kraemer, and Linden 2010). These data sources have several advantages over other alternative data, such as secondary data on publicly traded firms or primary data on a small sample of firms in an industry. Our data sources include a diverse set of firms in terms of industry, size, and location (in France) based on a representative sample of the population, and information on the firms’ innovation activities. Thus, while our data does not account for the increased challenges in value-chain caused by the extensive participation of China in the global market (Gereffi and Lee 2012) and by the COVID-19 pandemic (Gereffi, Pananond, and Pedersen 2022), it provides a unique opportunity to get a general view on the association between R&D investment and innovation performance in the presence and absence of vertical partner concentration.

Finally, to compute industry dynamism we collected data on industry value added from 2003 to 2008 from the OECD’s Structural Analysis Database (STAN).

Table 1 provides a description of the variables used in the econometric analysis and their sources.

Table 1. Description of Model Variables.

Variable Name	Description	Source
Innovation performance	the share of a firm’s turnover attributed to products new to it or the market.	CIS 2006–2008
R&D intensity	Logarithm of the ratio of firm’s R&D spend in 2008 including both internal and external R&D activities	CIS 2006–2008
Supplier Concentration	If the firm responds to the question ‘Do the 3 principal suppliers represent more than 50% of your activities’ as yes, then it is marked as 1, otherwise 0.	COI 2006
Customer Concentration	If the firm responds to the question ‘Do the 3 principal customers represent more than 50% of your activities’ as yes, then it is marked as 1, otherwise 0.	COI 2006
Joint Customer & Supplier Concentration	If the firm responds to both the questions ‘Do the 3 principal suppliers; and customers represent more than 50% of your activities’ as yes, then it is marked as 1, otherwise 0.	COI 2006
Size	It is the logarithm of the number of total employees, in 2006	CIS 2006–2008
Search	It is the sum of firm’s information sources that provided information regarding innovation projects or helped in completion of existing projects	CIS 2006–2008
Group	If the firm is part of an industrial group, then it is marked as 1, otherwise 0	CIS 2006–2008
International Market	If the firm has presence in international markets through its products then it is marked as 1, otherwise 0.	CIS 2006–2008
Knowledge Outsourcing	If firms acquired external knowledge through purchase or licencing that improved firm’s knowledge reserves in developing new and improved products, then it is marked as 1, otherwise as 0.	CIS 2006–2008
Personnel training for innovative activities	If the firm has engaged in internal or external training of its personnel focused towards development of new or significantly improved products, then it is marked as 1, otherwise as 0	CIS 2006–2008
IT Capabilities	If the firm has taken measures to safeguard its online information, then the variable takes the value 1 and 0 otherwise.	CIS 2006–2008
High/Low Dynamic industries	Industry dynamism is defined in terms of the volatility in annual industry sales over five pre-sample years, from 2002 to 2008. Upon regressing industry value added on year and then defining those with standard errors, above the average as high-dynamic industries (262 of 768 firms) and the remaining ones as low-dynamic industries (506 of 768 firms).	CIS 2006–2008

4. Data Variables

4.1. Dependent Variable

Innovation Performance: We follow the widely used approach in the literature (Grimpe and Kaiser 2010; Grimpe et al. 2017; Laursen and Salter 2006; Leiponen and Helfat 2010) of operationalising innovation performance as the share of a firm’s turnover attributed to products new to it or the market.

4.2. Explanatory Variables

R&D intensity: R&D intensity is defined as the ratio of a firm’s total R&D investment to sales. It is a commonly used proxy for firms’ new product development efforts and the development of their absorptive capacity (Garriga, Von Krogh, and Spaeth 2013; Laursen and Salter 2006). In line with the standard practice in the literature (Leiponen and Helfat 2010), we log transformed the variable to minimise skewness.

Vertical partner concentration: The COI data set provides an interesting information on whether a firm’s three main customers (suppliers) represent 50% or more of its total sales (input sourcing). We used this information to create three dichotomous variables that measure customer concentration, supplier concentration, and joint supplier and customer concentration. Customer concentration takes the value 1 if a firm’s three largest customers account for more than 50% of its total sales and 0 otherwise. Supplier concentration takes the value 1 if a firm’s three largest suppliers account for more than 50% of its total input sourcing expenses and 0 otherwise. Joint supplier and customer concentration take the value 1, if a firm’s three largest customers and suppliers account for more than 50% of its total sales and total sourcing expenses and 0 otherwise. Although we make no explicit prediction about the influence of this last variable, we expect simultaneous concentration of supplier and customers to be equally or more hazardous as concentration in any one of the two dimensions, for the value creation and value capture processes (Langlois 2006; Shin, Kraemer, and Dedrick 2017). Lack of information on sales of firms’ suppliers and customers prevented us from constructing more standard indexes of concentration such as Herfindahl index (Elfenbein and Zenger 2017). Nevertheless, our measure of concentration is an improvement over more restrictive measures of concentration used in the previous research, such as selling to (buying from) only one customer (supplier) (Patatoukas 2012; Wagner and Bode 2006).

Industry Dynamism: We followed prior literature and defined industry dynamism in terms of the volatility in annual industry sales over six pre-sample years, from 2003 to 2008 (Girod and Whittington 2017; Richard et al. 2019). We follow the approach of Richard et al. (2019) of regressing industry value added on year and then defining those with standard errors above the average as high-dynamic industries (262 of 768 firms) and the remaining ones as low-dynamic industries (506 of 768 firms).

4.3. Control Variables

We employ several variables to control for factors that may influence a firm’s innovation performance. See Table 1.

Firm size might influence the ability to access resources and develop organisational capacity for innovation. We followed prior literature and control for firm size (Size), measured as the logarithm of the total number of a firm’s employees (Grimpe and Kaiser 2010; Grimpe et al. 2017; Laursen and Salter 2006).

Firms that are part of a group may have better access to resources to support innovation activities (Lhuillery and Pfister 2009) and may have vertical exchange partners belonging to the same group, which may influence the nature of value creation through innovation as well as the sharing of innovation gains. We control for whether a firm is part of a business group (Grimpe et al. 2017; Laursen and Salter 2014) using the dichotomous variable Group which takes the value 1 if the firm is part of a group and 0 otherwise.

