The influence of patent ownership rights on academic invention: Evidence from a natural experiment

Transcription

1 The influence of patent ownership rights on academic invention: Evidence from a natural experiment Dirk Czarnitzki, KU Leuven & ZEW Mannheim Thorsten Doherr, ZEW Mannheim Katrin Hussinger, U Luxembourg, KU Leuven & ZEW Mannheim Paula Schliessler, ZEW Mannheim & KU Leuven Andrew A. Toole, USDA & ZEW Mannheim Abstract In this paper we consider the example of Germany to examine how the patent rights assignment influences the incentives for academic researchers to produce and publicly disclose new knowledge. We establish a causal relationship between patenting output by individual German professors and the abandonment of the professor s privilege by exploiting the institutional structure of the public research system in Germany along with the researcher-level exogeneity of the 2002 change in Germany s employment law. Our results show that German professors, particularly those with some experience in commercializing their IP, have reacted to the abolishment of the professor s privilege by reducing their patenting activity. The results question the effectiveness of the law change. 1

2 1. Introduction In Germany, ownership rights of university inventions have been subject to a legal change about a decade ago. In February 2002, the German government changed clause 42 in the employer invention law, 1 the so-called professor s privilege. Once established based on Article 5 of the German constitution which protects the freedom of science and research this clause presented an exemption from the employer invention law by granting university scientists - as the only occupational group in Germany - the privilege to retain the ownership rights to their inventions. The law change of 2002 transferred the ownership rights of university inventions from the inventor to the university with the intention to increase the exploitation of university inventions towards commercial ends. There was the belief that the efforts, costs and financial risks involved would prevent university scientists from patenting. The transfer of the ownership rights was intended to increase the number of patents that originate from universities. Whether the abolishment of the professor s privilege would reach its goal was discussed controversially at the time of its introduction (Kilger and Bartenbach 2002, Rösler 2001). Now, more than a decade later, it is time to investigate the actual effects of the law change on patenting activities of German professors. Intellectual property (IP) policies are among the most powerful instruments available when considering increasing the incentives for the commercialization of inventions. In particular, the U.S. Bayh-Dole Act of 1980 was perhaps the most influential and far-reaching of these IP policies. For U.S. universities, this legislation clarified and facilitated the institutional ownership of inventions discovered by researchers who were supported by federal funds. The key component of the Bayh- Dole model is granting the university, not the government, ownership rights to patentable inventions discovered using public research funds (Crespi et al. 2006, Geuna and Nesta 2006, Kenney and Patton 2009). Many observers credit the Bayh-Dole Act with spurring university patenting and licensing that, in turn, stimulated innovation and entrepreneurship (The Economist 2002; OECD 2003; Stevens 2004). Based on this perceived success, the Bayh-Dole Act has evolved into a model of university IP 1 42 ArbnErfG of the employer invention law ( Arbeitnehmererfindergesetz (ArbnErfG)). 2

3 policy that has been emulated and debated in many countries around the world including Germany, Denmark, Japan, China, and others (OECD 2003; Mowery and Sampat 2005; So et al. 2008). However, the incentive effects on academic researchers of different IP ownership models are not well understood. This is surprising because academic researchers are the most important players in university-industry technology transfer. These individuals perform the research that creates new discoveries, decide how and if those discoveries are publicly disclosed, often participate in further development of patented discoveries, and may participate in spin-off companies as a form of commercialization (Thursby and Thursby 2008). In this paper we consider the example of Germany to examine how the patent rights assignment influences the incentives for academic researchers to produce and publicly disclose new knowledge. Other than in the Bayh-Dole model of the US, where patent rights have been transferred from government ownership to the university, Germany has decided to transfer patent rights ownership from the individual researchers to the university starting in 2002, the law change commonly referred to as the abolishment of the professor s privilege. Our central contribution is establishing a causal relationship between patenting output by individual German professors and the abandonment of the professor s privilege by exploiting the institutional structure of the public research system in Germany along with the researcher-level exogeneity of the 2002 change in Germany s employment law. Using a unique database covering the lifetime patenting record of individual researchers, we implement difference-in-difference estimators of patenting output measures using treated and control groups of academic researchers before and after Germany s revocation of the professor s privilege in The treatment group are university professors that are exogenously affected by the abandonment of the professor s privilege, while the control group consists of public research organization researchers that never had this professor s privilege. 3

