Endogeneity between internationalization and knowledge creation of global R&D leader firms: an econometric approach using Scoreboard data

1. Introduction

Among the very large number of stimuli a firm has for expanding its knowledge base, conducting R&D and engaging in exporting activities are identified as two especially prominent activities that offer the firm opportunities toward this direction (Johanson and Valhne, 1977; Nonaka, 1994). In this article we consider these two activities of firms as potential knowledge flows to augment their knowledge stock. Although their relationship has been investigated thoroughly by scholars (Lall and Kumar, 1981; Schlegelmilch and Crook, 1988; Wilmore, 1992; Lefebvre et al., 1998; Wakelin, 1998; Bleaney and Wakelin, 2002; van Dijk, 2001; Lachenmaier and Woessmann, 2006; Pla-Barber and Alegre, 2007), the existence of endogeneity between these two activities tends to be omitted in both empirical findings and theoretical predictions.

The article at hand, builds on the potential existence of endogeneity between technological change and foreign markets forces, as they are represented by R&D and Export Intensity of the Global R&D leaders. Unravelling such an endogenous relationship, is in line with the theoretical framework arguing that Technology Push and Market Pull forces act simultaneously (von Tunzelmann, 1995; Ruttan, 1997; Ito and Lechevalier, 2010) and are essential in evolving technological paradigms and shaping new technological trajectories (Dosi, 1988). The best equipped candidates to shape or redefine existing and new paradigms and trajectories are those firms that distinguish themselves from their peers, in that they spend a great deal of their resources in augmenting their knowledge base.

More specifically, we argue that the knowledge bases of these firms are fueled, in a particular mode, by both directions, i.e. from the R&D activities, and also from knowledge flows generated due to interactions with foreign markets. In this context however, the R&D process is characterized by great uncertainty and risk, and can be perceived as a search for more efficient methods of value creation. (Nelson, 1982; Castellacci, 2007). We demonstrate how Technology Push and Demand Pull knowledge flows are linked and how they affect firms’ decisions to invest in the creation of new technological knowledge. Specifically, for the global R&D leader (RDL) firms, Technology Push and Demand Pull knowledge flows are by no means equally important. Essentially, knowledge flows from foreign markets are employed in the service of improving the efficiency of Technology Push knowledge flows generation mostly through the inducement of Localized Technical Change (Atkinson and Stiglitz, 1969; Antonelli, 1998). This fact in turn, provides the opportunity to the RDL firms, that by their very nature are oriented in generating knowledge via the notable investments in R&D activities, to reinforce and sustain their competitive advantage.

Drawing information from the UK Government Department of Business Innovation and Skills (BIS), regarding RDL firms, we devised a pooled cross-section dataset for 2006–2007 period and tested for the presence of endogeneity with respect to R&D and export intensity. The econometric approach employed takes advantage of the intensity measures, instead of models of binary variables, exploiting the full information conveyed by the outcome of the RDL firms’ decision-making processes with respect to their R&D and exporting activities.

The rest of the article is organized as follows: in the next section, a brief review of the literature regarding the empirical and theoretical findings of the relationship between R&D and exporting activities will be conducted. Section 3 presents some modeling issues and the econometric strategy to be followed, while Section 4 provides the data and definition of variables. Section 5 is devoted to the discussion of the econometric results and Section 6 concludes the article and poses some further research directions.

2. Theoretical and empirical context

2.1 The drivers of innovation

The relevant literature has been quite extensively preoccupied with the identification of the drivers of technical change, as they are reflected in the decision to invest in R&D activities. In the form of hypotheses, different but, as we shall argue, complementary aspects have been expressed, and ever since, scholars have been arguing in favor or against them. In the heart of this discussion, the ‘Technology Push’ and ‘Market Pull’ (Schmookler, 1966; Rosenberg, 1990), along with the Schumpeterian hypotheses regarding bigness and fewness (Acs and Audretsch, 1987) have attracted the focus of the efforts trying to answer “what drives and determines innovation.”¹

Additionally, cumulativeness, appropriability conditions and technological opportunities have been nominated to be among the cornerstones of industrial evolution and dynamics (Scherer, 1965; Malerba and Orsenigo, 1993, 1996).

Innovation has been detected using a number of alternative measures. Firm's expenditures on R&D, patents and Intellectual Property Rights, along with innovation indicators—depicting whether a firm has introduced a process or product innovation or both (Peters, 2009), are the most commonly employed proxies of innovation. In this article, we focus on firms’ expenditures on Research and Development as an innovation indicator, having one distinctiveness: the R&D expenditures concern firms that have been globally announced to be those that spend the most in this area of investment. This constitutes the backbone of our analysis in the sense that these firms’ most prominent characteristic is the fact that they have all that it takes to be considered as carriers of technical change and new technological trajectories (Dosi, 1988). At the same time, this extraordinary investment on R&D, arms these firms with the necessary capabilities to cope with global competition (Chandler, 1986). To elaborate further, the intense and sometimes harsh global competition, in which these firms are destined to function and compete, imposes a one-way solution, that of continuous improvement on both products and processes, allowing, and determining to some extent, the scientific evolution and at the same time sustaining a firm’s competitiveness (Ernst and Kim, 2002). On the other hand, these firms need to meet a highly diversified global demand for their products and they need to do so, timely and effectively exploiting scale and scope economies, as well as dynamic returns to learning (Chandler, 1994).