Internationalisation provides access to new sources of organisational learning but may also cause unintended knowledge leakages and increase coordination costs (Kafouros et al. 2008; Martínez-Noya and García-Canal 2018). This might affect a firm’s value creation and value capture abilities from its innovation activities. We therefore account for internationalisation with the variable International market which takes the value 1 if the firm operates in international markets and 0 otherwise.

Firms routinely seek external knowledge to complement their internal knowledge pool. It is widely suggested that firms that engage in external knowledge development activities experience higher innovation performance (Belderbos et al. 2018; Grimpe and Kaiser 2010). Variety in sources of knowledge is an indication of the level of a firm’s openness to external sources, its broader efforts to access knowledge and ability to create and capture value from innovation (Garriga, Von Krogh, and Spaeth 2013; Laursen and Salter 2006; Leiponen and Helfat 2010). We use the variable Search to capture the breadth of a firm’s knowledge search activities. We follow previous studies and measure search breadth as the sum of nine different types of sources of information reported by a firm as either important or very important (Garriga, Von Krogh, and Spaeth 2013; Laursen and Salter 2006). In addition, we account for firms’ engagement in knowledge outsourcing with the variable Knowledge Outsourcing, which takes the value 1 if a firm outsources its knowledge development activities and 0 otherwise.

Trained and skilled staff will be better able to integrate new knowledge in a firm’s products and technologies and support swift adaptation of organisational processes, thereby contributing to the creation and appropriation of value (Ketata, Sofka, and Grimpe 2014). We use Personnel Training to account for firms’ investment in staff training. It is a dichotomous variable that takes the value 1 if the firm trains its employees and 0 otherwise.

Investments to protect technological assets from leakages could prevent loss of relevant knowledge to vertical exchange partners and allow greater capture of value from a firm’s innovation efforts. This may reflect in the firm’s information technology capabilities and ability to codify new knowledge that may as well enhance its value creation abilities (Melville, Kraemer, and Gurbaxani 2004). We then add as a control the dichotomous variable IT Capabilities, which indicates whether the firm has taken measures to safeguard its online information. This variable takes the value 1 if information technology protection measures are in place and 0 otherwise.

Finally, we include industry dummies to account for industry-specific features of the inventive and commercialisation processes.

Table 2 presents descriptive statistics of all the study variables. In our sample, on average 25% of firms’ turnover is due to sales of new or improved products (with a standard deviation of 0.24). R&D investment represents on average 3% of total sales. Only supplier (customer) concentration is observed in 11% (25%) of the sample and joint supplier and customer concentration is observed in 9% of the sample. 34% of the firms in the sample operate in dynamic industrial environments. On average, firms employ 948 employees, use 3.7 of the 9 sources of information surveyed and implement 4 of the 5 measures surveyed to safeguard their online information. 58% of firms outsource knowledge development activities.

Table 2. Descriptive Statistics and Pairwise Correlation of Key Variables.

Variable	Obs	Mean	Std. Dev.	Min	Max	1	2	3	4	5	6	7	8	9	10	11	12	13
Innovation Performance (1)	768	0.25	0.25	0.01	1.00	1
Size (2)	768	6.11	1.26	2.48	11.25	−0.0126	1
Search (3)	768	3.78	2.24	0.00	9.00	−0.0006	0.2234*	1
Group (4)	768	0.89	0.32	0.00	1.00	−0.0301	0.3974*	0.0941*	1
International Market (5)	768	0.91	0.28	0.00	1.00	−0.0043	0.2126*	0.1034*	0.1243*	1
Outsourcing (6)	768	0.58	0.49	0.00	1.00	0.0574	0.2503*	0.2189*	0.2046*	0.1266*	1
Personnel Training (7)	768	0.70	0.46	0.00	1.00	0.0763*	0.1736*	0.2643*	0.1171*	0.0378	0.2267*	1
IT Protection (8)	766	4.05	1.07	0.00	5.00	−0.0622	0.4029*	0.1322*	0.3092*	0.0871*	0.1157*	0.1212*	1
R&D Intensity (9)	768	0.03	0.05	0.00	0.33	0.1240*	0.1140*	0.1185*	0.0142	0.0508	0.1432*	0.1082*	0.0572	1
Joint Customer & Supplier Concentration (10)	768	0.09	0.29	0	1	0.0366	−0.0818*	−0.0641	0.0292	0.0045	−0.004	−0.0219	−0.0431	−0.0598	1
Supplier Concentration (11)	768	0.11	0.31	0	1	−0.0845*	−0.1264*	−0.0287	−0.0932*	−0.0591	−0.0825*	−0.0326	−0.0465	−0.0597	−0.1104*	1
Customer Concentration (12)	768	0.25	0.43	0	1	0.0372	0.0832*	−0.0579	0.0449	−0.0751*	−0.01	0.0366	−0.0109	0.0651	−0.1870*	−0.1996*	1
Industry dynamism (13)	768	0.34	0.47	0	1	−0.1558*	−0.1293*	0.0425	0.0291	−0.0695	−0.0322	−0.029	0.0001	−0.1474*	−0.0713*	−0.0057	0.0241	1

Significance at p < 0.05.

The low correlation coefficients (<0.4) and a variance inflation factor (VIF) of 1.15 suggest that multicollinearity is not a concern in our model.

4.4. Statistical Approach

The following equation defines the full model with all the hypothesis testing variables⁶

Innovation performance_i = α + β₁R&D intensity_{i +} β₂ supplier concentration_{i +} β₃ customer concentration_{i +} β₄ industry dynamism_{i +} β₅R&D intensity_i x supplier concentration_{i +} β₆R&D intensity_i x customer concentration_i + β₇R&D intensity_i x supplier concentration_i x industry dynamism_i + β₈R&D intensity_i x customer concentration_i x industry dynamism_i +β₉Controls_i $+{\vartheta }_i$

The two-way interactions of R&D intensity with supplier concentration and customer concentration test hypothesis 1a and hypothesis 1b, respectively. A negative coefficient for these interaction variables will support our prediction that concentration of suppliers and customers weakens the positive association between R&D investment and innovation performance.

The three-way interactions involving industry dynamism, R&D intensity and the two concentration variables test hypothesis 2. A positive coefficient for these variables will support our prediction that the negative moderating effect of concentration on the relationship between R&D investment and innovation performance is weaker in dynamic industries.