4 The aim of this study is to evaluate the impact of the policy change on the patenting behavior at the individual researcher level, to identify characteristics affecting the degree to which a researcher reacts to the new law, and to derive conclusions on the overall effectiveness of the policy change with respect to the goals pursued by the German policymakers. The remainder of this paper is structured as follows. Section 2 gives an introduction to the background and implementation of the law change in Germany. Section 3 summarizes the empirical studies evaluating the impact of changes in ownership regulations of university-invented IP. While most of this literature is US-based, some recent contributions also focus on European countries. We also review the empirical literature for Germany. Section 4 introduces the empirical approach, the data collection strategy and provides the empirical results. Section 5 provides several robustness checks. Section 6 concludes with a discussion of the results and implications for policymakers. 2. The Policy Change: The abandonment of the professor s privilege in Germany The abolishment of the professor s privilege was the result of a general shift of the German science and technology policy towards promoting industry-science interaction. In response to the ongoing perception that Germany - and Europe in general - was lagging behind in transforming basic research into commercial value 2 the German government launched a comprehensive action program called Knowledge creates markets.3 This program addressed various aspects of industry-science interaction covering science-industry knowledge transfer, science-based spin-offs, industry-science collaboration, and the exploitation of scientific knowledge in the private sector. The abolishment of the professor s privilege was the most drastic step undertaken to facilitate the commercialization of university generated knowledge. 2 The so-called European paradox (European Commission 1995). 3 Wissen schafft Märkte, see Bundesministerium für Bildung und Forschung and Bundesministerium für Wirtschaft und Technologie (2001). 4

5 The intention of the law change was to spur the patenting outcome of universities. Policy makers took the low patenting activities of universities prior to 2002 as an indication for missing commercialization activities of scientific knowledge (Becher et al. 1996). 90% of all inventions filed as patent applications before the German Patent and Trademark Office stem from employees of firms or non-university research institutes (Kilger und Bartenbach 2002, p. 1173). However, instead of applying for a patent themselves university researchers often relied on industry partners to carry out the commercialization part of the joint projects (Verspagen 2006). The low patenting activities of university professors were attributed to the costly and time-consuming patent application procedures that had to be carried out and financed by the inventor himself. The government s decision to transfer the ownership rights of academic inventions to the university implies that from February 7th 2002 onwards academic inventors have to report their inventions made after February 6th 2002 to the university, unless they decide not to make them public at all. The university then gets four months to consider the invention for patenting. In case the university does not see a possibility for commercialization the inventor should be notified and is then given the right to pursue commercialization himself. In case the university does claim the invention the inventor receives 30% of the revenues obtained from licensing or other types of commercialization. In turn, the university takes care of the patenting process and pays all expenses occurring within the application process of the patent, such as fees, translation costs and lawyer fees. The inventor has the right to disclose his research results within academia two months after communication with the university, entitling him to publish his research results in scientific publications. 4 As German universities had little experience regarding technology transfer activities the government decided to outsource the patent valuation process to regional patent valuation agencies (PVAs). The establishment of PVAs was supported with a budget of 46.2 million EUR to be used before the end of 2004 (Kilger und Bartenbach 2002). Up until now, there are 29 autonomous PVAs established in 4 If the university inventor had made prior contractual agreements with a third party such as an industry partner, there was transition period granted. Contracts made before July 18 th 2001 were to be treated under the old law until February 2003 ( 43 ArbnErfG). 5

6 Germany that each serves a regional university network and employs experts specialized in these universities research areas. The PVA is supposed to support the entire process from screening the invention and applying for patent protection to organizing its commercialization process and to promote collaboration between universities and industry. Nowadays, it is clear that the PVAs will not become profitable. After an initial surge, revenue growth stagnates since In 2010, PVAs earned 4.9 millionen EUR but had to be publicly supported by 8.5 million in order to be able to continue their operations (Cuntz et al. 2012). In addition to the financial aspect, industry partners and some universities are not satisfied with the services of the PVAs. Especially, for existing industry-science collaborations PVAs appear as a disturbing factor (Frank et al., 2007). Both the profitability of the PVAs and the possibility that the involvement of third party creates a burden rather than helpful was already feared when the law change was enacted (Kilger and Bartenbach 2002). When deriving hypotheses on how the German law changes alters incentives to pursue research activities that lead to patentable output the particularities of the German law have to be taken into account. Other than the US change in law (the Bayh-Dole Act) that transfers the IP right from the government to the individual universities, allowing them to pursue commercialization independently and to create remuneration schemes for the inventors, the German policy change transfers the IP rights from the individual inventors to the university. While in the US case it is straightforward that the policy change should increase licensing possibilities for inventors and thus incentives for patentable research, the German case is more complex. Whether the incentive to pursue patentable research activities increases with a transfer of IP rights from the inventor to the university depends on whether the situation of the particular scientist improves compared to the situation prior to the law change. The financial position of the scientists depends on both benefits and costs he has to bear. After the abandonment of the professor s privilege in Germany, an academic inventor has the right to only 30% of the revenues generated from his patent. The royalty share thus decreases compared to the pre-law change situation when the 6

