Academic patent licenses: Roadblocks or signposts for nonlicensee cumulative innovation?

Accepted Manuscript Title: Academic Patent Licenses: Roadblocks or Signposts for Nonlicensee Cumulative Innovation? Authors: Kyriakos Drivas, Zhen Lei, Brian D. Wright PII: DOI: Reference:

S0167-2681(17)30084-7 http://dx.doi.org/doi:10.1016/j.jebo.2017.03.018 JEBO 4012

To appear in:

Journal

Received date: Revised date: Accepted date:

28-9-2015 16-8-2016 20-3-2017

of

Economic

Behavior

&

Organization

Please cite this article as: Drivas, Kyriakos, Lei, Zhen, Wright, Brian D., Academic Patent Licenses: Roadblocks or Signposts for Nonlicensee Cumulative Innovation?.Journal of Economic Behavior and Organization http://dx.doi.org/10.1016/j.jebo.2017.03.018 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Academic Patent Licenses: Roadblocks or Signposts for Nonlicensee Cumulative Innovation? Kyriakos Drivas*, Zhen Lei**, and Brian D. Wright*** *Department of Economics University of Piraeus, Karaoli & Dimitriou 80. 18534 Piraeus, Greece ** Department of Energy and Mineral Engineering and the EMS Energy Institute The Pennsylvania State University, 110 Hosler Building, University Park, PA 16802 ***Department of Agricultural and Resource Economics 207 Giannini Hall #3310, University of California at Berkeley, Berkeley, CA 94720 HighlightsWe empirically address the effects of exclusive licensing of university inventions on subsequent patented innovation by nonlicensees. - We bring evidence from detailed files on patented invention disclosures, and their licensing activity, at three renowned National Research Laboratories, Los Alamos, Lawrence Livermore, and Lawrence Berkeley Laboratories, and nine campuses of University of California, as recorded by the UC Office of Technology Transfer. - We find that after exclusive licensing forward citations by private sector nonlicensees actually increase.

-

An unanticipated exclusive license appears to be a signpost pointing out commercially relevant innovation pathways that nonlicensees follow with successful patented research. - Results are robust to a series of alternative explanations. Abstract Academic inventions are key drivers of technical progress in modern economies, and exclusive licensing has become the dominant means of transfer to the private sector. However, the strong licensee incentives generated by exclusive academic licensing are generally assumed to come at the expense of discouragement or diversion of research by nonlicensees. Using data from university campuses and national research laboratories, we find that after exclusive licensing forward citations by private sector nonlicensees actually increase. An unanticipated exclusive license appears to be a signpost pointing out commercially relevant innovation pathways that nonlicensees follow with successful patented research. Tests using multiple pre-license information disclosures support this signaling hypothesis. JEL classification: O31, O32, O33, O34, O38

Keywords: intellectual property; ; ; ; ; , exclusive licensing, academic invention, innovation sequence, technology transfer, patent citations. 1. Introduction Academic inventions are a key source of technical progress (Jaffe, 1989; Adams, 1990; Berman, 2011). Since they tend to be embryonic, in need of further research and development,1 transfer of technology to the private sector is an essential step in a successful academic innovation program. But how is this transfer best achieved? In the fourth decade after the passage of the Bayh-Dole Act, the global increase 1

Schacht 2012 p. 4. In a survey by Jensen and Thursby (2001 Table 1 p. 243) university technology transfer managers reported that 48% of inventions are “proof of concepts but no prototype” and 29% had only a laboratory-scale prototype.

1

in academic patents, and exclusive licensing of those patents, remain controversial (Mowery and Sampat, 2005; Schacht, 2012; and Boldrin and Levine, 2013). There is concern that the incentivation of development by the licensee comes at the cost of reduction of positive externalities for nonlicensee innovators. In this paper we empirically address the effects of exclusive licensing of university inventions on subsequent patented innovation by nonlicensees. The general topic of sequential innovation has received much academic attention since the pioneering model of Green and Scotchmer (1995), yet there is little systematic evidence on the effect of university patenting and licensing on the rate and direction of further innovation beyond academia, apart from analyses of citations in academic publications, with conflicting conclusions (Murray and Stern, 2007; Sampat, 2004; Fehder, Murray and Stern, 2014; Thompson, Mowery and Ziedonis, 2016). In the absence of adequate evidence, there are strong a priori arguments supporting the consensus that academic patent licensing, predominantly exclusive,2 blocks, diminishes or diverts research by nonlicensees. Grant of an exclusive license implies a credible commitment to enforce the patent monopoly rights against others conducting research on innovations that might be infringing, or already practicing an infringing invention (Nelson, 2004; Cohen, 2005). In a classic paper Kitch (1977) makes the analogy to a mining claim. On the other hand some nonlicensees might divert their efforts to finding inventions in the same field that are valuable only as non-infringing substitutes for the licensed patent, implying wasteful duplication that reduces the overall productivity of the innovation sequence. The above arguments assume that all information relevant to follow-on researchers (other than whether and how it will be licensed) is common knowledge after the patent (or its application) is published. This assumption is unlikely to hold in practice. Indeed in a recent survey of scientists, Jensen and Webster (2014) find that licenses to patented research can affect third parties via restrictions on licensee publication of information relevant to the patented and licensed invention. On the other hand, it is plausible that news of licensing might be a positive signal to nonlicensees about the prospects for, or feasibility of, further non-infringing, useful innovation in relevant fields. Such a signal might encourage nonlicensees to explore ways to make use of the licensed patent or information revealed therein, leading to socially useful follow-on development or utilization of the invention. However, there is no empirical evidence in the literature regarding any informational effect of patent licensing on nonlicensee researchers. 2

See for example Henry et al. (2003) and Pressman et al. (2006).

2

For academic inventions, an informational effect of licensing might seem particularly dubious a priori. Information about the licensed invention is typically disclosed much earlier in some form such as a conference presentation, working paper, or academic publication. Nevertheless, academics know well that publication of a paper does not mean all researchers in the field are instantaneously aware of the implications of its findings for downstream research. Further, a patent may reveal key technical details for understanding or reproducing the invention, not included in related scientific papers.3 Although university patents or patent applications offer critical technical information about the invention, that information is costly to acquire. Indeed, even patent examiners who are experts in their fields may find locating relevant prior patents and papers to be a substantial and perhaps overwhelming challenge (Lemley and Sampat, 2013; Lei and Wright, 2015). Furthermore, researchers may be reluctant to search more diligently for prior patents in the United States because they are wary of charges of willful infringement based on that prior knowledge.4 Thus, news of a license might draw attention to a patent not previously identified as important by downstream innovators. Furthermore, a license reveals commercially relevant information not found in the patent or related publications. It affirms that an inventor other than the patentee found the invention to be sufficiently valuable to justify the substantial costs of negotiation and the financial obligations specified in the license (often including an upfront payment to reimburse the cost of patent prosecution by the academic licensor).5 Relative to an academic patent, a license is much more credible evidence of “commercial opportunity.” In the classic survey by Jaffe and Trajtenberg, a majority of citing patentees identify awareness of such opportunity as a significant influence on the development of the relevant cited invention; far fewer mention information in patents or technical literature (Jaffe and Trajtenberg, 2002, p. 388). Thus the net effects of exclusive academic licensing on further research by nonlicensees, and on the field and focus of that research, are empirical issues that have yet to be resolved. The evidence we bring to these questions includes files on patented invention disclosures at three renowned National Research Laboratories, Los Alamos, Lawrence Livermore, and Lawrence Berkeley 3

A recent survey of nanotechnology researchers revealed that 64% of those who had looked to patents to gain scientific knowledge found useful information. One industrial participant noted: “Usually the way a new technology is described is much more reliable and reproducible in a patent than in a scientific paper. Unfortunately many academic researchers purposely remove essential steps for reproducing data, for fear other researchers will catch up with them and publish first.” (See Ouellette, 2012, footnote 145 p. 560, and p. 561). 4 Cohen et al. (2002) find that researchers in Japan, a less litigious society, report a greater tendency to read patent documents to obtain useful information. 5 If instead the required payment were structured entirely as a running royalty, the announcement of an exclusive license would not necessarily imply any commitment by the licensee to a significant minimum evaluation of the patent. See Gallini and Wright (1990) for more on the informational implications of different contractual forms for an exclusive license.

3

Laboratories, and nine University of California (UC) campuses, as recorded by the UC Office of Technology Transfer (hereafter OTT).6 This confidential dataset is unique in that it contains licensing information and other confidential contractual information, and its size is large enough to permit econometric analysis. We use a forward citation to a patent from these UC campuses or National Labs (“UC/NL” patent) as an indicator of successful innovation that is informed or influenced by, or builds upon, that patent (Jaffe, Trajtenberg and Henderson 1993).7 Following exclusive licensing, citations of exclusively licensed UC/NL patents by their licensees (“licensee citations”) increase, although the direction of causality is not clear.8 These increased licensee citations come mainly from patents in the same narrow technology fields (according to the US patent classifications) as the licensed patent. In this paper we focus mainly on the responses of nonlicensees. We exploit variation in the timing of exclusive licensing, relative to the date of patent grant, to identify the impacts of licensing on nonlicensee citations. We find no support in the patent citation data for the consensus presumption that exclusive licensing has net negative effects on follow-on innovation by nonlicensees. In fact, we find that citations to licensed patents by nonlicensees (“nonlicensee citations”) actually increase significantly after licensing. Further, we find no evidence that the increased citations by nonlicensees come from increased patenting in the same narrow field as the licensed patent that could be interpreted as “inventing around” the licensed patent. Hence we conjecture that the net positive nonlicensee citation response is induced by the positive signaling effect of the exclusive license regarding prospects for related downstream research. To explore the signaling hypothesis we turn to OTT data on pre-licensing secrecy agreements, non-exclusive contracts that reveal confidential information about the invention disclosure to potential licensees. The existence of multiple secrecy agreements before licensing indicates that competing potential licensees have extra information about the invention, reducing the salience of information conveyed to the market by licensing. We find that the relation between licensing and subsequent nonlicensee citations becomes insignificant for patents with multiple prior secrecy agreements, 6

The OTT has recently been restructured into two departmental units, Innovation Alliances Services and Research Policy & Coordination Unit and, within the UC Office of President (http://www.ucop.edu/ott/about.html). However, consistent with usage during the period the sample was generated, here we still refer to the office in charge of technology transfer activities as the OTT. 7 Recent work has employed patent citations as a metric for cumulative innovation and knowledge flows (for instance Branstetter, 2001; Peri, 2005; Galasso and Schankerman, 2015). In particular, they have been employed at an academic context to measure the non-market channels of knowledge flows (Mowery and Ziedonis, 2001; Belenzon and Schankerman, 2013). 8 Winning an exclusive license could give the licensee an incentive to conduct follow-on research. Alternatively, the licensee might take the license because he is already engaged in research related to the invention.