Setting off to identify what drives RDL firms’ intense orientation toward R&D activities, it is essential to sketch first the contextual foundations of our analytical framework. As it was previously mentioned, being among those who invest the most in R&D sets out a signal that competition has been taken one step ahead of the current technological frontier. Investing on R&D boosts scientific research and constitutes a strategic choice (Dierickx and Cool, 1989) toward a sustainable competitive advantage (Rumelt, 1984). From that point of view, the development and expansion of internal firm-specific capabilities and routines (Nelson and Winter, 1982; Teece and Pisano, 1994) associated with the generation of new technological knowledge, as well as the identification and exploitation of technological opportunities, has been named in the relevant literature as the “Technology Push” hypothesis (Mowery and Rosenberg, 1979).

On the other hand, much attention has been given to the influence of the demand conditions in directing firms' innovation activities. The relevant literature has documented that demand conditions exert quite a significant impact on the decision and extent of a firm's innovative strategy. In other words, and according to the “Demand Pull” hypothesis (Schmookler, 1966; Scherer, 1982; Ruttan, 1997; Piva and Vivarelli, 2007), market signals drive the innovation possibility frontier of each and every firm (Dosi, 1988). As Scherer (1982) puts it, the demand-pull theory is inextricably linked with the ability to make profits. Therefore, market opportunities (Kamien and Schwartz, 1982: 35) are what matters in directing innovative activities.

Nonetheless, Mowery and Rosenberg, (1989: 8) noted that

“… the acquisition of knowledge for innovation is not an once-and-for-all matter. Rather than a unidirectional, one-time occurrence of transfer of basic scientific knowledge to application, the processes of innovation and knowledge transfer are complex and interactive ones, in which a sustained two-way flow of information is critical. The ability to adopt a new technology, to evaluate a new technique, or even to pose a feasible re-search problem to an external research group may require substantial technical expertise within the firm”.

2.2 The knowledge base formation

Dosi (1988) defined a firm's knowledge base as “the set of information inputs, knowledge and capabilities that inventors draw on when looking for innovative solutions.” With this definition as a starting point, later on, Kogut and Zander (1992) argued that creating new knowledge does not occur in abstraction from current abilities, but rather new learning, such as innovation, is the product of a firm's combinative capabilities to generate new applications from existing knowledge. As a result, the knowledge-based view of the firm (Grant, 1996) conceptualizes the firm as an entity that mainly creates knowledge in a unique way, which in turn constitutes its competitive advantage, and this is why the core competences of a firm are hard to imitate by competitors (Spender, 1996).

The generation of technological knowledge can be viewed as a complex system of interactions between actual production experience and formal R&D efforts (Kline and Rosenberg, 1986). Therefore, the development of new technologies can rely on the knowledge acquired by means of learning by doing and learning by using in the spectrum of techniques currently being used (von Tunzelmann, 2000). Deepening the above considerations, the process of building a firm's knowledge base is by nature, a highly recursive and cumulative process associated with path-dependent as well as past-dependent characteristics (David, 1985; Antonelli, 1998) and dynamic increasing returns of learning (Geroski et al., 1997).

More specifically, R&D activities can be directed toward the solution of any problem, but once it has been carried out, the resulting knowledge is specific to the problem addressed (Atkinson and Stiglitz, 1969). In other words, path dependence of R&D investments refers to irreversibility, as well as, indivisibility characteristics which determine, to a great extent, the direction and the potential of the firm's knowledge base augmentation (David, 1975). Past-dependent characteristics affect the generation of new technological knowledge due to differential initial conditions existing among industrial structures. These differentials translate in uneven rates of change with respect to factor prices, demand preferences and other industry-specific characteristics which in turn affect the firm's technological evolution (Caves and Porter, 1977; Antonelli, 1998). Dynamic increasing returns of learning accrue to the fact that the generation of new technological knowledge can be viewed as a quasi-economic good, in the sense that it is acquired and recombined through a joint process of internal and external learning, production, and communication (Geroski et al., 1997).