7 inventor was allowed to negotiate directly with a firm and did not have to split the revenues with the university. From this argument we expect a negative impact on patenting activity as the incentive for patenting reduces. At the same time, however, the costs for the patent application had to be covered by the individual researcher prior to the change in law while now the TTO or the newly established patent valuation covers all application fees and takes care of the patent application process. Costs Benefits the inventor royalty share drops from 100% (inventor ownership) to 30% of the revenues -> royalty share decreases need to involve the TTO in negotiations with industry partners third party in negotiations application fees covered by university technology transfer office manages the process and offers assistance for industry contacts/technology transfer less search costs for finding commercialization partners, see Hellman (2007) 3. Existing empirical studies on the effects of the law change in Germany Several publications have shown how the change in the employer-invention law as well as differences in the universities experience with technology transfer has affected ownership patterns of academic patents. Schmoch (2007) was the first to conduct a large scale quantitative analysis of the patenting activity at German universities after the law change in Classifying those patents as university inventions 7

8 that list a professor as inventor and assuming a constant share of non-professor inventors at universities he shows that the number of university-owned patent applications increased while inventions applied for privately by university inventors have decreased. He additionally shows that the share of industry assignments of university inventions has strongly decreased from 1999 onwards. Regarding the individual professors Schmoch shows that several highly active inventors have been discouraged by the change in law while others, previously non-patenting professors have been encouraged, such that he concludes a substitution effect has taken place. A follow-up study aimed at the evaluation of the overall effect of the law change and the interconnected establishment of patent valuation agencies in Germany (Cuntz et al. 2012) provides support for these results. Based on data provided by Fraunhofer ISI (similar source as in Schmoch 2007) this study shows that the share of university-owned university inventions has increased after 2002 while the share of individually or industry- owned university inventions has decreased. They further show that the overall number of patented university inventions has decreased starting in 1999, even prior to the abandonment of the professor s privilege. In a recent publication Von Proff et al. (2012) examine the patent assignment patterns of universityinventions by German professors prior and post the policy change. For this analysis they create a database of all professors that are listed at least once as inventor of a university owned patent between 1991 and 2006 and supplement this with information on the year of PhD, field of research and university-level data. For these professors a manual search of their entire patenting record has been conducted, including the patents assigned to entities other than the university. They find that the share of university-invented patents increases over time, but is not characterized by a structural break in They do however find that among all university-invented patents, the share of university-owned patents increases after 2002, which shows the expected effect of the law change on the assignment pattern. The authors further estimate multinomial logit regressions at the patent level to explain the ownership of university-invented patents, testing whether university ownership gained importance vis-à-vis firm, PRO or individual ownership, whether the patenting experience of 8

9 the university affects the assignment choice and whether the impact of the law change has a stronger effect on inexperienced researchers choices of assignment. They find that university ownership in general increases for all university-invented patents, and that inventions from universities with more patenting experience are more likely to be assigned to the university. They further find that first time patent inventors are affected more strongly by the law change: the likelihood of their patents being assigned to university increases more than for experienced researchers. In order to make sure that these results are not driven by the exclusion of professors never patenting with a university assignee the analysis is replicated using the population of professors at 6 universities, for which staff lists have been linked to patenting data. The results show a similar overall effect of the law change on the likelihood of university assignment, but no longer show an effect with respect to the researcher s experience. These results suggest that the overall number of patents from university inventors has not increased, but resulted in a shift of ownership of the patents. Czarnitzki et al (2011) provide preliminary results (due to a short time span with available data on patenting after 2002) on the impact of the change in law on the quality of academic patents, suggesting a drop in quality after the abandonment of the professor s privilege due to a reorientation towards scientific publication effort. Regarding the evaluation of the patent valuation agencies Cuntz et al. (2012) find the patent applications made by patent valuation agencies to be rather constant since 2004 while the revues have increased. Nevertheless the money spent to support these agencies still by far exceeds their revenues, which is why Cuntz et al. (2012) concludes that patent valuation agencies cannot yet be seen as a financing source for universities. In a study on university industry cooperation by Stifterverband für die deutsche Wissenschaft interviews with industry and university partners engaged in industry-science cooperation have revealed that patent valuation agencies are often seen rather critical as they introduce another party 9

10 in the negotiations regarding research output and commercialization, thereby complicating the commercialization process (Frank et al 2007, p. 20). 4. Empirical Implementation 4.1. Exploiting Germany s Natural Experiment The empirical analysis uses a natural experiment to identify the causal effects of transferring patent rights to the university. In particular, we implement difference-in-difference estimators using treated and control groups of academic researchers before and after Germany s revocation of the professor s privilege in Our natural experiment exploits the institutional structure of the public research system in Germany along with the researcher-level exogeneity of the 2002 change in Germany s employment law. Germany s public research enterprise consists of universities and a set of other public research organizations (PROs). Prior to 2002, university academic researchers retained the patent rights to their inventions and these rights allowed them to decide whether or not to seek a patent and how to commercialize their discoveries (e.g. by partnering with a firm or not). However, in Germany s PROs (Fraunhofer Society, Max Planck Society, Helmholtz Association, Leibniz Association), academic researchers did not have this professor s privilege. The patent rights to their inventions were owned by the PRO. While university employees were thus affected by the change in law the group of public research institute employees this constitutes the control group that was not affected by this exogenous change. 4.2 Data Collection As the focus of this research is on the incentive effects of the IP policy change on patenting we focus on the population of academic researchers (university and PRO) in Germany who have been an inventor of at least one patent at the German or European Patent Office between 1973 and

