Absorptive Capacity and the Strength of Intellectual Property Rights

Transcription

1 Absorptive Capacity and the Strength of Intellectual Property Rights Kira R. Fabrizio Goizueta Business School, Emory University, Atlanta, Georgia, March 14, 2008 Abstract A policy debate continues about the relative merits and and negative consequences of federal policies that encourage patent protection to university research results. Given the importance of university research for innovation in many industries, the effects of such policies are of critical importance to firms. This paper examines the impact of the increase in university patenting on the effectiveness of firm strategies historically associated with enhanced exploitation of university research. Results from a simultaneous equation estimation using a panel data set of biopharmaceutical firms suggest that connectedness to university scientists provides substantially less benefit for innovative outcomes of firms as university patenting increases. This has implications for understanding the impact of the policy change on technology transfer and also for firm managers seeking to take advantage of university-generated research. JEL Classifications: O31, O32, O33, L65, D83 Keywords: absorptive capacity, innovation, intellectual property policy, pharmaceutical and biotechnology, university research 1

2 1 INTRODUCTION It is well understood that the intellectual property environment affects knowledge transfer, as well as the allocation of inventive activity and appropriability of gains from innovation (Arrow 1962, Schumpeter 1950, Teece 1986, Gans et al. 2002). This work argues convincingly that stronger appropriability conditions encourage investments in innovative activity, facilitate the market for technology transfer, and tend to benefit the inventor. In part because of these benefits of intellectual property protection, U.S. universities have been encouraged by federal policy to pursue patent protection for university research results. The most recent and most studied such policy is the Bayh-Dole Act of 1980, although this Act was the culmination of many years of formal and informal policies relating to patenting by universities (Eisenberg 1996). The Bayh-Dole Act granted universities to the right to apply for and own patent rights to inventions based on federally funded research. The Act also allowed and encouraged universities to exclusively license these patented inventions, and allowed universities to collect revenues in return for such licenses. The purpose of this policy was to encourage the commercial development of university-generated technologies for the broader purpose of supporting and sustaining economic growth. In the absence of formal property rights, the argument was that firms would not be willing to invest in early-stage technologies that required substantial development in order to commercialize, and potentially valuable technologies could be left undeveloped on the lab shelves (Mazzoleni and Nelson 1998, Eisenberg 1996). The patent protection and possibility of exclusive license would allow investing firms to appropriate the returns to such an investment. Proponents of the Act point to the growing number of university-generated patents as evidence that the Act is a success (Economist 2002). There is substantial evidence that university patenting was increasing even prior to the Bayh-Dole Act, partly as a result of previous policies and partly due to both strengthening intellectual property rights in general and an increase in biomedical activity at universities (Mowery and Sampat 2001). However, there is little doubt that university patenting has increased exponentially since It is not clear from this increase what effect university patenting has had on knowledge transfer from universities to industry. On one hand, the logic behind the Act appears sound. University-generated technologies are in fact often early stage (Jensen and Thursby 2001, Cohen et al. 2002, Mansfield 1991) and require substantial investments by licensing firms. However, the dominant knowledge transfer channels by which university research reaches industry are those that characterize an open science environment characteristics of traditional academic norms, such as publication, conferences, labor mobility, and informal interactions (Dasgupta and David 1994, Cohen et al. 2002). Recent work has addressed the concern that university research, previously openly and publicly available through publications and conferences, is becoming fenced off by intellectual prop- 1

3 erty protection, thereby limiting a critical input to industrial innovation. Surveys of researchers indicate such concerns are valid (Heller and Eisenberg 1998, Blumenthal et al. 1986, Campbell et al. 2002). However, survey evidence presented by Walsh et al. (2003) suggests that, at least in the biomedical field, patents on upstream research tools may be delaying but are not preventing downstream innovation. Further research is required to evaluate the impact of university patenting on knowledge transfer to industry (Mowery and Sampat 2005). In the absence of formal property rights, incentives and norms in the university system reward public disclosure of research results (Dasgupta and David 1994). Therefore, research results existing in the public domain, available as spillovers to others who wished to make use of the results. Existing literature has documented the importance of university-generated research to industrial innovation (Cohen et al. 2002, Mansfield 1991, 1998, Cockburn and Henderson 2001, Zucker et al. 2002). However, the use of even publicly available research is hardly free and easy. Effectively monitoring the research environment, appropriately appraising relevant research, and making effective use of this research depend on the capabilities of the potential recipient (Cohen and Levinthal 1989). In particular, the accumulated research knowledge of the recipient, similarity between this knowledge and the publicly available knowledge, and connections with the community of researchers generating knowledge aid absorption of knowledge from this spillover pool (Cohen and Levinthal 1989, 1990, Cockburn and Henderson 1998, Cockburn et al. 2000, Fabrizio 2006). This paper explores the effect of university patenting on the benefits derived from firms research activities that have historically provided advantage in accessing university research. If patenting in fact facilitates the exchange of technologies between universities and firms, the benefits of investing in research activities that enhance the ability to exploit university science would be expected to increase. However, if university patenting is restricting the availability of university research results or otherwise shrinking the potential pool from which firms can draw, then the benefits of investing in such research activities would be expected to decrease. I explore these competing predictions using data on patenting and research activities for a sample of biotechnology and pharmaceutical firms for the period. The empirical strategy employed uses the variation in university patent propensity within technology areas across time and the variation in research activities across firms and over time to identify the effect of a change in university patenting on the relationship between firm research activities and the search process for new inventions. Results of a single equation OLS model are compared to a generalized least equations estimation of a systems of equation that explicitly models the endogenous nature of investments in firm research strategies. Results suggest that the value of firms absorptive capacity-building research activities is in fact changing as universities increasingly patent their research results. Previous research has 2