4

providing further evidence that an exclusive license on an academic invention is a signpost pointing out to previously uninformed nonlicensees novel and attractive innovation pathways that they might exploit with successful, patented innovations. An alternative interpretation of the observed increase in nonlicensee citations to a UC/NL patent following its licensing is that the timing of licensing is endogenous and a growing demand for (and interest in) the invention or its technical field leads to both the licensing event and an increase in nonlicensee citations. We provide several empirical tests that suggest this alternative “growing demand” hypothesis is unlikely to be the predominant rationale for the increase in nonlicensee citations after licensing. Our inference of a signaling effect of the licensing of academic patents very recently received support in a study of the effects of licensing on paper citations by Thompson, Mowery and Ziedonis (2016), who use an innovative method of automating the matching of patents with relevant research publications. A related signaling effect on competitors is implied in the finding of Austin (1993) that, in the biotech industry, grant of a valuable patent increases the market value of the patentee’s rivals. The next section of this paper presents the hypotheses. Section 3 presents the data. Section 4 discusses our empirical strategy and Section 5 presents the results, followed by a concluding section. 2. Conceptual Framework: Effects of Exclusive Licensing on Nonlicensee Innovation The major rationale for exclusive licensing is that it strongly motivates the licensee to pursue further research and development, protected by the patent. Furthermore, a university inventor who anticipates hefty royalties from an exclusive license might take extra care to minimize the availability of information related to, but not protected by, the patent, that might be of use to nonlicensee competitors (Hellman and Perotti, 2011), and the university might be more strongly motivated to sue infringers. On the other hand, if exclusive licensing directly signals, or motivates the licensee to signal, the existence of promising technology or market potential, that signal might induce nonlicensees to conduct research related to the licensed invention or to associated unprotected information revealed in the patent. It is noteworthy that although details of a university license are confidential, OTT personnel inform us that the licensing event tends to become known by other interested firms, including the licensee’s competitors. 9 9

Exclusive licensees may have incentives to disclose licensing deals if they think that it is beneficial to do so (for example, to dissuade others from contemplating entering, or to encourage researchers further downstream to pursue paths of innovation that might require a license to the licensee’s own innovation building on the academic patent). In particular, given UC’s prestigious status in research, licensees often announce that they have a license to a UC technology. Moreover, OTT publicly lists all available technologies. Hence removal of a technology from that list is an indirect indicator of exclusive licensing.

5

Thus, the net effect of exclusive licensing on nonlicensee innovation is an open empirical question. In this paper we focus on the impact of exclusive licensing of patented academic inventions on follow-on innovation by nonlicensees.10 To proxy for follow-on innovation, we employ citations in subsequent patents (“forward citations”) as indicators of knowledge flows to the citing innovators (Jaffe, Trajtenberg and Henderson, 1993). We provide a brief literature review and discuss in detail the merits and drawbacks of employing this proxy for our empirical strategy in Online Appendix A. 3. Data The University Technology Management Process The nature of our licensing data reflects the procedures of the OTT, which in the sample period also served three leading National Research Laboratories.11 Upon receipt of an invention disclosure from university researchers, this office or its counterpart at an individual campus assigns the disclosed invention to a licensing officer who then becomes responsible for handling the invention. If the invention is the result of research supported by a sponsor holding a “right of first negotiation” for a license, the officer is obligated to approach this sponsor first. In other cases, the officer might send out Non-Confidential Disclosures (NCDs)12 to firms that could be interested in the invention, to seek those willing to pay for patenting the invention in return for a (generally exclusive) license. If no potential licensee emerges, the OTT then decides whether to file a patent application for the invention itself and pay the patent prosecution fee. If the OTT makes such an “at-risk filing”13 the officer then keeps searching for a potential licensee, who might be asked to reimburse part or all of the patent prosecution cost. Hence a licensing agreement could be struck before filing for a patent application, between patent filing and patent grant, after patent grant or never. As part of the pre-licensing negotiation process, the OTT might sign one or more “secrecy agreements” with potential licensees. Upon signing a secrecy agreement, a potential licensee has access to critical information about the invention, for example the nature of a related patent application. Finally, after a technology is exclusively licensed, if OTT receives inquiries about it, the office would respond that it has been licensed to another party. 10 The null hypothesis we test is that exclusive licensing has no effect on follow-on innovation by nonlicensees, as measured by forward patent citations, because we had no strong prior on the sign of the effect of exclusive licensing. A null hypothesis that exclusive licensing has negative impacts on nonlicensees would be more consistent with the consensus on the effects of licensing on nonlicensees, discussed above. If the latter were the null, the significance tests we present below would be conservative. 11 For a more thorough description of UC technology management consult Graff, Heiman and Zilberman (2002). 12 NCDs contain very brief description of disclosed inventions (often one or two paragraphs). In our sample of patents, 13.1% (421 out of 3,232) are associated with NCDs. 13 An “at-risk” filing is more likely if inventors have already publicly disclosed an invention or wish to publish soon, given that patent applications need to be filed within twelve months of public disclosure in the U.S. (35 USC 102(b)).

6

Secrecy agreements are evidence of a firm’s interest in, and/or receipt of, greater information regarding the invention. Multiple secrecy agreements can exist simultaneously for a given invention disclosure. If a seriously interested firm needs more time to review the technology or to acquire funds for licensing fees, the OTT might offer an exclusive “letter agreement” (Letter of Intent) under which it temporarily takes the invention off the market, in return for a specified payment by the firm in addition to commitment to secrecy. Only one letter agreement can be in force for one disclosure at any point in time for a field of use. The term of a secrecy or letter agreement is typically six months. A feature of the licensing process that we take into account in our econometric estimation is that, in the majority of cases, a license covers more than one UC patent. An invention can result in more than one patent; a firm that licenses the invention will usually license all patents in the invention. On the other hand more than one patented invention can be included in a single license simultaneously, or other patented inventions can subsequently be added to an existing license.14 These inventions usually stem from the same research lab or part of the same overarching research project. Of our 853 patents that were licensed exclusively, the overwhelming majority (89%) are licensed as bundles. The average size of the bundle is 5.5.15 Data Overview The UC OTT graciously granted us, under appropriate confidentiality restrictions, access to information on patenting and licensing of a unique dataset of patents resulting from invention disclosures at the nine campuses that comprised the University of California throughout the sample period, and at three Department of Energy National Laboratories (Lawrence Berkeley National Lab, Lawrence Livermore National Lab and Los Alamos National Lab).16 The dataset includes 1,715 licensed US patents with grant years from 1977 to 2009.17 These patents are associated with 900 inventions disclosed to the OTT between February 19, 1975 and June 30, 1997. As we allow at least a twelve-year observation period for inventions to be issued a patent, our data are not likely to be censored with respect to patent activity. 18 Of these inventions, 552 (61.33%) resulted in a single patent.19

14

One of the first studies that uncovered this many-to-many relationship within the academic context is Pressman et al (2006). Uniquely licensed patents receive 1.47 private nonlicensee citations per year while bundled patents receive 1.06 (p-value=0). In our baseline group the difference is 1.71 vs. 1.46 with the difference statistically significant at the 10% level. 16 The UC Merced campus was opened in 2005, towards the end of our sample interval. Los Alamos National Laboratory and Lawrence Livermore National Laboratory became separate legal entities in 2005 and 2007 respectively. 17 Less than 10% of the patents are not assigned to the Regents of the University of California. These may be assigned to individuals, other universities or small firms (in the latter case they are co-assigned with individuals or universities). 18 In fact, only 1% of early inventions take more than twelve years to be issued a patent. 19 This finding is contrary to the notion that there is one-to-one correspondence between invention disclosures and patents. 15

7

The OTT dataset identifies three types of licenses: “exclusive,” “exclusive-with-limit” and “nonexclusive.” However, in discussions, OTT personnel suggested that the distinction between “exclusive” and “exclusive-with-limit” is not meaningful: an exclusive-with-limit license is usually a license exclusive within a field of use,20 which could be an entire industry, a nation or a set of nations. An “exclusive” license, on the other hand, may have clauses allowing the OTT to license the patent to another firm in another geographic area or for another field of use.21 Accordingly, we denote all license agreements coded with some form of exclusivity (“exclusive” or “exclusive-with-limit”) as “exclusive” and patents associated with these agreements as “exclusively licensed.” According to our chosen definition, the number of exclusively licensed patents is approximately seven times larger than that of non-exclusively licensed patents (1,253 vs.178). This overwhelming propensity for exclusivity in licensing of UC patents is consistent with other reports on licensing of academic inventions (Schissel et al. 1999; Henry et al 2003; Pressman et al 2006). Another 284 patents have been licensed both exclusively and nonexclusively; after the first license was abandoned by the licensee(s), a license of a different type was later awarded to the same or a different licensee. For the remainder of the paper, exclusively (nonexclusively) licensed patents refer to patents that have been licensed only exclusively (nonexclusively), though they may have been licensed more than once throughout their patent life.22, 23 4. Empirical Strategy Econometric Strategy Identifying a causal effect between exclusive licensing and nonlicensee citations requires a counterfactual: how many nonlicensee citations would have been made if a UC/NL patent had never been exclusively licensed? The baseline sample in our analysis includes patents that are exclusively

20

A generic example of an exclusive-with-limit license involves a technique that transfers DNAs in plant cells. It could be licensed exclusively to one firm for cotton research and another firm for corn research. A specific example involves Geron Corporation, which originally licensed exclusively-with-limit a human stem cell technology from the Wisconsin Alumni Research Foundation (WARF) for development of six tissue types. A few years later Geron tried to extend the license to cover exclusively twelve additional types, but WARF agreed to grant only a nonexclusive license. Finally WARF sued Geron and the ruling was in WARF’s favor (Stolberg, 2001). 21 Only 49 of 228 patents in our baseline sample have been licensed exclusively-with-limit, while the rest have been licensed exclusively. Of the 1,253 exclusively or exclusively-with-limit licensed patents in our data set, only 311 belong to the latter category. 22 Table B1 of Online Appendix B presents the distribution of the licensed patents in the sample, by licensing type and by timing of licensing in relation to timing of patent grant. 23 Table B2 of Online Appendix B shows summary statistics on characteristics of the subgroups of patents by licensing regime. Non-exclusively licensed patents seem to be quite different from those exclusively licensed. Hence they are not appropriate control groups for the latter. For example, exclusively licensed patents have, on average, more claims (18.91 vs. 14.31) and more International Patent Classifications (IPCs) (6.71 vs. 5.03) than nonexclusively licensed patents.

