In the innovation literature, little attention has been paid to technical efficiency as a measure of performance, despite the fact that technical efficiency is key to explaining firm productivity. In this study we analyze the influence of internal R&D, external R&D and R&D cooperation on the technical efficiency of the firm. The data used come from the Panel de Innovación Tecnológica 2016 for Spanish companies, from whose database we have selected the set of innovative manufacturing firms. To estimate the coefficients of interest we have constructed a knowledge production function, estimated using Stochastic Frontier Analysis (SFA) and the Heckman correction. The results indicate that both internal R&D and R&D cooperation have a positive and significant influence on technical efficiency, while external R&D has a significant negative relationship. Knowing this type of information in advance is very important for business managers and policy makers, since it facilitates decision-making on innovation strategies, as well as the formulation of public policies to support innovation leading to an efficient allocation of public resources.
The economic literature recognizes that firms that are able to sustain a continuous flow of innovations experience higher growth rates and a higher probability of survival (Aghion & Howitt, 1998; Audretsch, 1995). Likewise, the economic literature also recognizes that innovativeness depends on the stock of knowledge that each firm possesses (Caloghirou, Giotopoulos, Kontolaimou, Korra & Tsakanikas, 2021), i.e., on its respective knowledge base (Jin, Wang, Chen & Wang, 2015; Zhou & Li, 2012). Now, firms build their knowledge base through internal investments in R&D and by acquiring external knowledge and recombining different knowledge sources (Hervas-Oliver, Sempere-Ripoll & Boronat Moll, 2022; Kogut & Zander, 1992). Consequently, the innovativeness of firms is related to their ability to create, acquire, recombine and apply different types of knowledge (Cohen & Levinthal, 1990; Nguyen, 2022).
However, very often, when many companies try to undertake innovation processes, they realize that they do not have the necessary and sufficient knowledge to successfully solve the innovative challenges they have set themselves, i.e., they detect that they have a knowledge gap (Hall & Andriani, 2002). In a situation of this nature, the company must be able to detect which are the main gaps in its knowledge base. Gaps that it must try to fill, creating or incorporating the necessary new knowledge. To this end, it should be emphasized that companies have different strategic alternatives for accessing the technological knowledge they need to develop their innovation processes (Berchicci, 2013; Chesbrough & Bogers, 2014; Grimpe & Kaiser, 2010; Tsai & Wang, 2009).
On the one hand, it is well established that internal R&D activities tend to generate technological knowledge (Griliches, 1979; Mansfield, 1969). In this way, firms improve their innovation capacity (Afuah, 2002; Stock, Greis & Fischer, 2001) and their corresponding business performance (Griliches, 1986; Tsai & Wang, 2004). However, the accumulation of this type of knowledge often requires substantial internal investment in R&D, as well as the assumption of high risks and a long time horizon for its materialization.
On the other hand, they can try to acquire the knowledge they need through external R&D purchases (Veugelers & Cassiman, 1999). This method of acquiring external knowledge is quick to implement, its cost is known in advance, presents a reduced level of risk and improves the flexibility of the contracting company (Harrison, 1994), since it facilitates the transformation of fixed costs into variable costs. However, from the point of view of innovation, its contribution is the subject of discussion and controversy. Some authors point out that the purchase of R&D services helps to reduce the firm's costs but has little or nothing to do with improving innovation capacity (Bettis, Bradley & Hamel, 1992; Weigelt, 2009). However, other authors emphasize that it allows quick and easy access to the R&D knowledge that these external suppliers possess (Howells, Gagliardi & Malik, 2008) and also helps to increase the speed of the corresponding innovation processes (Ebrahim, Ahmed & Taha, 2009).
However, there are situations in which it is not possible to resort to the purchase of external R&D services, mainly when the knowledge to be purchased is of a strategic nature and belongs to the so-called innovation frontier. This type of knowledge is usually not available on the market.
Moreover, companies find themselves in a context where technology is increasingly complex and its territorial and sectoral origin is very heterogeneous, the life cycle of products is shorter, and the degree of globalization of markets and competition are increasingly intense. As a result, it is very difficult to generate new knowledge and advanced technologies from a company's own resources alone, given the diversity, complexity and cost of the sources of knowledge that need to be managed. In addition, R&D investments nowadays tend to increase the proportion of fixed over variable costs, leading to the need to shorten the recovery times of the corresponding investments in order to control risk (Bower & Hout, 1988; Yip, 1992).
Consequently, in such a scenario, it is difficult for companies to undertake the development of the corresponding innovation activities alone, especially those that require high volumes of investment or involve the assumption of high technological risk. This is why innovative companies tend to share investment expenses, operating costs, risks and profits to a greater extent than other companies (Reger & Kuhlmann, 2012). Therefore, these companies tend to establish R&D cooperation agreements (Das & Teng, 2001; Park, Mezias & Song, 2004) in order to complement their internal capabilities (Vanhaverbeke, Duysters & Noorderhaven, 2002) and access the advanced knowledge of interest possessed by the corresponding partners (Park, Chen & Gallagher, 2002; Rothaermel, 2001; Teece, 1981).