As a concomitant of the above, the generation of new technological knowledge can be viewed as a continuous technological search seeking to improve and diversify existing techniques and applications (Nelson, 1982). The outcome of this knowledge generating process induces Localized Technological Change (LTC), (Atkinson and Stiglitz, 1969; Antonelli, 1998), in the sense that the technological change induced, concerns a specific set of techniques and/or problem solutions, and therefore, possible spillovers to the entire production possibility set are limited. Localized technological knowledge should be perceived as the product of systemic bottom–up process of induction from actual experience, sharply contrasting with the top–down process of deduction from general scientific principles on which the received theory of knowledge as a public good is rested (Dosi, 1988). It is evident that such a specific targeting may evoke improvements in a firm's efficiency and productivity.

2.3 Endogeneity and findings from the literature

The existence of an endogenous relationship between R&D and exporting activities, even though recognized very early in the literature (e.g. Keesing, 1967), it has been neglected for almost 20 years and then only scarce and rather peripheral research has been undertaken, especially when it comes to empirical analyses. On the contrary, from a theoretical point of view, and specifically in the context of Product Life Cycle and Endogenous Growth theory, the endogenous relationship between innovation and exporting activities constitutes a pillar of the analysis. More specifically, the Product Life Cycle theory argues that innovation will eventually lead to exporting (Posner, 1961; Vernon, 1966; Krugman, 1979; Dollar, 1986) and that exports act as a channel for diffusing, as well as transferring knowledge and technology (Saggi, 2002). This theoretical strand is strongly interrelated with the Market Selection Hypothesis (MSH); (Wagner, 2007) which favors the argument that exporters have superior performance characteristics than nonexporters.

On the other hand, the Endogenous growth models of International Trade (Grossman and Helpman, 1989, 1990, 1991; Segerstrom et al., 1990; Young, 1991; Aghion and Howitt, 1998, ch. 11) consider that innovative activity is endogenously determined and predict some dynamic effects from its relationship with international trade. More specifically, it is argued that the exporting firms’ access to foreign markets provides them with feedback from their suppliers and/or customers, which gives them the opportunity to transform this knowledge into innovation. This theoretical strand has been recorded as opposite to the market selection hypothesis and is named Learning by Exporting Hypothesis (LEH); (Evenson and Westphal, 1995).

At an empirical level, the firms’ exporting activities have occupied the relevant literature from the early 1970s and in particular, the last two decades’ research on this topic has been intensified due to the globalization process. Although there is a substantial amount of literature investigating the determinants of exporting activities, this field of research has been guided, implicitly or explicitly, by the two above mentioned hypotheses, i.e. MSH and LEH, regarding the identification of determinants and their relationship with export intensity.

In particular, with respect to the extensive research on whether R&D activities have any influence on exporting propensity, results have been mixed. Indicatively, Schlegelmilch and Crook (1988) and Lefebvre et al. (1998), found that R&D intensity has no influence on export intensity. Lall and Kumar (1981), investigating a sample of Indian firms, revealed a negative relationship between export and R&D intensities. On the other hand, many empirical studies have revealed a positive and statistically significant impact of R&D on export propensity (Willmore, 1992; Wakelin, 1998; Bleaney and Wakelin, 2002; Lachenmaier and Woessmann, 2006; Pla-Barber and Alegre, 2007). These studies however, do not explicitly address the existence of endogeneity nor are they concerned with the causality of this relationship. In addition, exporting activities stemming from developing countries are treated differently from those deriving from developed countries in terms of what determines them respectively (van Dijk, 2001).

Toward dealing with the endogeneity issue between these two firm decisions, Hughes (1986), using cross-section country level data, applied a Hausman test and found that the relationship between R&D intensity and export intensity, was simultaneously determined. Toward the same direction, Clerides et al. (1998) applied causality tests in order to define the pattern of causality between R&D and export intensity and found that more productive firms chose to export. Smith et al. (2002), using a sample of Danish firms, tackled the issue of endogeneity and reported that R&D increases the probability that a firm will become an exporting firm. Even more interesting is the empirical work by Harris and Li (2008): using an UK sample of firms, they investigated the endogeneity between R&D and export propensity, and argue that a crucial factor for the lack of evidence on this endogenous relationship may be due to the lack of appropriate data and problems with econometric methods that allow testing for such an endogenous relationship. In a very recent article, Ito and Lechevalier (2010) applied a system of generalized methods of moments (GMM) estimation in a rich 10-year panel of Japanese firms, “treating” the endogeneity problem between R&D and export intensity with lagged variables of R&D intensity as instruments.

3. Methodological underpinnings

In this section, we present the methodological route we followed in order to ground the relationship between the export and R&D propensity of RDL firms. We focus primarily on the formulation and implementation of an efficient test of weak exogeneity between the outcomes of the two aforementioned decision-making processes of firms.

It is apparent that the variance of the error term of the RDINT equation depends on the parameter α and the variances of ⁠. The parameter α as reflected in equation (5) is the slope of the linear equation, or in other words reflects the degree of correlation between the two error terms.