4 found that the firms own basic science research and collaborations with university scientists are associated with a faster pace of knowledge exploitation for pharmaceutical and biotechnology firms (Fabrizio 2006). Results here indicate that the relationship between the firm s collaborations with university scientists and the pace of knowledge exploitation decrease as university patenting increases. The relationship between the internal basic research of the firm and the pace of knowledge exploitation is not significantly altered as university patenting increases. This research contributes to the on going debate about the role of intellectual property rights in the knowledge transfer process. From a policy perspective, it highlights the potential for negative consequences for knowledge transfer associated with strengthening intellectual property rights. For firms in a changing intellectual property rights environment, it suggests that the strategies that are effective for taking advantage of publicly available research knowledge may be less relevant as the research is increasingly subject to formal property rights protection. 2 Theory and Hypotheses This study draws from two distinct but related theoretical literatures. First, there is a considerable and growing literature describing and testing the importance of firm research activities in identifying, assimilating, and exploiting external knowledge. In the original seminal work in this area, Cohen and Levinthal (1989, 1990) coin the term absorptive capacity to describe the firm s ability to identify, assimilate, and exploit knowledge from outside the firm s boundary. The authors highlight the effect of the strength of intellectual property protection on the relationship between firm research activities and the knowledge base of the firm. Second, it is critical to consider what university research is contributing to the innovation process. Recent literature considering the role of scientific knowledge in the innovation process characterizes search for innovation as an uncertain process in which knowledge can provide a guide (Fleming 2001, Fleming and Sorenson 2004). This work demonstrates the importance of scientific knowledge in the search process for new innovations. Considering these literatures together generates the unambiguous prediction that firms investing in absorptive capacity-generating research activities will benefit more from university research, with some specific implications for the efficiency of search for innovation. However, the impact of the increase in university patenting on this relationship is not clear. The direction of the predicted effect turns on the expected impact of university patenting on the amount and availability of research results generated by universities. Accordingly, we begin by discussing the alternative expectations of the impact of university patenting. 3

5 2.1 University patenting and the availability of research results The Bayh-Dole Act was the product of a period of considerable concern in the United States about competitiveness and innovation in the the increasingly global marketplace. Prior to the Act, universities were granted patents, but the approval to do so was granted on a case by case basis and governed by a patchwork of policies. Exclusive licenses of patents granted for the results of federally funded research were not encouraged. Often, the government retained ownership of patents resulting from research supported by federal funds, and typically these were not exclusively licensed. The Bayh-Dole Act was an effort to formalize and homogenize patent policy regarding the results of federally funded research and, as a result, encourage industrial productivity and innovation (Eisenberg 1996). The two key aspects of the Bayh-Dole Act were that it replaced the inconsistent set of practices with a uniform policy applicable to all universities, and that exclusive licenses of these patent were encouraged, for which the universities were allowed to collect licensing revenues (Mowery and Sampat 2005). This was expected to facilitate commercialization of universitygenerated research results because it would attract the necessary commercial investment required develop the research into commercialization innovations. Because university result results are often early-stage, significant development is often necessary before commercialization is profitable, and firms would be hesitant to make such investments without an exclusive license to protect them from competition. It was argued that universities were in a better position to perform these technology transfer activities (than the government) because they had the knowledge and familiarity with the inventions required to identify potential licensees (Eisenberg 1996). In addition, providing universities with the right to collect licensing revenue provides an incentive for prompt and proactive disclosure and diffusion of the knowledge required to make use of the patented technology. In other words, an it created the incentive to advertise the patented technology rather than simply publish the results (Mazzoleni and Nelson 1998). 1 The Bayh-Dole Act passed overwhelmingly (Eisenberg 1996). Discussions leading up to its passage, however, did not include any discussion of the potential for negative consequences of the Act on other channels of knowledge transfer between universities and industry (Mowery and Sampat 2005). Given the relative importance of more informal knowledge transfer channels, such as publications, conferences, and personal meetings (Cohen et al. 2002), any potential negative consequences for these other means of knowledge transfer may have significant impact of industrial innovation. There is considerable evidence to suggest that patenting may lead university researchers to increase secrecy over non-patented research results, thereby restricting knowledge transfer via 1 See Eisenberg (1996) and Mazzoleni and Nelson (1998) for discussions regarding the merit of such arguments made in favor of university patent rights. 4