8

licensed after grant and that remain licensed throughout their patent life. We exploit variation in the timing of exclusive licensing to identify its effects on forward citations.24 In essence, we use patents that are licensed later as the counterfactual for those licensed early. For example, for two similar patents A and B, if patent A is licensed in the 5th year after grant and patent B in the 10th year, the nonlicensee citations that patent B receives during the 6th through 9th year after grant can be used as a counterfactual for citations to patent A during this period if patent A was not licensed at the 5th year. Most licensee researchers work in the same field as the licensed patent. Whether and when a patent is exclusively licensed likely depends on the awareness of firms in the field regarding the invention and their interest in the invention. The exact timing of licensing of the patent is somewhat exogenous from the viewpoint of citing nonlicensees, who are overwhelmingly in different fields. Potential licensees’ valuations are private information, and tend to be heterogeneous and highly uncertain, and negotiations between the OTT and potential licensees are confidential. We shall show below that indeed early- and late-licensed patents are similar in most patent characteristics. Moreover, when interpreting and discussing the results in the next Section, we consider in detail the concern that the timing of licensing could be endogenous, reflecting an increasing demand for the invention that explains increased citations. We also utilize another sample including patents that are exclusively licensed within two years prior to patent grant (licensed before grant) and remain licensed throughout their patent life, as a comparison group for the baseline patents. We employ a Dif-in-Difs strategy to identify the effect of exclusive licensing on nonlicensee citations.25 We shall argue below that these two sets of patents are likely comparable. This is not surprising as the processes of patenting and licensing of university inventions are often intertwined and the timing of licensing relative to the timing of patent grant is, to some extent, stochastic.26 See Online Appendix C for further discussion of these two econometric strategies. The econometric specification for testing the impact of exclusive licensing on nonlicensee citations is:

24

Note that the citation data have been collected up to December 31, 2009. A Dif-in-Difs strategy usually involves a control group that remains untreated during the study period, to control for changes over time caused by other factors that are applicable to the treatment and control groups. The comparison group discussed here involves patents that remained treated (exclusively licensed) during the study period, again to control for some changes over time shared by both the baseline and comparison group, enabling us to identify the average treatment effect. 26 For example, even if both parties are willing to sign a licensing agreement, it takes time for them to agree on all aspects of the contract and draft the final agreement. 25

9

(1)

Yi,s,t = β0 + β1*Exclusive_Licensei,s + Patenti + Periods+Yeart+εi,s,t

Yi , s ,t is the number of forward citations by nonlicensees that patent i receives s periods (years) after grant and in calendar year t, indexed by application year of citing patents. We are primarily interested in citations from private and for-profit nonlicensees (including firms and individuals), as discussions of university technology transfer and the Bayh-Dole Act are primarily focused on follow-on research and development by the private sector. However, results are similar when we examine all nonlicensee citations, including those from universities and non-profit research institutes. This is no surprise, given that private nonlicensee citations constitute 85-90% of total nonlicensee citations in our sample. The variable of key interest, Exclusive_Licensei,s, is an indicator taking the value of 1 if patent i is under an exclusive license agreement at period s since grant. We exclude period 0, the year of license, in the analysis since it is unclear whether that year should be classified as “before licensing” or “after licensing”.27 We include individual patent fixed effects (Patenti), years since grant fixed effects (Periods) to control for dynamics in forward citation patterns, and calendar year dummies (Yeart) to control for macroeconomic conditions that might affect forward citations to the licensed patent in a given calendar year. In the regressions reported in the text, we focus on only the first 15 years since grant for each patent, to avoid complications that may arise from patent expiration (Koo and Wright 2010). As the dependent variable is a count variable, we report results for a fixed effects Poisson estimator throughout. The fixed effects Poisson does not allow for overdispersion in the data but delivers consistent estimates as long as the mean of the dependent variable is correctly specified (Gourieroux et al 1984). We should note that the Poisson estimator by design drops patents for which the dependent variable takes the value of zero across all time periods. Another popular estimator for count data is the fixed effects Negative Binomial. However, the unconditional fixed effects Negative Binomial, while accounting for overdispersion, suffers from the incidental parameters problem, though Allison and Waterman (2002) show in simulation studies that the

In unreported analyses, we include and classify period 0 as “after licensing” (Exclusive_License=1). The results remain similar and, as expected, the coefficients for Exclusive_License become slightly smaller, suggesting that for some patents licensed late in a year, the license year is closer to “before licensing”. 27

10

resulting bias is small. 28 On the other hand, the conditional FE negative binomial has been shown not to be “a true fixed-effects model since it fails to control for all of its predictors” (Hilbe 2011, pg. 478). It is noteworthy that our main results are robust to the choice of estimator, including fixed effect Poisson, unconditional and conditional fixed effects Negative Binomial, and the correlated random effects Poisson model proposed by Wooldridge (2005). Our entire analysis has also been performed using a fixed effects OLS model, which allows us to keep patents for which the dependent variable is zero across all periods and has been employed by some well-regarded studies involving similar dependent variables (for instance, Williams, 2013). The results, available upon request, are similar to those reported here. Finally, since many patents in our data are licensed together as bundles, we cluster standard errors at the bundle level to account for possible serial correlation among patents that are licensed together (Bertrand, Duflo and Mullainathan 2002) Baseline Group: Patents Exclusively Licensed after Grant Our baseline sample for estimating the impacts of exclusive licensing, by comparing forward citations before and after the licensing event, consists of the group of 228 patents that were exclusively licensed after patent grant29 with licenses maintained in force until expiration.30 To examine the number of unique private citers of the licensed UC/NL patents, we draw data from Lai et al (2011) that list disambiguated patent owner names. 31 Of the 209 baseline (licensed after grant) patents that have obtained at least one private nonlicensee citation, the mean number of unique citers is 6.53 and the median 4, suggesting that for a majority of patents, there are at least a handful of follow-on innovators. Figure 1 shows the variation in timing of the exclusive license among these baseline patents, which our empirical strategy exploits to identify the impacts of exclusive licensing. Our baseline sample could induce bias if sample patents licensed late are systematically different from those licensed early.32 Panel A of Table 1 compares characteristics of these two subgroups in the baseline sample, where

28

Another issue with the unconditional fixed effects negative binomial model here is that when we consider licensee citations (not our main focus) it does not always converge, likely due to the large number of zeros. (As we shall discuss in the next sub-section, only 55 out of 228 patents receive licensee citations.) 29 Since we exclude citations from the year of grant, as discussed above, for patents licensed in the year of grant we do not have forward citations when they were unlicensed. We exclude these patents from the baseline group. 30 The reason for excluding patents whose exclusive licenses were revoked/abandoned is that they could be quite different from patents whose licenses had been maintained. Mixing these two groups of patents could add noise. Unless stated otherwise, licensed patents are referred to patents that have been licensed and the licenses have not been revoked. One robustness check in the Online Appendix includes patents whose licenses were revoked; the results in general still hold. 31 https://dataverse.harvard.edu/dataset.xhtml?persistentId=hdl:1902.1/15705 32 A common perception is that high quality technologies tend to be licensed early.

11

“licensed early” is defined as licensed within five years after grant and “licensed late” is defined as licensed on or after the sixth year since grant. Overall, we do not observe stark differences between these two subgroups. Only the number of IPC classifications and patent prosecution length are statistically different (at the 5% level). However, both differences might well be artifacts.33 The number of secrecy agreements is similar for both groups, suggesting that initial interest expressed by firms was also similar. The difference in the number of letter agreements is significant, at the 10% level, which is to be expected since use of a letter agreement might be associated with a delay in signing the licensing agreement. With respect to forward citations received by these UC/NL patents, we also do not observe a significant difference in overall non-licensee citations per year between early and late licensed patents. The number of licensee citations is greater for early licensed patents, which is expected as their licensees have more post-licensing years to do follow-on research, before the patent expires. Figure 2 shows citations by nonlicensees to early and late licensed patents before exclusive licensing. While unlicensed, both subgroups receive, on average, a similar number of private nonlicensee citations during the first five years after grant. Further, as shown in Table 1, focusing on years prior to licensing, the number of private non-licensee citations per year is 1.39 for early licensed and 1.47 for late licensed patents (the difference is not statistically significant). Further, in the year just before licensing, early licensed patents receive an average of 1.36 private non-licensee citations while late licensed patents obtain 1.62 (again, the difference is not statistically significant). These observations provide support for using one subgroup as a control for the other. Comparison Groups: Patents Exclusively Licensed Before Grant Augmenting the baseline group (patents exclusively licensed after grant), we also use patents exclusively licensed before grant as a comparison group. We exclude patents that are added to existing license agreements and thus bundled with patents licensed earlier, because for those patents the licensing dates recorded are the dates of initial licenses (not the dates those patents were actually licensed). We further restrict the comparison group to patents for which an exclusive license was issued within two years before grant and maintained to patent expiration.34 As noted, these patents are likely to be

33

For patents filed and licensed on the same day, those with shorter patent prosecution at the USPTO are more likely to belong to “late licensed” patents, as the definition of “early or late licensed” patents depends on the lag between timing of patent grant and licensing. Further, if more IPC classifications tend to imply longer prosecution, this could explain at least part of the difference in IPC classifications between the two groups. 34 Note that the results involving the comparison group are robust to different cutoff years. For instance, similar results are obtained using as the comparison group patents exclusively licensed within three years before grant.

12

comparable to the baseline patents that were exclusively licensed after grant, since the timing of licensing is to some extent stochastic. This comparison group consists of 86 patents exclusively licensed within two years before grant, not added to existing licenses and for which the licenses were maintained to expiration. Panel B of Table 1 shows that patents in this comparison group do not differ systematically from those in the baseline group with respect to most patent characteristics. The number of non-licensee citations per year received by the baseline patents is similar to that for the patents in the comparison group, though the former accrue more licensee citations. For the 286 patents (both in the baseline and comparison groups) that have obtained at least one private nonlicensee citation, the average number of nonlicensee citers is 6.18 while the median is 4. 5.

Results

Graphical Evidence Figure 3A displays private nonlicensee citations (from firms and individuals) -6 to +6 years around the year of exclusive licensing, for an average patent in the baseline group (licensed after grant).35 Nonlicensee citations significantly increase around two years after exclusive licensing, suggesting a lag between information flows and resulting patent applications. The increase in private nonlicensee citations after exclusive licensing is robust to the exclusion of outliers such as the top ten most highly cited patents. In Figure 3B we present coefficient estimates from a fixed effects Poisson model involving the sample of both the baseline patents and the comparison group (patents licensed within two years before grant). We regress the private nonlicensee citations on thirteen interaction terms between a dummy on whether a patent belongs in the baseline group and the number of years before or after the licensing event, in addition to year since grant dummies and citation calendar year dummies. The figure exhibits the coefficients for these interaction terms and their 95% confidence intervals, and thus is a graphic illustration of the results of the Dif-in-Difs estimation.36 This figure shows that prior to licensing there is no pre-trend in nonlicensee citations for the baseline patents that were licensed after grant. After licensing baseline patents exhibit a positive trend in nonlicensee citations compared to the patents licensed within two years before grant.

35

We run a Poisson regression where the dependent variable is nonlicensee citations and the regressors are patent fixed effects, years since grant fixed effects and calendar year dummies, as in equation (1) but excluding the Exclusive_License indicator. We then plot the residuals around the -6/6 year window of the licensing event in Figure 3A. 36 For a similar treatment see Azoulay et al (2015), p. 1129.

13

Main Results Table 2 presents estimates of the marginal effect of the key variable, Exclusive_License, indicating how private nonlicensee citations (from firms and individuals) change after award of an exclusive license. Columns 1-5 consider only the baseline patents. Column 1 is estimated via OLS, Column 2 via the correlated random effects Poisson model, Column 3 via the conditional fixed effects Negative Binomial, Column 4 via the unconditional fixed effects negative binomial and Column 5 via the fixed effects Poisson model. A coefficient from a non-linear estimator can be interpreted as follows: for instance in Column 5, citations increase by Exp(0.41)-1 = 1.51-1=51% after licensing compared to before licensing. The average marginal effects for the non-linear estimators are presented in brackets. All models show that private nonlicensee citations of the baseline patents significantly increase following exclusive licensing, suggesting that nonlicensees conduct more follow-on research after a patent is exclusively licensed, and that this result is robust to estimator selection. Hereafter, we focus on results from a fixed effects Poisson model. Column 6 includes, in addition to patents in the baseline group, patents in the comparison group (licensed within two years before grant). The coefficient for exclusive licensing remains positive and significant after this extension of the sample. Columns 7 and 8 of Table 2 address the change of private nonlicensee citations after exclusive licensing, for patents resulting from federally sponsored research, which are the focus of the Bayh-Dole Act.37 After exclusive licensing, nonlicensee citations exhibit a larger increase for this restricted sample. This might reflect the fact that federally sponsored research tends to be more basic and thus more likely to lead to embryonic inventions, for which exclusive licensing sends important signals of commitment to development and of the pathways to follow-on innovation opportunities. In Online Appendix D, we conduct a number of robustness checks by varying the analytical sample (such as excluding patents licensed to research sponsors, excluding patents licensed in bundles, including patents with licenses that were later relinquished or revoked, and redefining the Exclusive_License indicator to take into account the potential lag between initiation of nonlicensee follow-on research and their filing of patent applications. The results of these robustness checks are consistent with the main results in Table 2. In Online Appendix E, we investigate whether the effect of exclusive licensing on nonlicensee citations differs by industries. We find that an exclusive license

37

Our dataset includes a field in which inventors report the name and type of the sponsors (government, non-profit, corporation, etc.) of the research project that results in the disclosed invention.