Given the bounded nature of our dependent variable, innovation performance, taking values between 0 and 1, and ordinary least square estimates are not appropriate (Papke and Wooldridge 1996). We thus employ a fractional probit model, in line with the approach adopted by recent studies (Laursen and Salter 2014; Ramalho, Ramalho, and Murteira 2011).

5. Results

The results of the analysis are presented in Table 3. We ran five models, with the first model containing only controls (Model 1, column 1) and subsequent models sequentially adding the hypothesis-testing variables: Model 2 adds R&D intensity (column 2), Model 3 supplier concentration, customer concentration, and joint supplier and customer concentration (column 3), Model 4 the interaction between R&D intensity and the three vertical partner concentration variables (column 4), and Model 5 the three-way interaction between R&D intensity, vertical partner concentration and industry dynamism (column 5).

Table 3. Fractional Probit Estimations.

	(1)	(2)	(3)	(4)	(5)
VARIABLES	Controls	IV	Moderators	With Concentration	Three-way interactions with Industry Dynamism
Size	−0.0229	−0.0208	−0.0230	−0.0229	−0.0258
	(0.0274)	(0.0273)	(0.0272)	(0.0272)	(0.0271)
Group	−0.0410	−0.0147	−0.0335	−0.0462	−0.0267
	(0.107)	(0.109)	(0.109)	(0.108)	(0.109)
Search	−0.00176	−0.00438	−0.00325	−0.00280	−0.00126
	(0.0123)	(0.0123)	(0.0124)	(0.0125)	(0.0125)
International Market	−0.0587	−0.0511	−0.0513	−0.0480	−0.0194
	(0.106)	(0.107)	(0.108)	(0.107)	(0.107)
Knowledge Outsourcing	0.0717	0.0578	0.0538	0.0588	0.0606
	(0.0594)	(0.0597)	(0.0596)	(0.0594)	(0.0589)
Personnel Training	0.117*	0.112*	0.111*	0.114	0.108*
	(0.0656)	(0.0650)	(0.0647)	(0.078)	(0.0646)
IT capabilities	−0.0563*	−0.0592**	−0.0568*	−0.0580**	−0.0568**
	(0.0291)	(0.0291)	(0.0293)	(0.0291)	(0.0290)
R&D Investment intensity		1.433***	1.402***	2.269***	2.424***
		(0.528)	(0.537)	(0.763)	(0.833)
Joint Customer & Supplier Concentration			0.0781	0.216*	0.204
			(0.104)	(0.122)	(0.147)
Supplier Concentration			−0.184*	−0.185*	−0.181
			(0.0944)	(0.104)	(0.142)
Customer Concentration			0.0300	0.119	0.0355
			(0.0649)	(0.0812)	(0.102)
Joint Customer & Supplier Concentration x R&D Intensity				−4.635**	−5.479**
				(2.124)	(2.203)
H1a: Supplier Concentration x R&D Intensity				0.409	1.001
				(1.847)	(2.866)
H1b: Customer Concentration x R&D Intensity				−2.116**	−1.810
				(0.998)	(1.145)
H2: Joint Customer & Supplier Concentration x R&D Intensity x Industry Dynamism					23.71*
					(12.28)
Supplier Concentration x R&D Intensity x Industry Dynamism					0.650
					(3.779)
Customer Concentration x R&D Intensity x Industry Dynamism					0.738
					(2.257)
					(0.393)
Constant	−0.589***	−0.626***	−0.597***	−0.637***	−0.525**
	(0.199)	(0.199)	(0.205)	(0.206)	(0.237)
Observations	768	768	768	768	768
Log pseudolikelihood	−425.829	−424.731	−423.940	−422.930	−421.631
Wald Chi 2	53.77***	62.19***	75.00***	81.49***	100.63***
df	18	19	22	25	33

Robust standard errors in parentheses ***p < 0.01, **p < 0.05, *p < 0.1.

Among the control variables, the coefficient of personnel training is significant and positive across all models. Industry dummies, which are not reported, show significant coefficients.

The coefficient of R&D intensity is significant and positive in all models (β = 1.43, p = 0.007, model 2; β = 2.26, p = 0.003, model 4; and β = 2.42, p = 0.004, model 5), confirming our implicit baseline expectation that the more a firm invests in R&D, the greater its share of turnover due to new products.

The coefficients of the interaction variables between R&D intensity and customer concentration, and between R&D intensity and joint customer and supplier concentration, are negative and significant (β= −2.115, p = 0.034; β= −4.634, p = 0.029, respectively, model 4). However, the coefficient of the interaction between R&D intensity and the supplier concentration is not significant. As interpretations based on the coefficients in non-linear regression models may be misleading, we plot marginal effects. Figures 1b to 1c suggest that at low levels of R&D intensity, vertical partner concentration does not significantly influence firms’ innovation performance. However, at moderate and high levels of R&D intensity, firms that concentrate exchanges with either customers (Figure 1b) or jointly with customers and suppliers (Figure 1c) experience lower innovation performance than do firms with a more distributed exchange structure. No significant negative effect is observed for firms that concentrate exchanges with suppliers (without simultaneously with customers) at any level of R&D intensity (Figure 1a). These results, thus, support H1b but not H1a.

Figure 1. Predicted Innovation Performance across different levels of R&D Intensity with Supplier and/or Customer Concentration.

We next test H2 that proposed that the weakening effect of vertical partner concentration on the positive relationship between R&D investment and innovation performance is weaker in dynamic industries. Supporting this hypothesis, the coefficient of the three-way interaction involving the variables R&D intensity, joint customer and supplier concentration and industry dynamism reported in model 5 on Table 3, has a positive and significant coefficient, suggesting even a positive association between R&D investment and innovation performance for firms with vertical partner concentration in dynamic industries. The coefficient of the three-way interaction involving customer concentration is a positive sign but not significant, suggesting the absence of a limiting effect of vertical partner concentration on the positive relationship between and R&D investment and innovation performance in dynamic industries.