6 these other channels. Based on interviews and surveys, several researchers have found evidence of a reduction in sharing of research results associated with involvement in commercialization of academic research and an increased level of secrecy among faculty members with industrysource funding (Louis et al. 2001, Campbell et al. 2002, Blumenthal et al. 1996a, 1986). Adverse effects of this increased secrecy on academic research, such as delay of publications, abandoning promising research, and refusal or delay of sharing information with a group, are documented by Campbell et al. (2002). In that study, three quarters of the respondents felt that data withholding detracted from communication in their field and slowed the rate of progress in their field (in this case genetics). Companies that support academic research often require the academic institution to keep proprietary information confidential for longer than necessitated by the patent filing (Blumenthal et al. 1996b, Rahm 1995), and many academic researchers delay publications in order to keep information private, especially when sponsored by industry funding (Blumenthal et al. 1996a). However, there is also some evidence that suggests patenting is not associated with restricting knowledge transfer. Empirical studies of the relationship between faculty member patenting and the number of publications generated have not found any evidence of a decrease in publication activity associated with patenting (Azoulay et al. 2004, Lissoni et al. 2004, Fabrizio 2007). Further, survey evidence from both university and industry researchers indicates that researchers are able to gain sufficient access to the upstream research tools, even when they are patent protected, although negotiating access may delay their research (Walsh et al. 2003). Therefore, it is not clear whether university patenting facilitates, hinders, or does not affect the availability of university-generated research results to industrial researchers. The current debate includes support for all three possibilities (Boettiger and Bennett 2006). This work contributes to this debate by considering the effect of university patenting on the value of firm research activities that enhance firm innovative outcomes by allowing the firm to tap into university research knowledge. 2.2 Absorptive capacity and the IP environment The concept of absorptive capacity focuses attention on the fact that knowledge outside the boundaries of the firm is not freely and effortlessly absorbed by the firm, even if it is in the public domain. Instead, effort, expertise, and purposeful action on the part of firm researchers are required to identify, assimilate, and exploit this external knowledge (Cohen and Levinthal 1989). 2 2 Factors affecting the use of knowledge external to the firm in the innovation and development process have been discussed in the context of the monitoring role of R&D (Kline and Rosenberg 1986, Rosenberg 1990), the necessity of investment in absorptive capacity (Cohen and Levinthal 1990, Pavitt and Patel 1995), and the market for intellectual property (Arrow 1971, Williamson 1975, 1985, Teece 1981, 1989, 1996, Lamoreaux and Sokoloff 5

7 Firms with superior capabilities in this regard will be better able to identify, assimilate, and make use of the pool of relevant external knowledge. This relationship has been examined and confirmed in empirical studies of the marginal value of R&D investments (Cohen and Levinthal 1989) and the relationship between firm research activities and innovation outcomes (Cockburn and Henderson 1998, Fabrizio 2006). A central parameter in the foundational absorptive capacity literature is the strength of the appropriability environment. Cohen and Levinthal (1989) describe the effect of the strength of the appropriability conditions on the incentives for a firm to invest in generating knowledge asset. The strength of this appropriability condition cuts two ways; a strong appropriability environment provides more protection for the firm s own knowledge assets, and thus an incentive to invest, while also hindering access to other firms knowledge, dampening the incentives to invest in absorptive capacity-enhancing knowledge assets. With respect to the firm s own R&D, strong appropriability allows the firms to realize more of the value from their investment (because less leaks out to others), and therefore encourages R&D investment. However, with respect to potential absorption of other firm s R&D, strong appropriability decreases the effectiveness of the focal firm s absorptive capacity in capturing other firms R&D knowledge, and thereby decreases the value of the firm s own R&D in terms of absorptive capacity benefits. At the extreme of strong appropriability, excludability is protected, and knowledge does not spill from the knowledge source into the pool available to be absorbed into a firm. These opposing effects of appropriability conditions, along with hypothesized differences in the relative importance of the firm s own R&D to the ability to absorb external knowledge, generate testable predictions with regard to optimal investment in R&D, which (Cohen and Levinthal 1989) test and confirm. As Lane et al. (2006) note, most of the studies relating to absorptive capacity have either assumed away or ignored the role of the intellectual property environment. When the conceptual model employed does consider the intellectual property environment, as in Lane et al. (2006), it typically enters as a driver of incentives to develop absorptive capacity an antecedent in the model. This is only half of the story. The reason that the intellectual property environment affects the incentives to invest in research activities is precisely because the IP environment conditions the effectiveness of firm research investments, and the resulting absorptive capacity of the firm, in accessing (potentially) beneficial knowledge from outside the firm boundary. The IP environment therefore must intervene in the relationship between absorptive capacity and any measure of performance. This is completely consistent with the Cohen and Levinthal (1989) model of absorptive capacity; The absorptive capacity parameter that they propose describes the knowledge base of a firm as a function of the interaction of the firm s absorptive capacity and the intellectual 1999, Nelson 1959, 1982). 6

