Did Patenting Strategies Persistently Change After TRIPS?

Did Patenting Strategies Persistently Change After TRIPS? Jorge Lemus and Guillermo Marshall March 30, 2016 [PRELIMINARY AND INCOMPLETE DRAFT. COMMENTS and SUGGESTIONS ARE WELCOME. PLEASE DO NOT CITE.] Abstract The 1995 TRIPS agreement changed the patent term in the United States from 17 years from the patent issue date to 20 years from the patent application filing date. We look at a broad set of observable patent characteristics and construct measures of quality, disclosure, and scope of patents using proxies that have traditionally been used in the literature such as citations, number of independent and dependent claims, number of classifications, as well as novel indicators such as figures, number of words, and the complete tree of dependency of claims. Our results show that TRIPS mostly caused transitory changes in patent applicant behavior, most of which can be attributed to a set of applicants choosing to submit an application before the policy change. However, we find evidence of a pronounced and permanent decline in the use of continuations by patent applicants. These results suggest that penalizing the patent term of applicants using continuations was effective in reducing continuation use and shed light on which interventions may be effective in further reducing continuation (ab)use. 1 Introduction With the goal of facilitating international trade, members of the World Trade Organization (WTO) negotiated an agreement on Trade-Related Aspects of Intellectual We wish to thank Melissa Wasserman for helpful comments and suggestions. Department of Economics, University of Illinois Urbana Champaign. email: jalemus@illinois.edu Department of Economics, University of Illinois Urbana Champaign. email: gmarshll@illinois.edu 1

Property Rights (TRIPS) setting standards and regulations for patents, trademarks, and copyrights. If the law in a country member of the WTO differed from what the agreement was proposing, that country had to modify their intellectual property laws to comply with the agreement. For the United States, the TRIPS agreement had a direct impact on the scope of patent protection. Previous to the implementation of the agreement, in June 8, 1995, a patentee was granted 17 years of protection from the issue date of the patent. After June 8, 1995, the patentee obtained 20 years of protection from the filing date of the patent application. The prosecution time of a patent application the difference between the issue date of a patent and filing date is not exogenous: It corresponds to the time spent by the USPTO reviewing the application plus the time spent by the applicant responding to the USPTO. The TRIPS agreement had a direct impact on applicants incentives, since before June 8, 1995, there was no loss in the length of patent protection due to a longer prosecution time, while after June 8, 1995, longer prosecution time meant a shorter duration of patent protection. In this paper, we aim to understand the effect of TRIPS on the characteristics of patent applications and behavior of patent applicants. Despite the clear change in incentives introduced by TRIPS, only a few scholars have empirically looked into them. Sukhatme and Cramer (2014) argue that TRIPS caused a local change of behavior in response time to USPTO notices in some industries, notably the pharmaceutical and software industries. This evidence supports the previous analysis in Lemley (1994). Abrams (2009) shows that patent counts and citations increased after TRIPS. We look at a broader set of patent characteristics and construct several measures of quality, disclosure, and scope of patents using proxies that have traditionally been used in the literature such as citations, number of independent and dependent claims, number of classifications, as well as novel indicators such as figures, number of words, and the complete tree of dependency of claims. For most of our measures, our results show that TRIPS only caused transitory changes in behavior, which we mainly attribute to the anticipation of new policy. For these measures, we find no robust evidence that TRIPS caused a permanent change in the characteristic of issued patents or applicants behavior. However, we do observe an important and permanent change in the use of continuations. As Abrams (2009) explains, there are three relevant dates related to the change in patent term induced by TRIPS: April 15, 1994, when the agreement was signed; December 8, 1994, when the agreement was ratified by the U.S. Congress; and June 8, 1995, when the agreement was formally implemented in the U.S. For patent applications between December 8, 1994 and June 8, 1995, the patent length was the maximum between 17 years from grant date and 20 years from the application date. That is, patent applications during that period obtained the longest patent term between the 2