Therefore, there are situations in which neither of the two alternatives mentioned above (internal generation or purchase of R&D in the market) are feasible for companies. When this happens, many companies are forced to implement cooperation agreements, since this is the only practicable alternative for obtaining the resources and capabilities they need (Das & Teng, 2000; Eisenhardt & Scoonhoven, 1996; Markides & Williamson, 1996; Park et al., 2004). In this regard, as the evolution of technology has moved along a path of greater complexity and acceleration, R&D cooperation agreements have increased significantly since the 1980s (Caloghirou, Ioannides & Vonortas, 2003; Hagedoorn, 2002). This type of agreement gives companies rapid access to the knowledge possessed by other companies, thereby facilitating and accelerating their corresponding learning and innovation processes (Becker & Dietz, 2004).
Therefore, with the emergence of the open innovation paradigm (Chesbrough, 2003), the management of R&D processes within companies has become much more complex, as managers are now obliged to find a fine balance between the internal sources of knowledge they possess and the corresponding external sources they intend to incorporate (Berchicci, 2013). But the complexity is even greater, since, in turn, they must also carefully analyze which external sources best contribute to the improvement of their corresponding innovation capacity. The first of the above challenges has received significant contributions in the innovation literature, but the analysis of which external sources best contribute to the improvement of innovation capacity has received much less attention. This paper addresses the latter question, investigating to what extent the two main external sources of knowledge (R&D cooperation and R&D purchasing) contribute to the improvement of the firm's innovation capability. In this sense, this paper contributes to the literature in two ways.
First, most studies analyze the contribution of the different sources of knowledge to the improvement of the firm's productivity. However, productivity has multiple causes, so the evaluation of the contribution of the different sources of knowledge must be narrowed down to a measure that best represents the management of the corresponding innovation processes. In this sense, in this study we propose as a performance measure the distance of each firm to its corresponding innovation frontier, that is, we propose to use as a performance measure the so-called technical (in)efficiency (Aigner, Lovell & Schmidt, 1977; Meeusen & van den Broeck, 1977). Firms farther away from this frontier will be more inefficient from the innovative perspective. In this regard, it should be borne in mind that there are notable differences in inefficiency between firms in the management of their innovation processes (Korres, 2008; Mairesse & Mohnen, 2002), even between firms belonging to the same industry (Cohen & Levinthal, 1989; Thompson, 2001). Moreover, the literature on innovation has repeatedly pointed out that technical (in)efficiency has a decisive influence on productivity (Aigner et al., 1977; Miller & Upadhyay, 2000). The concept of technical (in)efficiency arises in the context of the so-called stochastic frontier analysis, which involves the use of a stochastic frontier production function. Therefore, in this paper we use a knowledge production function that relates inputs of the innovative process to an output of that process, namely sales from new or improved products (Bos, van Lamoen & Sanders, 2011; Gantumur & Stephan, 2010).
Second, the ultimate goal of incorporating external sources of knowledge into the company's innovation process is to achieve an improvement in the efficiency of the use of the innovation resources that the company manages and, consequently, to increase the company's economic profitability. Therefore, it would be of great benefit to know in advance the technical efficiency of the two main external sources of knowledge, which are R&D cooperation and R&D purchases. In short, it is a question of knowing if both sources of knowledge contribute to the improvement of the company's innovation capacity, or, on the contrary, if each of these external sources has a different field of specialization.
In what follows, this study is divided into the following sections. In section two we establish the theoretical framework and state the corresponding hypotheses. In Section 3, we describe the source of the data used, define the variables and detail the methodology employed. Then, in Section 4, we present the results and their discussion, and in Section 5 we draw the conclusions.
2Framework and hypothesisInternal R&D constitutes the cornerstone on which all other R&D alternatives used in the innovation process of firms pivot. R&D expenditures are strongly and significantly related to the probability of introducing new products in the market (Parisi, Schiantarelli & Sembenelli, 2006; Torii, 1992) and the synergies, positive and negative, that arise from the simultaneous implementation of two or more innovation alternatives are strongly conditioned by the amount of internal R&D previously built by firms (Catozzella & Vivarelli, 2014; Hagedoorn & Wang, 2012). Therefore, it can be assumed that internal knowledge generation can positively influence the innovative performance of firms (Katila & Ahuja, 2002).
In this sense, internal R&D has often been characterised as the main source of knowledge that contributes to the growth of firms' innovative capacity and productivity (Griliches, 1979; Scherer, 1982). In fact, there are numerous studies that have contrasted the existence of a positive and significant influence of internal R&D on different measures of firms' innovative performance (e.g., Becker & Dietz, 2004; Beneito, 2006; Guisado-González, Guisado-Tato & Ferro-Soto, 2014; Love & Roper, 1999, 2001). Furthermore, numerous studies have also analysed the relationship between internal R&D and firm productivity (e.g., Audretsch & Belitski, 2020; Griliches, 1979, 1986; Hall & Mairesse, 1995).