11 This is a subpopulation of all academic researchers in Germany. The broader population would include academic researchers who only publish. However, the IP policy change does not impact researchers who never show an interest in commercialization over the entire time period. The goal of the data collection is to generate a panel dataset of academic researchers, both professors and public research organization employees, in Germany that maps these researchers lifetime scientific output as measured by patents and publications 5, independent of their employer. This means that changes in residence and employers will be taken into account. Table 1 summarizes the data collection procedure. The resulting dataset contains the variables listed in table 2. The level of observation is a researcher-year. TABLE 1 DATA COLLECTION PROCEDURE 1. Search European Patent Office and the German Federal Patent and Trademark Office databases for every university faculty member that was an inventor on at least one patent. These individuals are identified by their professor doctor designation in the inventor name field. This method was validated and used in prior research (see e.g. Czarnitzki et al. 2007, 2009) Search European Patent Office and the German Federal Patent and Trademark Office databases for patents applied for by any of the German Public Research Organizations (PROs). The list of PRO institutes has been obtained from the Bundesberichts Forschung und Innovation 2012 by the Federal government. From these patents keep those that list the public research organization as only applicant. Take the inventor names of these patents. These inventors are assumed to be PRO inventors as we restrict the patents to those applied for only by the PRO (These are 70% of all 5 We collect publication data in order to be able conduct a conditional difference-in-difference estimation as robustness check and we use publications in our matching procedure. 6 One concern may be the assignment of the professor tag to honorary professors who are not comparable to regular university professors. However, the number of these honorary professors is quite low. In addition wrongly assigning the professor tag to these researchers will only strengthen our results. If we still find a significant treatment effect this indicates that our result even holds when adding some wrong professors. Therefore there is no need for concern here. 11

12 patents involving a PRO). We apply this restriction as we want to avoid having industry researchers cooperating with PROS to be classified as PRO researchers. 3. Pool the professors and PRO members obtained in step 1 and 2 and use the inventor mobility method to (Appendix A) obtain these inventors lifetime record of patenting activity. This method searches the entire patent databases from 1978 onwards to obtain similar names and to pool inventors with the same name as being the same person based on information on common applicant names and addresses, IPC classifications, co-inventors and citations to tackle the homonymy problem. Some of the professors will turn out to also appear as PRO researcher at some point in time. We drop those researchers associated with both institutions. By doing so, we exclude those researchers for which we do not know which IP policy is binding, the policy of the university or the policy of the PRO. In the end, we are able to identify single inventors with their lifetime patenting record. 4. List of inventors from step 3 is used to perform name searches in the Thomson Reuters Web of Science publication database, : First retrieve all publications from Web of Science that match with respect to the names in our inventor list, have at least one German affiliation. Second, stick to these authors from Web of Science, leaving aside the patent data, and harmonize them in a similar way as for patent data using information on journals, coauthors, citations and affiliations. 5. Link the list of unique authors to the list of unique inventors: Apart from the names we can only use the affiliations appearing as applicant of the patent and the affiliations listed in the published papers as matching criterion. 12

13 6. We generate a panel dataset for the unique inventor-researchers, including information on their patent, citation-weighted patent and publication count for each year. We count patents at the family level. The unit of observation is a researcher-year. We restrict the observation period from 1995 to We define entry into research as the year in which a researcher first appeared as a patent inventor or author of a publication. We drop researchers that enter after 2002 as these are not affected by the law change. This implies that we have an unbalanced panel. While this entry criterion seems reasonable, assuming that a researcher s career starts with his first public disclosures of research results, it is more difficult to define the end of a career, the exit from science, as we have no information on the researcher s real age. Exiting the researchers after they do not appear in publications or patents anymore is problematic, as this might be a result of the policy change. We therefore propose two different exit criteria: a. Assuming a career length of 35 years we exit the researchers 35 years after their first appearance b. We exit the researchers if they do not have a publication in 5 subsequent years. This is done in order to make sure that the researcher did not leave academia and to exit those that did. 4.3 Empirical Approach In order to establish a causal link between university ownership of IP and changes in the production and disclosure of academic research, we implement micro-econometric treatment effects models. In particular we apply difference in difference estimators in order to identify the causal effect of the law change. The difference-in-difference approach is based on estimating the counter-factual situation as the outcome of the treated firm in a period prior to the treatment. It is estimated as the difference between the outcome of the treated firm after the treatment and the outcome of the treated firm prior to its treatment. In order to control for general macroeconomic changes a second difference is 7 We do however keep those researchers that entered science before this date and also include their patenting and publication record prior to This implies that a researcher does not necessarily need to have a patent in this particular time period to appear in our sample. 13