14

especially encourages nonlicensees to undertake patented follow-on research in industries where patents are more important as means of appropriation, hence exclusive licenses are stronger indicators of patent value. For comparison, we also investigate the correlation between exclusive licensing and citations to the licensed UC/NL patent by the licensee (licensee citations).38 In our dataset, forward citations by licensees prior to licensing are virtually absent: only eight out of 228 patents in the baseline group receive at least one citation from their future licensees before exclusive licensing; overall, we observe only 13 licensee citations prior to licensing. After licensing, still a minority of the baseline patents (55 out of 228) receive licensee citations. Table 3 shows the results from the fixed effect Poisson estimation regarding licensee citations after vs. before exclusive licensing. The coefficients for Exclusive_License are positive but insignificant, reflecting the fact that a large majority of licensed patents receive no licensee citations both before and after licensing. The Informational Effect of Exclusive Licensing: Empirical Evidence Contrary to our prior expectations, informed by analogy to earlier studies finding negative or at best insignificant effects of patenting on paper citations, results thus far indicate that exclusive licensing of a UC/NL patent is followed by a statistically significant increase in nonlicensee patent citations. One interpretation is that on balance, the exclusive license is a signpost indicating to relevant researchers some promising paths through related research fields; this informational role outweighs the fact that the exclusivity erects a roadblock or a tollbooth on some already-visible routes. How can we test the hypothesis that exclusive licensing plays a positive informational role in innovation sequences initiated in academia? We found an answer in the practice of the OTT of offering secrecy agreements to give potential licensees access to confidential information39 regarding the

38

If an exclusive license incentivizes the licensee to pursue further research on the licensed patent, such follow-on activity is likely to lead to generation of additional subsequent patents by the licensee that likely cite the prior licensed patent (Thursby and Thursby, 2003). By this rationale, after an exclusive license is awarded, the number of licensee citations should increase. On the other hand, it is implicit in the Bayh-Dole Act that exclusive licensees might well engage in further development and commercialization of the licensed invention that do not involve further patentable innovation (Eisenberg 1996). There is also potential hazard of the licensee “shelving” the patent, suppressing its development (Dechenaux, Thursby and Thursby 2009). However, the OTT tries to minimize this hazard, using measures such as milestone payments that raise the cost of a shelving strategy. 39 Unlike NCDs, which offer a simple general description of inventions designed so as not to reveal critical information, information revealed through secrecy agreements is much more detailed and specific.

15

invention or the patent application.40 Forty-one percent of the patents in the baseline group have associated secrecy agreements and 22% have more than one.41 A secrecy agreement means that a potential licensee interested in an invention has sought or been offered extra information related to the invention, prior to licensing. In either case, the existence of an agreement indicates that the potential licensee became well informed about the invention prior to licensing. If there is only one secrecy agreement for a licensed invention then it is likely to involve the actual exclusive licensee, who is more likely to keep the information confidential. Existence of multiple secrecy agreements implies that at least one nonlicensee was well informed prior to licensing, reducing the informational effect of the exclusive license. Multiple agreements are also likely to increase the probability that information leaks out to other interested nonlicensees. If the informational effect is important, nonlicensee citations should be less responsive to licensing when there are multiple secrecy agreements prior to licensing. Figure 4 addresses nonlicensee citations before and after an exclusive license for patents in the baseline group, separating those with and without multiple secrecy agreements, and controlling for patent fixed effects, years since grant and calendar year.42 Nonlicensee citations for patents without multiple secrecy agreements substantially increase following licensing, whereas there is no evidence of an increase after licensing of patents with multiple secrecy agreements. The econometric results in Table 4 show that, for the subsample of patents with no more than one secrecy agreement, the effect of licensing on nonlicensee citations remains significant and the magnitude of the effect is even greater than in Table 2. For patents with multiple secrecy agreements, however, the coefficient of Exclusive_License is no longer statistically different from zero (Columns 3 and 4 of Table 4). Differences in the characteristics and technology mix of these two patent groups do not appear large enough to explain the differences in citation responses presented in Table 4. 43 40

Information about the patent application might be particularly important for applications prior to November 29, 2000, which were usually published at time of grant in the United States. Starting on that date, United States applications have been published 18 months after the priority date (with an opt-out option chosen by a small minority of applicants) in accord with the Intellectual Property and Communications Omnibus Reform Act of 1999. 41 For the comparison group (patents licensed within 2 year before grant), 41% have a secrecy agreement while 33% have more than one. 42 We first regress nonlicensee citations on patent fixed effects, years since grant fixed effects and calendar year dummies, as in equation (1) but excluding the Exclusive_License indicator. We then plot the residuals around the -6/+6 year window of the licensing event. 43 Table B3 in Online Appendix B compares patent characteristics between these two groups. They are similar in most of the patent characteristics, except for patent prosecution length and average grant year. For the baseline patents with multiple secrecy agreements, 40% belong to the Chemical, Drugs and Medical (CDM) technology categories and 58% belong to the Computer, Communications, Electronic and Mechanical (CCEEM) categories. For the baseline patents without multiple secrecies, the percentages are 34% in CDM and 63% in CCEEM. A Kolmogorov-Smirnov test for the equality of

16

We also examine the number of unique private citers for licensed UC/NL patents with and without multiple secrecy agreements. The means are 6 and 6.67 respectively. (The difference is not statistically significant.) Furthermore, patents without multiple secrecy agreements have on average 4.5 unique citers prior to licensing while patents without have 3.72. Hence, patents with multiple secrecy agreements have on average 1.5 new unique citers after licensing, while patents without multiple secrecies have 2.95 (the difference is statistically significant at the 10% level). These descriptive statistics indicate that: (1) the number of unique innovators that build on a UC/NL patent is typically small; (2) while both groups of patents ultimately reach on average roughly the same number of unique citing innovators (6 and 6.67), patents with multiple secrecy agreements reach more innovators prior to licensing. One might hypothesize that multiple secrecy agreements might imply a delay in licensing, and so be associated with patents that are late licensed; our results might be picking up a time dimension rather the informational effect of secrecies. Two observations can help alleviate this concern. First, as mentioned earlier, secrecy agreements can be signed simultaneously with many firms. Second, of the patents licensed after grant, 20.6% of the early licensed patent and 23.5% of the late licensed patents are associated with multiple secrecy agreements. Further, of the 50 patents in the baseline group that have more than one secrecy, 26 are early licensed and 24 late licensed. These figures show that multiple secrecy agreements do not imply longer delays in licensing. Hence, when we distinguish patents by the number of secrecies, we do not artificially select on the timing of licensing. These results support our hypothesis that the information role of exclusive licensing, as a signpost pointing to promising follow-on research paths, increases forward citations by nonlicensees. For inventions with multiple prior secrecy agreements, this informational role virtually disappears. We learn something further from this result: there is no evidence that our anticipated roadblock effect dominates when the informational effect is diminished

Technological Fields of Citing Patents Although licensing is recognized as a key step in the utilization of academic research, we have negligible systematic information about the effect of licensing on the paths followed in subsequent

distributions on the number of secrecy agreements between these CDM and CCEEM patents fails to reject the null hypothesis that they are similar. Thus the differential effects of exclusive licensing on nonlicensee citations between patents with multiple secrecy agreements vs. those without multiple secrecy agreements is unlikely due to the statistically insignificant difference in the technology mix of patents in these two subsamples.

17

research. To explore the direction of the first steps taken by follow-on innovators, we examine technology fields of the citing and cited patents. If a citing patent has a primary 3-digit US class different from that of the licensed patent, we further distinguish whether the two patents share the same field among the 37 broader technology subcategories presented in Hall, Jaffe and Trajtenberg (2001),44 denoted as HJT subcategories. As shown in Panel A of Table 5, the number of nonlicensee citations from the same US class as the licensed patent does not significantly increase after licensing. The significant increase in nonlicensee citations following licensing comes from different US classes, in particular from patents beyond the broader HJT subcategory of the licensed patent. A similar analysis of licensee citations yields starkly contrasting results, as shown in Panel B of Table 5. The increase in licensee citations after licensing is mainly attributable to increased licensee citations in the same US class. The number of licensee citations in different US classes (either those in different US classes but in the same broader HJT subcategory, or those in different HJT subcategories) does not change significantly. We conclude that exclusive licensing encourages nonlicensees to respond on research paths that, for the most part, lead to innovations in fields outside the broader HJT subcategory of the licensed UC/NL patent. Such induced innovations, being in distinct fields, are unlikely to infringe the academic patent. If “inventing around” a licensed patent is likely to yield patents that are in the same US class, it is not apparent in the data. 45 The finding that exclusive licensees, by contrast, respond with more research in the same narrow field might indicate that firms most likely to conduct research closest to the licensed patent tend to be willing to bid the most to become the exclusive licensee. This might suggest that researchers in the same narrow field tend to be more aware of the licensed research before licensing. It might alternatively indicate that exclusive licensing has a negative effect on competition by nonlicensees in the same field that offsets the informational effect. To the extent that the discouraged research in the same field would have constituted wasteful duplication, this might well be efficient. To the extent that it reduces complementary research that would have generated externalities to others, it might not.

44

Note that Hall, Jaffe and Trajtenberg (2001) classified patents into 36 technology subcategories. The NBER update added one more technology subcategory (https://sites.google.com/site/patentdataproject/). 45 This less directly related research might include projects consistent with a broad definition of “inventing around” the patent. If so, we conjecture that such research might be less likely to be wastefully duplicative than if it were in the same narrow field.

18

In Online Appendix F, we provide two illustrative examples from our data of responses of a licensee and a non-licensee to exclusive licensing.