In general, the economic literature recognises that internal R&D is a fundamental input for the development and improvement of technical efficiency within productive organisations (Barasa, Vermeulen, Knoben, Kinyanjui & Kimuyu, 2019; Kumbhakar, Ortega-Argilés, Potters, Vivarelli & Voigt, 2012). Technical inefficiency reflects the difference between the observed output and the maximum output obtainable, so in order to determine it, it is necessary to measure the deviation of each firm in relation to its production frontier. In general, the literature on innovation recognises that productivity is driven by technical efficiency (Aigner et al., 1977), which has prompted a series of studies to try to find out what its main determinants are. In this sense, there are studies that have analysed the main sources of efficiency using different variables and firm-specific variables (e.g., Caves, 1992; Caves & Barton, 1990; Green & Mayes, 1991). In this respect, the innovation literature argues that the closer a firm's position is to the innovation frontier, the more likely it is to use internal R&D as opposed to other innovation strategies (König, Lorenz & Zilibotti, 2016). In fact, there are studies that indicate that more innovative firms achieve higher technical efficiency (Dilling-Hansen, Madsen & Smith, 2003; Sánchez-Pérez & Díaz-Mayans, 2013) and that those that invest more in R&D obtain higher levels of technical efficiency (Kim, 2003; Kumbhakar et al., 2012; Pisa-Bo & Sánchez-Pérez, 2017; Sheu & Yang,2005; Torii (1992). Consequently, several studies have found that there is a positive and significant relationship between R&D and technical efficiency (e.g., Bonanno, 2016; Kumbhakar et al., 2012). Therefore, in line with the above, we hypothesize the following:
Hypothesis 1 Internal R&D has a positive and statistically significant influence on technical efficiency among innovative firms in the Spanish manufacturing sector.
In the economic literature the use of R&D outsourcing is controversial. In this respect, it is debated whether R&D outsourcing is a suitable instrument to improve the innovative performance of firms or whether it is only appropriate to control or reduce costs in R&D. In this respect, many studies point out that outsourcing is primarily a way to reduce costs and increase the flexibility of firms (Ellram, Tate & Billington, 2008). This is because R&D outsourcing involves the purchase of R&D results from specialised external suppliers (Grimpe & Kaiser, 2010), which possess advanced technologies and enjoy economies of scale that allow them to perform the same R&D tasks more efficiently and effectively than the internal departments of the contracting companies. (McCarthy & Anagnostou, 2004).
However, other studies highlight that R&D outsourcing allows access to R&D knowledge from external suppliers (Bunyaratavej, Hahn & Doh, 2007; Howells et al., 2008; Lewin & Peeters, 2006), which makes it easier for firms to improve their innovation activities (Bertrand & Mol, 2013; Calantone & Stanko, 2007; Howells et al., 2008). Moreover, R&D outsourcing allows synchronising internal R&D activities with those that are outsourced, which facilitates increasing the speed of R&D processes (Ebrahim et al., 2009; Langlois, 2003) and, consequently, improving the innovative performance of firms in terms of product variety and time to market (Ebrahim et al., 2009; Eisenhardt & Schoonhoven, 1996; Schoonhoven, Eisenhardt & Lyman, 1990). In this sense, and in relation to the Spanish manufacturing sector, there are some studies that have reported a positive influence of R&D outsourcing on some measure of innovative performance (e.g., García-Vega & Huergo, 2019; Tojeiro-Rivero & Moreno, 2019).
However, many other studies point out that the use of R&D outsourcing has a negative impact on innovative performance. Thus, some authors argue that inducing firms to the combined use of ready-to-use external technologies leads to a gradual and persistent neglect of in-house R&D development (West, Vanhaverbeke & Chesbrough, 2006), leading to an overall impoverishment of the firm's innovative performance (Bettis et al., 1992; Weigelt, 2009).
Furthermore, other studies argue that knowledge builds on itself (Geroski, 2005), making it extremely difficult to generate advanced R&D activities in the absence of prior experience (Griffith, Redding & Reenen, 2004), i.e. in the absence of a high absorptive capacity. Therefore, the extensive use of R&D outsourcing may hinder the internal creation of new knowledge and lead to a reduction of the corresponding absorptive capacity (García-Vega & Huergo, 2019), as in this situation the firm tends to abandon the internal development of its corresponding knowledge base (Manning, Massini, Peeters & Lewin, 2018). Therefore, there are authors who propose a negative relationship between R&D outsourcing and innovation capacity (Becker & Zirpoli, 2017; Grimpe & Kaiser, 2010). In this sense, Zirpoli and Becker (2011) point out that the more R&D firms outsource, the less knowledge they have of the components that make up the architecture of their products (Ciravegna & Maielli, 2011), and, consequently, the lower their innovation capacity.
