Voluntary technological disclosure as an efficient ... - CiteSeerX

10 downloads 1023 Views 500KB Size Report
A firm with high R&D intensity, from a high-tech sector, ... between the theoretical treatment of knowledge spillovers and business practice ..... ATVEF on television, IETF on Internet), and research consortia (such as SEMATECH or MCC).
Voluntary technological disclosure as an efficient knowledge management device: an empirical study Stéphane LHUILLERY3 Université Paris Nord, UFR d’économie, 99, av. J.B. Clément, 93430 Villetaneuse, France [email protected]

Revised draft - January 2003

Abstract This paper investigates three questions related to endogenous information and knowledge disclosure by firms: Which industry sectors are more apt to disclose information and knowledge? Why is such knowledge released? Is knowledge disclosure an efficient strategy? An empirical analysis on four French data sets that focus on appropriation, the practices of innovation, and the related payoffs suggests answers to these questions. A firm with high R&D intensity, from a high-tech sector, participating in R&D partnerships is found to be more likely to engage in disclosure. Firms in the sample were found to "leak" their knowledge to public laboratories to a greater degree than to other private sector firms. Leakage also was found to be associated with improved innovation performance. This research helps broaden the literature on knowledge management practices to include not only the pursuit of formal intellectual property rights such as patents but less formal inter-organizational knowledge transmission mechanisms.

Keywords: innovation, endogenous spillovers, cooperation, appropriation JEL classification: O32, L21, C81 3 I thank B. Hall, F. Sachwald, C. Le Bas, L. Miotti, E. Pfister, an anonymous referee, and the participants in the 2001 seminar of the doctoral school on Economics and Management of Innovation and Knowledge (Measurements and econometrics of innovation) in Strasbourg (BETA) for helpful comments and suggestions. The Ministry of Industry (SESSI) and the ministry of research and higher education (DPD C3) are also acknowledged for the availability of data. The opinions and remaining errors are the author’s alone.

1

I. Introduction Since Arrow (1962), empirical and theoretical studies of the innovation process have underlined the importance of knowledge spillovers. Many empirical works have been made to assess the existence and the importance of the external available stocks of knowledge on R&D decisions and productivity (See Klette et al. (1999) for a survey). Some authors have developed a theoretical oligopoly framework where spillovers may deter R&D investments (see De Bondt (1996)) for a review) or even the impact of spillovers on economic growth. This literature considers that spillover levels are exogenous. A first step towards the concept of endogenous spillover is to consider that firms develop costly strategies in order to obtain external knowledge. While the usual models assume that firms absorb knowledge without significant costs, several empirical studies (Cohen & Levinthal (1989), Henderson & Cockburn (1996), Adams (2000)) show that the capacity to absorb external knowledge from other firms’ knowledge depends on their own R&D efforts. Theoretical models are also available to ground the empirical findings (Cohen & Levinthal (1989), Kamien and Zang (2000)). A second step towards this concept is to consider that firms are also able to design appropriation mechanisms in order to prevent spillovers to competitors (See Levin et al. (1985)). Nevertheless, there is an important feature of spillovers that these previous theoretical and empirical models have failed to take into account. Empirical evidence suggests that firms let their knowledge flow either at the level of the innovation process or at the innovative product or process level. This purposeful knowledge sharing behavior is now quite well documented in case studies and industrybased empirical studies, which provide useful insights into the different means of disclosure and the characteristics of the firms engaging in transmission. A variety of research works suggest that sharing knowledge cannot be restricted to a research joint venture framework or technical consortia. This paper is an attempt to bridge between the theoretical treatment of knowledge spillovers and business practice through an empirical investigation involving a large number of firms across multiple industries. The empirical inquiry focuses on three questions: What motivates firms to disclose information and knowledge? Which industries are more apt to engage in leakage? Does disclosure enhance a firm's innovation productivity? Using data compiled from four different data sets on innovation, appropriation, and innovation success, this research explores answers to these questions. The surveys include information on either knowledge sharing mechanisms or disclosure practices implemented by innovative firms. We examine the impact of industry sectors, firm size, R&D intensity, and R&D partnering on four dichotomous proxies of knowledge disclosure. An innovation production function is also considered to investigate the impact of endogenous leakage on innovation performance. Finally, the relationship between a firm's disclosure strategy and its pursuit of formal intellectual property rights (IPRS) is examined.

2

The paper is organized as follows. In Section II, we survey the theoretical literature that examines the rationale behind disclosure strategies as well as the previous empirical work done in this area. In Section III, we develop a framework of analysis and measurement that incorporates both the strategy of the transmitting firm to leak and appropriate its knowledge and the ability of the receiving firm to absorb it. Section IV presents the survey data and the empirical methods employed. Section V discusses the results, and the final section concludes.

II. Knowledge disclosure: theory and evidence A. Theoretical models Several recent theoretical works suggest that endogenous leakage occurs not only in a cooperative setting but in a competitive setting as well. Firms can establish knowledge management systems to retain information and knowledge but also to boost the release of knowledge. In demonstrating these outcomes many authors use a two-step oligopoly model with cost-reducing R&D to link the R&D investment and subsequent profit level to an exogenous externality rate (for a survey, see De Bondt (1996)). When comparing non-cooperative and cooperative R&D investments, a first natural hyptothesis is that different spillover levels occur for cooperating groups and competitive groups. The models sometimes assume that spillovers are endogenous when cooperation occurs (Katz (1986); Kamien et al. (1992); Poyago-Theotoky (1995), Beath et al. (1998)). These cooperative models suggest that complete knowledge disclosure is required to maximize profits when R&D expenditures are coordinated.1 In fact, an RJV will obtain any given R&D output more cheaply than in a non-cooperative equilibrium. Indeed, compared to the non-cooperative equilibrium it may achieve greater R&D output with a smaller total expenditure on R&D. An RJV may achieve more R&D output than the noncooperative equilibrium, especially when the products are differentiated, independent or even complementary. Here the disclosure rate is subordinate to R&D cooperation. Katsoulakos & Ulph’s paper (1998) presents a model in which the R&D and information sharing are endogenously and independently set even for a non-cooperative equilibrium. They find that it is possible that noncooperative equilibrium achieves maximal information sharing, especially when firms conduct complementary research and operate in different industries. In an RJV, information sharing is more likely to occur. Two general conclusions seem to be found in the oligopoly literature dealing with R&D investment. First, information or knowledge disclosure is more likely to occur in R&D cooperative projects. A more precise prediction is that the more the products are differentiated between R&D partners, the higher the level of R&D investments and the higher the leakage rate. Secondly, primarily when products are differentiated, firms may be able to achieve full research design and information

3

sharing, and thus they are more likely to increase their profitability. More precisely, these theoretical results suggest the two following hypotheses : H1