Causality between Exclusive Licensing and Nonlicensee Citations To interpret an increase in nonlicensee citations following an exclusive license as information effects, we need to address whether the event of exclusive licensing is exogenous to nonlicensees. An alternative story is that interest in the UC/NL patent arises from many firms at a certain point in time. After one of them wins the license, other firms might decide to continue their follow-on research on the licensed patent, in which case an increase in nonlicensee citations could be observed after exclusive licensing, but there is no causal relationship between these two. In other words, a growing demand for (and interest in) the UC/NL patent might lead to both the licensing event and the observed increase in non-licensee citations after licensing. Given the embryonic and early stage nature of university inventions, this “growing demand” story might well be less plausible than it could be for private-sector inventions. We are informed by OTT that typically only a small number of potential licensees are interested in a particular invention. Even among them, the valuation of the invention tends to be heterogeneous and highly uncertain. Nevertheless, to test whether a growing demand for a UC/NL patent determines the timing of exclusive licensing of the patent, we conduct a survival analysis, using a Cox proportional hazards model (Cox 1972), to estimate the impact of the number of forward citations to a UC/NL patent in the baseline group up to period t after grant on the hazard rate of licensing in period t. To control for a differential effect of citations by technology field we include five technology field dummies following Hall et al (2001)46. To control for the quality and experience of researchers, we extract prior patenting experience of inventors of a UC/NL patent, using the disambiguated data from Lai et al. (2011). We include a dummy variable which takes the value of 1 if any of the inventors of UC/NL patent i has at least one prior patent.47 The results, shown in Table 6, suggest that there is no correlation between the number of forward citations (in various prior time windows) and the hazard rate and timing of exclusive licensing. We also address the possibility that increased interest in the technological field of an exclusively licensed patent induces its licensing, even if it does not induce extra citations to the patent before the

46

Hall et al (2005) show that different technology fields exhibit different forward citation patterns which can have a differential impact on firm value. 47 Azoulay et al (2010) and Li and Agha (2015) have shown that accounting for researcher quality can account for a large part of variation in the data.

19

event. To test this conjecture, we use two metrics for the general interest in the field of the licensed patent: (1) the number of patents filed at the USPTO in a given year that are in the same 3-digit US patent class as the licensed patent; and (2) the number of patents filed at the USPTO in a given year that cite patents in the 3-digit US patent class of the licensed patent. We run similar survival analyses, using the two measures of the general interest in the field of the UC/NL patent during certain prior time windows, rather than forward citations that a UC/NL patent receives during the time windows. The results, shown in Table 7, indicate that there is no correlation between the general interest in the field of a licensed patent and the timing of licensing. Together, the results in Tables 6 and 7 offer no support for an inference that licensing of the UC/NL patents in the baseline group is caused by an increase in general interest in these patents, or in the fields to which these patents belong. Several other results that we have obtained also cast doubt on the plausibility of the “growing demand” story as the predominant explanation for the increase in nonlicensee citations following licensing. First, nonlicensee citations after licensing do not significantly increase for patents with multiple secrecies. The number of secrecy agreements that an invention receives likely reflects its commercial potential and the degree of interest from the industry, as suggested by OTT staff and implied by the data. 48, 49 The information and signaling effect rationale can explain why the increase in nonlicensee citations following licensing is more significant for patents without multiple secrecies, which might well have received less interest from the industry prior to licensing. The event of licensing sends signals about the commercial potential in fields related to the invention. One might argue that the lag between “demand” for the invention and the licensing event could be greater than the lag between “demand” and follow-on citations, particularly for patents with higher “demand” and multiple secrecies. In this case, nonlicensee citations would likely occur before licensing, which could explain no significant increase in nonlicensee citations after licensing for patents with multiple secrecies. To test this possibility, we focus on periods before licensing and re-run Equation (1) using a dummy indicating a certain time window before the event of licensing, to estimate if there is any significant increase in nonlicensee citations before licensing. The results in Table 8 suggest that there is no significant uptick in nonlicensee citations either one year (Columns 1 and 4), or two years (Columns

OTT staff informed us that “more secrecy agreements is usually an indication of a higher commercial potential.” (quote from authors’ communication with OTT staff). 49 Among our analysis sample (patents in the baseline and comparison groups), patents with multiple secrecy agreements are licensed sooner than patents without, in terms of licensing lag from invention disclosure, as suggested by regressions of licensing lag on an indicator of multiple secrecies, controlling for patent characteristics, disclosure years and technology fields. 48

20

2 and 5), or three years before licensing (Columns 3 and 6), for patents with multiple secrecies and for those without. Second, such a “growing demand” story is inconsistent with the fact that the increase in nonlicensee citations after licensing comes mostly from patents in US patent classes beyond the broader HJT subcategory of the licensed patent, whereas the increased licensee citations are made from patents in the same class. If the “growing demand” story led to both the exclusive licensing of the UC/NL patent and the observed post-licensing increase in nonlicensee citations, we would not expect to see a difference between the technology fields of the increased nonlicensee and licensee citations. Third, we observe differences in the timing of nonlicensee and licensee citations following licensing. In Table 9, we show the time path of the effect of exclusive licensing in each of the fifteen years that follow licensing. The results in Columns 1 and 2 show that the increase in nonlicensee citations mainly occurs in the second, third and fourth year after licensing, consistent with the lag reflected. This result is perhaps more obvious when we bundle such dummies in Columns 3 and 4. As can be seen the increase occurs mainly within the first two to five years after licensing with some additional increase in years six to nine. By contrast, the increase in licensee citations not only occurs sooner (starting from the first year after licensing), but also covers a more extended period (through the thirteenth year following licensing). Even though our licensee result should be interpreted with caution, given the small sample, the observation that the increase in nonlicensee citations lags behind the initial increase in licensee citations once again casts doubt on the “growing demand” story. If nonlicensee citations increase after licensing because there were many firms interested in the invention, among which only one got the license but other firms (nonlicensees) still conducted follow-on research on the invention, then we would not expect nonlicensee citations to lag relative to licensee citations unless the licensee conducts and succeeds in follow-on research faster than nonlicensees. Nor would we expect increased nonlicensee citations to be in fields different from that of the licensed patent. On the other hand, such time lag in nonlicensee citations is consistent with our signaling story that the event of licensing draws nonlicensees’ attention to the invention. Note that although we consider a growing demand hypothesis to be implausible for the licensed UC/NL patents overall, there will no doubt be cases where the patents are licensed as the fields experience a sudden surge in interest.

Signaling of Exclusive Licenses vs. Signaling of Licensee Patents 21

We next consider the possibility that the information effect of licensing is generated not by the licensing event itself, but indirectly by the licensee’s follow-on patents. That is, nonlicensees do not realize the potential of the licensed UC/NL patent until they notice the fruit of the licensee’s follow-on research, as made public via the grant of the licensee’s patent citing the UC/NL patent (in the years prior to the introduction of application publication 18 months after priority filings). As mentioned earlier, more than three quarters of licensees did not obtain patents citing the licensed UC/NL patents after licensing. Thus, it is likely that the event of licensing, rather than the licensee’s successful follow-on research, and patenting, is the dominant source of signals motivating nonlicensees’ follow-on research. In Table 10 we examine whether there is an additional change in nonlicensee citations following the grant of the licensee’s citing patent. Column 1 shows the coefficient of the Exclusive_License indicator for the baseline patents, and Column 3 shows the result when the sample is restricted to the baseline patents that have licensee citations, i.e. patents for which the licensees have conducted successful follow-on research and obtained subsequent patents citing the licensed patents.50 For the latter subsample, the increase in nonlicensee citations following exclusive licensing is even greater in magnitude, though less significant due to the small sample size. In Columns 2 and 4, an indicator for whether the licensee’s first citing patent has been issued or not (Licensee_First_Cite) is added in the specifications to estimate the impacts on nonlicensees of successful licensee follow-on research, controlling for the effect of the licensing event.51 As shown in Columns 2 and 4 of Table 10, the coefficients for the Exclusive_License indicator remain similar, but the estimates for the Licensee_First_Cite indicator are insignificant, though positive. The results suggest that it is primarily the event of exclusive licensing, not the follow-on patenting by the licensee, that sends positive signals to other firms.52

Does Exclusivity Discourage Nonlicensee Citations? It is possible that the net positive effect of exclusive licensing hides a negative effect of exclusivity that is dominated by a positive informational effect. While the overall effect is informative,

50

We drop 4 patents that are only cited by the (future) licensee prior to licensing. Therefore, for these regressions we use 51 baseline patents that receive licensee citations. Results do not change if we include these 4 patents. 51 Note here that the year when the first licensee citation took place is excluded from the analyses, to be consistent with our treatment of the year of license. In unreported analyses, we include both the year of license and the year of first licensee citation and the results are similar. 52 We do not have enough information on “first sale,” i.e., successful commercialization, of licensed patents, to test whether successful commercialization of a licensed patent sends signals to nonlicensees.

22

the isolated effect of exclusivity also has policy relevance. We have already observed that the evidence from cases with multiple secrecy agreements do not support this conjecture. Here we consider a separate issue, whether nonexclusively licensed patents have a significantly different change in nonlicensee citations after first license. Table 11 involves regressions specified as Equation (1), except that the sample includes both exclusively and nonexclusively licensed patents and we use an interaction term, License*NonExclusive, to estimate the differential impact of licensing on nonlicensee citations if the license is nonexclusive.53 The results show the difference between nonexclusive licensing and exclusive licensing to be insignificant. They not only imply that it is the event of licensing that sends signals to the market, but suggest that the exclusivity of a license does not significantly discourage follow-on research by nonlicensees. If anything, the negative coefficients of the interaction term, License*NonExclusive, indicate that exclusive licensing might send more positive signals than nonexclusive licensing. The insignificant coefficients could be due to the relatively small sample size of the nonexclusively licensed patents.54 Future studies with larger samples of nonexclusively licensed patents could shed more light on this issue. Note also that these results need to be interpreted with caution, as the characteristics of nonexclusively and exclusively licensed inventions are likely to differ.

6.

Conclusion For the research laboratories and university campuses included in our empirical study, our results

strongly reject the consensus that exclusive licensing on balance discourages follow-on innovation by nonlicensees or diverts their efforts to research aimed at less socially useful “inventing around”. We find that citations to licensed patents in patents of nonlicensees increase significantly after exclusive licensing. We see no evidence that this increase is mainly generated by attempts to invent around the licensed patent. We find several types of evidence against the alternate hypothesis that a growing demand for (and interest in) the invention generates both the licensing event and this increase in nonlicensee citations. Given the strong positive changes in citations after issue of an exclusive license, we posit that an exclusive license sends a signal about the increased prospects for, or feasibility of, further innovation in It could be argued that a subset of potential “nonlicensees” in the case of nonexclusive licenses have signed licenses and therefore their citations will show up as licensee citations in our data. For this reason, we re-run specifications of Table 8 by pooling licensee and nonlicensee citations for the nonexclusively licensed patents. The results remain similar. 54 For example, Column 1 of Table 8 includes 56 non-exclusively licensed patents and 228 exclusively licensed patents, licensed after grant, with licenses maintained through patent expiration. 53

23

relevant fields. We test this signaling interpretation using the fact that some licenses in our sample are preceded by multiple secrecy agreements with potential licensees. Licensing after such information releases to multiple potential licensees is not associated with any significant increase in nonlicensee citations Our findings indicate that exclusive licensing of university patents and related technology transfer activities generates information externalities that increase innovation by nonlicensees. They offer support for claims that university patenting and technology transfer generates social benefits in addition to, and perhaps more important than, licensing royalties. More generally, they furnish an expanded empirical basis for arguments by defenders of the Bayh-Dole Act that systematic university patenting and technology transfer helps overcome a significant informational divide between the world of academia and the world of industry (Colyvas et al, 2002).55 An exclusive license is a signpost pointing out novel, attractive and unblocked innovation pathways through different research fields. Nonlicensees follow these pathways with successful, patented research. A recent study by Thompson, Mowery and Ziedonis (2013) finds positive effects of academic patent licensing on citations of journal publications, supporting our signaling hypothesis. A positive signaling role for exclusive licensing, so important that it dominates undoubted negative exclusionary effects on nonlicensees, runs counter to a highly intuitive consensus. However, an exclusive academic license establishes that another innovator has made a credible financial commitment, indicating perception of a substantial commercial opportunity. Such a commitment might well have much more influence on innovators than information revealed in the patent or academic publication. Some new institutional economics writers have seen the patent as analogous to a mining claim, which allows commercial development by preventing expropriation by competitors (Kitch 1977) or acts as a focal point or beacon that facilitates coordination of activities of commercialization including developers, investors, laborers and marketers (Kieff, 2005; Demsetz, 2002). Here we find evidence that exclusive licenses act as signposts pointing out, among a plethora of technical alternatives, innovative paths at the end of which nonlicensee innovators perceive commercially valuable opportunities.