Finally, it is difficult for companies to build a solid and lasting advantage from the knowledge acquired from their external suppliers, since many other companies have access to the knowledge of these same external suppliers (Grimpe & Kaiser, 2010). Therefore, firms that resort to R&D outsourcing may experience, momentarily and circumstantially, an improvement in their innovation capabilities, but this improvement is not sustainable over time, because the advantage acquired is ephemeral, and easily surpassed by other competitors. Consequently, in accordance with the arguments set out above, we put forward the following hypothesis:
Hypothesis 2 R&D outsourcing has a negative and statistically significant influence on technical efficiency among innovative firms in the Spanish manufacturing sector.
The high level of technological complexity is causing an increasing number of firms to be unable to obtain, from their own internal R&D, the new knowledge they need to compete in the market, which pushes them to establish R&D cooperation agreements with other organisations (Chesbrough, 2006). But the increase in such agreements does not diminish the efforts and investments in internal R&D. Quite the opposite is true. Finding out who possesses the advanced knowledge needed, assimilating it, and integrating it into the corresponding innovation process requires the existence of powerful in-house R&D (Dingler & Enkel, 2016). This is because in order to assimilate new knowledge it is not enough to maintain close and prolonged contact with the knowledge holder (Escribano, Fosfuri & Tribó, 2009), but one must also have the ability to "understand" this new knowledge, i.e. firms need to make high intensity investments in R&D (Gnyawali & Park, 2009) in order to create a high absorptive capacity for the knowledge that other organisations generate (Cohen & Levinthal, 1990; Zahra & George, 2002).
Moreover, technology today is not only more complex, but also changes at a faster pace, requiring higher investments than in the past and entailing a higher level of risk (Gnyawali & Park, 2011), resulting in much shorter product life cycles. This means that, in order to survive, companies have much less time to access new knowledge than in the past, which pushes them to establish cooperation agreements with other organisations, even with rival companies (Bengtsson & Kock, 1999). In short, we could summarize by pointing out that R&D cooperation makes it easier for firms to internalize spillovers and knowledge transfers between partners (Guisado-González, González-Blanco, Coca-Pérez & Guisado-Tato, 2018), generates economies of scale at the R&D level, facilitates access to complementary knowledge and technology services, and allows sharing costs and risks, avoiding duplication of R&D efforts (Kokkinou, 2010).
However, the continued use of R&D co-operation agreements by a growing number of firms can only be sustained over time if this innovation strategy exhibits a positive impact on firm performance, because if the impact is negative, its persistence over time would not be sustainable. In this sense, it is worth underlining that there is a wide range of studies that have corroborated a positive impact on different measures of firm performance (e.g., Becker & Dietz, 2004; Belderbos, Carree & Lokshin, 2004, 2006; Faems, Van Looy & Debackere, 2005; Guisado-González, Rodríguez-Domínguez, Vila-Alonso & González-Vázquez, 2021; Hernández-Trasobares & Murillo-Luna, 2020; Love & Mansury, 2007; Nieto, Santamaria & Fernández, 2015; Pekovic, Grolleau & Mzoughi, 2020; Zeng, Xie & Tam, 2010).
However, the performance of firms also depends to a large extent on the efficiency of their corresponding productive systems, since, for a given productive technology and environment, firms can exhibit significant differences in terms of efficiency. Therefore, efficiency in the management of productive resources has an important influence on different measures of firm productivity and performance. Moreover, innovation and technology are an important source of efficiency and competitiveness, so that access to the development of new technologies and innovations is crucial for firms. Consequently, to the extent that R&D cooperation makes it easier and cheaper to access and use new knowledge and technologies, it is conceivable that such cooperation can have an important influence on the technical efficiency of firms. Such an influence has not received much attention in the literature. Exceptions are Charoenrat and Harvie (2014) and Charoenrat, Harvie and Amornkitvikai (2013) who find that cooperation has a positive and significant influence on technical efficiency in the Thai manufacturing industry. Consistent with the above arguments, we hypothesize the following:
Hypothesis 3 R&D cooperation has a positive and statistically significant influence on technical efficiency among innovative firms in the Spanish manufacturing sector.
The data used in this study come from the Panel on Technological Innovation 2016 (PITEC 2016). PITEC is a survey on innovation activities of Spanish firms based on the Community Innovation Survey (CIS). In PITEC, companies answer questions related to their innovation activities, and also report on some of their most relevant economic characteristics. It is a firm-level panel database. From PITEC 2016 we select manufacturing firms, and after eliminating observations with some kind of impact on the variables of interest that invalidates the corresponding observation, we obtain a database with 3858 observations. From these observations, we select firms that have introduced product or process innovations and that, at the same time, exhibit positive innovation expenditures during the period of analysis, finally obtaining a set of 2218 innovative manufacturing firms.