old system and the new system, providing incentives to submit applications instead of waiting for the new system to be in place. This transition periods offered the best of both worlds, encouraging inventors to submit all the applications that they could possible file. Our results shows that applicants not only filed more patent applications during the six months prior to the TRIPS implementation, but they submitted their better ideas. This suggest that applicants may have private information regarding the quality of their application at the time of filing. Finding whether TRIPS produced a permanent change in the filing strategies of applicants is relevant, since the effective patent length which refers to the time between the patent issue date and the first time a product protected by the patent is commercialized could have effects on the research strategies of firms. Budish et al. (2013), using data from cancer clinical trials, show that patents could under-incentivize firms to engage in long-term research. 2 Literature Review Our paper relates to three areas of empirical patent research. The first one studies the relation between patent characteristics and patent value; the second one studies the incentives the patent system provides to applicants; and the third studies the challenges faced by examiners during the examination process. Patent quality is not a well-defined concept. As pointed out in Graf (2007), patent quality can be defined in different ways: by how well a patent meets the patentability requirements (patentable subject matter, utility, novelty, non-obviousness, good description and enablement); by the validity of the patent and the scope of its claims (holds valid even after re-examination in trial and its boundaries clearly defined); economic value (allows commercialization of inventions, generates private returns to the patent holder). Allison et al. (2003), for example, assumes that litigated patents are at least a subset of the most valuable patents and studies their characteristics. The paper finds that litigated patents are often issued to small companies or individuals, cite more prior art and are more cited than non-litigated patents, and present a higher prosecution time. Lanjouw and Schankerman (2004) build a factor model with observable indicators correlated to unobservable quality: the number of claims, forward citations to the patent, backward citations in the patent application, and family size. This method reduces the dimensionality of the problem and generates a single index of patent quality. In a sample of patents, Lerner (1994) shows that patent scope defined as number of classifications in which a patent is allocated to has an economically and statistically significant impact on the valuation of firms. 3

Jean O. Lanjouw (2001) show that patents with more claims and more citations by subsequent patentees are substantially more likely to be involved in litigation in each of the four technology fields we study. Bronwyn H. Hall (2005) shows there is a positive and significant relation between patent citations and firms market value. Mann and Underweiser (2012) investigate legal validity as a metric of quality and use the textual features of the patent (how well the specification and claims are aligned) and detailed information from the prosecution histories. Instead of looking at the correlation between patent characteristics and outcomes such as patent validity or firms market value, some authors have used a revealed preference approach to estimate the value of patents. Schankerman (1998) uses data from France to estimate parametric and non-parametric models of patent renewal to determine the private return of patents. Pakes (1986) develops a structural model to estimate the distribution of value of patents using renewal data from Europe. The problem with these approaches is that individual patents might not have a large value in isolation, but their value could derive from their contribution to a firm s patent portfolio as explained in Hall and Ziedonis (2001). The design of the patent system not always provides correct incentives to patent novel, non-obvious and useful inventions and institutional details sometimes can create perverse incentives for inventors. Jaffe (2000) studies the relevant changes to the US patent system from 1980 to 2000, focusing mainly on the creation of the CAFC, the Bayh-Dole act, the expansion of patentability of software, and the TRIPs agreement. Quillen Jr and Webster (2009), for example, point out patent applications rose significantly after the Court of Appeals for the Federal Circuit (CAFC) was created, however this increase in patent applications did not have a positive impact on growth rate in the U.S or on firms R&D expenditures. Similarly, Jaffe and Lerner (2011) argue both the creation of the CAFC and the financing of the USPTO weakened the patent system overall. Perhaps one of the strangest features of the USPTO is its inability to decisively reject a patent application, since applicants can always file for continuations to prolong the examination process. Lemley and Moore (2004) describe in detail the practice of patent continuations in the US patent system, and they conclude that there are abuses in the use of continuations although they are not so severe to justify the abolition of continuations. Related work by Hegde et al. (2009) studies the motives for firms to use continuations in the prosecution of their patents and shows that firms in different technology areas use different continuation strategies. Finally, in 2000 a new type of continuation, called request for continued examination (RCE), was created and Cotropia et al. (2013) shows that the number of refiled applications since 1996 have increased over time, specially RCEs which began in the year 2000. Apart from not having a uniform measure of quality, and to have a system that provides 4