The probability to disclose is higher when a firm implements R&D cooperative

agreement(s); H2

The probability to disclose is higher for R&D partners as clients or customers

(strategic complements), than for cooperation with public labs (considered here to be neither strategic complements nor subsitutes) and much more than partnerships with competitors (strategic substitute). The previous RJV models are simplistic since the sharing is found to be an all or nothing proposition. However incomplete sharing occurs for different reasons. Imperfect information sharing may come up against an upper limit. Such a ceiling can exist for technical or economic reasons. In the case of the former, complete codification of the knowledge for transmission to another firm is simply considered impossible to achieve because the nature of the technology (complexity, tacitness); De Bondt and Wu (1997) suggest also that knowledge sharing may be imperfect because of technological or cultural distance between firms. In this case, increasing codification costs or transmission costs would lead to an intermediate level of information sharing (as suggested by David and Foray (1995)). Of course, firms can also hide some of their research results from their partners. Adverse selection in knowledge sharing and moral hazard problems can also be considered (See d’Aspremont et al. (1998), for a survey of the literature on R&D teams). The voluminous literature, however, requires data on codification costs and asymmetric positions in RJV that are very difficult to obtain for a large representative sample. This topic is out of the scope of the present paper. The concept of disclosure strategy is not restricted to the RJV literature, and voluntary leakage can also be implemented in a pure R&D competitive framework with races for invention. Using such a framework, Lichtman et al. (2001) show that an efficient strategy for a laggard firm may be to narrow the scope of or fully preempt a rival’s patent. In this case, the laggard discloses any research results it had in common with the rival and can compete in a duopoly sub-market instead of a monopoly market. From the leader’s perspective, the first stage of information disclosure is a means of driving laggards from the race. However a leader will not adopt a disclosure strategy at the innovation output level (second stage). Laggards and leaders do not necessary disclose the same kind of information. This kind of model seems thus very hard to test because it detailed descriptions of the information and knowledge flows as well as the strategic position of the competitors. De Fraja (1993) obtains different and more tractable results employing a similar modeling approach: first, firms tend to let their knowledge leak when they have to share the market with imitators, when the loss of the laggard is small compared to the

1

Kamien et al. (1992) then refer to these arrangements as Research Joint Ventures (RJVs).

4

gain to the leader if he wins, and when both firms choose to shorten the invention race (See also Von Hippel (1987), Haroff et al. (2002)). Second, De Fraja shows that the level of R&D efforts experiences a similar outcome: a high required R&D intensity tends to induce disclosure in order to shorten the length of the invention race. A higher disclosure rate is thus associated with a higher R&D intensity. De Fraja’s results are consistent with the conclusions from the litterature on R&D cooperation. These theoretical models yield the following hypothesis: H3

Information disclosure may be complementary to appropriation strategies rather

than a substitute. De Fraja’s framework, combined with oligopoly models with R&D cooperation2 between firms with differentiated product suggest an additional hypothesis: H4

Information disclosure is higher for R&D intensive firms.

B. Empirical evidence At the empirical level, a number of studies provide direct evidence of intentional knowledge sharing by firms and the means implemented to disclose information and knowledge. A number of different works suggest that the disclosure strategy is much more frequent and important than usually considered among scholars; that it occurs using a variety of spillover channels and that the phenomenon can be informally or formally organized. Even if the empirical evidence is more extensive than the theoretical literature, the empirical findings are consistent with the theoretical predictions. A first step to understanding the knowledge disclosure possibilities is to recognize that a market for knowledge exists. Arora et al. (2000) suggest an increased roll for market mediated technology transactions covering licensing and or contract R&D. The rise is sometimes presented as a general movement of codification of knowledge that facilitates knowledge transaction and absorption at lower costs. The emerging information and communication technologies make it easier to codify, transport and trade knowledge (Antonelli (1992)). Voluntary knowledge disclosure transactions are made possible thanks to the implementation of property rights. Market transactions include labor mobility as an important source of knowledge transmission: in this setting, a firm may refuse to bid up an employee’s wage in order to retain her. This result is found at the local level (See Saxenian (1994)) even if mobility may also found to be associated with knowledge flows regardless of geographic proximity (Rosenkopf & Almeida (2001)). Even if a departing employee is required to sign a nondisclosure statement that

2

R&D intense firms are firms with a high R&D level compared to their turnover. The relation of R&D intensity with the R&D disclosure rate is however not simple in R&D cooperative models. In D’Aspremont & Jacquemin (1989) framework, it is possible to show that R&D intensity is higher for a cooperative R&D firm with complete knowledge sharing compared to a competitive firm with a lower R&D externality rate.

5

prevents her from revealing company secrets, the possible cross-fertilization through mobility seems to weaken its enforcement (Saxenian (1994); Gilson (1999)). At the same time, many case studies document that knowledge is revealed outside the market in a more informal way. Allen (1983) provided historical evidence, based on the British blast furnace industry in the nineteenth century, that innovative firms release of technological information and performance to their competitors. From the mini-mill firms of the US steel industry, Von Hippel (1987) and Schrader (1991) show that many employees frequently give technical information to colleagues from competitors (and from other industries as well) through informal networks.3 Contrasting Silicon Vallye, California with Route 128 in Massachusetts, Saxenian (1994) shows that such disclosures by Silicon Valley engineers can even become social norms. Powell et al. (1996) reminds us that informal information and knowledge trading often occurs during the course of a formal knowledge exchange agreement (See also Arora (1995) on licenses). The last references underline the importance of spontaneous and even involuntary face-to-face communications. Lynn et al. (1993) or Appleyard (1996) failed to find a statistically significant difference in knowledge-sharing using personal contacts between U.S. and Japanese engineers. Several recent papers provide evidence that the dissemination of information and knowledge by firms is also made through publications: corporate researchers published about 8 percent of U.S. scientific papers in 1994 (Hicks (1995). Henderson & Cockburn (1996)) underline that researchers in the pharmaceutical industry are often aware of their rival’s research thanks to a high rate of open scientific publications. Recent studies point to the striking number of disclosures among competitors. Freely revealed information between users is also common especially when network effects occur (See Harhoff et al. (2000), for the case of IBM or Brand’s interview (2001), on the DoD and internet)). Furthermore, knowledge transfers and communications are often encouraged among affiliates within a group. The knowledge sharing is expected to be especialy high when a specialised R&D affiliate takes the place of internal R&D department. Knowledge sharing may be different for affiliates of foreign multinational enterprises. Foreign firms may encounter difficulties in cooperating or entering local networks in host countries. To overcome such difficulties, knowledge disclosure may be an efficient strategy (See Blomström & Kokko (1998)). Taken together, the arguments leads to the following hypotheses : H5

Information disclosure is higher for affiliates who are encouraged to share

knowledge inside their group; H6

Information disclosure is higher for foreign affiliates relative to host country

affiliates.

3

See also Harhoff et al. (2002) for examples.

6

Many studies suggest that disclosure is often embedded in a more formal framework such as official R&D cooperative agreements. Following the theoretical literature, several econometric studies give clues to the existence of endogenous spillovers in bilateral cooperation agreements or larger R&D consortia (See Irwin & Klenow (1996); Siebert (1996)). Endogenous spillovers are however rarely directly measured. The existence of endogenous leakage is evoked as a means to interpret in the different empirical results. Branstetter & Sakakibara (1998) take a more direct approach and consider than “the degree of R&D spillovers realized within a consortium may be affected by the organizational structure of the consortium” (p 4). Using a sample of 226 Japanese firms, observed from 1983 to 1989, the authors find that firms participating in RJVs have higher R&D investment and productivity, related to an increase of potential endogenous spillovers. The potential spillovers are defined as a positive function of the technological proximity of participants or the degree of basic technological orientation of the research. The paper makes an interesting distinction between knowledge sharing at the input level due to the characteristics of the knowledge production process from that at the output level of the innovation process. The paper concentrates on the first category to explain the potential knowledge sharing in Japanese consortia, but it downplays the subsequent outcome of the phenomenon that firms may also decide to disclose information and knowledge concerning their innovation output. Cassiman and Veugelers (2002) distinguish incoming spillovers and outgoing spillovers. Using a sample of 411 Belgian firms, the authors find that firms with higher incoming spillovers are more likely to be actively engaged in cooperative R&D agreements. The author's results suggest that the level of knowledge leakage and absorption is not exogenous to the firm, which manages their absorption and appropriation capabilities during the innovation process. In this framework, they find however that firms which are more effective in appropriating the results from their innovation process are also more likely to cooperate in R&D. The result seems at odds with theoretical models where information sharing and R&D cooperation are linked. More generally, consortia are likely to organize the access to information and knowledge in ways that make it acceptable to the members. Three different formal collective structures can be found in the literature: patent pools (Merges (1999)), private official and unofficial standardization consortia (e.g., ATVEF on television, IETF on Internet), and research consortia (such as SEMATECH or MCC). Considering different types of structures, Rosenkopf and Metiu (2001) suggest that the kind of technical committee activity facilitates the entry of less-established firms into alliance networks where knowledge sharing occurs. Intermediation can also be implemented by independent firms as joint venture capitalist firms (See Aoki (1999)) or contract research units (See Arora & Merges (2001)).