3.2VariablesModels assessing different innovation strategies typically use a measure of firm performance as the dependent variable (Cassiman & Veugelers 2006). Often, the most commonly used dependent variable has been one of the many existing measures of firm productivity or innovative performance. However, these indicators may not adequately reflect the combined influence of different innovation strategies. For example, using patents as a measure of innovative performance can lead to biased influences, as patents can vary significantly in terms of quality and innovative content (Bzhalava, 2015; Griliches, 1990). Consequently, in this study we use as a performance measure the distance of each firm to its corresponding innovation frontier, i.e. we propose to use as a performance measure the so-called technical (in)efficiency (Aigner et al., 1977; Meeusen & van den Broeck, 1977), extracted from a firm's knowledge production function, estimated using stochastic frontier techniques (Coelli, Rao, O'Donnell & Battese, 2005). Firms closer to their corresponding frontier will be more efficient from an innovation perspective, as a consequence of the combined action of the external and internal sources of knowledge that each firm uses. Moreover, the literature on innovation has repeatedly pointed out that technical (in)efficiency has a decisive influence on productivity (Aigner et al., 1977; Miller & Upadhyay, 2000).
Internal R&D, External R&D and R&D cooperation are the determinants of inefficiency that we use in this study. These variables take the value 1 if the firm conducts internal R&D, buys ready-to-use external R&D and cooperates in R&D with other organisations, respectively; and take the value 0 in all other situations.
Also, in line with the innovation literature we incorporate the following control variables in our inefficiency model: R&D subsidy, Hightech and Labour. R&D subsidy takes the value 1 if the firm receives some kind of public support for innovation, while Hightech takes the value 1 if the firm belongs to a medium-high or high technology-intensive industry (OECD, 2003). In all other cases, both variables take the value 0. Finally, we use the Labour variable as a proxy for firm size, as innovative performance might be influenced by the corresponding economies of scale and scope (Henderson & Cockburn, 1994). This variable is related to the number of employees of the firm. In our analysis we use the natural logarithm of this variable.
In the basic specification of the knowledge production function we use the classical variables Knowledge and Labour as independent variables and SalesNewProducts as dependent variable, all expressed in natural logarithms (Bos et al., 2011). The variable Labour has already been defined above and the variable SalesNewProducts represents sales from new or improved products. The variable Knowledge represents the accumulated knowledge stock. PITEC does not directly provide data on this stock, but provides information on total innovation expenditure. Thus, based on these expenditures and using the perpetual inventory method (Guellec & van Pottelsberghe de la Potterie, 2004; OECD, 2009), the net knowledge stock at the beginning of period t (Knowledget) can be represented as a function of the net knowledge stock at the beginning of the previous period (Knowledget-1), plus the total innovation expenditures during previous period (It-1), minus the corresponding depreciation of the knowledge stock in t-1 (δKnowledget-1), assuming geometric depreciation at a constant rate δ (Berlemann & Wesselhöft, 2014):
Repeatedly substituting the above equation for the stock of knowledge in each year, we obtain:
Consequently, the stock of knowledge in period t is the weighted sum of the total expenditure on innovation in the past. Therefore, to know the stock of knowledge in period t it is necessary to know the complete time series of total innovation expenditure, and this is rarely possible. To overcome this handicap, Harberger (1978) uses neoclassical growth theory, assuming that the economy under consideration is in its steady state. Under this assumption, output grows at the same rate as the knowledge stock (Berlemann & Wesselhöft, 2014):
Combining (1) and (3):
In line with many previous studies, we assume that the depreciation rate of accumulated knowledge is 15% and the growth rate of total innovation expenditures, prior to the sample period, of 5%. (e.g., Bos et al., 2011; Hall & Mairesse, 1995). It should also be taken into account that total innovation expenditures may vary strongly from year to year. In order to minimize this kind of bias, in this study we use the five-year average value (measured in constant monetary units), rather than a single year (Kowledget-1). In this way, we generate more stable and reliable knowledge stock estimates (Harberger, 1978).
Likewise, in the empirical development, it must be taken into account that the natural logarithm of zero does not exist and the natural logarithm of 1 is zero. Therefore, to the variables SalesNewProducts and Knowledge, which have observations with zero values and are expressed in natural logarithms, we add a 1 before calculating the corresponding logarithm, as is usual in this type of situation (e.g., Escribano et al., 2009).
On the other hand, as the sample used in this study may generate sample selection biases, we will apply the Heckman correction. To this end, we will use the innovator variable and a set of variables related to the presence of obstacles to innovation. The innovator variable used takes the value 1 if the firm has introduced product or process innovations in the period and, at the same time, has positive innovation expenditures, and take the value 0 in all other situations.