14 calculated between the outcomes of a non-treated control sample observed for the same time period. The DiD estimator is then calculated as the double difference. The advantages of the DiD approach are straightforward: both common macroeconomic trends and constant individual-specific unobserved effects can be controlled for. In our case the treatment (law change) is exogenous such that we do not need to worry about individuals self-selecting into the treatment group. It is however necessary to control for individual characteristics of the researchers in order to account for general differences in the composition of the researchers of the two groups. The difference-in-estimator is implemented estimating count data fixed effects panel models of patenting output. We chose count data models in order to account for the fact that patenting counts can only take nonnegative values. We choose fixed effects Poisson models with robust standard errors in order account for overdispersion. As demonstrated by Gouriéroux et al. (1984), due to the fact that the Poisson model is in the linear exponential class, the Poisson coefficients estimates are consistent as long as the mean is correctly specified and the robust standard errors are consistent even under misspecification of the distribution. The unit of observation is a researcher-year observation and the dependent variable is the patent and citation-weighted patent output per year. We generate a dummy for the time period prior and post the policy change. This dummy named New Policy takes value 1 for the years 2002 and onwards in which the new IP was active, and value 0 for the years prior to this new policy. We interact the professor dummy with this policy variable and this interaction effect captures the treatment effect of the law change. 8 We control for career age, the time elapsed since the first patent or publication, and its square and we include 2-year time dummies to capture the general time trend. 8 Note that both the professor dummy and the new policy dummy do not appear in our regression output as such as the professor dummy is captured by the fixed effect and the New Policy dummy is absorbed by the time dummies. Therefore only the interaction term will appear in our output. 14

15 4.4 Descriptives The analysis in the main part of the paper uses exit rule a) as described in table 1. The sample consists of researcher-year observations. We restrict the years from 1995 to 2008 such that we have a long enough period prior and post the policy change. We do however take into account the researchers entire patenting and publication record such that the career age is calculated over the entire history of the researcher. This also allows us to use all past characteristics of patenting and publications activity to calculate measures of experience. The sample consists of researchers, featuring 5268 professors and 7839 public research organization researchers 9. The variables generated and used in our analysis are displayed in Table 2. TABLE 2 DEFINITION OF VARIABLES Variable name Definition # Patents Nr of patents applied for in year by researcher # Citation-weighted patents Nr of citations received by patents applied for in given year in the three subsequent years Professor Career age New policy Professor*New policy Inventor was professor at some point in his career Year-year of first patent or publication Dummy for years>=2002 Interaction of professor dummy and New Policy Figure 1, display a growth index 10 of patenting output over time for the group of PRO researchers as well as the group of professors. This figure is supposed to show the development of patenting activity over time and to give a first insight on whether we can expect to find a treatment effect of the change in law in The general negative trend in patenting output per researcher over time 9 As stated above, we have excluded those researchers for which the data indicates that they were employed both at a PRO and were a professor at a university as it is not clear which regulations are applied to these researchers. 10 The index is set at 1 in 1995 and then shows the development over time. 15

16 that can be observed from 1998 onwards has been noted in the prior empirical literature (e.g. Cuntz et al. 2012). As this strong decrease cannot be observed for overall patenting it cannot be attributed to a general trend. Previous literature has suggested several reasons for the observed downward trend of patenting in public science. These can be the increased focus on publication as performance measure for academia, the end of the New Economy boom, changes in patenting laws for the field of biotechnology, in which a large share of research is carried out at public research organizations and universities, or the non- mentioning of university and PRO researchers in patent applications arising from joint research of academia and industry (Cuntz et al. 2012, p.22). More importantly however, we observe that, while the downward development of treatment and control group is parallel until 2002, it starts diverging in 2002/2003, with patenting output for professors decreasing more strongly than for the control group. This indicates a negative treatment effect of the law change on the group of professors. Given that the legal rule does not kick in immediately for all researchers as contracts made before July 18th 2001 were to be treated under the old law until February 2003 ( 43 ArbnErfG), this result is consistent with the timing of the law change. FIGURE 1: GROWTH INDEX OF PATENTING OUTPUT PER YEAR Control Professor 16