Again, we plot marginal effects for a more meaningful interpretation of the findings. Figures 2a–c reveal that the relationship between innovation performance and R&D investment is negative for firms with customer concentration in low-dynamic industries. In these industries, the predicted innovation performance at low (10^th percentile) and high (90^th percentile) levels of R&D intensity are, respectively, 24 and 30% of total sales without customer concentration compared with 25 and 26% of total sales with customer concentration (Figure 2a). This reflects a shift in the marginal effect of R&D under customer concentration from positive to negative (from 1.0 to −3.8 percentage points) as R&D intensity increases from low to high levels (Figure 2b). In dynamic industries, the predicted innovation performance at low and high levels of R&D are, respectively, 19 and 24 (without customer concentration) and 25 and 28% of total sales (with customer concentration) (Figure 2a). This reflects a relatively smaller decline in marginal effects of R&D under customer concentration from 6.5 to 4.2 percentage points, as R&D intensity increases from low to high levels (Figure 2c).

Figure 2. Predicted Innovation Performance across different levels of R&D Intensity in High/Low dynamic industries with Customer Concentration.

Similar results are true for firms with joint customer and supplier concentration. In low-dynamic industries, the predicted innovation performance at low (10^th percentile) and high (90^th percentile) levels of R&D intensity are, respectively, 24 and 29% of total sales without joint customer and supplier concentration compared with 30 and 21% of total sales with joint customer and supplier concentration (Figure 3a). In dynamic industries, the corresponding values are, respectively, 21 and 21 (without joint customer and supplier concentration) and 19 and 74% of total sales (with joint customer and supplier concentration) (Figure 3a). These differences reflect opposing directions of change in the marginal effects of joint customer and supplier concentration, across industrial environments: in low-dynamic industries, the difference of marginal effects of R&D as it increases from low to high levels is 6.7 and −8.1 percentage points (Figure 3b) and in high-dynamic industries −2.4 to 50.4 percentage points (Figure 3c).

Figure 3. Predicted Innovation Performance across different levels of R&D Intensity in High/Low dynamic industries with Joint Customer & Supplier Concentration.

Figures 4a–c confirm that supplier concentration has no noticeable effect on the relationship between R&D (Figure 4a) and that this is irrespective of whether firms are in high- or low-dynamic industrial environments (Figures 4b,c).

Figure 4. Predicted Innovation Performance across different levels of R&D Intensity in High/Low dynamic industries with Supplier Concentration.

In sum, customer concentration, on its own or jointly with supplier concentration, is associated with reduced innovation performance for medium-low to high levels of R&D intensity. This reducing effect is, however, asymmetric across firms operating in low- and high-dynamic industries. For firms that operate in high-dynamic industries, the negative moderating effect of customer concentration or joint customer and supplier concentration, on the relationship between R&D investment and innovation performance is, respectively, attenuated and reversed. At very low levels of R&D intensity, the reducing effect of customer concentration, on its own or jointly with supplier concentration, is not observed irrespective of the industrial environment the firm operates in. Supplier concentration exerts no influence on the relationship between R&D intensity and innovation performance, be it in low or highly dynamic industrial environments or at different levels of R&D intensity.

5.1. Robustness Checks

We carried out several tests to ensure the robustness of our findings (results are available in the Appendix). First, we tested our hypotheses using alternative estimation models (Table A, Appendix): Tobit, OLS, and a two-part model. Although typically appropriate when the dependent variable is censored, studies have extensively used the Tobit model to estimate equations involving bounded proportion-dependent variables (see, for e.g. Leiponen and Helfat 2010). Even though OLS is not suitable when the dependent variable is of bounded proportion, it has the advantage of providing coefficients that are easy to interpret. The results of both Tobit and OLS align with those reported in this paper. As some firms in our sample do not innovate, it is important to know if different processes underlie the decision to innovate and the intensity of innovation. Towards this, we followed Wulff’s (2019) procedure that compares the one-part model in which the dependent variable includes both zero and non-zero values (our original model) with a two-part model (with exclusion restriction) where the decision to innovate and innovation intensity are modelled separately. The Reset test does not reject that model specifications in both one-part and two-part models are correct. A likelihood ratio test suggests that the error terms in both equations of the two-part model are independent, underscoring the choice of the one-part model. Further, the one-part fractional probit model provides a better fit than the two-part model; the AIC and BIC statistics for the one-part model are, respectively, −9.49 and −9.33 compared with −9.38 and −9.05 for the two-part model.

Second, we conducted an alternative test for H1a and H1b, without using the interactions but adopting a subsample analysis that permits more effectively accounting for the cross-sectional nature of our data and thus checking the robustness of our results (Hoetker 2005). If differences exist in the relationship between R&D investment and innovation performance for firms that differ in terms of vertical partner concentration, then the sign and/or significance of the coefficient of R&D intensity should be different for groups of firms with different vertical exchange structures. By including in each sub sample, firms with identical structure of vertical partners, accounts for potential unobserved heterogeneity, for example, associated with the choice of the structure of vertical partners and R&D intensity. We thus split our sample into four categories of firms: Firms with (1) no customer or supplier concentration, (2) supplier concentration, (3) customer concentration, and (4) joint customer and supplier concentration. We ran separate estimation in each of these four subsamples. The results of this exercise corroborate our original findings. For the subsample of firms with no customer or supplier concentration, and those with only supplier concentration, the coefficient of R&D intensity is positive, respectively, at the 1% and 5% levels. Instead, for the subsamples with joint customer and supplier concentration, and customer concentration, the coefficient of R&D intensity is not significant; a Wald test (Chi2 = 0.39, df = 1) suggests an insignificant difference between the coefficients of R&D for these two samples. See Table A in Appendix.

Third, we checked whether our results are affected by potential non-linearities between R&D intensity and innovation performance. We thus re-estimated our models by including the square term of R&D intensity. The coefficient of the square term of R&D intensity is not significant, while the sign and significance of the coefficient of the interactions between R&D intensity and partner concentration persist. See Table A in Appendix.

Fourth, we tested our results with a more restrictive definition of innovation, with the share of turnover due to products that are new-to-market only. These results, consistent with our original findings, are presented in Table B of the Appendix.

Fifth, while our dependent and key explanatory variables originate from different data sources and are temporally separated (with vertical partner concentration referring to 2006 and R&D investment and innovation performance to 2008), our results may be affected if earlier performance may condition R&D investment and the structure of the vertical exchanges. We test for this, following Wooldridge (2009), by estimating the effect of the share of turnover due to new products (innovation performance) in 2006 on R&D investment and on the concentration of exchanges (with suppliers or customers) in 2006. The coefficient of innovation performance in 2006 is not significant in both models, suggesting that reverse causality may not be an issue (Reeb, Sakakibara, and Mahmood 2012). See Table C in Appendix.