Case studies also show that the transmitter is far from naive. Aoki (1999) asserts that firms in Silicon Valley do not disclose specific information crucial for their own competitive niches but instead only

7

information and knowledge relevant to diminishing the turbulence of the evolving technological frame. Using data on industry-financed research, Cohen et al. (2000) find that organized leakage by firms through academic publications is, in more than half the cases, less complete or slower than in the usual academic setting. In addition, several case studies mention the refusal to allocate the best researchers to RJVs (e.g. Bhattacharya et al. (1992)). Opportunism seems however not to be a compulsory trait of information and knowledge sharing. Innovation through an informal, entirely open and decentralized epistemic community such as Linux is a well-documented workable model (See Von Hippel (2002)). Three aspects remain quite hazy in the literature surveyed. First, there is little evidence of the effect of size on the disclosure phenomenon. If the disclosure process can be seen as negotiated by two monopolists, one could conclude that large firms have stronger bargaining power than their small counterparts and are thus more likely to receive than give. However, one can argue that large firms are more likely to protect their monopoly rents (through credible threats or higher litigation capacities as in Lerner (1995) and that the leakage rate is thus positively influenced by the size of the firms. Another argument would be that large firms are more likely to induce standardization effects than small ones and are thus more likely to freely disclose knowledge to their users. These arguments leads to the following hypothesis: H7

Information disclosure is higher for large firms which are more able to manage

their environment and to enforce networking effects. Second, the degree of novelty of knowledge is not directly interpreted as a determinant of knowledge disclosure. Even if the phenomenon is documented in low-tech industries (as in Von Hippel (1987) or even Merges (1999)), most of the case studies deal with high tech sectors, such as semi-conductors, telecommunication or pharmaceuticals (See Saxenian (1994); Aoki (1999); Cockburn & Henderson (1996)). Many studies are then largely focused on manufacturing industries with a rapid pace of technological change, suggesting that industries with a slow pace of technological change are less likely to share knowledge, yet some counter arguments can be found (See Appleyard (1996)). This observation underpins the following hypothesis: H8

Information disclosure is higher for High Tech sectors, especially in ICT

industries where network effects and or technical comity exist. Finally few general results are available on the effects of knowledge disclosure on firm performance. Some work based on specific industries suggest thats informal trading knowledge networks have a positive effect on firm performance (Schrader (1991)); that a provision such as the publication of academic papers is an efficient long-term incentive for researchers (Cockburn et al. (1999)); that RJV firms participating in RJVs have higher R&D productivity (Branstetter & Sakakibara (1998)).

8

III. Knowledge disclosure: further building blocks Unlike much of the past literature, this study underscores the potential importance of knowledge leakage that goes beyond formal intellectual property rights. Few theoretical and even fewer empirical studies provide rationale for disclosure behavior broadly defined or deal with measurement issues. In order to develop a coherent and workable framework to analyze and explain knowledge sharing, three additional building blocks must be first put in place: the place of voluntary knowledge disclosure in a theory of endogenous spillovers, the incentives to disclose, and the issue of measuring the disclosures.

A. Integrating disclosure strategy While establishing the endogenous nature of absorption strategies, the economics literature has tended to assume that leakage is "unintentional" (Mishan (1971)) or relies on the nature of technology. In this view, a firm can control and endogenize the level of leaking knowledge only through its R&D investment decision. A first step in an analysis of endogenous outgoing spillovers is to consider that firms make costly investments in order to limit the level or the usefulness of leaking knowledge. The different appropriation strategies are now well documented (Levin et al. (1985); Cohen et al. (2001)). A complementary view is to assume that the share of basic research in R&D is a means to control outgoing knowledge (Vonortas (1994), Yin & Zuscovitch (1994)). The addition of appropriation strategies to absorptive strategies give a more complete and factual view of endogenous externalities (Cassiman & Veugelers (2002); Gersbach & Schmutzler (2003)). Purposeful knowledge sharing remains however outside the scope of these papers giving only a partial view of the leakage assoicated with knowledge externalities. Two important features of an endogenous theory of knowledge flows are missing: First the leakage rate should be divided into an appropriation strategy and a disclosure strategy that both require different costly investments. Second, the link between the two strategies needs to be clarified. On one side, since Arrow (1962), the two strategies are considered to be antagonistic (Cassiman & Veugelers (2001)). On the other side, empirical investigations of the electronics and pharmaceutical industries suggest that informal knowledge dissemination occurs even when firms follow an aggressive patent application strategy and/or are competitors (Appleyard (1996), Hall and Ziedonis (2001), Henderson and Cockburn (1996)). Usually assumed as at odds with the appropriation mechanisms, the case studies on patent pools and research or standardization consortia suggest also that disclosed knowledge can be a complementary and not a strategic substitute with patent strategies. It also suggests that the kind of knowledge freely disseminated by firms is not the same as knowledge that is protected or sold (See Von Hippel (1987), Merges (1999), Aoki (1999)). Formally, the technological externality rate χ ij between two firms i and j can be defined as the product of a leakage rate λi of the transmitter i and an absorptive capacity α j of the receptors j. Often the

9

externality rate between firms is considered exogenous. In this case, a given externality rate is consistent with an infinite combination of disclosure and absorptive rates (See the iso-externality curve for χ ij = 0.1 in Figure 1). The understanding of a given externality rate in an economy thus relies on the identification of the disclosure capacity as well as the absorptive capacity. [Figure 1 about here] Past research has emphasised this dual requirement both at a theoretical and empirical level. With respect to incoming spillovers, the endogeneity of χ ij often relies on the endogeneity of α (e.g. Cohen and Levinthal (1989); Adams (2000)). Figure 1 shows that χ ij rises with α ( α1 < α 2 < α 3 ) for a given exogenous leakage rate λ 4. α depends on internal absorptive capacities. We propose in this paper to focus on the characteristics and endogeneity of the λ parameter considering that the absorptive capacity of firms is independent of disclosure strategies. Following Merges (1999) or Aoki (1999), we consider that knowledge that is intended for retention (RK) may differ from knowledge that is meant for disclosure (DK). Thus the knowledge potentially useful to receptor firm j from firm i is a weighted sum of the two public knowledges:

K j = [( 1 − ρi ) RK i + δi DK i ] α j , i ≠ j , where ρ i , δ i are respectively the rate of retention and disclosure of firm i, and α j the rate of absorption of firm j.5.This paper attempts to clarify the determinants of the disclosure coefficient and its link with the appropriation coefficient considering that the organisational practices dedicated to disclosure and retention activities are at the core of the two parameters. For tractability, in order to test empirically the hypotheses from the large body of theoretical and empirical literature, it is necessary to neglect this important aspect of appropriation.

B. The impact of knowledge sharing on performance The decision to disclose knowledge or not depends on whether the expected benefits from the leakage outweigh the costs resulting from disclosure. The costs may include the decline of market power attributable to the knowledge plus the transaction costs of disclosure (See Appleyard (1996)). The latter costs can be broken down into the codification costs associated with the diffusion of knowledge and the communication costs. Of course, like the retention strategy, the disclosure activity is imperfect. The degree to which the disclosure capacity is really the choice of the firm depends also on the location and

4

In this case, the disclosure rate becomes the limiting component of χ.