3.3MethodologyAs noted above, in this study we will use technical (in)efficiency (related to the firm's ability to generate the maximum output for a given set of inputs), extracted from the knowledge production function using stochastic frontier analysis (SFA) techniques. The SFA model has been proposed by Aigner et al. (1977) and Meeusen and van den Broeck (1977). The SFA specification we use (e.g., Battese & Coelli, 1995; Kumbhakar, 1990) involves a production model with two error components: the first one accounts for the random effect (νi), and the second one for the technical inefficiency (ui). The SFA specification can be expressed as follows:
where, Yi is the logarithm for the output of the i th firm, Xi is the (k × 1) vector of explanatory and control variables associated with the i th firm, and β is the (k × 1) vector of unknown parameters to be estimated. Vi is a random error assumed to be iid (0, σv2). Ui is a non-negative random variable, represents the technical inefficiency, and is assumed to be independently distributed as truncations at zero of the (μ,σu2) distribution.
With the assumption of a linear functional relationship, the mean of the distribution of ui is a function of a set of explanatory variables (zi) and can be specified as:
where, zi is a (p × 1) vector of explanatory variable of inefficiency, and δ is a (1 × p) vector of unknown parameters to be estimated.
Two approaches are typically used to estimate the parameters of models (5) and (6): the two-step approach and the one-step approach. The two-step approach performs two separate estimations, estimating first the parameters of Eq. (5), in order to obtain the inefficiency values, and then the parameters of Eq. (6). However, many authors have pointed out that this method generates incorrect standard errors (e.g., Amsler, Prokhorov & Schmidt, 2016; Kumbhakar, Parmeter & Zelenyuk, 2017; Kutlu, 2010). To address these errors, the one-step approach implements the estimates of models (5) and (6) in a single step. This estimation strategy is the one that enjoys the highest predictability (e.g., Diaz & Sanchez, 2008; Kumbhakar et al., 2012), and therefore the one we use in this study.
In this study we estimate three models. In model 1 we estimate a Cobb-Douglas production function and in model 2 a Translog production function, both without the endogenous explanation of the inefficiency components (table 2). The variables of interest in both models are Knowledge, Labour and SalesNewProducts, expressed in natural logarithms. First, it is necessary to check that the SFA estimation is adequate. To this end, it will be necessary to detect the existence of inefficiency in both specifications. In case it is not detected, it would be more appropriate to use an ordinary least-quadratic estimation. For this purpose, after performing the corresponding estimations, we implement a likelihood-ratio test assuming the null hypothesis of no technical inefficiency (H0: σu = 0).
In addition, it is also necessary to test which of the two basic specifications best fits the data, in order to finally use the most appropriate basic specification. To assess the fit of both specifications we use the likelihood ratio test and the wald test, in order to check that both tests lead to the same endpoint.
Model 3 has the functional form selected in the previous step (Cobb-Douglas or Translog) and also incorporates Eq. (6), i.e. the variables that could potentially affect inefficiency (Internal R&D, External R&D and R&D cooperation), as well as the corresponding control variables (R&D subsidy, Hightech and Labour, the latter expressed in natural logarithms). This model is estimated using the one-step approach. The estimation will allow us to test whether the three hypotheses are fulfilled.
Finally, it should be noted that our analysis focuses on innovative manufacturing firms (2218), selected from the total number of Spanish manufacturing firms (3858). Consequently, the sample is not strictly random, so that sample selection biases may exist. In order to correct for this bias, we implement the Heckman correction. This methodology is based on a two-step procedure. The first step uses a probit selection equation, estimating this equation on all manufacturing firms (3858) and using as dependent variable whether or not the firm performs innovation activities (Innovator). From this estimation, the inverse of the Mills ratio is calculated. In the second stage, the outcome equation is estimated. The inverse of Mills' ratio, calculated in the first stage, is incorporated into this equation. The inverse of Mills' ratio corrects for sample selection bias. In this second stage, only innovative firms are considered, i.e. 2218 observations. Wooldridge (1995) points out that the independent variables in the outcome equation should be a subset of the independent variables in the selection equation. In this study, the selection equation has the same independent variables as the outcome equation, plus a set of variables representing different barriers to innovation.
4Results and discussionThe basic descriptive statistics for the variables used in this study are shown in Table 1. In this table it can be seen that 87% of innovative manufacturing companies carry out internal R&D activities, 46% cooperate in R&D with other organisations, 45% receive some kind of public aid for innovation and 51% are medium-high and high technology-intensive companies.
Basic descriptive statistics.
Table 2 shows the coefficients of the SFA estimates of the two basic specifications, Cobb-Douglas and Translog. The same table shows the likelihood-ratio test for the existence of inefficiency in both specifications. The results of this test indicate that the null hypothesis of no inefficiency (H0: σu = 0) must be rejected at a level of 1%. Therefore, since there is inefficiency in both specifications, the SFA estimation is adequate.
Basic models without explanation of inefficiency.
Standard errors are in parentheses
∗ Significant at 10%, ∗∗ significant at 5%, ∗∗∗ significant at 1%.
Next, we proceed to select the basic specification that best fits the available data. For this purpose, once both specifications have been estimated by SFA, we implement the likelihood ratio test and the Wald test. These tests are commonly used to evaluate the difference between nested models. Therefore, the implementation of either of these two tests would be sufficient, but we apply both tests to verify that both lead to the same selection result.