4. Data and variables definitions

The constructed dataset is drawn from the R&D Scoreboard provided by the British Department for Business, Innovation and Skills (BIS)—former Department of Innovation, Universities and Skills (DIUS)—and covers two time periods. More specifically, the data report financial and other basic economic characteristics of firms for the years 2006 and 2007.

Attention should be drawn to the fact that the reporting firms were selected on the basis of their R&D expenditures that are funded by themselves. R&D undertaken under contract for other agents such as governments or other companies, as well as the firm's share of any associated firm or joint venture R&D investment, are excluded. Furthermore, the financial statements used in this case are the consolidated group financial statements of the ultimate parent company. Firms which are subsidiaries of any other company are not ranked separately. The specific data handling procedure incorporates the view that the crucial strategic decisions are taken by the central management of the firm and the degree of freedom which remain for the “peripheral management” is rather limited to operational aspects (Penrose, 1959).

From the 2007 R&D Scoreboard (DIUS, 2007) which provides information regarding the financial year 2006, 595 RDL firms have reported their exporting activities. Among these, 164 firms lacked crucial information and for that reason, they were excluded from that year's sample. Thus, for the year 2006 our sample consists of 431 firms. From the corresponding 2008 R&D Scoreboard (BIS, 2008), which in turn provides information regarding the financial year 2007, export sales information was available for a total of 667 RDL firms, from which 270 firms were short of other important information and thus were excluded, leaving 397 firms in the specific year sample. In sum, the 2-year dataset, 2006–2007, consists of 828 observations, out of that 42 firms are considered to be entrants in 2007 and 65 firms, even though they appeared in the 2006 sample, exited in 2007. The structure of our dataset is determined by 360 firms present in both years and 54 firms appearing either in 2006 or 2007. The final composition of the dataset used along with the basic distribution characteristics of the two crucial variables, that is EXPINT and RDINT, is presented in Table 1.

Additionally, in Figure 1, the kernel density plots of RDINT and EXPINT are presented. Specifically, RDINT kernel density function depicts the well-known regularity of a left skewed distribution (Cohen and Klepper, 1992). Of course, in the case where the RDL firms are the unit of analysis, it is not surprising that the entire distribution is shifted rightwards. We should also point out that, contrary to what might be expected, RDINT variable is not censored at 1. This fact may be interpreted by the particularities of a small number of firms that exist in our sample and they invest more in R&D expenditures than they can presently afford, apparently using intensively external financial resources (Bougheas, 2004). Essentially, this admittedly small subgroup of firms discounts that their competitive advantage will arise from the inflation of their knowledge base through the exploitation of a technological opportunity. Thus, combining the non-censored nature of the RDINT variable neither to zero nor to unity, we have been led not to use a Tobit model for describing the RDINT function.

For the total of the excluded firms, either because they failed to report export sales or because they had missing values in some variables, an issue of selection bias may arise. For that reason we performed nonparametric Mann–Whitney U statistic in order to test whether statistically significant differences exist in the distribution characteristics of our dataset and in those of the excluded firms. The tests were performed upon a series of firm-specific variables (R&D intensity, firm size, industrial and location distribution, profitability, profit margin, and labor productivity). The performed nonparametric tests did not reveal any statistically significant differences between the two samples.³

The selected explanatory variables of the EXPINT equation are drawn from what the relevant literature of MSH and LEH dictate. The corresponding selection of independent variables for theRDINT equation is guided by the relevant literature on Schumpeterian hypotheses of bigness and fewness along with the conditions for firms’ dynamic evolution, i.e. appropriability, cumulativeness, and technological opportunities. The full set of the employed variables, their corresponding definitions, and basic descriptive statistics are presented in Tables 2 and 3 respectively.

5. Results and discussion

In this section we present and discuss the empirical estimates of the econometric model as it was previously formulated and the accompanying tests which evaluate the validity of the empirical results. The structure of this section consists of three subsections. In the first subsection we present the treatment of a number of issues that arise in the econometric context we have adopted. The second subsection is concerned with the discussion about the empirical findings regarding the endogeneity between R&D and export intensity. The FIML estimates of the export and R&D intensity determinants are discussed in the final subsection.

5.1 Econometric issues⁴

On the basis of previous empirical findings and theoretical argumentations regarding the determinants of R&D and export intensity, we have included a meaningful and informed set of explanatory variables among the available economic and financial variables. Estimation results of the above two-equation model are presented in Table 4.

In the context of the variables’ selection, for each equation, an important issue needs to be addressed. Despite the fact that the main focus for the potential existence of endogeneity lies in the first equation, where R&D intensity is a suspected endogenous variable, the relevant economic theory dictates that export intensity is also a determinant of R&D intensity (Bhattacharya and Bloch, 2004).