55

Our paper, however, does not address the concern that university patenting and licensing might have deleterious effects on academic norms and practices including commercially motivated or related secrecy, delays in publication of research results, academic freedom, redirection of research efforts away from curiosity-driven topics toward applications with the prospect of financial returns, among others (Cohen et al., 1998; Washburn, 2005; Aghion, Dewatripont and Stein, 2009; Lach and Schankerman, 2008; and Merrill and Mazza, 2011). For the institutional role of the university on knowledge generation and diffusion consult Dasgupta and David (1994).

24

Indeed, our results imply that an exclusive license to a patent, like an exclusive miner’s claim, can have social value far beyond the exclusionary right emphasized by Kitch (1977). An early mineral claim in a given region might (like the exclusive licenses studied here) have a valuable signaling effect, in addition to and perhaps more valuable than the private exclusionary effect. For example, Edward Hammond Hargraves, commonly considered the founder of the gold rush in mid-nineteenth century Australia, made mining claims on land near Bathurst, New South Wales, that he recognized as goldbearing by analogy to the geology and landscape he had observed in the previous year in the neighborhood of successful Californian gold claims. The governments of the colonies of New South Wales and later Victoria revealed that they valued this signaling effect to be far higher than the expected private value of early mining claims, by offering prizes of ten thousand and five thousand pounds respectively for discovery of commercial quantities of gold. Hargraves, who won both prizes, openly stated that his aim was to make money from the prizes, not from his mining claims (See Blainey 1960). When governments are in a better position than private individuals to assess the full social value of an innovation, prizes can be particularly efficient (Wright 1983). Our finding of positive licensing externalities is consistent with the fact that academic inventors may be rewarded in prizes, promotions and prestige, in addition to (and in some cases far more valuable than) any share of any royalties from licensing. Our results support this practice with evidence that exclusive licenses, on balance, encourage further unlicensed innovation, the benefits of which are not captured in royalties. Finally, we must note that our findings come with several caveats. First, even though patent citations are in our view better measures of innovation responses than citations in publications, both are noisy metrics for follow-on research achievements. Second, secrecy agreements are just one imperfect measure of pre-license information releases and/or prior interest of potential licensees. Perhaps, following the lead of Roach and Cohen (2013), complementary surveys could further elucidate how exclusive licensing impacts nonlicensees’ decisions to conduct follow-on research related to licensed academic patents. Third, although our study includes three national research laboratories as well as nine research university campuses in the nation’s largest university system, we suspect that the results would generalize to other large public and private research universities in the United States. In particular, given that our data come from a top tier institution, one might speculate that the signaling effect of exclusive licensing of inventions from less prestigious research institutions could be even stronger. Replication in other institutional settings is clearly important, although achieving the collaboration and coordination

25

necessary to assemble an appropriate dataset is clearly a major challenge. Finally, whether our conclusions generalize to non-academic licensing is another interesting and important question.

References Adams, J. "Fundamental Stocks of Knowledge and Productivity Growth." Journal of Political Economy 98(1990): 673-702. Aghion, P., M. Dewatripont, and J. Stein. "Academic Freedom, Private Sector Focus, and the Process of Innovation." Rand Journal of Economics 39, no. 3 (2009): 617-635. Allison, P. D., & Waterman, R. P. (2002). Fixed–effects negative binomial regression models. Sociological methodology, 32(1), 247-265. Austin, D. H. "An Event-Study Approach to Measuring Innovative Output: The Case of Biotechnology." American Economic Review Papers and Proceedings 83, no. 2 (1993): 253-58. Azoulay, P., J. L. Furman, J. L. Krieger, and F. Murray. "Retractions." Review of Economics and Statistics 97, no. 5(2015): 1118-1136. Azoulay, P., J. S. Graff Zivin, and J. Wang. "Superstar Extinction." Quarterly Journal of Economics 125, no. 2(2010): 549-589. Belenzon, S., and M. Schankerman. "Spreading the Word: Geography, Policy and University Knowledge Diffusion." Review of Economics and Statistics. 95, no. 3 (2013): 884-903. Berman, E. P. Creating the Market University: How Academic Science Became an Economic Engine. Princeton University Press, 2011. Bertrand, M., E. Duflo, and S. Mullainathan. "How Much Should We Trust Differences-in-Differences Estimates?" Quarterly Journal of Economics 119, no. 1 (2004): 249-275. Blainey, G. “Gold and Governors.” Historical Studies: Australia and New Zealand 9, no. 36 (1961): 337-350. Boldrin, M., and D. K. Levine. "The Case Against Patents." Journal of Economic Perspectives 27, no. 1(2013): 322. Branstetter, L. G. "Are Knowledge Spillovers International or Intranational in Scope? Microeconometric Evidence from the U.S. and Japan." Journal of International Economics 53(2001): 53-79. Cohen, W. M. “Patents and Appropriation: Concerns and Evidence.” Journal of Technology Transfer, 30 (2005): 57-71. Cohen, W. M., R. Florida, L. Randazzese, and J. Walsh, “Industry and the Academy: Uneasy Partners in the Cause of Technological Advance,” in R. Noll, ed., The Future of the Research University, Brookings Institution Press, 1998. 26

Cohen, W.M., A. Goto, A. Nagata, R.K. Nelson, and J. P. Walsh, “R&D Spillover, Patents and the Incentives to Innovate in Japan and the United States.” Research Policy, 31 (2002), 1349-1367. Colyvas, J., M. Crow, A. Gelijns, R. Mazzoleni, R. R. Nelson, N. Rosenberg, and B. N. Sampat. "How Do University Inventions Get Into Practice?" Management Science 48, no. 1(2002): 61-72.

Cox, D. R. “Regression Models and Life-Tables.” Journal of the Royal Statistical Society. Series B (Methodological) (1972): 187-220. Dasgupta, P., and P. A. David. "Toward a New Economics of Science." Research Policy 23, no. 5(1994): 487521. Demsetz, H. "Toward a Theory of Property Rights II: The Competition between Private and Collective Ownership." The University of Chicago Law Review 31, no. S2(2002): 653-672. Eisenberg, R. S. "Public Research and Private Development: Patents and Technology Transfer in GovernmentSponsored Research." Virginia Law Review 82, no. 8(1996): 1663-1727. Elfenbein, D. W. "Publications, Patents, and the Market for University Inventions." Journal of Economic Behavior & Organization 63, no. 4(2007): 688-715. Fehder, D. C., F. Murray, and S. Stern. “Intellectual Property Rights and the Evolution of Scientific Journals as Knowledge Platforms.” International Journal of Industrial Organization 36 (2014): 83-94. Furman, J. L., and S. Stern. "Climbing atop the Shoulders of Giants: The Impact of Institutions on Cumulative Research." American Economic Review 101, no. 5(2011): 1933-63. Galasso, A and M. Schankerman. “Patents and Cumulative Innovation: Causal Evidence from the Courts.”

Quarterly Journal of Economics, 130, no. 1 (2015): 317-369. Gourieroux, C., Monfort, A., & Trognon, A. (1984). Pseudo maximum likelihood methods: Applications to Poisson models. Econometrica: Journal of the Econometric Society, 701-720. Graff, G. D., A. Heiman, and D. Zilberman. "University Research and Offices of Technology Transfer." California Management Review 45, no. 1(2002): 87-116 or 88-115. Green, J. R., and S. Scotchmer. "On the Division of Profit in Sequential Innovation." RAND Journal of Economics 26, no. 1(1995): 20-33. Green, W. H. "The Behavior of the Fixed Effects Estimator in Nonlinear Models." The Econometrics Journal 7, no. 1(2004): 98-119. Hall, B. H., A. B. Jaffe, and M. Trajtenberg. "The NBER Patent Citation Data File: Lessons, Insights and Methodological Tools." NBER Working Paper 8498 (2001). Hall, B. H., A. Jaffe, and M. Trajtenberg. "Market Value and Patent Citations." Rand Journal of Economics 36, no. 1(2005): 16-38. Heller, M. A., and R. S. Eisenberg. "Can Patents Deter Innovation? The Anticommons in Biomedical Research." Science 280, no. 5364(1998): 698-701. 27

Hellmann, T. F., and E. C. Perotti. "The Circulation of Ideas in Firms and Markets." Management Science 57, no. 10(2011): 1813-1826. Henry, M. R., M. C. Cho, M. A. Weaver, and J. F. Merz. "A Pilot Survey on the Licensing of DNA Inventions." Journal of Law, Medicine & Ethics 31(2003): 442-49. Hilbe, J. M. (2011). Negative binomial regression. Cambridge University Press. Jaffe, A. "Real Effects of Academic Research." American Economic Review 79(1989): 957-970. Jaffe, A., M. Trajtenberg, and R. Henderson. "Geographic Localization of Knowledge Spillovers as Evidenced by Patent Citations." Quarterly Journal of Economics 108, no. 3(1993): 577-598. Jaffe, A. B., M. Trajtenberg. Patents, Citations & Innovations: A Window on the Knowledge Economy: MIT Press, Cambridge MA, 2002. Jensen, P. H. and E. Webster. “Patents, Transaction Costs, and Academic Research Project Choice.” Economic Record 90 no. 289 (2014), 179-196. Jensen, R., and M. Thursby. "Proofs and Prototypes for Sale: The Licensing of University Inventions." American Economic Review 91, no. 1(2001): 240-259. Kieff, F. (2005) IP Transactions: On the Theory & Practice of Commercializing Innovation, Stanford Law and Economics Olin Working Paper. Kitch, E. “The Nature and Function of the Patent System.” Journal of Law and Economics, 20, no. 2 (1977), 265290. Koo, B., and B. D. Wright. "Dynamic Effects of Patent Policy on Sequential Innovation." Journal of Economics & Management Strategy 19, no. 2(2010): 489-512. Lach, S., and M. Schankerman. "Incentives and invention in universities." Rand Journal of Economics 39, no. 2(2008): 403-433. Lancaster, T. "The Incidental Parameter Problem Since 1948." Journal of Econometrics 95, no. 1(2000): 391-413. Lei, Z., R. Juneja, and B. D. Wright. "Patents Versus Patenting: Implications of Intellectual Property Protection for Biological Research." Nature Biotechnology 27, no. 1(2009): 36-40. Lei, Z. and B. D. Wright, “Why weak patents? Examiner ignorance or pro-applicant procedural rules?” Working paper, 2015. Lemley, M. A. and B. Sampat, “Examiner Characteristics and Patent Office Outcomes.” Review of Economics and Statistics, 94, no. 3 (2012): 817-827. Lerner, J. "The Importance of Patent Scope: An Empirical-Analysis." Rand Journal of Economics 25, no. 2(1994): 319-333. Levin, R. C., W. M. Cohen, and D. C. Mowery. "R & D Appropriability, Opportunity, and Market Structure: New Evidence on Some Schumpeterian Hypotheses." The American Economic Review 75, no. 2(1985): 20-24. 28