applicants with incentives to file applications strategically, the USPTO has been criticized for issuing too many bad quality patents. The criterion for what is patentable is widely diverse among different patent offices. Using data from 1990 to 1995, Jensen et al. (2005) finds that about 38% of all patents granted by the USPTO were also granted by the JPO and the EPO; the remainder (62.3%) were either rejected, withdrawn, or were still awaiting a final decision. The patent examination itself is a complex process. Popp et al. (2004) presents a detail description of the examination process at the USPTO. Using data on patents from 1976-1996, the authors study prosecution time as a function of patents characteristics such as number of references, claims, and figures. Carley et al. (2014) computes patent allowance rates at the USPTO at different stages of the examination process, using data from 1996-2005. Harhoff and Wagner (2009) analyze the duration and outcomes of patent examination at the European Patent Office as a function of applicant characteristics, indicators of patent quality and value (request for accelerated examination, international family size, citations, generality index), and determinants that affect the complexity of the examination task (claims, number of patent references, and non-patent references, originality, and number of IPC classifications). The paper finds evidence that applicants accelerate the application for their most valuable patents and fight longer if those patents are at risk of being rejected. These results support previous conclusions in Régibeau and Rockett (2007). Examiners also have a hard time reviewing prior art. Ideally, a patent should cite all relevant prior art, but this is unfeasible due to search costs and strategic motives. By citing more prior art, an applicant can obtain a stronger patent it would be harder to invalidated ex-post by citing prior art. However, applicants can obtain broader patents by not citing all prior art. Sampat et al. (2005) finds evidence that patent examiners have a comparative disadvantage versus applicants in searching for nonpatent prior art and foreign patents. They conclude that patents are likely to be of lower quality for technological areas for which most prior art is not embodied in U.S. patents. As far as we are concerned there are just a few studies of the effect of TRIPS on patents characteristics. Sukhatme and Cramer (2014) shows that different industries have different sensitivities to patent length. In particular, they find that the pharmaceutical industry cares more about patent term compared to other industries. These results support the analysis in Lemley (1994). Abrams (2009) studies the effect of TRIPS on the number of patent applications and their citations after they are issued. The author finds that there was a statistically significant change in the number of patents applications following the TRIPS agreement, but no significant difference in mean citations per patent. Our results corroborate these findings but add the effect of TRIPS on other indicators of quality, disclosure and scope of patents. Even more, we argue that these results are only transitory and did not induce a permanent change in behavior. 5

3 Data We combine data from different sources to construct our dataset. First, we scraped the website https://www.patexia.com/ from which we obtained information for all issued patents between 1990-2000 including: patent number, the text for all the claims and their sequence, application and priority dates, type of entity (small or large), patent renewal data, references, citations over time, number of classifications, and prosecution time. We also collected all the available information for patent applications dating from 1989 to 2000, from the USPTO bulk downloads of PAIR. For each patent application we obtained the transaction history, the continuation data, and the documents for foreign applications. Additionally, we used the USPTO Historical Patent Data files in Marco et al. (2015) and data on maintenance fees and patent classification from http://www.reedtech.com/. 4 TRIPS and Observable Patent Characteristics A patent is a legal document that can be viewed as a bundle of observable characteristics, some of them associated with patent quality, others with patent disclosure, and others with patent scope. There is no agreement on exactly what constitutes patent quality, disclosure and scope, although many papers have used some of the observable characteristics as proxies for those variables. In this section, we present evidence of the effect of TRIPS on prosecution time, references, citations, number of independent and dependent claims, number of words per claim, number of classifications, number of figures, number of words in the patent description, renewal information, and the use of continuations. To determine the effect of TRIPS on these observable characteristics, our empirical strategy is to run OLS regressions for several versions of the following model: y ijt = α + small i + λ j + γ t + f(pending jt, pending t ) + ε ijt, (1) where i corresponds to an application of an issued patent, 1 j the NBER industry code of the application, t is the month year of application, y ijt is the outcome variable of interest (citations, renewals, number of figures, claims, etc.) small i is an indicator of whether the applicant is a small entity, λ j is an industry fixed effect, γ t is a month year time effect, f(pending jt, pending t ) is a polynomial function of pending applications at time t (trying to control for the backlog) and finally ε ijt is an error term robust against heteroskedasticity. 2 1 Unfortunately, we do not observe quality measure of patent applications that are never issued. 2 Figure 3 shows that the USPTO does not change their patent approval behavior after TRIPS. 6