Finally, in the absence of appropriate instruments for vertical partner concentration variables, we used the Lewbel’s (2012) method of estimating regression models with endogenous regressors. Results confirm that both customer only and joint customer and supplier concentrations dampen the positive relationship between R&D investment and innovation performance. The Sargan test (47.843, df 63, p-value = 0.921) does not reject the null hypothesis that instruments are valid and correctly excluded. The Stock and Yogo test of the null hypothesis that instruments are weak are also rejected (F = 39.2, with 10 being the rule of thumb for rejecting the null hypothesis) (Andrews, Stock, and Sun 2019). See Table D in Appendix.

Overall, these sensitivity tests confirm our main results presented in Table 3.

6. Discussion

Although there is wide recognition that firms’ vertical partners are integral to the successful creation and commercialisation of innovative products and services (Adner and Kapoor 2010; Afuah 2000; Kang and Afuah 2010), scholarly attention remains scarce on the form in which the concentration of exchanges with a few vertical partners influence the firms’ innovation processes. In particular, whether similar or different innovation returns to R&D investment arise in concentrated and in distributed exchanges with suppliers and customers is still unknown. Building on insights from transaction cost and evolutionary economics theories, our study aimed to shed light on this issue.

Our findings can be summarised as follows. Customer concentration significantly limits the effect of R&D investment on firms’ innovation performance. This negative effect of customer concentration duplicates when it occurs simultaneously with, rather than without, supplier concentration in low dynamic industries at high R&D levels. However, supplier concentration, by itself, does not significantly limit the effect of R&D investment on innovation performance. The negative effect of customer concentration, either with or without supplier concentration, does not exist for firms that are active in highly dynamic industries, i.e. it is prevalent only in low dynamic industries.

Results on customer (on its own or jointly with supplier) concentration support the evolutionary economics’ expectation that vertical partner concentration limits firms’ search focus and consequently the pursuit of more distant but eventually higher-value innovation directions (Laursen 2011; Zhong et al. 2021). Results also support the transaction cost economics’ expectation that appropriation conditions in the industry environment influence the opportunism risk due to value partner concentration (Elfenbein and Zenger 2014, 2017; Langlois 2006; Mawdsley and Somaya 2021) and thus limiting the likelihood of appropriating value from R&D investment.

The fact that results for concentration of exchanges with suppliers only are not significant suggest that the perceived needs of customers may be the leading driver of firms’ innovation activities (Hsu et al. 2021; Leonard-Barton 1992; Shin, Kraemer, and Dedrick 2009; Zhong et al. 2021). In addition, firms tend to constantly seek efficiency gains (Elfenbein and Zenger 2014), which potentially lowers their perceived opportunity costs in switching suppliers. Therefore, firms may favour contracting with a few suppliers in order to maximise efficiencies. In contrast, since technological advances in the upstream markets derive significantly from downstream markets (Adner and Kapoor 2010), a wider customer base may be much more important for broadening the search directions for new products (Laursen 2011). Thus, the limiting effect on the innovation search directions may be less accentuated in the case of upstream concentration only than in the case of downstream concentration only.

These results on the relationship that vertical partner concentration has with a firm’s innovation performance to R&D investment highlight mainly three things. First, they seem in line with, but also complement, previous findings on the influence of core customers on firms’ innovation process. Although customers play a key role in bringing ideas for product development (Chatterji and Fabrizio 2014; von Hippel 1998), too much reliance on them could reduce the variety in the knowledge the firm uses for ideation and problem-solving, leading the firm’s search towards more familiar domains (Laursen 2011; Thomke, von Hippel, and Franke 1998; Zhong et al. 2021). This can result in organisational inertia and learning traps whereby the firm becomes blind to promising new technological trajectories or get stuck in technologies and products with reduced value and applications (Adner and Kapoor 2010; Afuah 2000). Customer concentration could then restrict the application of a firm’s R&D outputs and the chances of wider commercialisation of its innovations (Helfat and Raubitschek 2000). Our findings on customer concentration highly resonate with this view that firm’s capabilities developed around some few customers could turn into core rigidities (Leonard-Barton 1992; Mawdsley and Somaya 2021), especially when the firm operates in less dynamic industries; a context in which appropriation conditions are relatively low and identifying alternative customers may be more difficult than identifying alternative suppliers (Arruñada and Vázquez 2006).

Second, our findings about the egregious effect of joint customer and supplier concentration may be related to the paradox of embeddedness (Laursen 2011; Lokshin, Hagedoorn, and Letterie 2011; Uzzi 1996), which refers to moulding the firm’s innovation activities around its vertical partners’ knowledge. This carries the risk that the firm defines innovation search in known domains, reducing the likelihood of identifying solutions that could lead to better product designs or better applications and functionalities for existing products, especially when faced with technological changes (Laursen 2011; Zhong et al. 2021). Our findings add an important clarification to the embeddedness paradox by suggesting that it may not exist universally but is likely prevalent in certain relations (e.g. with customers) and in certain contexts (e.g. industries with low dynamism).

Third, our findings also connect with the notion that vertical exchange partners may engage in the encapsulation of knowledge, meaning that they conceal critical knowledge elements from the firm (e.g. Langlois 2006, 1396). Encapsulation of knowledge may make the firm vulnerable to their partners expanding into its activities, leaving the firm with lower bargaining power and lower ability to appropriate the value created by its R&D investments (Langlois 2006; Shin, Kraemer, and Dedrick 2017; Wagner and Bode 2006).

Our results attest to these risks of vertical partner concentration, especially when the firm operates in low dynamic industries. Technological advancements in these industries are mostly cumulative, rather disruptive, and often lead to process innovation (Cohen and Klepper 1996; Malerba and Orsenigo 1997; McGahan and Silverman 2001; Nelson and Winter 1982), increasing the relational value of vertical partners (Elfenbein and Zenger 2014). Our study highlights that, in these industries, the ability to appropriate the value derived from R&D investments is more closely connected with the structure of the value-chain than in highly dynamic industries.