Levin, R. C., A. K. Klevorick, R. R. Nelson, W. A. G., R. Gilbert, and Z. Grilliches. "Appropriating the Returns from Industrial Research and Development." Brookings Papers on Economic Activity 3(1987). Li, D., & Agha, L. (2015). Big names or big ideas: Do peer-review panels select the best science proposals?. Science, 348(6233), 434-438. Merrill,S. A. and A. M. Mazza. “Managing University Intellectual Property in the Public Interest.” Committee on Management of University Intellectual Property: Lessons from a Generation of Experience, Research, and Dialogue. National Academies Press, 2011. Mowery, D. C., R. R. Nelson, B. N. Sampat, and A. A. Ziedonis. "The Growth of Patenting and Licensing by US Universities: An Assessment of the Effects of the Bayh-Dole Act of 1980." Research Policy 30, no. 1(2001): 99119. Mowery, D. C., and A. A. Ziedonis. "The Geographic Reach of Market and Non-Market Channels of Technology Transfer: Comparing Citations and Licenses of University Patents." National Bureau of Economic Research Working Paper Series No. 8568(2001). Murray, F. "The Oncomouse That Roared: Resistance and Accommodation to Patenting in Academic Science." Sloan School of Management Working Paper, (2006). Murray, F., and S. Stern. "Do Formal Intellectual Property Rights Hinder the Free Flow of Scientific Knowledge? An Empirical Test of the Anti-Commons Hypothesis." Journal of Economic Behavior & Organization 63, no. 4(2007): 648-687. Nelson, R. R. "The Market Economy, and the Scientific Commons." Research Policy 33, no. 3(2004): 455-471. Ouellette, L. "Do Patents Disclose Useful Information?" Harvard Journal of Law and Technology 25, no. 2(2012): 531-593. Peri, G. "Determinants of Knowledge Flows and Their Effect on Innovation." Review of Economics and Statistics 87, no. 2(2005): 308-322. Pressman, L., R. Burgess, R. M. Cook-Deegan, S. J. McCormack, I. Nami-Wolk, M. Soucy, and L. R. Walters. "The Licensing of DNA Patents by US Academic Institutions: An Empirical Survey." Nature Biotechnology 24, no. 1(2006): 31-39. Roach, M. and W. M. Cohen, “Lens or Prism? A Comparative Assessment of Patent Citations as a Measure of Knowledge Flows from Public Research.” Management Science, 59, no. 2 (2013): 504-535. Sampat, B. N. "Genomic Patenting by Academic Researchers: Bad for Science?" (2004). Sampat, B. N., and A. A. Ziedonis. "Patent Citations and the Economic Value of Patents." Handbook of quantitative science and technology research (2005): 277-298. Santos Silva, J. M. C., and S. Tenreyro. "On the existence of the maximum likelihood estimates in Poisson regression." Economics Letters 107(2010): 310–312.

29

Schacht, W. H. "The Bayh-Dole act: selected issues in patent policy and the commercialization of technology." Library of Congress Washington DC Congressional Research Service , DTIC Document (2012). Schissel, A., J. F. Merz, and M. C. Cho. "Survey Confirms Fears About Licensing of Genetic Tests." Nature 402, no. 6758(1999). Stolberg, S. G. “Suit Seeks to Expand Access to Stem Cells,” New York Times, August 14, 2001. Thompson, N. C., D. C. Mowery, and A. A. Ziedonis. " Does University Licensing Facilitate or Restrict the Flow of Knowledge and Research Inputs among Scientists?" UC Berkeley and University of Oregon, manuscript (2013). Thursby, J., and M. Thursby. "University Licensing and the Bayh-Dole Act." Science 301(2003): 1052-52. Trajtenberg, M. "A Penny for Your Quotes: Patent Citations and the Value of Innovations." RAND Journal of Economics 21, no. 1(1990): 172-187. Washburn, J. University, Inc.: The Corporate Corruption of Higher Education. 2005. Walsh, J.P., C. Cho and W.M. Cohen, “View from the Bench: Patents, Research and Material Transfers,” Science, September 23 (2005), pp. 2002-2003. Williams, H. L. "Intellectual Property Rights and Innovation: Evidence From the Human Genome." Journal of Political Economy, 2013, 121(1), 1-27. Wooldridge, J. M. (2005). Simple solutions to the initial conditions problem in dynamic, nonlinear panel data models with unobserved heterogeneity. Journal of applied econometrics, 20(1), 39-54.

30

Figure 1: Distribution of baseline patents (licensed exclusively after grant), by timing of license

Figure 2: Private nonlicensee Citations to baseline patents before licensing: early vs. late licensed patents

31

Figure 3A: Private nonlicensee citations to baseline patents, by years from licensing

Notes: Residuals from regressing via Poisson nonlicensee citations on patent fixed effects, year since grant dummies and calendar year dummies.

Figure 3B: Effects of the licensing event on private nonlicensee citations

Notes: The solid line represents coefficient estimates from a fixed effects Poisson regression in which the Private nonlicensee citations are regressed on year since grant dummies and citation calendar year dummies. The sample includes the baseline patents and the patents licensed within two years before grant. The regression further includes thirteen interaction terms between a dummy on whether a patent belongs in the baseline group and the number of years before after the licensing event. The 95% confidence interval (corresponding to robust standard errors, clustered around case codes) around these estimates is plotted with dashed lines.

32

Figure 4: Private nonlicensee citations to baseline patents, by years from licensing: patents without multiple secrecy agreements vs. patents with multiple secrecy agreements

Notes: Residuals from regressing via Poisson nonlicensee citations on patent fixed effects, year since grant dummies and calendar year dummies.

33

Table 1. Patent characteristics of exclusively licensed patents, by timing of license Panel A Patents early Patents late licensed licensed within at or after 6th 5 years after year after grant grant Patent Characteristics Claims

P-Value

Patents licensed after grant

Panel B Patents licensed within two years before grant

17.81 (14.28) 5.28 (3.69) 6.60 (6.32)

20.03 (17.34) 5.29 (3.90) 8.05 (7.56)

0.25

9.85 (13.52) 1995.18 (5.44) 2.46 (1.13) 1.70 (3.46) 0.42 (0.66)

11.51 (17.71) 1996.91 (4.91) 2.90 (1.46) 2.31 (3.65) 0.55 (0.78)

0.38

228

86

P-Value

18.71 (14.78) 5.45 (4.15) 7.41 (7.44)

16.70 (13.64) 5.07 (3.04) 5.59 (4.41)

0.29

8.40 (12.17) 1995.73 (5.51) 2.62 (1.32) 1.98 (4.17) 0.35 (0.60)

11.64 (14.89) 1994.49 (5.29) 2.26 (0.80) 1.36 (2.26) 0.50 (0.73)

0.07

#of patents Forward Citations Per Year

126

102

Total Forward Citations

2.19

1.92

0.068

2.14

2.06

0.61

(0.11) 1.55

(0.09) 1.46

0.47

(0.14) 1.62

(0.07) 1.51

0.41

(0.08) 0.32 (0.05)

(0.07) 0.08 (0.02)

(0.06) 0.21 (0.02)

(0.11) 0.05 (0.001)

US Classifications IPC Classifications Backward Patent Citations Grant Year Prosecution Duration Secrecy Agreements Letter Agreements

Private nonlicensee citations Licensee Citations # of patent-year obs

0.44 0.03

0.09 0.02 0.18 0.09

0

0.98 0.09

0.01 0.01 0.17 0.14

0

34

Private nonlicensee citations (Before licensing) # of patent-year obs

1,594

1,391

1.39 (0.14)

1.47 (0.09)

371

913

2,985

1,046

0.62

35

Table 2: Change in private nonlicensee citations after exclusive licensing

VARIABL ES

Dep. Variable: Private nonlicensee citations Panel A: All patents Panel B: Patents from federally sponsored research projects (1) (2) (3) (4) (5) (6) (7) (8) Baseline patents (patents exclusively licensed Baseline patents Baseline Baseline patents after grant) and patents patents and patents exclusively (patents exclusively licensed within exclusively licensed within two years before licensed two years before grant. after grant) grant. Linea CRE CNB UNB Fixed Effects Poisson (Columns 5-8) r Poisson Mode l

Exclusive_L 1.087 0.581** 0.379 0.394 icense ** * *** *** (0.48 (0.209) (0.126 (0.142 6) ) ) [0.55] [0.38] [0.20] Observation 2,545 s Number of 209 p

2,545 209

209

0.410***

0.474***

0.595***

0.578***

(0.153)

(0.146)

(0.187)

(0.178)

[0.21]

[0.24]

[0.19]

[0.24]

2,545

2,545

3,500

698

957

209

209

286

64

86

Notes: Patent fixed effects, year since grant dummies and citation calendar year dummies are included. Column 1-5 consider the “baseline patents”, defined as patents exclusively licensed after grant with licenses that continued until expiration of the full patent term and each Column employs a different estimator. Column 1 employs a linear model (OLS); Column 2 the CRE (correlated random effect) Poisson (it is estimated via Poisson Random Effects by further including a set of dummies that define at which period each patent was licensed – 13 dummies in total); Column 3 the CNB (conditional fixed effects negative binomial); Column 4 the UNB (unconditional fixed effects negative binomial); and Column 5 the fixed effects Poisson model. Columns 6-8 also use the latter model. Dependent variables are yearly private nonlicensee citations, cited by firms and individuals. Average marginal effects are in brackets for straightforward comparison with the linear model. Standard errors clustered by bundled patents in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%.

36

Table 3: Change in licensee citations after exclusive licensing Dep. Variable: Licensee citations (1) (2) (3) (4) Panel A: All patents Panel B: Patents from federally sponsored research projects Baseline patents Baseline patents and Baseline patents Baseline patents (patents patents exclusively (patents and patents exclusively licensed within two exclusively exclusively licensed years before grant. licensed licensed within after grant) after grant) two years before grant. Exclusive_License

# of Obs # of patents

0.507 (0.616)

1.629 (1.037)

0.266 (2.590)

0.512 (1.019)

693

1,003

143

236

51

76

13

21

Notes: All columns are estimated via Poisson. Patent fixed effects, year since grant dummies and citation calendar year dummies are included. Standard errors clustered by bundled patents in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Columns 1 and 3 include the “baseline patents,” defined as patents exclusively licensed after grant with licenses that continued until expiration of the full patent term. Columns 2 and 4 include the baseline patents and the comparison patents, patents exclusively licensed within two years before grant with licenses that continued until expiration of the full patent term. Dependent variables are yearly counts of licensee citations to exclusively licenses UC/NL patents.