We also run several versions of the following alternative model, y ijt = α+small i +λ j +1{After June 1995}π+γ year(t) +f(pending jt, pending t )+ε ijt, (2) where 1{After June 1995} is an indicator for the month years after June 1995, γ year(t) is a year of application fixed effect, and all other variables are defined as above. An important difference between equations (1) and (2) is that in equation (2) we incorporate year of application fixed effects instead of month year of application fixed effects. By only including year of application fixed effects in equation (2), we use within-1995 variation in outcomes to study local effects caused by TRIPS. We will later argue that this strategy will be present issues as the data suggests that a large number of applicants anticipated the policy change and chose to submit their applications before June 1995. This selection issue will prevent us from interpreting the estimates of equation (2) as causal. 4.1 Patent Quality First, as we have already mentioned, the concept patent quality is not well defined, as there are several criteria to evaluate patents. From the social perspective, a good patent is one that discloses novel knowledge and with boundaries clearly defined. From a private perspective, a good patent is one that increases the private returns either by signaling, as part of a patent portfolio, or by creating rents through licensing. From the legal perspective, a good patent is one that meets all the legal requirements and it is consider valid in court. The number of citations a patent received have been associated with quality, in the sense that more cited patents are litigated more often and they increase the market value of the firm, as shown by Bronwyn H. Hall (2005). The number of references in the patent have also been associated with higher quality, because citing more prior art makes harder to invalidate a patent claiming its subject matter is obvious. Finally, information about the renewal of patents has been used to estimate the private value of a patent, for example in Pakes (1986), which is an alternative proxy for quality. We estimate equations (1) and (2) using as quality outcomes citations after 5 and 10 years, renewal probability after 4, 8 and 12 years. Figure 1 and Table 1 in the appendix show estimates for equations (1) and (2), respectively. Table 1 shows a small but significant negative effect of TRIPS on all these quality measures. However, the first 5 panels of Figure 1 show that these results have to be interpreted carefully. We observe no long-term effect of TRIPS on the citations dimension or the renewal probabilities. However, it is interesting to notice that applications filed between December 1994 and June 1995, are more likely to be renewed after 4 years than applications at any other date. This could be explained by two confounding effects. First, there may be adverse selection and patent applicants know the quality of their patents at the time of application. If that is the case, it is likely that applicants submitted their most 7