The likelihood ratio test compares the log-likelihoods of the two basic specifications, and tests whether the difference is statistically significant. If it is, this implies that the Translog specification is a better fit to the data than the Cobb-Douglas specification. Table 2 shows that the value of the likelihood ratio test is 9.84, and that it is statistically significant at the 5% level. Therefore, we can affirm that the two basic specifications are not nested and that the Translog specification fits the data better than the Cobb-Douglas specification.
Wald test examines the model that has more parameters, in our case the Translog specification, and evaluates whether restricting the extra parameters relative to the Cobb-Douglas specification seriously impairs the model fit. Therefore, the aim is to test whether the parameters that the Translog specification adds, relative to the Cobb-Douglas specification, are significantly different from zero. Table 2 shows that the Wald test takes the value of 10.74, and that it is significant at a level of 5%. Therefore, we cannot state that the variables that the Translog specification adds, with respect to the Cobb-Douglas specification, are irrelevant, so it is the Translog specification that best fits the available data.
Likewise, at the end of the aforementioned Table 2, the ratio (σu2/σu2+σv2) of each of the two specifications is shown. In this regard, we note that the Translog specification attributes 96.09% of the variance to inefficiencies in the innovation process, while in the Cobb-Douglas specification this attribution is 95.94%.
Table 3 presents the results of the SFA regression, using the one-step approach, of the translog basic specification together with the determinants of inefficiency. As our interest is focused on the evaluation of the coefficients of these determinants, we omit any comment on the coefficients of the translog basic specification, as these do not constitute the object of our investigation. The coefficients of the inefficiency determinants will allow us to test the three hypotheses we have put forward.
SFA regression coefficients with inefficiency determinants.
Standard errors are in parentheses
∗ Significant at 10%, ∗∗ significant at 5%, ∗∗∗ significant at 1%.
Regarding the interpretation of the coefficients of the determinants of inefficiency, it should be borne in mind that since the hypotheses have been formulated in terms of efficiency, the interpretation of the estimated parameters should be made exactly in the opposite sign of the corresponding coefficients in Table 3. Consequently, through the results shown in Table 3, we find that the internal R&D and R&D cooperation variables have a positive and statistically significant influence on efficiency, and that the external R&D variable has a negative and statistically significant influence on efficiency. Therefore, the results support hypotheses 1, 2 and 3.
The results in Table 3 are consistent with the findings of most of the innovation literature regarding the influence of the three variables analyzed on innovative performance. For example, regarding the influence of internal R&D on innovative performance, Love and Mansury (2007) consider that internal R&D is the most important determinant of innovation while Berchicci (2013) suggests that internal knowledge accumulation strongly influences the firm's innovative performance. In general, there are many studies that have shown that internal investment in R&D improves the innovative performance of the firm (e.g., Cockburn & Henderson, 1998; Fabrizio, 2009; Malva, Kelchtermans, Leten & Veugelers, 2015). Likewise, many authors have also stressed that internal R&D affects technological development from a double aspect (e.g., Leten, Kelchtermans & Belderbos, 2022; Lokshin, Belderbos & Carree, 2008); on the one hand, it strengthens the firm's absorptive capacity to take advantage of external knowledge, and, on the other hand, it constitutes a direct source of innovation. In this sense, Leten et al. (2022) find that the positive relationship between internal R&D and innovative performance is significantly mediated by this double play.
The relationship between internal R&D and technical efficiency is complex (Tang & Wang, 2019), which to some extent explains why ambiguous results have been found in the innovation literature. Thus, a few studies have found that the relationship between R&D expenditures and technical efficiency is negative and significant (e.g., Amornkitvikai & Harvie, 2010; Barasa et al., 2019; Barasa, Kimuyu, Kinyanjui, Vermeulen & Knoben, 2015; Fahmy-Abdullah, Sieng & Isa, 2021; Gumbau & Maudos, 2002; Kim, 2003). Other studies have found that the relationship is not statistically significant (Amornkitvikai, Harvie & Charoenrat, 2014; Fahmy-Abdullah, Sieng & Isa, 2018; Ismail, Noor & Abidin, 2014; Kim, 2003; Noor & Siang, 2014; Suntherasegarun & Devadason, 2023). However, there are even more studies that have found a positive and statistically significant relationship (e.g., Amornkitvikai & Harvie, 2010; Bhat & Kaur, 2022; Bonano, 2016; Driffield & Kambhampati, 2003; Fahmy-Abdullah et al., 2021; Hu, Yang & Chen, 2014; Ismail et al., 2014; Kim, 2003; Kumbhakar et al., 2012; Liik, Masso & Ukrainski, 2014; Peng & Zhang, 2023). As our empirical results also confirm the existence of a positive and statistically significant relationship, hypothesis 1 of our study is endorsed, in line with the mainstream innovation literature.