8 property environment applicable to the related firms from whom external knowledge may be gleaned. The mathematical statement of this relationship proposed by Cohen and Levinthal (1989) is as follows: z i = M i + γ i (θ j i M j + T ) where z i is the knowledge base of firm i and γ i represents the firms absorptive capacity with respect to external knowledge. The external knowledge pool is made up of other firm s related research, j i M j, conditioned by the strength of intellectual property protection, θ, and the pool of public research knowledge generated by universities and government labs, T. Importantly, γ i is described as a function of the firms own research investments as well as the characteristics of the external knowledge that make the firm s own expertise more or less important for absorption. Note that the firm s knowledge base (z i ) is a function of the interactions between absorptive capacity (γ i ) and the IP environment (θ). Although Cohen and Levinthal (1989) do not explicitly consider the possibility that the availability of the pool of public science (T ) could be affected by intellectual property restrictions, it is a simple extension of their model to do so. 3 As is well documented, formal intellectual property protection (i.e. patenting) of university research as grown exponentially since the U.S. Congress passed the Patent and Trademark Act Amendments (P.L ), known as the Bayh- Dole Act, in 1980 (Mowery and Ziedonis 2002, Henderson et al. 1998a). This Act facilitated patenting and licensing of research supported by federal funding, which made up approximately 70% of university research at the time (Henderson et al. 1998a). Congress explicitly regarded the transfer of knowledge from universities to the private sector as a desirable outcome of federally-funded research (Jaffe 2000), and passed this law in response to a perceived need for a reliable technology transfer mechanism to increase the commercialization of university-based technologies and to provide uniform rules for patenting and licensing of federally-funded research (The Council on Governmental Relations, 1993). With the addition of a spillover parameter on the public research term, analogous to the parameter conditioning spillovers from other firms research due to appropriability conditions, it is possible to make predictions regarding the knowledge base of the firm (holding other parameters fixed). 4 Borrowing notation from the original model, with the modification to include 3 The authors note that, for simplicity, they have assumed that all research outside of industry (i.e. university and government labs) is made public, so that the implicit spillover parameter on such research is equal to one (see footnote 7). 4 Of course, the model also implies that the optimal level of R&D investment by the firm changes with changes in such a parameter. I return to this in the empirical section. 7

9 the parameter κ, representing the appropriability conditions with respect to university research, the model is as follows: z i = M i + γ i (θ j i M j + κt ) Analogous to θ, κ ranges between 0 and 1. The debate about the effect of university patenting on the availability of university research results to industry can be framed as competing expectations about how κ changes with increasing university patenting. If one believes that university patenting is associated with more technology being available for transfer from university to industry, then κ increases toward 1 as university patenting increases. In this case, for a given level of firm absorptive capacity (γ), the level of knowledge available to the firm (z i ) increases as appropriability with regard to public research increases (κ moves toward 1). However, if one believes that the pool of available university research results is in fact being constricted by university patenting, then increasing university patenting is reflected in κ decreasing toward zero. When κ is zero, none of the university research is in the pool of potential spillovers available to firms via their investments in absorptive capacity. In this case, for a given level of firm absorptive capacity, the level of knowledge available to the firm decreases as appropriability with regard to public research increases (κ moves toward 0). Therefore, the model generates competing predictions, depending on the effect of university patenting on the availability of research knowledge to the firm via channels mediated by absorptive capacity. We take these competing predictions to the data. In order to so, however, we must first consider the role of scientific knowledge in the industrial innovation process and the appropriate measure of γ, the firm s absorptive capacity. 2.3 Science and the search for innovation Industrial innovation in a number of fields relies heavily on public science (i.e. research performed in universities or government labs) as an input to the innovation process. Universities are an important source of research results for firms in many industries, and this importance has been growing over time (Narin et al. 1997). Several researchers have described industry use of university-based basic scientific research in the development of new products and processes (Cohen et al. 2002, Mansfield 1991, 1995, Mansfield and Lee 1996, Mansfield 1998, Grossman et al. 2001, McMillan et al. 2000, Narin and Olivastro 1992, Cockburn and Henderson 2001). Public science is particularly important as an input to innovation in the industries studied here: the biotechnology and pharmaceutical industries (McMillan et al. 2000). Firms in the drugs and medical products industry report that their innovations draw heavily from academic 8