valuable patents at the transition period to get the best of both worlds. Second, it may be that with a shorter effective patent term, the value of renewing the patent is lower and patentees are better off not renewing them. 4.2 Patent Disclosure Disclosure is one of the main motivations of the patent system. The patent system grants the inventor an exclusion right in exchange for the disclosure of the invention. This disclosure has to be sufficiently clear so any person skilled in the art can make use of the invention (once the patent has expired). As in the case of quality, measuring disclosure is not a simple task. Observable measures of that may relate to disclosure are the number of words in the description and the number of figures. It is not unreasonable to think that longer descriptions are related to more disclosure of the invention. If inventors are worried that prosecution time can take too long, hence lowering the effective patent term of the patent, they could try to shorten the prosecution time by changing their disclosure efforts. By submitting a shorter document with less figures, the patent may be issued quicker. We estimate equations (1) and (2) using as disclosure outcomes the number of figures and the number of words as the outcome variable. Figure 1 and Table 2 in the appendix show estimates for equations (1) and (2), respectively. Table 2 in the appendix show the regression results. They show a significant effect indicating a negative effect of TRIPS on all these disclosure measures. Again, looking at Figure 1, we find that most of the effect is explained by an anticipation effect. It is really striking to see that the number of figures during the anticipation period (Dec1994-Jun1995) increased six times compared to the previous (and subsequent) six months. A similar pattern is observed for the outcome measure number of words. Inventors clearly had incentives to submit their most lengthy patents during the transition period. However, we do not see a permanent decrease in these disclosure measures after June1995, indicating that most of the effect is local selection. 4.3 Patent Scope The patent scope is arguably the most important part of a patent. Patent claims set the boundaries to separates what part of the invention is protected and what is not. Independent claims are the broadest statements, while dependent claims are subsets of independent claims. Dependent claims are stronger, since they are harder to invalidate in court. Also, the number of words per claim are an important measure of breadth. An independent claim with more words will typically be more specific and less vague, making the claim stronger. A different measure of patent breadth, used by Lerner (1994), is the number of technological classifications that the patent is allocated in. The more technological classifications a patent is allocated in, the broader the patent. 8

We estimate equations (1) and (2) using as disclosure outcomes the number of independent claims, as well as the number of dependent claims, and the ratio between dependent over independent claims. Figure 2 and Table 2 in the appendix show estimates for equations (1) and (2), respectively. Table 2 shows a significant negative effect of TRIPS on all these breadth measures. Looking at Figure 2, we can notice that the effect is mostly due to the anticipation of the change. Similar to the reasoning in the case of more figures and words, inventors could have submitted all the applications with a large number of claims during this transition period. 3 5 Effect of TRIPS on Other Applicant s Actions Apart from the effect that TRIPS could have on the patent application document itself, the policy change could have affected other actions that are taken by patent applicants. For example, Sukhatme and Cramer (2014) shows that patent applicants rush their actions in light of shorter patent term when they delay patent prosecution. In this section, we look at two actions that patent applicants can take during the patent application: Use of continuations and the delay in response to first communication. 5.1 Delay in Response to First Communication When a patent application is reviewed by the USPTO, there is communication between the patent office and the applicant. Using the data from the transaction history, we look at how long it takes for the applicant to respond to the USPTO s first communication with the applicant. In Figure 2, we find that during the transition period there is a significant increase in the applicant s response delay time. After the change, we do observe a reduction in the applicant s response delay time, indicating that patent applicants are minimizing the part of the prosecution time that they can directly affect. Sukhatme and Cramer (2014) finds that the software and drugs industries are more likely to respond faster. 5.2 Use of Continuations At the time of the implementation of TRIPS, applicants could choose to delay their patent prosecution by employing continuations. There are three type of actions: Continuations, Continuations-in-Part, and Divisional Continuations. Continuations allow for further examination of the original patent application, while continuations-in-part allow the application to add new subject matter. A Divisional application, literally divides the patent application into several different applications that could potentially lead to different patents. 3 We also studied as measure of scope the number of classifications and results are similar. 9

Submarine patents is the terminology used to describe patent applications that take a long time to issue. One motive for the applicant to delay the patent prosecution is to wait for less uncertainty in an industry and therefore have clear infringement targets once the patent is issued. Industry participants would not be aware of these patent applications, since before 1999 patent applications remained secret until their issue and publication. 4 The TRIPS legislation decreased applicants incentives to file submarine patents, but did not necessarily end the use of continuations. In the next sections, we analyze the effect of TRIPS on different types of continuations which differ by industries, as explained by Hegde et al. (2009). 5.2.1 Divisional Continuations As shown in Figure 2 in the appendix, the use of divisional continuations, which is a decision taken by the USPTO does not have an effect after the implementation of TRIPS after 1995. We can observe a very significant anticipation effect 6 months prior to the implementation. One reason could be precisely that patent applications were sending longer patent applications and, as a response, the USPTO was forced to use more divisions. 5.2.2 Continuations As shown in Figure 2 in the appendix, the use of continuations, which is a decision taken by the applicant, has a very significant reduction after June 1995. We can see a very significant decrease after June 1995 and no significant anticipation effect 6 months prior to June 1995. This drop in the use of continuations may capture the increased cost of filing a continuation (and restarting the patent examination) after June 1995. In 1997 there was another change that caused a large decrease in the use of continuations http://www.uspto.gov/web/offices/com/sol/notices/aa97.html. 5.2.3 Continuations in Part Continuations in part are continuations that add new subject matter to the patent application. In Figure 2, we can see a clear decrease in the use of continuations-inpart after June 1995 and also a large anticipation effect before the implementation of TRIPS. 4 American Inventors Protection in 1999 required patent applications be published after 18-months of the filing date. 10