However, due to the necessary recursiveness of the model, it is not possible to introduce export intensity as a control variable in the equation where R&D intensity is the dependent variable.⁵ Therefore, we use export Intensity with one time lag ⁠, as an Instrumental Variable (IV). Later on, we address this issue in more detail as it is crucial for disentangling the relationship between these two activities and their underlying economic intuition.

In order to select the model with the best econometric properties among alternatives, a forward selection process was followed. This implies that some variables with no statistically significant coefficients have been included in the final model as they are considered to be an important finding and because such an inclusion does not worsen the overall econometric performance of the model. In the econometric framework adopted in the article at hand, and presented in the previous section, a plethora of econometric issues arise which may result in inconsistent estimates of the parameters, standard errors and diagnostic statistics of the system of equations (1) and (3).

A potential and rather serious drawback of the Blundell–Smith procedure may be that the Tobit model assumes that the two decisions of whether and how much to export are affected by the same set of factors. If this is not the case, severe heteroscedasticity, due to Tobit misspecification, would be present. To cope with such a possibility, Cragg (1971) introduced a general model which permits us to test whether the initial decision of EXPINT > 0 versus EXPINT=0 should be separate from the decision of how much to export, given that the firm has decided to become an exporter. Cragg's approach requires the estimation of three models, namely the decision whether to export, the decision of how much to export and the restricted version of the latter. In order to avoid this triple estimation procedure, Lin and Schmidt (1984) devised a Lagrange Multiplier (LM) test, equivalent to Cragg's test, which requires the estimation of the Tobit model only (Greene, 2007: E25–7).

In our case, the estimated value of the LM statistic is equal to 19.137, which is smaller than the critical value of ⁠. Therefore, the null hypothesis that the two decisions are driven by the same factors is not rejected and thus, the Tobit model may be considered as an acceptable specification (Table 5).

Given the result of the Lin and Schmidt (1984) test, one would expect that the possibility of occurrence of the heteroscedastic Smith and Blundell disturbance term is substantially reduced (Wooldridge, 2002: 533). Nevertheless, we have conducted some additional tests regarding the correlation of the major independent variables of the EXPINT equation and the variance of the Smith and Blundell residuals. In all cases, we cannot reject the hypothesis of the homoscedastic error term since the influence of each independent variable on the variance of the Smith and Blundell residuals is statistically nonsignificant (Table 5).

Regarding the normality assumptions of the u_EXP and u_RD it should be mentioned that Wooldridge (2002: 531) proves the Smith and Blundell endogeneity test being valid without any distributional assumptions on the reduced form of the endogenous variable, in our case the RDINT. In order to test the normality assumption, which is employed in the case of EXPINT equation, we have performed a simple Jarque–Berra test based on the residuals of the Smith and Blundell specification. The estimated value of the corresponding JB-statistic does not permit us to reject the null hypothesis of the normality of the u_EXP error term (Table 5). In order to check for the presence of multicollinearity among the regressors, we have also computed the Variance Inflation Factor (VIF) matrix. Results indicated that no serious multicollinearity problems arise (see note 5) (Table 5).

Although the adopted econometric strategy tests for possible endogeneity of RDINT with respect to EXPINT, one could point out that additional endogeneity issues may be present regarding other crucial variables of the estimated model. In order to deal with such a possibility, the results of a number of simple Davidson and MacKinnon (1993) endogeneity tests are presented in the fifth part of Table 5. We have tested for endogeneity regarding the variables of SIZE, PRFAB, PRMRG, LBPRD, MSEURand MSNAM. In all cases, we do not find statistically significant influence of each one of the above variables on the Smith and Blundell residuals and thus, the null hypothesis of exogeneity for all the above variables, is not rejected. The most problematic case is the one of the PRFAB variable where the corresponding P-value is just over the 10% level.

Finally, some issues concerning the potential use of alternative econometric strategies need to be addressed. One could argue that a dynamic panel GMM-IV estimator should be employed as a better solution in order to cope with the possible endogeneity between RDINT and EXPINT variables. In the context of GMM-IV, the presence of endogeneity is only indirectly detected by the corresponding Hansen's statistic (1982), which is concerned with testing whether the employed instruments are valid. In addition, several issues, related to the available data arise, and in particular, panel data GMM, in the case of small number of time periods and large cross-sections, requires the Arellano–Bond (1991) difference GMM estimator (Holtz-Eakin et al., 1988). This manipulation leads to the inclusion of lagged explanatory variables into the model, which due to the nature of our dataset is not feasible to do since only two time periods are available for all of the used variables except RDINT. It should be noted that it would be possible for one to derive and estimate a GMM model where one dependent variable is censored and the other one is a linear model by employing Quasi-ML estimation techniques (Meyerhoefer et al., 2005; Crino, 2010). Nevertheless, this procedure is beyond the scope and intentions of this article.