17 TABLE 3 DESCRIPTIVES FOR TREATMENT AND CONTROL GROUP PRIOR AND POST THE POLICY CHANGE Professors prior to law change (N=33966) Professors after law change (N=36876) Variable Mean Std. Dev. Min Max Mean Std. Dev. Min Max # Patents # Citation-weighted patents Career age Control group prior to law change (N=45078) Control group after law change (N=54873) Variable Mean Std. Dev. Min Max Mean Std. Dev. Min Max # Patents # Citation-weighted patents Career age Table 3 provides descriptive statistics for the treatment and control group for the period prior and post the policy change. While the professors show a decline of their patenting activity by about 43 percent, it is only about 30 percent for the control group, indicating that while patenting declines for both groups, there is still a difference in magnitude. A similar difference can be found for citationweighted patenting which decreases by 58 percent for the treatment group and by 45 percent for the control group. The difference in career age of the two groups indicates that it makes sense to control for career age in our estimations. While the comparison of the professor (treatment) and the PRO group (control) over time gives some first insight into differences in the development of patenting activity, we need to conduct a difference-in-difference analysis that allows us to control for the particular characteristics of the individuals order to be able to derive conclusions on the impact of the change in law. 17

18 4.5 Estimation results Patenting output The first model estimates the researcher s patenting output. We split the sample into two periods only, the pre-law change period from 1995 to 2001 and the period from 2002 to 2008 that covers the period after the policy change. The dummy variable new policy captures these two periods. By interacting this dummy with the professor dummy we can directly identify the treatment effect on the treated, the professors. We control for career age, its square, as well as 2-year time dummies to capture the general time trend. The results are displayed in Table 4. The treatment effect of the law change on the professors, as captured by the interaction of the new policy dummy and the professor dummy, is negative and significant. On average, the number of patents of university professors decreases by 22% as consequence of the new law. This indicates that the group of professors, those directly affected by the law change, show a significant decrease in their patenting activity compared to the control group of PRO researchers. This provides support for the hypothesis that the law change discourages patenting by reducing the (monetary) incentives for the researchers. This treatment effect shows that, while the overall patenting trend as captured by the time dummies is also negative, there is an additional significant effect that only hits the professors 11. Column 2 repeats the estimation with the subsample of researchers that had at least one patent in the period prior to 2000, such that we consider the subgroup of researchers with some patenting experience prior to the law change only. We find patenting this group of researchers to be affected even more strongly by the law change: They reduce their patenting activity by about 30 percent. The results in column 3 and 4 are based on a conditional difference-in-difference estimation for which the treatment and control group have been matched in 1998 (4 years prior to policy change) 11 We have also estimated a model in which we generate an interaction term of each time dummy with the professor dummy in order to check whether patenting output indeed only differs between treatment and control group in the years after the policy change. This is confirmed. The regression output can be found in the appendix. 18

19 based on the researchers prior publication count, publication subject field and career age. We apply caliper matching (nearest neighbor, caliper threshold = 0.005). We then use the treatment group (professors) and their selected control observations and restrict the panel from 1999 to As in previous models we estimate Poisson count models of patenting output, but only include the treatment dummy and the time dummies. Our sample thereby reduces to reduces to 5668 researchers, 2910 treated and 2758 control researchers. The results remain stable, but reduce slightly in magnitude. For the overall sample the effect is now -19 percent while it is -27 percent for the sample of experienced researchers Citation-weighted patenting output Table 5 repeats the analysis for citation-weighted patenting output which is supposed to capture the quality dimension of the patents. Overall citation-weighted patenting output reduces by about 27 percent for the overall group of professors due to the law change and by about 28 percent for the subsample of researchers with at least one patent prior to The effect is even stronger than when looking at simple patent counts, suggesting the drop in quality to be even stronger than the drop in overall patenting. The conditional difference-in-difference yields an overall treatment effect of -17 percent, and an effect of -18 for the subgroup of experienced researchers. 19

20 TABLE 4 POISSON MODELS OF PATENTING OUTPUT # Patents Overall With at least 1 patent prior 2000 Conditional Diff-in-Diff overall Conditional Diff-in-Diff with at least one patent prior 2000 Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Professor*NewPolicy *** (0.044) *** (0.050) *** (0.056) *** (0.061) Career age ** (0.012) *** (0.014) Career age squared (0.000) 0.001** (0.000) Year dummies (baseline: 1995/1996) 1997_ *** (0.033) 0.250*** (0.035) 1999_ *** (0.051) 0.227*** (0.055) 2001_ (0.071) (0.077) ** (0.040) *** (0.041) 2003_ (0.090) (0.099) *** (0.049) *** (0.053) 2005_ (0.111) (0.122) *** (0.050) *** (0.054) 2007_ (0.133) (0.147) *** (0.055) *** (0.060) Observations Robust standard errors. Significance: * p < 0.1, ** p < 0.05, *** p <