6 Conclusion (In Progress) In this paper we analyzed the effect of the change in patent term caused by TRIPS on several observable dimensions of patents. We find that TRIPS did not cause a permanent change on several observable measures of quality, disclosure, and scope. However, we found a significant anticipation effect indicating that patent applicants cared about the policy change and they change their behavior in response to the eventual policy change. Interestingly, despite that applicants did not sacrifice the quality or disclosure of their patents documents, TRIPS caused an important change in the use of continuations and a smaller but still significant effect on the time applicants took to respond to the first communication sent by the USPTO. Our preliminary results suggest that patent applicants are aware of the quality of their patent applications. References Abrams, David S (2009) Did TRIPS spur innovation? An analysis of patent duration and incentives to innovate, University of Pennsylvania Law Review, pp. 1613 1647. Allison, John R, Mark A Lemley, Kimberly A Moore, and R Derek Trunkey (2003) Valuable patents, Geo. Lj, Vol. 92, p. 435. Bronwyn H. Hall, Manuel Trajtenberg, Adam Jaffe (2005) Market Value and Patent Citations, The RAND Journal of Economics, Vol. 36, pp. 16 38. Budish, Eric B, Benjamin N Roin, and Heidi Williams (2013) Do Fixed Patent Terms Distort Innovation?: Evidence from Cancer Clinical Trials, Evidence from Cancer Clinical Trials (September 5, 2013). Chicago Booth Research Paper. Carley, Michael, Deepak Hegde, and Alan C Marco (2014) What is the Probability of Receiving a US Patent?. Cotropia, Christopher A, Cecil D Quillen Jr, and Ogden H Webster (2013) Patent Applications and the Performance of the US Patent and Trademark Office, Fed. Cir. BJ, Vol. 23, p. 179. Daniel K. N. Johnson, David Popp (2003) Forced out of the Closet: The Impact of the American Inventors Protection Act on the Timing of Patent Disclosure, The RAND Journal of Economics, Vol. 34, pp. 96 112. Graf, Susan Walmsley (2007) Improving patent quality through identification of relevant prior art: approaches to increase information flow to the patent office, Lewis & Clark L. Rev., Vol. 11, p. 495. 11