10 research and that new products and processes would have been delayed without access to this research (Mansfield 1991, 1998, Collins and Wyatt 1998). Patents in the drugs and medicine category cite significantly more scientific publications than patents in other fields (Narin et al. 1997), and these patents more heavily cite basic research journals (Narin and Olivastro 1992). Firms rely on basic science developments in biology and biochemistry, and many new drugs and delivery systems have their origins in discoveries at universities or government labs. 5 How does existing knowledge, and in particular scientific knowledge, contribute to the innovation process? As described in the search models of innovation presented by Evenson and Kislev (1976), Nelson (1982), and Kortum (1997), existing knowledge facilitates the search for new innovations by allowing researchers to focus their search in the most likely areas of opportunity (David et al. 1992b, Fleming and Sorenson 2004). The knowledge base of the firm can improve this search process by shifting the mean, changing the variance, or opening new areas of the distribution to exploration (Evenson and Kislev 1976). While scientific knowledge certainly does not provide straight forward direction to certain innovation success, understanding and application of scientific principals guide the search behavior of researchers in a way that improves the time and resource efficiency of the search process. The scientific knowledge generated through basic research is particularly useful for guiding the search for innovation because basic research provides knowledge about the underlying phenomenon. As described by Fleming and Sorenson (2004), this deeper understanding of a fundamental problem or the properties of a system guides the search process in several ways. First, it provides the ability to better assess potential projects, thereby avoiding the least promising avenues of search a priori. Second, it may provide tools and techniques to evaluate alternatives more efficiently. Finally, by suggesting which areas of search are theoretically most likely to produce positive results, science may provide some guidance as to how long researchers should spend on a particular area of search before concluding that the failures that they are experiencing should rule out that area of search. These contributions of the knowledge derived from basic research promote efficiency and productivity in the industrial search for innovations. This conceptualization of search for innovation provides predictions relating to a performance measure of firm innovation not typically examined. A superior knowledge base should provide greater search efficiency for new innovations. If firm research activities enhance the absorptive capacity of a firm and thereby provide benefits in terms of the level of external knowledge available and the speed at which firms are able to identify and acquire the knowledge, and a greater knowledge base provides for more efficient search, then these research activities should improve the efficiency of search for new innovation. This expectation is supported by results 5 Cockburn and Henderson (1998) summarize case studies of many important drug developments with their origins in public sector science. 9

11 in existing literature (Fabrizio 2006). The question to be explored here is how the prevailing IP environment affects the search benefits associated with firm research strategies to acquire external scientific knowledge. For this, we turn to a discussion of these absorptive capacitybuilding research activities. 2.4 Absorptive capacity and search While in the original absorptive capacity research Cohen and Levinthal (1989) focused on the firm s R&D expenditures as a means of building absorptive capacity, later work in this area has explored more nuanced aspects of a firm s research activities as they relate to absorptive capacity. I focus here on two aspects of firm research that have been associated with building absorptive capacity: internal basic science research and external linkages with university scientists. Existing literature has demonstrated a positive relationship between these firm research activities and the number and importance of patents generated (Cockburn and Henderson 1998, Gambardella 1992, Zucker et al. 2002, Baum et al. 2000). I consider these two categories of absorptive capacity building research activities as distinct, and consider differences in the impact of the recent increase in university patenting on the effectiveness of each type of absorptive capacity to provide performance benefits with respect to search for new inventions. Internal basic research generates search benefits in two ways. It provides a direct benefit by contributing directly to the knowledge base of the firm. This is not affected by a change patent protection to university research. Internal research also provides firm researchers with the ability to scan the knowledge environment, identify and understand relevant research outcomes, and assimilate this new knowledge into the firm s knowledge base to facilitate search for new inventions. Some degree of commonality between the firm s internal research and the external research may be necessary for successful knowledge transfer (Lane and Lubatkin 1998, Prager and Omenn 1980). Basic research performed internally by the firm creates a bridge of familiarity between firm and university researchers and provides a common vocabulary that facilitates communication. This common knowledge base assists firm researchers in identifying and exploiting university science and also allows allows for more effective communication, understanding, and, consequently, knowledge transfer between the university and firm researchers. This absorptive capacity benefit of internal science depends on being able to identify, evaluate, and exploit the external knowledge. Identification of research results may become easier as patenting increases, if patenting increases the codification, disclosure, and relative visibility of research results. Increased disclose may made the firms ability to evaluate and exploit knowledge more critical. On the other hand, the ability to make use of the patented university research results may be negatively affected by the increase in property rights protection. Follow-on use of the patented research results and any strictly complementary knowledge and materials, will 10