Likewise, in table 3 we confirm that between external R&D (R&D outsourcing) and technical efficiency there is a significant negative relationship. This results support hypothesis 2. We are not aware of studies that have analyzed the relationship between R&D outsourcing and technical efficiency, but the economic literature reports some studies that analyze the relationship between outsourcing and technical efficiency. In this respect, the results found are ambiguous, although more numerous are those that report a significant negative relationship or a non-significant relationship. In line with our finding, the study by Phillips (2013) finds that outsourcing staff has a negative and significant impact on technical efficiency of Japanese water utilities, while Ludwig, Groot and Van Merode (2009) find that outsourcing does not significantly affect the technical efficiency of most Dutch hospital utilities. On the other hand, Mishra, Sinha and Thirumalai (2009) find that outsourcing project organization is associated with significantly lower levels of technical efficiency compared to collocated insourcing project organization, and Mishra, Sinha, Thirumalai and Van de Ven (2020) find that the technical efficiency of an IT project subject to offshore outsourcing is lower than that of a domestic-insourcing project. At the other extreme, Lin, Jin and Guo (2023) note that the results of their study suggest that outsourcing technology-intensive services has a positive and significant impact on technical efficiency, while Kim (2011) finds that outsourcing services has a positive and significant impact on the technical efficiency of Malaysian Hotels.
Finally, the results shown in Table 3 also support our hypothesis 3, since they reflect that there is a significant positive relationship between R&D cooperation and technical efficiency. In general, empirical studies on the relationship between the two variables are very scarce, and, to the best of our knowledge, the relationship between R&D cooperation and technical efficiency is not the main focus of most of these studies. In general, these studies show that cooperation has a positive and significant influence on technical efficiency. For example, out of a total of six studies we have analyzed, four show a positive and significant relationship (Charoenrat & Harvie, 2014, 2013; Charoenrat et al., 2013; Hu et al., 2014) and two show a significant negative relationship (Ng & Li, 2003; Peng & Zhang, 2023). On the other hand, there are studies that have as their main objective the comparative analysis of R&D cooperation agreements between groups of companies. These studies estimate the technical efficiency of each group and disentangled how each collaboration group affects R&D efficiency, using a combination of stochastic frontier analysis and meta-frontier analysis in the empirical analysis (e.g., Chung, Jo & Lee, 2021; Lee, Lee & Shon, 2020; Na, Lee, Hwang & Lee, 2021).
5ConclusionsThe company has multiple alternatives to carry out its innovation processes. The literature on innovation has abundantly and reliably shown how these different alternatives have contributed to improve/worsen different measures of company performance, whether this performance is evaluated at the economic, financial or innovative level. However, there are few empirical studies that analyze the performance of the company through the corresponding technical efficiency, despite the fact that such efficiency is key to explaining both the productivity and the economic and financial profitability of the company. This is so to the extent that technical efficiency depends on the effectiveness with which the firm manages its resources, since the level of technical efficiency achieved depends on the position of the firm in relation to its corresponding technological frontier. Precisely, one of the main contributions of this study is that it presents an analysis of the influence of the firm's internal R&D, R&D outsourcing and R&D cooperation agreements on technical efficiency. For this purpose, we use a knowledge production function, and by means of stochastic frontier analysis techniques we estimate, in a single step, the innovation frontier and the corresponding technical inefficiency, as well as the influence of internal R&D, external R&D and the R&D cooperation on this inefficiency.
Internal R&D constitutes the main ingredient of all possible combinations of the different innovation alternatives available, since ultimately all knowledge flows coming from outside the company have to be valued and integrated by the corresponding internal R&D. And in this sense, the results of the empirical analysis show that internal R&D has a positive and significant influence on the technical efficiency of the company.
We have also found that knowledge flows from R&D cooperation agreements are effectively integrated by the corresponding internal R&D, since their influence on technical efficiency is also positive and significant.
However, the influence of outsourcing R&D on technical efficiency is negative and significant. Consequently, there is a high consensus in the economic literature that outsourcing constitutes a strategy aimed at cost savings, and therefore has the capacity to improve different measures of economic or financial performance. However, in the field of innovation, the contribution of outsourcing is not so clear. Our finding indicates that its contribution is negative, and it also consumes resources that could be used to increase internal R&D capacity or to establish R&D cooperation agreements with other organizations, which do have a positive influence on the company's innovation capacity. In short, R&D outsourcing and R&D cooperation fall into different areas of specialization. While R&D cooperation contributes directly to the improvement of the firm's innovation capacity, R&D outsourcing seems to contribute to cost savings, but not to the improvement of the firm's innovation capacity.
Knowing this type of information in advance is of utmost importance, since it allows company managers to make the right decisions regarding the strengthening and extension of their corresponding innovation capacity. It is also very important for political decision-makers, since it provides them with vital information for the design of their corresponding innovation promotion policies, thus helping to avoid the implementation of policies that lead to an inefficient allocation of public resources.