5 All the three parameters can be considered as bounded between 0 and 1. A ceiling value other than unity can be introduced to include the increasing difficulties of the three different strategies.

10

nature of the technology. Tacit knowledge can therefore be embodied at the individual level as tacit human skills or may be at the organizational level, embedded in collective routines (Winter (1987)). The benefits expected by a "leaking" firm span at least three categories. First, a firm with a lengthy invention race may disclose knowledge to shorten the development process (De Fraja (1993)) especially when there is a small competitive advantage to be the leader. In this context, there is a direct effect on costs and a possible indirect effect induced by the competitive firm which also shares its knowledge. The case studies also suggest that sharing knowledge is done to secure the innovation process. Thus, even without any formal R&D agreement, sharing knowledge can be considered as a wide long-run precompetitive cooperation among firms. In this context, Haroff et al. (2002) suggest that the innovator may create or discharge through knowledge sharing “generalized reciprocity”obligations combined with reputation effects. Second, firms may, benefit from intentionally leaking the knowledge through network effects (with or without externalities) resulting from the pervasive adoption of technology sometimes influenced by standards (Katz & Shapiro (1994)). Third, following Katz (1986), many models have also shown that R&D partners can improve their profits if they maximize their (symmetric) disclosure rate. The disclosure strategy has therefore several possible impacts in terms of firm performance. First, sharing knowledge can lower innovation costs, with or without formal RJV agreements. Second, free disclosure of information and knowledge with reciprocity may increase the quality of innovation process or product. Third, it may protect the existence of a market for innovation in a oligopoly market where a monopoly position by others would be damaging. Finally, knowledge disclosure may widen the market for new products or processes. This rough summary, combined with theoretical results and empirical evidence, suggests three additional hypotheses: H9

Innovation productivity is higher for firms sharing knowledge.

The assumption is thus not restricted to the R&D cooperative framework. However, the oligopoly models on R&D cooperation yield to the two following hypotheses: H 10

Innovation productivity is higher for firms sharing knowledge through R&D

partnership(s), because of internalized spillovers. H 11

Innovation productivity is higher for firms belonging to a group where the

R&D costs and knowledge are more likely to be spread among the affiliates acting as a natural consortium. Here innovation productivity can be measured by the share of innovation sales or, if available, the innovation sales divided by innovation costs.

C. Measuring knowledge disclosure 11

Several surveys deal with the disclosure of knowledge by firms. However only a few aspects of the disclosure process are covered by questionnaires prior to those considered here. For example, the R&D survey harmonization resulting to the Frascati manual allows us to obtain the level of annual external R&D expenditures and receipts and thus to identify the level of R&D market transactions. One difficulty with these data is in determining the R&D financial flows earmarked for R&D subcontracts versus R&D cooperation. RJV activities are also hard to identify since, even if the R&D firm is surveyed, their receipts from partners are not well defined. Furthermore, financial flows may not accurately reflect the knowledge flows for groups with varied transfer pricing. Finally, R&D receipts are hardly precise when the kinds of partners are considered. The data may distinguish between private integrated firms, public labs and specialized R&D firms, but rarely are there identifiers that indicate whether the organizations are competitors or customers or suppliers.. A more general shortcoming is that the quantitative R&D data do not capture the different kinds of knowledge flows occurring as a consequence of an R&D contract or an informal cooperative relationship. Examination of patents is also useful to identify the level of information released. The value of knowledge codified in a patent is hard to measure even if patent citations can improve the measure of the quality of knowledge contained in the patent. Unfortunately, many patent systems do not include meaningful information on quality, so it is difficult to induct a systematic assessment of the quality of individual patents and the degree of access to the knowledge. From a methodological point of view, the measurement of voluntary disclosure requires great care. The concept of leakage mixes voluntary and involuntary disclosure strategies. The different agents (the firm, the group, and the employee) involved in disclosure need to be taken in to account. The distinction between formal and informal instances of disclosure cannot be made solely based on channel of disclosure (lunch, phone call…). A challenge arises when attempting to classify exchanges because multiple disclosure categories may apply to single exchange (e.g. a paper published by a member of a research consortium). The prior research surveyed here deals with several aspects of the disclosure phenomenon (market, consortia, informal transfer) but hardly exhausts the possible relationships between two parties engaged in knowledge sharing. In Figure 2, we depict the possible relationships and disclosure decisions (See also Appleyard (1996)). [Figure 2 about here] A horizontal reading makes us aware of the number of useful distinctions that need to be drawn between instances of knowledge disclosure. The modes at the right side of the figure are numerous and similar to each other. Thus, even if the bulk of disclosed knowledge is covered by the different sources or surveys, an indicator across the different channels may reflect imperfectly the knowledge sharing strategies of a firm. An indicator can also overlap another one and measure the same amount of knowledge sharing.

12

Taking note of these possible shortcomings, this paper seeks to test the different hypotheses outlined above. The following section presents the variables collected as well as the econometric strategies implemented to make operational the eleven research hypotheses. An additional section presents the results of the different estimations.

IV. Variables and econometric models A. The explained (dependent) variables The data set in this paper is constructed from five separate surveys: the first and second French community innovation survey (CIS1 and CIS2) survey, the first appropriation survey (PACE), the survey on innovation competencies (COMPET), and the annual R&D survey. The CIS1, PACE, and COMPET surveys deal with knowledge acquisition and knowledge disclosure. We focus in our study on the last aspect of endogenous externalities that is rarely considered in the empirical literature (see however Cassiman & Veugelers (2001)). These three surveys are combined with the annual mandatory R&D survey (R&D survey) as well as the non-innovative firms from the second CIS sample.6 Of the surveys used to collect the data in this study, the first CIS survey is particularly well adapted to pin down the means of knowledge transmission. A question is asked concerning the different means of transferring knowledge: “Did your enterprise transfer to a third party (including a firm within the enterprise group) new technologies by the following means? 1) R&D carried out for a third party ; 2) the rights to use a third-party invention (among them patent or license to sell) (LICEN) ; 3) consulting for other firms (CONSU) ; 4) equipment sales (EQUIPT) ; 5) the departure of qualified employees (LEAV) ; 6) communication with other firms (COMM) ; 7) the sale of a part of the firm (SELLING) ; and 8) strategic joint venture alliances.” The survey then asks the identification of the type(s) of receptor: within country, within the EU, in the USA, Japan, or Other). As mentioned, the listing of the different means of transfer proposed shows a bias toward market-mediated trade of knowledge and informal or temporary transfer is not included as an option. An alternative way characterize knowledge transmission is to obtain information on the knowledge management practices at the firm level. Here two different indicators are available. An initial step is to focus on the organizational rules surrounding information disclosure: “Are your engineers allowed to diffuse non-confidential aspects of your technologies (publishing articles, conferences)?” asked in the 1994 PACE questionnaire, or “Do you monitor communications on your strategic knowledge?” asked in the 1997 survey on Competencies, offer two examples contained in this research. The last question may

6

See the appendix on data sources for more information on the various surveys used in this paper.