12 be limited to those firms that secure a license to the patent. This limitation implies that even if internal expertise allows firms to identify and assimilate research knowledge, they are limited in their use of the knowledge. In addition to the firms own internal research, evidence suggests that connections and collaborations with external scientists are critical for successful knowledge transfer (Liebeskind et al. 1996, Zucker et al. 1994). Informal interactions are particularly important avenues for knowledge transfer between universities and industry (Cohen et al. 2002), and not only help to identify relevant scientific research, which may or may not be published, but also provide the firm with access to the tacit knowledge complementary to published research results. 6 In many cases, codified research results do not provide a researcher with enough knowledge to utilize the results in the absence of the related tacit knowledge (Dasgupta and David 1994, David et al. 1992a). As described by von Hippel (1994), research-related knowledge often resides with the researcher, and is sticky and difficult to transfer. Social networks and informal interactions provide an important avenue for exchange of such tacit, research-related knowledge (Liebeskind et al. 1996, Siegel et al. 2003) and are important to the development of the absorptive capacity necessary to internalize available knowledge and improve the research performance of a firm (Cockburn and Henderson 1998). Knowledge transfer among a group of researchers with an on going pattern of collaborations is more fine-grained, tacit, and cooperative than otherwise possible (Uzzi 1996, 1997). This knowledge transfer channel depends, in part, on the historical norms favoring disclosure and openness among university researchers (David et al. 1992a, David 1998, Dasgupta and David 1994). To the extent that collaborations with university scientists aid in identification of relevant codified research knowledge, the effect of an increase in university patenting on collaborationbased absorptive capacity is the same as the effect on internal research-based absorptive capacity. However, there are also reasons to expect that an increase in university patenting will have more of an effect of the benefits associated with collaborations. If patenting of research results is increasing the transfer of technology and knowledge from universities to industry, firms with the capabilities to better identify and assimilate this knowledge, due to connections with the scientific community, will be at an increasing advantage. Well connected firms will be at an additional advantage when licensing opportunities depend in part on connections to university researchers. However, if increases in patenting and commercialization activity are in fact decreasing university researchers willingness to share the tacit knowledge 6 In a survey of firms in the information technology, scientific instruments, and new materials sectors, industry researchers report that linkages to universities provide substantial advantage in terms of keeping abreast of university research, gaining access to the expertise of university researchers, and receiving general assistant with problem solving (Rappert et al. 1999). Similarly, Thursby and Thursby (2000) find that personal contacts of firm researchers are often the source of information about technologies available from university research. 11

13 typically conveyed through informal interactions and collaborations, this will have a detrimental effect of knowledge transfer based on collaborations with university scientists. Whereas in an environment of open sharing, these collaborations may have tapped into a network with a wealth of scientific knowledge, if the researchers in that network become less willing to discuss their research this connection has less value as an access point for university knowledge. 3 Empirical Methodology The purpose of the empirical analysis is to test the effect of a change in the degree of patent protection to university research on the value of firm research activities in accessing and exploiting knowledge from universities. In order to do so, it is possible to use the variation in increases in university patenting activity across various technological areas. While the fields of biology, chemistry, and biochemistry have all experienced some of the largest increases in patenting by university researchers (Nelson 2001), there is considerable variation in patenting activity across particular technological fields within these areas. For example, in the patent class C12N, Micro-organisms or Enzymes 8.2% of the patents in 1980 were generated by universities. By 1995, university patents represented 19.9% of the patents in this class. In contrast, in the patent class C08G, Macromolecular Compounds obtained Otherwise Than By Reactions Only Involving Carbon-to-Carbon Unsaturated Bonds 1.6% of the patents applied for in 1980 were from universities, and this grew to 2.7% in Differences in the growth of university patent propensity across the technological areas is a result of the degree to which university research in the field is applied in nature, and the perceived licensing opportunities in the field (Shane 2002). When a university technology licensing office experiences success in patenting and licensing in a particular area, they are more inclined to seek out and encourage disclosures in that particular area. The identification strategy in this empirical analysis is based on the assumption that the varying increases in university patenting across technology areas are exogenous in the sense that prior changes in university patent propensity are not a function of the current relationship between firm research activities and search for innovation. 4 Data I construct a panel data set of 83 firms in the biotechnology and pharmaceutical industries during the period. I explore the moderating effect of intellectual property (i.e. patent) protection of university research results on the relationship between firm research activities and the speed with which the firm is able to exploit existing knowledge in new patented inventions. I exploit variation along two dimensions to identify this effect: First, the changes in the intensity 12