7. Implications

7.1. Theoretical Implications

We believe that our study has implications for the literature in several ways. First, our research adds to the previous literature on concentrated vertical exchanges that have highlighted the risk of opportunistic behaviour by partners in affecting such outcomes as prices charged (Elfenbein and Zenger 2014, 2017) or firm growth (Mawdsley and Somaya 2021). We complement these studies by shifting the focus to the firm’s innovation process, as well as by highlighting that there might be important boundary conditions that need to be recognised for a fuller understanding of the effect of the structure of vertical exchange relationships on firm’s innovation performance. In addition, our study is perhaps the first to simultaneously consider the effect of joint customer and supplier concentrations on innovation performance, as well as to identify a boundary condition associated with these relationships. In doing so, this study demonstrates the need for future research to simultaneously consider both the nature of exchanges with customers and suppliers for a comprehensive understanding of the effects of the structure of vertical exchanges on firms’ innovation activities and their returns.

Second, our study contributes to the literature on innovation. As innovation activities are increasingly organised within value chains, it is crucial to understand how value-chain structure might influence firms’ innovation performance. In this study, we stress that the relationship between R&D investment and innovation performance differs according to the value-chain structure of the firm and the industrial context of the firm’s innovation activities (McGahan and Silverman 2001; Nelson and Winter 1982). Our study shows the crucial role of a distributed value-chain structure for mobilising a large variety of knowledge and the pursuit of more distant search paths, and consequently for larger innovation returns on R&D investment in low dynamic industries. A distributed value-chain structure is less relevant for firms’ appropriation of innovation returns to R&D investment in highly dynamic industries, where the role of network redesign for unlearn and exploration (Chen, Coviello, and Ranaweera 2021), the degree of dis-synergy between R&D and marketing investment (Grimpe et al. 2017) and the potential for product-services offering for value creation (Visnjic, Ringov, and Arts 2019) is particularly accentuated. Future theorising and inquiry of firms’ innovation activities should not neglect the structure of the vertical partners linkages, especially as manufacturing and innovation activities are increasingly organised in global value-chains (Cano-Kollmann, Hannigan, and Mudambi 2018; OECD 2022).

Third, we contribute to research on the paradox of embeddedness in value-chain exchanges, which pointed out that while relational embedding with a few vertical partners may offer the important benefits of repeated interactions and better knowledge exchanges, it also creates numerous disadvantages (Laursen 2011; Lokshin, Hagedoorn, and Letterie 2011; Uzzi 1996). Our study adds to this discussion on the paradox of embeddedness, especially in the situation of value-chain partner concentration (Elfenbein and Zenger 2014, 2017; Mawdsley and Somaya 2021). We point out and empirically test that the firms’ ability to appropriate value from their R&D investment is reduced for firms that concentrate their exchanges with a limited number of partners and the contingency of industry dynamism on these effects. High dynamic industries characterised by high uncertainties but also with higher market opportunities and possibilities for value appropriation (Cohen and Klepper 1996; McGahan and Silverman 2001) create conditions to mitigate the effect of vertical partner concentration on firms’ innovation performance. Future conceptualisation of paradoxes of embeddedness may need to take into consideration not only the extent of exchanges concentration but also the appropriability conditions in the firms’ environment.

Finally, our study has implications for the efforts to mend bridges between transaction cost and capability theories (Argyres and Zenger 2012; Mahoney and Qian 2013). Prior research in this regard point to firms’ efforts to mitigate transaction hazards through contracts, either based on their core capabilities (Weber and Mayer) or the type of knowledge exchange involved (Eapen and Krishnan 2019). Our study complements these attempts by suggesting that mutual learning in the innovation process between a firm and its partners can be a critical factor that can help minimise the risk of opportunism. Such a situation seems to arise in contexts with frequent and disruptive market and technological advancements, which may necessitate inter-dependent learning between firms and their partners to develop new technologies and products, reducing the risks of lock-ins with less interesting markets and technologies and of partner opportunism (Laursen 2011; Tatarynowicz, Sytch, and Gulati 2016). Future research may seek to garner additional insights, such as the potential role of tacit knowledge exchange for reducing the risk of opportunism in embedded relationships.

7.2. Managerial Implications

The findings of this paper may help managers improve their understanding of the implications of the structure of their firms’ exchanges with suppliers and customers. They may thus take steps to prevent the potential vulnerability of their firms to a lack of variety of knowledge sources and to opportunistic threats from partners, thereby enhancing their ability to generate returns from their R&D activities. In particular, our research cautions managers of the risks of customer concentration, especially when occurring in conjunction with supplier concentration and in industries characterised by low dynamism. In such situations, anticipating the prospects of severely reduced bargaining power or the threat of takeover of their business by vertical exchange partners may help managers devise strategies to mitigate these potential pitfalls of concentration. They could, for example, seek additional knowledge channels and resources within their ecosystems through horizontal exchanges, or explore ways to repurpose innovations developed for major customers. Overall, our results point to significant long-term challenges for managers in a world where vertical exchange partners are central to firms’ innovation processes.

8. Limitations and Future Research

Our study has certain limitations that offer interesting directions for future research. First, the lack of detailed data on vertical exchanges did not allow us to consider the degree of partner concentration or possible non-linear moderating relationships involving vertical partner concentration. Future research may explore these and other related questionsin particular, whether the length of the relationship or the type of governance arrangements between a firm and its vertical exchange partners could mitigate the detrimental consequences of concentrated vertical exchanges on the positive relationship between R&D investment and innovation performance (Argyres, Bercovitz, and Mayer 2007).

Second, we considered vertical concentration along a single direction, as lack of data prevented us from taking into account the consequences of suppliers or customers concentrating their exchanges with the focal firm. Future research could consider the unidirectional vs bidirectional aspects of concentration, and how it may affect the process of learning and knowledge development, the threat of opportunism, and, ultimately, firms’ returns to R&D investments.

Third, previous evidence suggests that firm density and governance structures in upstream or downstream industries bear a significant influence on firm performance (De Figueiredo and Silverman 2012). We were unable to integrate these dimensions into our framework but examining how these characteristics of the downstream and upstream industries may moderate the effect of vertical partner concentration on firms’ innovation performance could be an interesting avenue for future research.