13

be broader than in the PACE formulation since “strategic knowledge” can include non-confidential as well as confidential information or knowledge. Using the CIS1, PACE and Competencies surveys, four proxies of knowledge disclosure are computed and used as variables to be explained. From CIS1, we compute a dichotomous variable LEAK that is one when at least one means of transfer among the proposed seven items is used during the period.7 Only 26% of innovative firms8 note such a technology transfer. As mentioned, the indicators are biased toward market-mediated disclosure. To investigate the degree of bias, we also included a simpler measure of disclosure based on COMM ( “Communication with other firms”), which is used by 12% of the innovative firms from CIS1. [Table 2 about here] The 1994 PACE survey’s questions on diffusion are used directly to build a dichotomous variable called AUTHO, equal to one when engineers are allowed to diffuse non-confidential aspects of technologies, otherwise zero. Of 983 innovative firms from PACE with more than 49 employees forty-four percent allow such communication (weighted using population weights). Similarly, the 1997 Competencies survey is used to compute an indicator PERMI (standing for “permission”), equal to one when an innovative firm allows their employee to communicate strategic knowledge to outside parties. Despite the bias associated with the question, the results are very close to the results for the PACE survey: 45% of the innovative firms give such permission to their employees. The share of product innovation turnover (ITi) in total sales is also available9. Differences exist across the two different CIS questionnaires in terms of how innovation turnover is measured. In CIS1, the part of product innovative sales of the total 1992 sales (ITiCIS1 thereafter) is classified and ordered into four classes ( 0 to 10%, 10 to 30%, 30 to 70%, 70% and above). In CIS2, the level of the 1996 innovative turnover is available (to get ITiCIS2 the figure is divided by the 1996 total sales). The PACE and CIS2 surveys contained a dichotomous variable PATENT. The indicator is one, when a patent application was

7

The different items are those listed above. LEAK = Max (LICEN, CONSU, EQUIPT, LEAV, COMM, SELLING). Thus the first alternative (“R&D carried out for a third party”) is not included because information disclosure is supposed to be included in the transaction. 8

An innovation, as defined in the CIS and Competencies surveys, is a new or significantly improved product (good or service) introduced to the market or the introduction within your enterprise of a new or significantly improved process. The innovation is based on the results of new technological developments, new combinations of existing technology or utilization of other knowledge acquired by your enterprise. The R&D manager, the general manager, or the firm director is supposed to answer to the question. 9 In CIS2, innovative turnover is computed from the answers to the following question: “Please estimate, in percentage, how your turnover in 1996 was distributed between new or significantly improved products (goods or services) introduced during the period 1994–1996 and unchanged or only marginally modified products (goods or services) during the period 1994–1996.” Innovative turnover is the sum of turnover attreibuted to new or significantly improved products. The definition was similar in CIS1.

14

submitted on the period (null otherwise) which is the case in 52% in PACE and 49% in CIS2 (or 38% for all sizes as shown in Table 2).

B. The independent variables Four sets of independent variables are included as explanatory variables: size, industry, group membership, and R&D intensity. In the CIS1 and Competencies survey, firm size is a categorical variable based on the number of employees. Size 50-99 is taken as the reference group since the PACE survey concerns only firms with more than 49 employees. Thus, when the dependent variable is AUTHO, only five size dummies are introduced. As size increases, the coefficients of these dummy variables are expected to increase, with more disclosure for larger firms. At the same time, the relative effort of R&D, measured by the average R&D intensity (the logarithm of the ratio of R&D to sales, measured between 1988 and 1992 for CIS1 and PACE and between 1992 and 1996 for Competencies), is also expected to be positive and significantly correlated to the probability of disclosure. However, size is expected to be independent of the share of innovative sales as typically stated in the literature on the Schumpeterian conjecture on size (See Cohen & Klepper (1996)). Within-industry determinants of knowledge disclosure are controlled for by fixed effects defined at the 2-digit level of the NACE. Taking the Clothing, leather, and footware industry as a reference, a set of 13 dummies denoted INDm (m = 1,..,13) is included. Following Geroski’s (1998) argument, such sector dummies capture differences in technological opportunities facing firms located in different sectors. We expect that greater technological opportunities are linked to technological uncertainty as well as to network effects. The expected sign is thus expected to be positive and the impact on disclosure higher in the high tech sectors.10 The industry dummies are also expected to greatly improve the explantory power of the regression. The group dimension is available only for the Competencies questionnaires through CIS2. FRGROUP and FORGROUP are dummies for belonging to French and foreign groups respectively. In our sample, 49% of innovative firms belong to a group whereas 21% belong to a foreign group. The expected signs of FRGROUP and FORGROUP dummies are positive and the two parameters are expected to be jointly significant. Following the H6 hypothesis, we also expect that the FORGROUP parameter is positive and higher than the FRGROUP parameter when explaining disclosure behavior. In terms of innovative productivity, the two parameters are expected to be positively linked with IT (H11). However, the relative ranking of their values is unclear in this case.

10

The high-tech, low-tech dichotomy is here is based on the OECD classifications which take the R&D-intensity of different industries as a measure of their knowledge-intensity, strategic importance and future potential. Pharmaceuticals, electric & electronic supplies, electric & electronic components, and to a lesser extent aeronautics, train, ships and motor vehicles industries are thus considered to be high tech industries.

15

In order to test hypothesis 3, we assume that the determinants of patenting (PATENT) are the same as those for PERMI. Using size, R&D intensity and industry dummies as explanatory variables, the model is very similar to the usual specification of a patent propensity model (e.g. Brouwer & Kleinknecht (1998)). The expected signs of the second (patent) equation are predicted to be similar to the first despite previous contradictory results on patent propensity (e.g. on the effect of size, see Duguet & Kabla (1997), Brouwer & Kleinknecht (1998)). Furthermore, the different types of R&D partners are available through CIS1 and CIS2. 69% of the innovative firms claim to be engage in R&D partnerships (COOP = 1, null otherwise) , and the ratio falls to 39% in the CIS2 survey. We aggregated the different types of R&D partners to three indicators: VERTICAL, which is one when an innovative firm declares at least one partner among its customers, material or equipment suppliers (null otherwise); GRPE, which is one when at least one R&D link is used with the group laboratory of another sister firm (null otherwise); and lastly, PUBORG., which is one when partners such as universities, public laboratories or technical centers are encountered (null otherwise). Dummies for partner firms designated as consulting partners (CONSULT=1, null otherwise) as well as competitors (HORIZONTAL=1, null otherwise) are included without any modification. The inclusion of the R&D partner variables is possible only for the estimates based on CIS1 and CIS2 (merged with Competencies) samples, that is for the LEAK, COMM and PERMI endogenous variables. As stated in H2, a hierarchy of the R&D partners dummies is expected: the Vertical coefficient is expected to be significantly higher than the PUBORG and HORIZONTAL coefficients. Overall, engaging in R&D partnerships is expected to be positive for disclosure (hypothesis H1). We can summarize the multiple variables and the expected signs of parameters in two simple equations. The first one explains the disclosure strategies of firms: l =5

k =8

DISCi = ∑ τ l RDPARTli + γ 1 INTENS i + γ 2 NORDi + ∑ γ k SIZE ki l =1 > 0

>0

0

>0

k =3 > 0

m = 21

∑γ m =9

m

(1)

INDmi + γ 22 ONE + µ i

In this case, the various endogenous variables DISC are LEAK, COMM, AUTHOR or PERMI. Moreover, R&D partners are not available for the PACE questionnaire, and FRGROUP and FORGROUP are included only for the PERMI variable. Different hypotheses (H1, H2, H4, H5, H6, H7 and H8) can be tested using these specifications. The second equation is an innovation production function, implemented to test the H9 to H11 hypothesis:

16

l =5

k =8

ITi = δ DISCi + ∑ τ l ( RDPARTli × DISCi ) + γ 1 INTENS i + γ 2 NORDi + ∑ γ k SIZE ki >0