In any case, and keeping in mind the restrictions imposed by the employed dataset, i.e. the absence of a dynamic panel and the censored nature of EXPINT variable, we have implemented a GMM system estimation for EXPINT and RDINT exploiting the advantage of being an endogeneity-free problems method with which to juxtapose our results from the Smith and Blundell econometric approach. Furthermore, we have separately estimated the EXPINT equation using a Tobit model with endogenous covariates (TEC) and the RDINT equation using OLS. In the final part of Table 5, a Hausman (1978) specification test is presented which does not indicate inconsistency in the Smith and Blundell estimators used in this article compared with the corresponding system GMM estimators. On the contrary, the analogous Hausman tests between GMM and Tobit model, with endogenous covariates for the EXPINT equation and GMM and OLS estimators for the RDINT equation, do not permit us to accept the hypothesis that the estimators of TEC and OLS are consistent with the GMM class estimators. The econometric results, especially with respect to the relationship of the crucial variables i.e. EXPINT and RDINT, proved to be robust in several dimensions. More specifically, signs and statistical significance of the corresponding estimated parameters remain unaltered when the used dataset is reduced to only one cross-section and when outliers and influential observations are excluded.

Overall, we could argue that the adopted Smith–Blundell modified with IV econometric procedure presents in an explicit manner the endogeneity between RDINT and EXPINT on the one hand, and produces consistent estimators of the crucial variables of the model on the other. Of course, it should be stressed here that this is the case for the specific dataset and one should be very cautious, econometrically speaking, when using the Smith and Blundell estimation procedure since a plethora of econometric tests, regarding the assumptions, should be undertaken and the corresponding necessary corrections may not be always feasible.

5.2 The EXPINT and RDINT relationship

In this section, the focus of attention is drawn on the relationship between the two endogenously determined RDL firms’ decisions based on the estimation results presented in Table 4. At the bottom of the same Table, the estimated value and the corresponding asymptotic standard error of the parameter ψ, designed to test for the presence of endogeneity, are also displayed. According to the performed test, the hypothesis of weak exogeneity of RDINT with respect to EXPINT is not accepted. In the context of the theoretical framework developed previously, Technology Push and Demand Pull mechanisms as they are represented by RDINT and EXPINT, respectively, are interrelated and constituting a particular operational mode of knowledge generation.

Further below, we argue that in this mode of operation, there exists a crucial element, that of the latent in econometric terms, knowledge base of the firm. In order to further explore how this mode operates, we have devised an informative sketch as presented in Figure 2. Some clarifications are in order here. Arrow lines represent Knowledge flows, which are either due to Technology Push (Top–down direction) or Demand Pull (Bottom–up direction). RDINT and EXPINT represent observed firm decisions. The gray oblong represents the RDL firm stock of knowledge. Ellipses represent effects related to the functional operation of the firm's knowledge base.

Econometric results indicate that the R&D intensity positively affects a firm's exporting propensity. In terms of Figure 2, which illustrates the theoretical framework already discussed in a previous section of this article, the Technology Push mechanism as it is illustrated in R&D intensity, generates knowledge flows, which inflate the knowledge base of the RDL firm. This enlarged knowledge base grounds the RDL firms’ competitive advantage, which in turn results in superior performance in the global market. Following the bottom–up direction of Figure 2, export intensity positively affects R&D intensity, however, not in a linear form. The knowledge flows arising from the Demand Pull source implicitly augment the firm's knowledge base, following a quite different approach. In particular, information flows from foreign markets permit filtering and feedback mechanisms to occur and thus, to induce LTC (Atkinson and Stiglitz, 1969). LTC targets at a specific technology set, as it is embedded in the exported goods, and forces the RDL firm to make some targeted technological improvements as the latter are dictated by foreign market forces. This fact alone has two immediate implications. On one hand, the RDL firm is obliged to invest more in R&D resources in order to capitalize the information flows from the foreign markets. On the other hand, the knowledge outcomes of such an investment are prolific, in the sense that the firm has minimized the required search costs (Nelson, 1982) for tracking down the optimum technique, among all the available in the known technological space. According to von Tunzelmann (1995), the optimization of this search process, with respect to technical change, can be perceived as an increase of the R&D efficiency of the firm.⁶

We could argue that the early stages of export engagement, in which an RDL firm engages in exporting activities, have a greater impact, via the channels of filtering and feedback information, on expanding the firm's knowledge base. However, these channels are limited in their contribution and the more engaged one becomes, the less are the learning benefits from such a particular activity. Based on the anticipated effects of LTC on RDL firms’ knowledge base, we can rather safely assume that the technical change induced upon a specific set of techniques is not sufficient to evoke generalized technical change given the specific characteristics of RDL firms’ knowledge base. For the RDL firm to pursue a more generalized technical change and draw new trajectories (Dosi, 1988, 1997), i.e. in order to sustain its role of leadership in R&D investment, it needs to rely on additional forms of internationalization such as foreign direct investment (FDI), forms of alliances, etc.