14 of university patenting in a given technology area over time, and second, the variation across firms and over time in the level of absorptive capacity-enhancing research activities Sample The sample of firms is composed of companies listed as major pharmaceutical firms and those listed as biotechnology firms in Standard & Poor s Industry Surveys for the years 1979, 1985, 1990, 1995 and For each company, I collect firm-specific data covering the period from several sources. I rely on Corptech, Hoovers Online, Mergent, Lexis-Nexis article searches, and the Capital Changes Reports for corporate structure information detailing the founding date, geographic location, and mergers and acquisitions for each firm. 9 I was able to locate the required firm level data for 83 companies. 10 Note that use of public data restricts the sample of firms to those that went public, indicating at least some level of success (although some went on to fail or be acquired). Therefore, all results should be interpreted as conditional on at least achieving public status Pace of knowledge exploitation I rely on information found on the front page of each of the sample firms patents to develop a measure of the efficiency of search, which I call the pace of knowledge exploitation. In order to compare similar technologies, my sample of patents is limited to the sixteen 4-digit international patent classes most closely associated with pharmaceutical and biotechnology firms. 12 dependent variable measuring the pace of knowledge exploitation is calculated based on each patent s prior art citations; that is, the U.S. patents listed in the References Cited section of the patent. These prior art citations indicate the existing knowledge on which the new patent 7 Note that this identification strategy depends on firms participation in multiple technology areas, because the estimations include controls for the technology area. The innovation patterns of all of the firms in this sample meet this criteria, and the mean number of technology classes per firm is greater than 5. 8 These industries were not surveyed in the 1980 edition. 9 When a firm merges with or acquires another firm, the surviving firm remains in the sample and publications with either original firms name are used for the resulting firm in subsequent years. 10 I could not include companies with no identifiable patents during the sample period, so all results should be interpreted as conditional on having at least one patent. 11 Many of the firms do not span the entire 21 year window. A firm enters my sample in the first year for which Compustat reports employee data, and exits when this data item is no longer reported. Unfortunately, Datastream provides data only for 1980 and subsequent years, so the six foreign firms in the sample that are not covered by Compustat do not enter until The international patent classes that I include are A01N, A61B, A61F, A61K, A61M, C07C, C07D, C07F, C07H, C07K, C08F, C08G, C12N, C12P, C12Q, and G01N. These classes account for over 80% of all patents associated with the sample companies. The 13

15 builds. 13 The grant dates of the patents cited as prior art provides an indication of the age or vintage of the knowledge being built upon. The time lag between the cited patents and the new invention therefore represents the speed with which that prior knowledge is utilized in the new invention, with longer lags indicating a slower pace of knowledge exploitation. 14 Other studies, such as Sorenson and Fleming (2004), have used forward citation lags similarly to measure the speed of diffusion. I am interested in the backward citation lag as a measure of the age of prior art being built upon Firm research activities To test the hypotheses described above, I develop proxies for the the internal basic research performed by the firm and the collaborative linkages between the firm and university scientists. The ratio of the number of the annual scientific publications by firm researchers to the firm s R&D expenditures is used as a proxy for the internal basic science expertise of the firm, as it was in Cockburn and Henderson (1998). 16 I proxy for the firm s relative focus on collaboration with university scientists with the percentage of the firm s publications in a year that are co-authored with university scientists, also generated based on counts from the Science Citation Index. 17 Both of these publication-based measures reflect more than the firm s internal basic science research activity. Firms that openly publish research may have a more academic organizational structure, promoting interaction, disclosure, and individual inquiry (Henderson and Cockburn 1994). Firms with more university-coauthored publications, such as Genentech, may also have a significant number of Ph.D. scientists or founders that were university-affiliated. The 13 The information contained in patent citations has been used to trace the transfer of knowledge across inventors, institutions, geographic locations, and technology classes, and to develop measures of importance, generality, and originality (Trajtenberg et al. 1997, Henderson et al. 1998a,b, Hall et al. 2001). See Narin (1994) for a discussion of the use of patent citations generally. 14 Similarly, Narin (1994) defines the technology cycle time of an industry as the median age of the patents cited in other patents. As an example, he states that electronics, which is a relatively fast moving area, has a much shorter cycle time than slower moving areas such as mechanical areas. Deng et al. (1999) use the technology cycle time at the patent level to proxy for how quickly firms are innovating. Trajtenberg et al. (1997) describe the average backward citation lag of a patent as a measure of the remoteness in time of the patent, where a longer lag corresponds to drawing from older sources. 15 For the patent prior art citations, I exclude self-citations, i.e. citations where the assignee of the cited patent is the same as the assignee of the citing patent, following Mowery et al. (2002). Patent citation lags are grant year to application year. 16 Similarly, Gambardella (1992), Arora and Gambardella (1994), Lim (2004) use the count of firm publications as a measure of investment in basic research. As in these existing studies, I use the address field for the researcher to identify the company or university affiliation. 17 A similar measure of connectedness to university science was used in Cockburn and Henderson (1998), although those authors allow a single publication to count as more than one co-authorship if more than one outside institution is involved. 14