>0

l =1 > 0

+ ϕ 1 FRGROUPi + ϕ 2 FORGROUPi + >0

>0

0, DISCi = 0 if DISC*i ≤ 0, and DISCi represents LEAKi or COMMi or AUTHOi or PERMIi. The four equations are estimated using maximum likelihood estimation. As a first step, model (1), which includes the R&D partner variables is estimated without the cross effects, which are then entered one by one. Based on the theoretical models of RJVs, a positive cross effect between R&D intensity and VERTICAL or PUBORG is expected while the cross effect with HORIZONTAL is expected to be smaller. Thus five separate Wald tests, for the five different types of R&D partners, are conducted to test the nullity of the interaction term, along with a Wald test of the joint hypotheses that all the interaction terms are simultaneously non different from zero. In order to test the H3 hypothesis, we estimate the first equation, together with an equation explaining the probability of filing at least one patent (available in CIS2, PATENT=1, null otherwise). Considering the observed variables as indicators of latent variables measuring the IPR appropriation strategy, we use a simple bivariate probit model in which the first equation is the same as the previous: PERMI*i = γ1z1i + ξi, where PERMIi = 1 if PERMI*i > 0, and PERMIi = 0 if PERMI*i ≤ 0. For patenting there is a simultaneous second equation:

17

PATENT*i = γ2z2i + ζi, where PATENTi = 1 if PATENT*i > 0, PATENTi = 0 if PATENT*i ≤ 0.11 One way to test the interdependent hypothesis is by assuming that the error terms of both decisions are correlated in the form Cov[ξi, ζi] = ρ (See Gourieroux (2000)). In this case, if ρ>0, the complementarity between the endogenous variables is found as expected. Independence is rejected if the ρ variable is significantly different from 0. A third case may be that IPR and disclosure are substitutes (ρ 0, and INNO

i

= 0 if INNO*i ≤ 0 and INNOiCIS. is the observable

declaration in CIS1 and CIS2 of the latent variable INNO*iCIS.. We use similar selection equations for the two CIS samples. The group dimension is however available only for CIS2. Thus the selection equation can be summarized as the following: 14

INNOi* = α 1 SIZEi + ϕ1 FRGROUPi + ϕ 2 FORGROUPi + ∑ α k INDki + α 15 ONE + ei >0

>0

>0

k =2

We expect the likelihood of being an innovator to be a function of the conditions of technological opportunity: it is easier to innovate in high tech sectors than in others. We also introduce a size effect (SIZE) with a positive expected sign (See Cohen & Klepper (1996)). Because the share of innovation sales IT is reported by indicator variables (brackets) in CIS1, in the first selection model, this variable is modeled by a latent variable IT*i where ITi is the observed counterpart of the variable only if INNOi = 1. Since the observed variable is an ordered variable, the first selection model is an ordered logit model where IT*i = β’X2i + ei, with 0% ≤ ITi* ≤ 100%. The four IT brackets are reduced to three due to the limited number of firms in the last bracket. Thus, ITi = 0 if ITi* ≤ 10%, = 1 if 10% ≤ ITi* ≤ 30%, = 2 if 30% ≤ ITi* ≤ 70% and = 3 if ITi* ≥ 70%.

11

A simultaneous equations model with two binary dependent variables was not considered in this first stage of our investigation. Even if independence between the two error terms is rejected, the introduction of one endogenous variable in the set of independent variables is hard to justify because the causality between PATENT and PERMI is not well defined in the literature. The positive effect of PATENT on PERMI may be due to the particular case of standardization consortia: a patent is often presented as a prerequisite to bargain and disclose information and knowledge.

18

A generalized Tobit model is applied in a second selection model used to explain the continuous endogenous variable ITiCIS2 (See Mohnen & Dagenais (2000)). The ordered Logit with sample selection as well as the generalized Tobit model are estimated by a two-step procedure whose last step uses the full information maximum likelihood method. Compared to Xli, X2i includes the R&D intensity, the different leakage proxies (LEAK, COMM) and the kind of R&D partners. The efficiency of an innovative firm is likely to be higher when disclosure is implemented. To test the H10 hypothesis, a Wald test is conducted to test the joined hypotheses that all the interaction terms between disclosure and R&D partnerships are simultaneously non different from zero.

V. Results The estimates of model (1), shown in Table 3, confirm that on average, the R&D intensity is positively and significantly linked to the probability of transferring knowledge or of allowing people to release it. The results are found for the four different measures of knowledge disclosure. The probability of disclosing knowledge is positively associated with the size of the firms. The estimated parameters suggest that nonlinearities occur: size dummies are found to be positive and significant only for firms with more than 250 employees (using the LEAK, COMM and AUTHOR variables). The effect of size seems weaker when PERMI is considered. [Table 3 about here] The coefficients of the sector dummies confirm that there are indeed substantial differences between sectors. Mainly high tech sectors are found to be more open to knowledge disclosure. Textiles, and wood and paper are not significantly more likely to disclose information and knowledge than the clothing sector. In contrast, pharmaceutical industries, electric and electronic equipment, ships and aeronautics as well as stone and glass are more systematically open than firms from the clothing industries. The results do confirm that firms inside high tech sectors (as electric and electronic supplies industry) are likely to release information or knowledge.12 The result for the pharmaceutical industry is much less pronounced than in Cockburn and Henderson’s work (1996). In terms of low tech industries, as shown by Von Hippel (1987) and Schrader (1991), a sector such as basic metal and metal products seems to engage in some “communication with other firms,” at least for the CIS1 sample. Compared to the available literature based on sector studies, the results highlight the importance of knowledge disclosure in car and aeronautical industries. Lastly, to belong to a group is significantly linked to the probability of releasing knowledge. A Wald test indicates that the two coefficients are different from

12

The French E3 set (electric and electronics) in the French NACE classification includes computers (CITI 300) and telecommunication products (CITI 322)

19

zero (Wald is 4.88), however no evidence exists for the hypothesis that French groups are less likely to disclose knowledge that foreign groups (the Wald test is 0.17). The same results on knowledge disclosure are found when it is considered jointly with the patent strategy. Table 4 presents the results of the bivariate probit model and confirms that patents, like knowledge disclosure, are used by large firms and those with high R&D intensity. Differences occur when the sectors are considered. Even if the firms from the high tech sectors declare both filing patents and releasing knowledge, the IPR strategy is not as significant as expected for the pharmaceutical industry when the patent variable is taken from CIS2. This result may be due to the group effect in pharmaceuticals where the patent is owned by the holding company and not by the affiliate. The negative sign of the Group dummies, even if they are not significant, may reflect such an effect. The core of the result is however provided by the correlation between residuals. Controlling for observables, the error terms in the PERMI and PATENT are positively related in the sample matching CIS2 and competencies surveys: ρ = 0.12 in Table 4 and is significant at the 1% level whereas the correlation is still positive in the PACE survey but not significantly different from 0. Even if the results do not provide a strong proof as expected for positive interdependence among the two strategies, the results, at least, suggest that they are not substitutes. This finding is at odds with many standard conceptions of appropriation where patents are one possible way to control knowledge and where knowledge retention is always considered as a first best strategy. The interpretation of the results would be that information or even knowledge disclosure is useful for risk-averse firms in the face of sunk costs. In some industries, patents would be used less as a way to protect the returns from innovation and more as a way to be able to coordinate with other firms on information and knowledge disclosure. Disclosure is possible because patents protect firms from imitation through a credible threat of infringement suits. The recent rise of patenting would be interpreted as an increase in the needs for coordination between firms corresponding to the growing disclosure of knowledge (See Cohen et al. (1998), Hall & Ziedonis (2001), for other arguments on the use of patents and Cassiman & Veugelers’ comments (2001)). [Table 4 about here] The estimates of model (1) with variables on R&D partnership participation appended are presented in Table 5. The results suggest that cooperation explains a large part of the probability to disclose knowledge even if non-cooperative firms can also release information. The two-step oligopoly models seem to correctly focus on the incentive to release information within a cooperative agreement. The tests also examine the oligopoly models which predict that firms are more likely to disclose knowledge when they implement vertical R&D links. Although the results show that positive parameters are found for all three dependent variables, a Wald test suggests that the vertical and horizontal parameters are not