5.3 The determinants of EXPINT and RDINT

Beside the main relationship between R&D and export intensity that was analytically presented and discussed above, each set of control variables for each equation of the system will be the interest of this section. Considering the EXPINT equation, we have incorporated the explanatory variables capturing firm size, labor productivity, profit margin, profitability and location which are dictated by the MSH. Accordingly, we have included the destination of exporting activities as an explanatory variable in order to test for the LEH.

The role of SIZE is investigated since the relevant literature has identified it as being considerably important in explaining export intensity variation. In light of the above, significant quadratic effects of parabolic type are observed. This finding is in accordance with the empirical evidence provided by several authors (Schlegelmilch and Crook, 1988; Kumar and Siddhartan, 1994; Wagner, 1995; Wakelin, 1998). Wagner (1995) argued that firm size advantages are present up to a certain threshold due to coordination costs and bureaucratic issues, while Schlegelmilch and Crook (1988) argued that this nonlinearity is due to the fact that above a certain size, large firms find it more efficient to proceed to FDI rather than exporting. These explanations can be complementary rather than contradictory. More specifically, it can be argued that these firms are in a position to absorb and therefore, exploit export-oriented demand-pull knowledge flows until a certain size threshold. After they reach a certain size, alternative sources of demand-pull knowledge flows may become more attractive, i.e. FDI, intra-firm trade. Such being the case, the interrelated decisions to invest on R&D and penetrate foreign markets become more complicated.

The PRMRG variable which constitutes an indicator of the market power a firm may possess and exert, when included in the EXPINT equation, is positive and statistically significant. According to this empirical finding, the argument that international trade contributes to the increase of the welfare due to international competitive forces does not seem to apply in this case. Market power having a positive influence on export intensity suggests that the competition in foreign markets does not resemble the perfect competition model but on the contrary the oligopolistic model is more suitable to explain export behavior of firms (Krugman, 1979). As far as the LBPRD, PRFAB, EURD, NAMD, and the time TIMED are concerned, they were found to exert no statistically significant influence on the extent of Demand Pull knowledge flows.

Moving on to the Demand Pull knowledge flows originated from LEH, we should mention that they are not homogenous but instead may be dependent on their geographical origination (UNCTAD, 2005). The results indicate that both MSEUR and MSNAM are negative and statistically significant. The negative influence of the two variables has to be interpreted in relation to the excluded variable, MSROW. The RDL firms that mainly orient their exporting activities toward Europe and North America, exhibit overall, a lower degree of export commitment in relation to those that mainly participate in ROW markets. Therefore, it could be argued that any Demand Pull knowledge flows deriving from European and North American markets are expected to be relatively small and consequently any filtering and feedback mechanisms inducing LTC will be rather limited in scope and extent.

The discussion about the determinants of the R&D intensity equation, i.e. the Technology Push knowledge flows, concerns the Schumpeterian hypotheses regarding firm size and industrial market structure (Schumpeter, 1942), and also the three conditions that define the dynamic evolution of the industries, namely appropriability, technological opportunities, and cumulativeness (Malerba and Orsenigo, 1993, 1996).

Regarding the role of size in determining the extent of R&D intensity, the econometric results reveal statistically significant quadratic effects of the hyperbolic type. Bearing in mind that these firms have already spent such a considerable amount of resources as to be characterized as R&D leaders, the interpretation of such a nonlinear relationship draws, on one hand, from the fact that the production and distribution of new innovative products or processes is characterized by sharp economies of scale (Shy, 2004: 53) and on the other hand, from the seminal work of Cohen and Klepper (1996) and their notion of cost spreading advantage. More specifically, and with respect to the declining part of the U-shaped relationship between firm size and R&D intensity, we argue that these firms have invested a significant amount of resources in the development of a new technology as the latter is embedded in the new improved products or processes. The fruits of such an investment are capitalized via the disproportionately short-term increase of sales and therefore, increasing turnover spreads the cost of invention. However, this short-term supranormal profits cannot be sustained diachronically. Eventually, for the RDL firms to sustain their competitive advantage and thus their technological leadership, they have to evolve the existing technological paradigm by investing proportionally more resources into the development of a new process or product than their size changes. This process in turn, creates additional Technology Push knowledge flows.