21 TABLE 5 POISSON MODELS OF CITATION-WEIGHTED PATENTING OUTPUT # Citation-weighted patents Overall With at least 1 patent prior 2000 Conditional Diff-in-Diff overall Conditional Diff-in-Diff with at least one patent prior 2000 Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Professor*NewPolicy *** (0.070) *** (0.077) * (0.091) * (0.097) Career age *** (0.027) *** (0.030) Career age squared (0.000) (0.000) Year dummies (baseline: 1995/1996) 1997_ *** (0.065) 0.435*** (0.068) 1999_ *** (0.113) 0.354*** (0.121) 2001_ (0.153) (0.166) *** (0.076) *** (0.081) 2003_ (0.200) (0.215) *** (0.084) *** (0.088) 2005_ (0.248) (0.270) *** (0.088) *** (0.093) 2007_ (0.288) (0.309) *** (0.098) *** (0.106) Observations Robust standard errors. Significance: * p < 0.1, ** p < 0.05, *** p <

22 5. Robustness check: Exit based on scientific publications In order to make sure that we exit a scientist when he leaves academe before retirement, for example in order to go work at a firm, we need to find a way to distinguish between those researchers staying in academe and those pursuing a different career. We therefore suggest to re-run the regressions with the following exit rule: A person leaves the sample if we do not observe any publication for 5 consecutive years. This reduces the number of researcher-year observations to Our results (Table 6 and Table 7) remain robust to this alternative exit rule. On average patenting output reduces by 19 percent and citation-weighted patenting output reduces by 27 percent. Again, we provide the results for a conditional difference-in-difference and we find our results to be robust to this specification as well. 22

23 TABLE 6 ROBUSTNESS CHECK: POISSON MODELS OF PATENTING OUTPUT # Patents Overall With at least 1 patent prior 2000 Conditional Diff-in-Diff overall Conditional Diff-in-Diff with at least one patent prior 2000 Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Professor*NewPolicy *** (0.049) *** (0.057) *** (0.069) *** (0.076) Career age (0.014) *** (0.015) Career age squared ** (0.000) (0.000) Year dummies (baseline: 1995/1996) 1997_ *** (0.035) 0.214*** (0.037) 1999_ *** (0.054) 0.180*** (0.058) 2001_ (0.076) (0.083) ** (0.047) *** (0.049) 2003_ (0.098) (0.109) *** (0.064) *** (0.070) 2005_ (0.121) 0.06 (0.135) *** (0.068) *** (0.075) 2007_ (0.146) 0.07 (0.163) *** (0.079) *** (0.091) Observations Robust standard errors. Significance: * p < 0.1, ** p < 0.05, *** p <

24 TABLE 7 ROBUSTNESS CHECK: POISSON MODELS OF CITATION-WEIGHTED PATENTING OUTPUT # Citation-weighted patents Overall With at least 1 patent prior 2000 Conditional Diff-in-Diff overall Conditional Diff-in-Diff with at least one patent prior 2000 Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Professor*NewPolicy *** (0.081) *** (0.091) ** (0.116) ** (0.127) Career age ** (0.027) *** (0.029) Career age squared (0.000) (0.001) Year dummies (baseline: 1995/1996) 1997_ *** (0.065) 0.387*** (0.067) 1999_ ** (0.108) 0.303*** (0.115) 2001_ (0.153) (0.164) ** (0.091) *** (0.097) 2003_ (0.199) (0.213) *** (0.109) *** (0.115) 2005_ (0.251) (0.275) *** (0.112) *** (0.120) 2007_ (0.307) (0.334) *** (0.126) *** (0.138) Observations Robust standard errors. Significance: * p < 0.1, ** p < 0.05, *** p <

25 6. Discussion and Conclusion In this paper we have analyzed the impact of the abolishment of the professor s privilege in Germany on the patenting behavior of university professors. Our results show that German professors, particularly those with some experience in commercializing their IP, have reacted to the abolishment of the professor s privilege by reducing their patenting activity. This indicates that they consider the costs of the law change - truncated royalties and the need to involve a third party (university TTO) into the negotiations to be more severe than the potential benefits of the law change through the support of a university TTO or a patent valuation agency. This result is robust across numerous specifications and robustness checks. We further find researchers with patenting experience before the law change to exhibit an even stronger decrease in patenting activity than the overall group of professors, indicating that they are even less attracted by the combination of lower royalties with the support of a TTO. Overall, the decrease in professor patenting after the change in law suggests that the transfer of the costs and risk did not compensate the professor for the loss in revenue and the increased transaction costs and even decreased his incentives to innovate. From a policy point of view our results raise a concern that the law change did not achieve its goals of promoting university-science technology transfer and raising university patenting revenues. 25