Finally, our empirical analysis focused on a sample of French firms over a specific time period. Country-specific changes in institutional or wider economic factors during the period of analysis could affect managerial decisions and the relationships analysed in our study. These include the French R&D tax credits aimed at incentivising innovation and the low inter-firm labour mobility in France (Bodas Freitas et al. 2017; Mason, Beltramo, and Paul 2004). Although we do not have an a priori reason to expect that context-specific factors may change the relationships identified, testing our findings in other contexts might reveal important nuances in the contextual influences on the innovation process in firms. The period of analysis is 2006–2008, when the phenomena of value-chain partners’ role and impacts on firms’ innovation were already relevant and pervasive but certainly less than today. Thus, while we would expect that our results to be stronger in contemporary contexts, we call for future research to test the generalisation of these results for different countries and time periods.

This paper highlighted how a firm structures its vertical exchanges is critical to its ability to create and capture value from its R&D investments and how this relationship is subject to the contingent influence of the dynamism of the firm’s industry.

Disclosure statement

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

Notes

¹ Qualcomm’s lawsuit against Apple alleged that ‘Apple developed and carried out an intricate plan … to steal vast swaths of Qualcomm’s confidential information and trade secrets and to use the information and technology to improve the performance of non-Qualcomm chipset solutions … ’. The two companies settled their dispute with Apple agreeing to pay a one-time payment of about $4.5 to $4.7 billion (‘Qualcomm Pegs Payment From Apple at $4.5 Billion to $4.7 Billion’. The New York Times. May 1, 2019.)

² Exchanges with customers provide valuable feedback on how firms’ products perform and how they are used, helping to trigger the needed adjustments to products’ design and functionality and generating ideas for new products (Katila et al., 2017; Laursen 2011; von Hippel 1998). Customer interaction thus enables firms to develop technologies and products that are better suited to the market needs, enhancing the innovation potentials of firms’ R&D investments (Christensen and Bower 1996; Leonard-Barton 1992).

³ Technological spillovers originating from suppliers, resulting in heightened understanding and assimilation of technological inputs, can accelerate the integration of inputs into the innovation process and the time-to-market, and thereby enhance firms’ development of innovative products (Adner & Kapoor, 2010; Helfat and Raubitschek 2000; McEvily & Marcus, 2005; Uzzi 1996).

⁴ The COI was conducted only twice (1997 and 2006), so we matched the 2007 COI data with the CIS 2008 survey.

⁵ When compared to the manufacturing firms in the original CIS dataset, firms in our final sample are significantly larger in size, more likely to be part of a group, engage more extensively in search and in outsourcing R&D activities. However, the intensity of R&D investment is not significantly different. Food, wood and paper and metals industries are slightly over-represented in our sample, while chemicals, electrical equipment and motor transport are slightly less represented.

⁶ For brevity, the equation does not include two-way interactions of industry dynamism with R&D, supplier concentration, and customer concentration. The effects of these interactions are not hypothesised.

Appendix

Table A. Robustness tests: Tobit, OLS, two-part, sub sample & R&D intensity-square analysis.

	Tobit Analysis	OLS Model	Two-part model test with exclusion criteria		Sub Sample Analysis				Model with R&D Intensity square
VARIABLES	Interactions	With Concentration	Innovation Performance	Innovative vs non-innovative	No Concentration	Joint concentration	Supplier Concentration	Customer Concentration
Size	−0.00740	−0.00740	−0.00911	0.137**	−0.0168	−0.143	−0.00936	−0.00367	−0.0227
	(0.00874)	(0.00889)	(0.0304)	(0.0692)	(0.0380)	(0.117)	(0.0913)	(0.0472)	(0.0271)
Group	−0.0147	−0.0147	−0.0497	−0.168	−0.0833	1.583***	−0.183	−0.166	−0.0477
	(0.0350)	(0.0355)	(0.141)	(0.131)	(0.153)	(0.465)	(0.270)	(0.200)	(0.109)
Search	−0.000999	−0.000999	0.00460	0.0616**	−0.00577	−0.0544	0.0391	0.00289	−0.00291
	(0.00391)	(0.00397)	(0.0205)	(0.0278)	(0.0164)	(0.0373)	(0.0469)	(0.0241)	(0.0125)
International Market	−0.0172	−0.0172	−0.0244	0.149	0.0825	−0.845*	−0.297	0.0241	−0.0493
	(0.0338)	(0.0344)	(0.106)	(0.187)	(0.191)	(0.434)	(0.323)	(0.150)	(0.108)
Knowledge Outsourcing	0.0177	0.0177	0.0877	0.188*	0.119	−0.144	−0.108	−0.0312	0.0579
	(0.0187)	(0.0190)	(0.0858)	(0.113)	(0.0818)	(0.259)	(0.208)	(0.113)	(0.0597)
Personnel Training	0.0344*	0.0344*	0.161	0.334***	−0.0167	0.385	0.496***	0.112	0.114*
	(0.0197)	(0.0200)	(0.105)	(0.101)	(0.0891)	(0.263)	(0.176)	(0.144)	(0.0651)
IT Capabilities	−0.0181*	−0.0181*	−0.0676***	−0.0803	−0.0793**	0.111	−0.00309	−0.0347	−0.0577**
	(0.00927)	(0.00942)	(0.0169)	(0.0522)	(0.0387)	(0.0905)	(0.0988)	(0.0612)	(0.0292)
R&D Intensity	0.778***	0.778***	2.569***	5.893***	2.180***	2.133	4.012**	0.531	2.563*
	(0.278)	(0.282)	(0.762)	(2.224)	(0.780)	(2.457)	(1.724)	(0.727)	(1.339)
R&D Intensity squared									−1.139
									(3.904)
Joint Customer & Supplier Concentration	0.0684*	0.0684*	0.168**	−0.130					0.219*
	(0.0399)	(0.0405)	(0.0715)	(0.120)					(0.123)
Supplier Concentration	−0.0475*	−0.0475*	−0.202***	−0.0641					−0.183*
	(0.0280)	(0.0285)	(0.0509)	(0.135)					(0.104)
Customer Concentration	0.0363	0.0363	0.105	−0.144					0.119
	(0.0258)	(0.0262)	(0.113)	(0.196)					(0.0814)
H1a: Supplier Concentration x R&D Intensity	−0.000868	−0.000868	−0.240	−7.321***					0.379
	(0.625)	(0.636)	(1.688)	(2.531)					(1.835)
H1b: Customer Concentration x R&D Intensity	−0.695**	−0.695*	−2.397**	−5.183*					−2.106**
	(0.351)	(0.357)	(1.048)	(2.729)					(1.006)
H2: Joint Customer & Supplier Concentration x R&D Intensity	−1.523**	−1.523**	−4.978***	−8.394**					−4.732**
	(0.643)	(0.654)	(1.399)	(3.329)					(2.148)
									−0.638***
Constant	0.275***	0.275***	−0.866***	−0.155	−0.742**	−1.179*	−1.165**	−0.453	(0.205)
	(0.0657)	(0.0668)	(0.257)	(0.295)	(0.303)	(0.656)	(0.454)	(0.358)	0.379
Observations	768	768	952	952	420	72	81	195	768
Log pseudolikelihood	18.2600				−229.6245	−38.7411	−35.9489	−111.0911
Wald Chi (2)					56.90***	93.11***	335.32***	578.67***
atanhrho			0.389
			(0.622)
rho			0.37
			(0.537)