20

significantly different. The main detailed predictions of the theoretic models are thus not corroborated here. A positive effect of collaborating within the group is found using the CIS1 sample, but the effect vanishes when the PERMI variable is included: the introduction of the variables for group membership explain this result. More interesting is the positive effect found of PUBORG on LEAK and PERMI.13 Thus here the results seem strong even if the magnitude of the (marginal) effect seems to be lower than the effect of the vertical or horizontal variables. While not included in the typical oligopoly modeling, the results appear suggestive of the literature on S&T networks where basic research or publications are the only means of gaining “admission” in high status networks (Pavitt (1991)). To expand further on this last point, we test the significance of the cross effects between R&D intensity and the different kinds of R&D partners. The higher the disclosure inside an R&D agreement, the greater the importance for firms that are more likely to increase their R&D investments. A cross effect was expected to be significant, particularly for vertical R&D agreements where the market between R&D partners is complementary. The tests are disappointing as just one among the possible 15 combinations is significantly different from 0 even if a significant positive effect is found for vertical R&D partnership on PERMI (See Table 6). [Tables 5, 6 about here] Lastly, the innovation production function is estimated. The idea defended by Von Hippel (1987) and Schrader (1991), as well as Cockburn et al. (1999), that free disclosure has a positive effect on innovative performance is affirmed for the manufacturing industries. Tables 7 and 8 show (first column) that the effect on innovation turnover (IT) depends on disclosure, whether measured by LEAK, COMM, or PERMI. Thus, if the capacity of innovation depends on the absorptive capacity of a firm, it also seems to depend on a disclosure capacity. The results are at odds with current statements on appropriation strategies at the private level where every leakage is seen as negative, even if such leakage is an efficient strategy within R&D cooperations. [Tables 7, 8 about here] We introduce a general cross effect with the variable COOP that is 1 if a R&D partnership is declared by the innovative firm. The effect is found to be insignificant for the LEAK and COMM variables but a significant positive parameter is found for the CIS2-Competencies sample. When the COOP variable broken down into the different types of cooperation, no significant cross effects are found for the CIS1 sample. For CIS2, the intersection of PERMI with vertical R&D cooperation is found to be positive and significant for the production of innovative sales. In this case, Table 7 suggests that an innovative firm

13

The COMM variable stands for “communication with other firms” and is thus not likely to be significantly correlated to the ORGPUB.

21

improves its performance through the implementation of vertical R&D relationships that boost knowledge release. An inverse effect is found when the R&D partner is a consulting firm. A joint Wald test of the five different cross effects introduced at the same time confirms that the cross effects do not improve the model (the Wald statistic is 3.9 and 4.0 in Table 7 and 9.1 in Table 8.

VI. Conclusion The understanding of disclosure capacity is as important as that of absorptive capacity in comprehending how and why knowledge is diffused. A strong absorptive capacity may be useless if there is no satisfying external knowledge to use. The analysis of the “take off” and subsequent “landing” of knowledge is necessary to understand, a century after Alfred Marshall initially posed the problem, why “the secrets of industries are in the air.” The results presented here are an initial econometric answer finding evidence of knowledge disclosure strategies on a large multi-sector sample of firms. They suggest that a firm with high R&D intensity, from a high tech sector, belonging to a group, and involved in R&D partnerships is more likely to implement such a practice. It is also found that strategies promoting knowledge disclosure are a knowledge management device for innovating firms that leads to greater innovation performance. Lastly, the knowledge disclosure strategy seems to be a knowledge management practice that is implemented at the same time as an IPR strategy such as patenting. Several effects compete here and are very difficult to identify without the benefit of detailed case studies. The measurement issue is an interesting topic for future research on the understanding of externalities. At the same time, the link between the disclosure and absorptive practices merit analysis. A particular area for future research is the link between knowledge disclosure and IPR strategies of firms. The nonnegative relation found between these two decisions needs to be confirmed and justified. A more precise look into industry sub-samples may give further clues to the results. Lastly, the simple question of “How does knowledge disclosure occur?” cannot be limited to market transactions or cooperative agreements without taking into account the other modes of transmittals. Such a simple question requires also further data collection.

22

VII.

References

Adams, J.D. (2000) Endogenous R&D Spillovers and Industrial Research Productivity. NBER, N° 7484, January. Allen, R.C. (1983) Collective Invention. Journal of Economic Behavior and Organization, 4, 1-24. Antonelli, C. (1992) The economic Theory of Information Networks. in Antonelli, C., editor, The economics of information network. North-Holland, 5-27. Aoki, M. (1999) Information and Governance in the Silicon Valley Model. MITI/RI Discussion Paper Series #99-DOF-31. Stanford University, July. Appleyard, M. (1996) How Does Knowledge Flow? Interfirm Patterns in the Semiconductor Industry. Strategic Management Journal, Volume 17: Winter, 137-154. Arora, A. (1995) Licensing Tacit Knowledge: Intellectual Property Rights and the Market for KnowHow. Economics of Innovation and New Technology 4: 41-59. Arora, A., Fosfuri, A. and Gambardella A. (2000) Markets for Technology and Their Implications for Corporate Strategy. Heinz School, Working Paper, N°00-02, January, 40p. Arora, A., Merges R.P. (2001) Property Rights, Firm Boundaries, and R&D Inputs. Working Paper. Arrow, J.K. (1962) Economic Welfare and the Allocation of Resources for Invention. in R.R. Nelson (ed.), The Rate and Direction of Inventive Activity, Princeton University Press, N.B.E.R.. Beath, J., Poyago-Theotoky, j and Ulph, D. (1998) Organization Design and Information-Sharing in a Research Joint Venture with Spillovers. Bulletin of Economic Research, 50, Issue 1, January, 47-59. Bhattacharya, S., Glazer, J., and Sappington, D.E.M. (1992) Licensing and the Sharing of Knowledge in Research Joint Ventures. Journal of Economic Theory. 56:43-69. Blomström, M., Kokko A. (1998) Multinational Corporations and Spillovers. Journal of Economic Surveys, 12, 3:, 247-277. Brand, S. (2001), Founding Father. Wired, March, 145-153. Branstetter, L.G., Sakakibara, M. (2000) When do research consortia work well and why? Evidence from Japanese Panel Data. NBER, Working Paper, N°7972, October. Brouwer, A., Kleinknecht, A. (1999) Innovative output, and a firm's propensity to patent. Research Policy, (28) 6, 615-624. Cassiman, B., Veugelers, R. (2002) R&D Cooperation and Spillovers: Some Empirical Evidence from Belgium. American Economic Review, Vol. 92, No. 4, September, 1169-1184.