26 7. References Becher, G., Gering, T., Lang, O. and Schmoch, U. (1996), Patentwesen an Hochschulen: Eine Studie zum Stellenwert gewerblicher Schutzrechte im Technologietransfer Hochschule-Wirtschaft, BMBF (Hrsg.), Bonn. Bundesministerium der Justiz (1957), Gesetz über Arbeitnehmererfindungen, ArbnErfG. Bundesministerium der Justiz (2001), Gesetz zur Änderung des Arbeitnehmererfindergesetzes: Referentenentwurf 2001, (last retrieved ). Bundesministerium für Bildung und Forschung and Bundesministerium für Wirtschaft und Technologie (2001), Wissen schafft Märkte - Aktionsprogramm der Bundesregierung. Crespi, G.A., Geuna, A. and Verspagen, B. (2006), University IRPs and knowledge transfer: Is the IPR owernership model more efficient? SPRU Working Paper No Cuntz, A., Dauchert, H., Meurer, P. and Philipps, A. (2012), Hochschulpatente zehn Jahre nach Abschaffung des Hochschullehrerprivilegs, Studien zum deutschen Innovationssystem Czarnitzki, D., Glänzel, W. and Hussinger, K. (2007), Patent and Publication Activities of German Professors: An Empirical Assessment of Their Co-Activity, Research Evaluation 16(4), Czarnitzki, D., Glänzel, W. and Hussinger, K. (2009), Heterogeneity of Patenting Activity and its Implications for Scientific Research, Research Policy 38, Czarnitzki, D., Hussinger, K. and Schneider, C. (2011), Commercializing Academic Research: The Quality of Faculty Patenting, Industrial and Corporate Change 20(5), European Commission (1995), Green Paper on Innovation. Frank, A., Meyer-Guckel, V. and Schneider, C. (2007), Innovationsfaktor Kooperation. Bericht des Stifterverbands zur Zusammenarbeit zwischen Unternehmen und Hochschulen. r_kooperation/innovationsfaktor_kooperation_zusammenfassung. Geuna, A. and Nesta, L.J.J. (2006), University Patenting and its Effects on Academic Research: The Emerging European Evidence, Research Policy 35, Gouriéroux, C., Montfort, A. and Trognon, A. (1984), Pseudo Maximum Likelihood Methods: Application to Poisson Models, Econometrica, 52 (3),

27 Hellmann, T. (2007), The role of patents for bridging the science to market gap, Journal of Economic Behavior & Organization 63, Kenney, M. and Patton, D. (2009), Reconsidering the Bayh-Dole Act and the Current University Invention Ownership Model, Research Policy 38, Kilger, C. and Bartenbach, K. (2002): New Rules for German Professors, Science 298 (5596), Mowery, D.C., and Sampat, B.N. (2005), The Bayh-Dole Act of 1980 and University-Industry Technology Transfer: A Model for Other OECD Governments? Journal of Technology Transfer 30 (1/2), OECD (2003), Turning science into business: Patenting and licensing at public research organizations, Paris: OECD. Rösler, H. (2001): Bestand, Reform oder Abschaffung des 42 ArbNErfG, Universität der Bundeswehr München, Dissertation. Schmoch, U. (2007), Patentanmeldungen aus deutschen Hochschulen - Analysen im Rahmen der Berichterstattung zur Technologischen Leistungsfähigkeit Deutschlands, Studien zum deutschen Innovationssystem Nr , Berlin. Schoen, A. and Buenstorf, G. (2013), When Do Universities Own Their Patents? An Explorative Study of Patent Characteristics and Organizational Determinants in Germany, Industry and Innovation 20 (5), So, A.D., Sampat, B.N., Rai, A.K., Cook-Deegan, R., Reichman, J.H., Weissman, R. and Kapczynski, A. (2008), Is Bayh-Dole Good for Developing Countries? Lessons from the US Experience, PLoS Biology 6 (10), Stevens, A.J. (2004), The Enactment of Bayh-Dole, Journal of Technology Transfer 29, The Economist (2002), Innovation's golden goose -The reforms that unleashed American innovation in the 1980s, and were emulated widely around the world, are under attack at home, Technology Quarterly, Q Thursby, J.G. and Thursby, M.C. (2008) Are faculty critical? Their role in university-industry licensing, Contemporary Economic Policy 22(2), Verspagen, B. (2006), University Research, Intellectual Property Rights and European Innovation Systems, Journal of Economic Surveys 20 (4). Von Proff, S., Buenstorf, G. and Hummel, M. (2012) University Patenting in Germany before and after 2002: What Role Did the Professors' Privilege Play? Industry & Innovation 19 (1),

28 8. Appendix TABLE 8 POISSON MODEL OF PATENT COUNT, INTERACTION OF EACH TIME DUMMY WITH THE PROFESSOR DUMMY Coef. Std. Err. Interaction of Professor with time dummies (baseline: Professor* 1995_1996) Professor*1997_ (0.053) Professor*1999_ (0.061) Professor*2001_ * (0.068) Professor*2003_ *** (0.075) Professor*2005_ *** (0.074) Professor*2007_ *** (0.087) Career age *** ( Career age squared ) Time dummies (baseline:1995/1996) 1997_ *** (0.037) 1999_ *** (0.055) 2001_ ** (0.075) 2003_ (0.093) 2005_ (0.113) 2007_ (0.134) Observations