Robust standard errors in parentheses; ***p < 0.01, **p < 0.05, *p < 0.1.

Table B. Robustness test: Model with more restrictive definition of innovation: share of products that are new-to-market only.

	(1)	(2)	(3)	(4)
VARIABLES	Controls	IV	Moderators	With Concentration
Size	−0.0233	−0.0216	−0.0212	−0.0221
	(0.0319)	(0.0318)	(0.0315)	(0.0316)
Group	−0.0536	−0.0361	−0.0504	−0.0625
	(0.114)	(0.115)	(0.114)	(0.113)
Search	−0.0179	−0.0194	−0.0184	−0.0177
	(0.0143)	(0.0144)	(0.0146)	(0.0147)
International Market	0.102	0.107	0.109	0.118
	(0.130)	(0.131)	(0.130)	(0.129)
Knowledge Outsourcing	0.0672	0.0586	0.0574	0.0649
	(0.0648)	(0.0653)	(0.0654)	(0.0648)
Personnel Training	0.138*	0.134*	0.134*	0.137*
	(0.0705)	(0.0700)	(0.0699)	(0.0706)
IT capabilities	−0.0662**	−0.0683**	−0.0673**	−0.0682**
	(0.0314)	(0.0315)	(0.0315)	(0.0313)
R&D Investment intensity		0.922*	0.919*	1.721**
		(0.540)	(0.551)	(0.680)
Joint Customer & Supplier Concentration			0.0842	0.249**
			(0.114)	(0.125)
Supplier Concentration			−0.0903	−0.116
			(0.113)	(0.122)
Customer Concentration			0.0247	0.118
			(0.0730)	(0.0924)
Joint Customer & Supplier Concentration x R&D Intensity				−5.951***
				(2.280)
H1a: Supplier Concentration x R&D Intensity				1.126
				(1.716)
H1b: Customer Concentration x R&D Intensity				−2.177*
				(1.156)
Constant	−1.047***	−1.073***	−1.072***	−1.116***
	(0.241)	(0.239)	(0.243)	(0.246)
Observations	768	768	768	768
Log pseudolikelihood	−285.13	−284.80	−284.57	−283.54
Wald Chi (2)	48.40***	48.74***	57.18***	67.89***
df;	18	19	22	25

Robust standard errors in parentheses; ***p < 0.01, **p < 0.05, *p < 0.1.

Table C. Bivariate probit regression to test reverse causality.

VARIABLES	Supplier Concentration	Customer Concentration	R&D Intensity (2008)
Innovation Performance (2006)	−31.68	32.82	63.91
	(23.65)	(26.46)	(62.21)
R&D Intensity (2006)	0.183	−2.402
	(1.163)	(1.690)
Size (2006)	0.106	−0.0637	−0.0614
	(0.0800)	(0.0969)	(0.260)
Group (2006)	0.355	0.550*	–
	(0.245)	(0.286)	–
Search (2006)	−0.0217	−0.0424	0.100
	(0.0286)	(0.0335)	(0.0746)
International Market	−0.573**	−0.236	2.237**
	(0.252)	(0.295)	(0.876)
Knowledge Outsourcing (2006)	0.0855	−0.194	0.399
	(0.145)	(0.159)	(0.435)
Personnel Training (2006)	0.00516	−0.262	−0.617
	(0.149)	(0.161)	(0.430)
IT capabilities	−0.142**	−0.0217	−0.501*
	(0.0679)	(0.0784)	(0.270)
Supplier Concentration			−0.00704
			(0.262)
Customer Concentration			0.135
			(0.456)
Constant	−0.119	−0.226	2.497*
	(0.545)	(0.679)	(1.472)
Observations	471	471	250
Log pseudolikelihood	−484.49681		−13.438033
Wald Chi2	1759.95***		39.47***
df	38		12

Robust standard errors in parentheses; ***p < 0.01, **p < 0.05, *p < 0.1.

Table D. Robustness test using Lewbel’s method, Sargan test.

VARIABLES	Innovation Performance
Size	−0.00729
	(0.00855)
Group	−0.0167
	(0.0308)
Search	−0.000760
	(0.00419)
International Market	−0.0148
	(0.0325)
Knowledge Outsourcing	0.0183
	(0.0189)
Personnel Training	0.0341*
	(0.0200)
IT Capabilities	−0.0180**
	(0.00904)
Joint Customer & Supplier Concentration	0.0851**
	(0.0392)
Supplier Concentration	−0.0363
	(0.0350)
Customer Concentration	0.0569*
	(0.0291)
R&D Intensity	0.851***
	(0.242)
Supplier Concentration x R&D Intensity	−0.0932
	(0.578)
Customer Concentration x R&D Intensity	−0.868**
	(0.386)
Joint Customer & Supplier Concentration x R&D Intensity	−1.689**
	(0.776)
Constant	0.379***
	(0.0710)
Cragg-Donald Wald F statistic	39.210
Stock-Yogo weak ID test critical values
5 % maximal IV relative bias	20.79
10% maximal IV relative bias	10.79
20% maximal IV relative bias	5.69
30% maximal IV relative bias	3.95
Sargan statistic	47.843
Chi-sq(63) P-val =	0.922
Observations	768
R-squared	0.082

Standard errors in parentheses; ***p < 0.01, **p < 0.05, *p < 0.1.