23

Cockburn, I., Henderson R., and Stern S. (1999) Balancing Incentives: The Tension Between Basic and Applied Research. NBER Working Paper N° W6882, January. Cohen, W., Klepper, S. (1996) A reprise of Size and R&D. The Economic Journal, 106, July, 925-951. Cohen, W.M., Levinthal D.A. (1989) Innovation and Learning: The Two Faces of R&D. The Economic Journal, Vol. 99, n° 3, September, 569-596. Cohen, W.M., Nelson R.R., and Walsh, J.P. (2000) Protecting Their Intellectual Assets: Appropriability Conditions and Why U.S. Manufacturing Firms Patent (or Not). NBER Working Paper No. W7552, February. D'Aspremont, G., Bhattacharya, S. and Gérard-Varet, L.A. (1999) Bargaining and sharing knowledge. CORE Discussion Papers, 32p. David, P. A. & Foray D. (1995) Accessing and Expanding the Science and Technology Knowledge. STI Review, N° 16. De Bondt, R. (1996) Spillovers and innovative activities. International Journal of Industrial Organization, 15, 1-28. De Fraja, G. (1993) Strategic Spillovers in Patent Races. International Journal of Industrial Organization, 11, 139-146. Duguet, E., Kabla, I. (1997) Appropriation strategy and the use of the patent system: an econometric analysis at the firm level. CEME N°45. Geroski, P. (1998) An Applied Econometrician's View of Large Company Performance. CEPR, Working Paper, DP1862, April. Gersbach, H., Schmutzler A. (2003) Endogenous spillovers and incentives to innovate, Economic Theory, 21, 1, Springer Verlag, 59-79. Gilson, R. (1999) The Legal Infrastructure of High Technology Industrial Districts: Silicon Valley, Route 128, and Covenants Not to Compete, 74 New York University Law Review. 575. Gourieroux, C. (2000) Econometrics of Qualitative Dependent Variables. Cambridge University Press. Hall, B., Ziedonis, R.H. (2001) The Determinants of Patenting in the U.S. Semiconductor Industry, 1980-1994”, Rand Journal of Economics, 32: 101-128. Harhoff, D., Henkel, J., Von Hippel, E. (2002) Profiting from voluntary information spillovers: How users benefit by freely revealing their innovations. LMU, Working Paper. Henderson, R., & Cockburn, I. (1996) Scale, Scope, and Spillovers: Determinants of Research Productivity in the Pharmaceutical Industry. RAND Journal of Economics, Spring, 27(1), 32-59.

24

Hicks, D. (1995) Published Papers, Tacit Competencies and Corporate management of the Public/Private Character of Knowledge. Industrial and Corporate Change, 4, 401-424. Irwin, D. & Klenow, P. (1996) High-tech R&D subsidies. Estimating the effects of SEMATECH. Journal of International Economics, 40, 323-344. Kamien, M.I. & Zang, I. (2000) Meet me halfway: research joint ventures and absorptive capacity. International Journal of Industrial Organization, Volume 18, Issue 7, October, Pages 995-1012. Katsoulakos, Y., Ulph, D. (1998) Endogeneous spillovers and the performance of research joint ventures, The Journal of Industrial Economics 46, September, pp. 333-357. Katz, M.L. (1986) An Analysis of Cooperative Research and Development. Rand Journal of Economics, Vol. 17, n°4, Winter, 527-543. Katz, M.L., Shapiro, C. (1994) Systems Competition and Network Effects. Journal of Economic Perspectives, 8(2), 93-115. Lerner, J. (1995) Patenting in then shadow of competitors. Journal of Law and Economics, 38, 463-495. Levin R.C., Klevorick A.K., Nelson R.R., and Winter S.G. (1987) Appropriating the Returns from Industrial R&D. Brookings Papers on Economic Activity, n° 3, pp.783-820. Lichtman, D., Baker, S. and Kraus, K. (2001) Strategic Disclosure in the Patent System. Working paper, 40p. Lynn, H.L., Piehler H.R., Kieler M. (1993). Engineering Careers, Job Rotation, and Gatekeepers in Japan and the United States. Journal of Engineering and Technology Management. 10, pp. 53-72. Merges, R.P. (1999) Institutions for Intellectual Property Transactions: The Case of Patent Pools. August, Working paper. Mishan, E.J. (1971) The postwar literature on externalities: an interpretative essay. Journal of Economic Literature, 9, 1-28. Mohnen, P., Dagenais, M. (2000) Towards an Innovation Intensity Index: The Case of CIS 1 in Denmark and Ireland. Working Paper, CIRANO, N°20. Pavitt, K. (1991) What makes basic research economically useful? Research Policy (20)2, pp. 109-119. Poyago-Theotoky, J. (1995) Equilibrium and Optimal Sizes of a Research Joint Venture in an Oligopoly with Spillovers. Journal of Industrial Economics 43, 209–226. Powell, W.W., Kogut, K.W., Smith-Doerr, L. (1996) Interorganizational collaboration and the locus of innovation: Networks of learning in biotechnology. Administrative Science Quarterly, March, Volume: 41, Issue: 1, 116-141.

25

Rosenkopf, L. and Almeida, P. (2001) Overcoming Local Search through Alliances and Mobility. University of Pennsylvania and Georgetown University, Mimeo. Saxenian, A.L. (1994) Regional Advantage. Harvard University Press. Schrader, S. (1991) Informal technology transfer between firms: Cooperation through information trading. Research Policy, (20), 2, 153-170. Siebert, R. (1996) The impact of Research Joint Ventures on firm performance: an empirical assessment. WZB working paper, FS IV 96 – 3. Suzumura, K., Goto, A. (1997) Collaborative R&D and competition policy. in Waverman, L., Comanor, W.S., and Goto, A., editors. Competition Policy in the Global Economy. London and New York: Routledge. Von Hippel, E. (1987) Cooperation between rivals: informal know-how trading. Research Policy, 16, N°6, December, 291-30. Von Hippel, E. (2002) Horizontal innovation networks by and for users. MIT Sloan School of Management Working Paper No. 4366-02, June. Vonortas, N.S. (1994) Inter-firm cooperation with imperfectly appropriable research. International Journal of Industrial Organization, Volume 12, Issue 3, September, Pages 413-435. Winter, S.G. (1987) Knowledge and competence as strategic assets. in Teece, D.J. (ed.) The Competitive Challenge. pp. 159-183, Harper and Row. Yin, X., Zuscovitch, E. (1994) Some Economic Consequences of the Limitation of Technology Transferability. working paper, BETA, N°9404, Mars, 33 pages.

26

Appendix: Data Sources Table 1: Surveys on technological innovation at the firm level Name

Year

Period

Sampling Method

Method

R&D Survey

Each Year

19731999

1993

19901992

1993

19901992

1997

19941996

1997

19941996

Exhaustive All size, All sectors Sample >19 employees Manufacturing ind. Sample >49 employees Manufacturing ind. Sample >19 employees Manufacturing ind. Sample >19 employees Manufacturing ind.

Mail Mandatory Mail Non Mandatory Mail Non Mandatory Mail Non Mandatory Mail Non Mandatory

CIS1 First Community Innovation Survey PACE Appropriation Survey COMPETENCIESS urvey on Innovation Compétencies CIS2 Second Community Innovation Survey

Number of individuals

Resp. rate

3000-4000

79%

3917

83%

1797

N.A

4150

83%

3917

85%

COMPETENCIES ∩ CIS2: The matching of samples from the last two survey gives us a sample of more than 1538 innovative firms where organizational practices and R&D partners are observable. 1439 are technologically innovative. The sample is completed with the rest of non-innovative CIS2 firms to minimize the possible bias toward large firms. After weighting, the final sample of 3539 firms represents 20,566 French firms in the manufacturing industries.1 OTHER DATA SOURCES : Two other data sources are used in this paper. • The mandatory annual R&D survey between 1988 and 1996 from the Ministry of Education includes the annual budget of R&D of firms. • The 1996 “GROUP” database is managed by the SESSI and gives information on firm groups and their nationality.

1

Weights are computed here at the three-digit level of the NACE using seven strata for size (number of employees). 1

Table 2: Descriptive Statistics (weighted mean) Survey

PACE PACE CIS1

All firms/ Innovative firms Sample Size (not weighted)

CIS1

CIS2

ALL INNO ALL INNO ALL 1769 983 3648 1590 3539

Dummy for technological innovation Dummy for authorization to communicate (AUTHO) Dummy for technological transfer (LEAK) Dummy for communication (COMM) Dummy for “permission” to communicate (PERMI) Dummy for patent deposit (PATENT) Share of innovative turnover in total sales (ITCIS2) Dummy when 0≤IT CIS1