The Schumpeterian hypothesis concerned with the effect of industry concentration, with respect to innovation, was proxied with the Herfindahl index ⁠. According to the econometric results, increased concentration of an industry strengthens the R&D efforts of individual firms. Following Levin et al. (1985) industrial concentration as a determinant of R&D intensity may lose its interpretive power when one accounts for inter-industry systematic differences, such as appropriability conditions, technological opportunities, and cumulativeness of knowledge. Furthermore, and in order to control the presence of effects due to appropriability conditions, we included the RDL firms’ profit margin as an indicator (Kamien and Schwartz, 1982: 28). Our results indicate that this index exerts negative and statistically significant influence. Appropriability conditions, as they are captured by the RDL firms’ profit margin, seem to provide the RDL firms with the necessary safety, in order to crop the fruits of their investments on one hand, and not to decide hastily when and how much to invest in their next project, at least temporarily, on the other (Winter, 2006). At this point, one could reasonably argue that the firm's profit margin can only partially depict appropriability conditions especially with respect to RDL firms. Unfortunately, relevant information regarding detailed appropriability policies is not present in this research. In general, however, such information is quite difficult to be depicted in innovation surveys and firm's financial statements.

The next influential component of Technology Push knowledge flows is related to technological opportunities. These effects are tested through industry-specific variables⁷ (Scherer, 1965) as they are defined in Table 2. Results of the estimated coefficients show that all industry dummies are statistically significant and negatively affect R&D intensity. This is not a surprising result if one considers that the omitted industry dummy variable, which is used as a reference point, is the ICTHW variable (i.e. the hardware branch of ICTs).

The third and final element, cumulativeness, which determines the propensity of R&D activities, is approximated with the RDIN1 variable, which according to our empirical estimations is positive and statistically significant. This finding comes to support the role of path dependency (David, 1985) on R&D activities as it has been recorded in Malerba and Orsenigo (1993). Finally, it should be noted that the variable NAMD is positive and statistically significant, confirming the superiority in terms of knowledge creation flows of the North American firms.

6. Conclusions

Although RDL firms may be perceived as important carriers of technological change, with respect to the evolution of technological paradigms and trajectories, theoretical considerations are lacking, in terms of a unifying framework, about the interaction of sources and mechanisms with which knowledge is generated. In this article, we follow a data-driven research process, which allows us to construct a theoretical framework of the RDL firms’ knowledge base creation. Emphasis is given on the conspicuous relationship between Technology Push and Demand Pull as mechanisms of knowledge creation.

In this context, Technology Push knowledge flows are captured by R&D intensity and Demand Pull knowledge flows by degree of penetration to foreign markets respectively. In the process of building our theoretical framework, central role has been assigned to the endogenous relationship between export intensity and R&D intensity of the RDL firms.

Drawing from what has been recorded in the relevant literature and employing an appropriate econometric methodology, we demonstrate that in the process of augmenting the RDL firms’ knowledge base, Technology Push and Demand Pull knowledge flows are operating in conjunction. However, we find that they are not of equal importance. More specifically, the very nature of RDL firms indicates that Technology Push knowledge flows are crucial for them in order to maintain their leadership position through the exploitation of technological opportunities. In terms of access to foreign markets, the RDL firms are provided with knowledge inputs, which have mainly feedback and filtering character, and are thus, rather limited in scope, capable of inducing LTC. More specifically, decreasing marginal learning benefits, with respect to knowledge flows arising from exporting performance, are present. In other words, for the firm to pursue a more generalized technical change and draw new trajectories, i.e. in order to sustain its role of leadership in R&D investment, it needs to rely mainly on its own ability to produce new technological knowledge and/or in additional forms of internationalization.

Exploring deeper each knowledge generation mechanism, we conclude that the Technology Push side is ruled by the conditions of industries’ technological evolution, i.e. cumulativeness, appropriability, and technological opportunities as they have been recorded in the relevant literature. No exception is made regarding the RDL firms. As far as the validity of the Schumpeterian conditions is concerned, the picture is not so clear. However, the reader should keep in mind the obscurity, that goes hand in hand with the approximation of the Schumpeterian hypotheses, in terms of measurements issues. Regarding the Demand Pull knowledge generation mechanism, we outline a mixed pattern of influence, sourcing from the market selection and learning by exporting paradigms. RDL firms’ individual performance characteristics, together with the geographical distribution of targeted foreign markets are the main channels through which the knowledge flows from exporting activities.

The theoretical framework identified in this data-driven research process produces some further research directions. More specifically, it would be quite interesting for one to explore the role of exports’ geographical distribution on the RDL firms’ knowledge base observed heterogeneity. A consequent question is related to the unobserved RDL firms’ heterogeneous knowledge bases, and is directly linked with the causal ambiguity principle. Finally, we should not ignore whether other forms of internationalization behave in the same pattern when it comes to Demand Pull knowledge flows creation.

Acknowledgements

The authors would like to thank two anonymous referees, William Greene, Marco Vivarelli and participants in European International Business Academy (EIBA) Conference in December 2009 for their valuable comments and suggestions on earlier versions of this manuscript. We would also like to thank Janet Butt and Andrew Cullins from BIS (formerly DIUS) for their helpful assistance in the provision of the necessary information. The usual caveat applies.

References