16 analysis here does not differentiate between these aspects of firm research strategy, and so the coefficients on the publications and co-authorships variables may partially reflect other unobserved, correlated factors. The variables used here should be interpreted as indicator variables, and recognized as a reflection of the larger underlying strategy and organizational focus of the firm. 4.4 University Patenting The independent variable of primary interest is the interaction of the strength of the appropriability environment with respect to university research and each of these two measures of absorptive capacity-related research activities. I measure the strength of the appropriability environment using the percent of patents that are assigned to universities in each technology class - year observation, U niversityp atenting. Since a change in university patenting today may influence industrial innovation for many years to come, I measure university patenting in a way that takes account of the cumulative increase in university patenting from 1975 up to the year of the patent being considered. 18 For patents in patent class k applied for in year t, the cumulative percentage of university patenting is calculated as follows: Tt=1975 #universitypatents k,t Tt=1975 #totalpatents k,t 4.5 Controls In order to control for differences in the pace of knowledge exploitation across technological areas, I include sixteen technology class level dummy variables. To control for common changes over time, I include year dummy variables. To allow for the possibility of important unobserved firm-level heterogeneity, I also include firm-level fixed effects in the prefered specification. The results should therefore be interpreted as changes over time within a firm, controlling for common technology and year effects. 18 Ideally, I would measure the change in the percentage of university research results that are patented in each field in each year. Unfortunately, the total number of university research results is impossible to know. The percentage of patents in a field assigned to universities may increase for several reasons. First, it would increase when the percent of university research being patented increased. Holding the percent of university research being patented constant, the percent of patents assigned to universities would increase if university research increased relative to other research in a field. Finally, the percent of patents to universities could also increase if patenting by others decreased. To the extent that these other attributes of a technology area are changing, the university patenting measure will include these effects as well. However, it has been well established that patenting of university research is increasing, and it is very doubtful that all of the increase can be due to increased university research relative to other research. 15

17 I control for the firm s research intensity using the ratio of annual R&D expenditures to the number of employees at the firm level, as did Cohen and Levinthal (1989). I control for firm size with the natural log of the annual number of employees. Mowery et al. (1996) found that large firms did not demonstrate as much absorptive capacity with respect to alliance partners relative to smaller firms. In some specifications, I also include a dummy variable Biotech equal to one for companies listed as biotechnology companies in the Standard & Poor s Industry Surveys. 19 I control for two additional patent characteristics expected to be significantly related to the backward citation lag. First, I control for the average number of citations to prior art in the firm s patents in the given ipc class, which serves two purposes. This controls for the fact that since the distribution of the citation lag has a long right tail (there are very few lags of a very long length) and patent citations come only in whole numbers (i.e. it is not possible to have 0.5 of a citation), the more citations there are, the greater the likelihood of having a citations drawn from the long right tail of the distribution, and thus a higher mean lag. It also controls for the complexity of the patents, where more complex patents may cite more prior art and may also take longer to develop. Second, I control for the average number of claims in the firm s patent in the pc class and year to control for the complexity and breadth of the patent, which may be associated with longer development time and thus longer backward citation lags. I also control for two characteristics of the patent class: the total number of patents in the same technology class and application year and the growth in the number of patents in the technology class. 20 These control for citation inflation, or the fact that backward citation lags are getting shorter due to the increasing number of patents each year, resulting in more recent patents relative to the number of older patents available to cite. Summary statistics are reported in Table 1. There is considerable variation within the sample for all of the variables of interest. These summary statistics reinforce the strong science base in these fields. The mean percentage of publications co-authored with university scientists is 32.5%, and the range is from zero to 100%. Comparing the firm-year and firm-class-year summary statistics, it is not surprising that firms patenting in more classes are larger, older, and more likely pharmaceutical firms (not biotech). Other than these characteristics, the summary statistics at the two levels are quite similar. Table 2 reports the simple correlation coefficients for these variables (at the firm-class-year level of observation). Biotech firms tend to be smaller and younger. As expected, the average backward citation lag is positively correlated with the average number of citations made. 19 The number of such biotech companies grows over the sample period, and by the end of the period is approximately equal to the number of pharmaceutical firms. 20 Growth in the number of patents is measured as the increase (or decrease) in the number of patents in the class in prior 5 years. 16