37

Table 4: Multiple prior secrecy agreements and signaling effect of licensing on nonlicensee citations Dep. Variable: Private nonlicensee citations (1) (2) (3) (4) Panel A: Patents without multiple Panel B: Patents with multiple secrecy secrecy agreements prior to agreements prior to licensing licensing Baseline patents Baseline patents and Baseline patents Baseline patents (patents patents exclusively (patents and patents exclusively licensed within exclusively exclusively licensed two years before licensed licensed within after grant) grant. after grant) two years before grant. Exclusive_License

# of Obs # of patents

0.476*** (0.178)

0.540*** (0.168)

-0.00476 (0.153)

0.146 (0.179)

2,091 169

2,764 222

514 44

812 69

Notes: All columns are estimated via Poisson. Patent fixed effects, year since grant dummies and citation calendar year dummies are included. Standard errors clustered by bundled patents in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Columns 1 and 3 include the “baseline patents” defined as patents exclusively licensed after grant with licenses that continued until expiration of the full patent term – their difference is that Column 1 includes only patents that have no or one secrecy prior to licensing while Column 3 has multiple secrecies prior to licensing. Columns 2 and 4 include the baseline patents and the comparison patents, patents exclusively licensed within two years before grant with licenses that continued until expiration of the full patent term – their difference is the same as above. Dependent variables are yearly counts of private nonlicensee citations, cited by firms and individuals.

38

Table 5: Technological fields of citing patents, relative to licensed patents (1) From same US class as licensed patent

(2)

(3) Citations From US classes From different US different from classes but same HJT licensed patent subcategory as licensed patent

(4) From different HJT subcategories

Panel A: Private nonlicensee citations Exclusive_License

# of Obs # of patents

0.177 (0.270)

0.365** (0.147)

0.211 (0.133)

0.442** (0.175)

2,074 169

2,380 192

1,895 149

2,052 166

2.608* (1.468)

-0.0756 (0.754)

-0.265 (0.922)

0.337 (0.581)

484 36

464 33

260 18

365 26

Panel B: Licensee citations Exclusive_License

# of Obs # of patents

Notes: All columns are Poisson estimates. Patent fixed effects, year since grant dummies and citation calendar year dummies are included. Standard errors clustered by bundled patents in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. The data involve the baseline patents, patents exclusively licensed after grant. See also note to Table 2.

39

Table 6: Survival analyses: forward citations and timing of exclusive licensing VARIABLES Citations in periods t, t-1

(1)

(2)

(3)

(4)

0.0123 (0.0154)

Citations in periods t, t-1, t-2

0.0103 (0.0119)

Citations in periods t, t-1, t-2, t-3

0.0115 (0.00939)

Citations in periods t-1, t-2

0.0125 (0.0160)

Citations in periods t-1, t-2, t-3

Chemical Computer & Communications Drugs & Medical Electrical & Electronic Mechanical Dummy_PriorPatExp

Observations

(5)

0.0133 (0.0118)

-0.310 (0.598) 0.137 (0.532) 0.159 (0.570) -0.235 (0.535) -0.156 (0.636) 0.446*** (0.165)

0.353 (0.356) 0.881*** (0.305) 0.788** (0.361) 0.364 (0.290) 0.502 (0.525) 0.435** (0.174)

0.0208 (0.297) 0.706*** (0.253) 0.469* (0.265) 0.0819 (0.291) 0.0570 (0.522) 0.461** (0.194)

0.359 (0.353) 0.889*** (0.301) 0.793** (0.360) 0.378 (0.281) 0.538 (0.503) 0.438** (0.173)

0.0290 (0.297) 0.714*** (0.253) 0.475* (0.265) 0.100 (0.289) 0.102 (0.506) 0.463** (0.193)

1,284

1,056

858

1,056

858

Note: The regressions include additional controls: # of claims, # of patent references, # of nonpatent references, # of IPC classes, # of USPC classes, application year dummies, five technology field dummies per Hall et al (2001) and a dummy that is equal to 1 if any of the inventors in the patent had any prior patenting experience and 0 otherwise; they are omitted for the sake of space. Standard errors clustered in patent level in parentheses, *** p<0.01, ** p<0.05, * p<0.1. We use the Cox proportional hazards model and estimate the probability that a UC/NL patent i is licensed at period t (counted by years here), given that it has not been licensed up until t-1. A patent enters the dataset at the year of grant and exits with the first license. The key explanatory variables are the number of forward citations that the patent receives during a certain time window up to period t.

40

Table 7. Survival analyses: Patents in fields or citing fields of licensed UC patents and timing of exclusive licensing (1) # of patents in field of licensed patent in periods t, t-1 in periods t, t-1, t-2 in periods t, t-1, t-2, t-3 in periods t-1, t-2 in periods t-1, t-2, t-3

# of patents citing field of licensed patent in periods t, t-1 in periods t, t-1, t-2 in periods t, t-1, t-2, t-3 in periods t-1, t-2 in periods t-1, t-2, t-3

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

-12.7 (22.4) 13.5 (14.8) 17.7 (12.0) 20.2 (22.2) 24.2 (15.9)

-0.66 (1.11) -3.23 (7.15) 7.59 (5.56) 6.25 (0.1) 0.13 (7.24)

Observations 1,284 1,056 858 1,056 858 1,284 1,056 858 1,056 858 Notes: Both coefficients and standard errors are expressed in millions (i.e. they have been multiplied by 10^6). Standard errors clustered at patent level in parentheses, *** p<0.01, ** p<0.05, * p<0.1. We use the Cox proportional hazards model and estimate the probability that UC/NL patent i is licensed at period t (counted by years here), given that it has not been licensed up until t-1. A patent enters the dataset at the year of grant and exits with the first license. The regressions include additional controls: # of claims, # of patent references, # of nonpatent references, # of IPC classes, # of USPC classes and application year dummies The key explanatory variables are the number of patents filed in the same field (based on 3-digit primary US class) as a UC/NL patent in the baseline group, and the number of patents citing patents in the same field as the UC/NL patent, during a given time window up to period t.

41

Table 8. Private nonlicensee citations immediately before exclusive licensing Dep. Variable: Private nonlicensee citations (1) (2) (3) (4) (5) (6) Panel A: Patents without multiple secrecy Panel B: Patents with multiple secrecy agreements prior to licensing agreements prior to licensing Before_Licens e[-1]

-0.0341

0.00532

(0.152)

(0.176)

Before_Licens e[-2,-1]

-0.0593

-0.0277

(0.146) Before_Licens e[-3, -2,-1]

# of Obs # of patents

889 138

889 138

(0.152) -0.0463

0.0302

(0.183)

(0.252)

889 138

217 37

217 37

217 37

Notes: All columns are Poisson estimates. Patent fixed effects, year since grant dummies and citation calendar year dummies are included. Standard errors clustered by bundled patents in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. We run Equation (1) looking only at periods before licensing and using a dummy variable Before_License indicating a time window before the licensing, in place of the Exclusive_License dummy. Before_License[-1] takes the value of 1 for one period before licensing and zero otherwise. Before_License[-2,-1] takes the value of 1 for the two periods before licensing and zero otherwise. Before_License[-3,-2,-1] takes the value of 1 for the three periods before licensing and zero otherwise. Dependent variables are yearly private nonlicensee citations, cited by firms and individuals.

42

Table 9: Time path in the effects of exclusive licensing on nonlicensee citations

(1) Baseline patents

ExcLic_1 ExcLic_2 ExcLic_3 ExcLic_4 ExcLic_5 ExcLic_6 ExcLic_7 ExcLic_8 ExcLic_9 ExcLic_10 ExcLic_11 ExcLic_12 ExcLic_13 ExcLic_14

0.132 (0.139) 0.438** (0.197) 0.485** (0.206) 0.587** (0.241) 0.469* (0.275) 0.566* (0.323) -0.0551 (0.358) 0.135 (0.372) -0.127 (0.426) -0.210 (0.493) -0.0493 (0.679) -0.376 (0.844) -0.0133 (0.748) 0.397 (1.076)

(2) (3) (4) Panel A: Private nonlicensee citations Baseline patents and Baseline Baseline patents patents exclusively patents and patents licensed within exclusively two years before licensed within grant two years before grant 0.254* (0.149) 0.587*** (0.178) 0.657*** (0.174) 0.802*** (0.181) 0.683*** (0.201) 0.843*** (0.231) 0.301 (0.248) 0.621** (0.256) 0.386 (0.324) 0.407 (0.364) 0.600 (0.598) 0.441 (0.731) 0.796 (0.640) 1.226 (0.970)

ExcLic_2_5 ExcLic_6_9 ExcLic_10_1 4

Observations Number of patents

2,605 209

3,562 286

(5) (6) Panel B: Licensee Citations Baseline Baseline patents patents and patents exclusively licensed within two years before grant

0.147 (0.140)

0.244 (0.154)

1.280* (0.776)

1.721* (0.966)

0.505*** (0.184) 0.269 (0.284) -0.0111

0.650*** (0.161) 0.570*** (0.211) 0.532

0.782 (0.608) 1.325** (0.626) 1.792**

1.733** (0.862) 2.509** (1.004) 3.189***

(0.465)

(0.409)

(0.756)

(1.111)

2,605 209

3,562 286

708 51

1,012 76

Notes: the specifications are the same as in Table 2. ExcLic_i (i=1,2,.. 14) is an indicator for the ith year after exclusive licensing. * significant at 10%; ** significant at 5%; *** significant at 1%.

43

Table 10: Effect of licensee cites on nonlicensee citations Dep. Variable: Private nonlicensee citations (2) (3) (4) Panel A: Panel B: All baseline patents Baseline Patents with at least (patents exclusively licensed one licensee citation after after grant) licensing (1)

Exclusive_License

0.413*** (0.153)

0.405*** (0.151) 0.303 (0.194)

0.576* (0.315)

0.564* (0.315) 0.290 (0.236)

2,500

2,450

674

624

206

206

50

50

Licensee_First_Cite

Observations Number of patents

Notes: All columns are Poisson estimates. Patent fixed effects, year since grant dummies and citation calendar year dummies are included. Standard errors clustered by bundled patents in parentheses. Licensee’s First Cite takes the value of 1 for every period after the licensee’s first patent that has cited the licensed UC/NL patent has been issued and zero otherwise. * significant at 10%; ** significant at 5%; *** significant at 1%. Dependent variables are yearly counts of nonlicensee citations to an exclusively licensed UC?NL patent.

44

Table 11: Change in private nonlicensee citations after licensing, nonexclusive vs. exclusive Dep. Variable: Private nonlicensee citations (1) (2) (3) (4) Panel A: All patents Panel B: Patents from federally sponsored research projects Patents exclusively Patents exclusively or Patents Patents exclusively or nonexclusively nonexclusively exclusively or or nonexclusively licensed, licensed, nonexclusively licensed, after grant after grant or within licensed, after grant, or two years before grant after grant within two years before grant License

License*NonExclusi ve

# of Obs # of patents

0.453*** (0.145)

0.407*** (0.148)

0.756*** (0.190)

0.743*** (0.190)

-0.141

-0.130

-0.215

-0.227

(0.276)

(0.271)

(0.301)

(0.285)

3,153 257

4,426 358

841 77

1,162 105

Notes: All columns are Poisson estimates. Patent fixed effects, year since grant dummies and citation calendar year dummies are included. Standard errors clustered by bundled patents in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Columns 1 and 3 include the “baseline patents,” defined as patents exclusively licensed after grant with licenses that continued until expiration of the full patent term. Columns 2 and 4 include the baseline patents and the comparison patents, patents exclusively licensed within two years before grant with licenses that continued until expiration of the full patent term. Dependent variables are yearly counts of private nonlicensee citations, by firms and individuals.

45

46