Hall, Bronwyn H and Rosemarie Ham Ziedonis (2001) The patent paradox revisited: an empirical study of patenting in the US semiconductor industry, 1979-1995, RAND Journal of Economics, pp. 101 128. Harhoff, Dietmar and Stefan Wagner (2009) The duration of patent examination at the European Patent Office, Management Science, Vol. 55, pp. 1969 1984. Hegde, Deepak, David C Mowery, and Stuart JH Graham (2009) Pioneering inventors or thicket builders: Which US firms use continuations in patenting? Management Science, Vol. 55, pp. 1214 1226. Jaffe, Adam B (2000) The US patent system in transition: policy innovation and the innovation process, Research policy, Vol. 29, pp. 531 557. Jaffe, Adam B and Josh Lerner (2011) Innovation and its discontents: How our broken patent system is endangering innovation and progress, and what to do about it: Princeton University Press. Jean O. Lanjouw, Mark Schankerman (2001) Characteristics of Patent Litigation: A Window on Competition, The RAND Journal of Economics, Vol. 32, pp. 129 151. Jensen, Paul H, Alfons Palangkaraya, and Elizabeth Webster (2005) Disharmony in international patent office decisions, Fed. Cir. BJ, Vol. 15, p. 679. Lanjouw, Jean O and Mark Schankerman (2004) Patent quality and research productivity: Measuring innovation with multiple indicators*, The Economic Journal, Vol. 114, pp. 441 465. Lemley, Mark A (1994) An empirical study of the twenty-year patent term, AIPLA Quarterly Journal, Vol. 22, p. 369. Lemley, Mark A and Kimberly A Moore (2004) Ending abuse of patent continuations, BUL rev., Vol. 84, p. 63. Lerner, Joshua (1994) The importance of patent scope: an empirical analysis, The RAND Journal of Economics, pp. 319 333. Mann, Ronald J and Marian Underweiser (2012) A new look at patent quality: Relating patent prosecution to validity, Journal of Empirical Legal Studies, Vol. 9, pp. 1 32. Marco, Alan C, Michael Carley, Steven Jackson, and Amanda F Myers (2015) The USPTO Historical Patent Data Files: Two Centuries of Innovation, Available at SSRN 2616724. Miller, Shawn P (2013) What s the Connection Between Repeat Litigation and Patent Quality? A (Partial) Defense of the Most Litigated Patents, Stan. Tech. L. Rev., Vol. 16, pp. 313 833. 12

Pakes, Ariel (1986) Patents as Options: Some Estimates of the Value of Holding European Patent Stocks, Econometrica, Vol. 54, pp. 755 84. Popp, D, T Juhl, and D Johnson (2004) Time in purgatory: Determinants of the grant lag for US patent applications. Topics Econom. Anal, Policy, Vol. 4. Quillen Jr, Cecil D and Ogden H Webster (2009) Continuing patent applications and performance of the US Patent and Trademark Office-One more time, The Federal Circuit Bar Journal, Vol. 18, pp. 379 404. Régibeau, Pierre and Katharine Rockett (2007) Are more important patents approved more slowly and should they be? Vol. (2010) Innovation Cycles And Learning At The Patent Office: Does The Early Patent Get The Delay? The Journal of Industrial Economics, Vol. 58, pp. 222 246. Sampat, Bhaven et al. (2005) Determinants of patent quality: an empirical analysis, Columbia Univ., New York. Schankerman, Mark (1998) How valuable is patent protection? Estimates by technology field, the RAND Journal of Economics, pp. 77 107. Sukhatme, Neel U and Judd NL Cramer (2014) Who Cares About Patent Term? Cross-Industry Differences in Term Sensitivity, Cross-Industry Differences in Term Sensitivity (August 19, 2014). Trajtenberg, Manuel (1990) A penny for your quotes: patent citations and the value of innovations, The Rand Journal of Economics, pp. 172 187. 13

7 Appendix 14

Table 1: Local effect of TRIPS on observed outcomes: OLS Regressions (I) (1) (2) (3) (4) (5) Cites by Year 5 Cites by Year 10 Renewal Year 4 Renewal Year 8 Renewal Year 12 After June 1995-0.029-0.556*** -0.024*** -0.039*** -0.045*** (0.040) (0.105) (0.002) (0.003) (0.003) Small Entity -0.748*** -1.818*** -0.152*** -0.219*** -0.217*** (0.011) (0.033) (0.001) (0.001) (0.001) NBER Category FE Yes Yes Yes Yes Yes Year of Application FE Yes Yes Yes Yes Yes Pending Applications Controls Yes Yes Yes Yes Yes N 1,061,841 1,061,841 1,061,841 1,061,841 1,061,841 R 2 0.093 0.093 0.056 0.067 0.065 Notes: Robust standard errors in parentheses. * p < 0.05, ** p < 0.01, *** p < 0.001. Table 2: Local effect of TRIPS on observed outcomes: OLS Regressions (II) (1) (2) (3) (4) (5) Figures Words (000s) Ind. Claims Dep. Claims Dep. Claims/Ind. Claims After June 1995-2.798*** -2.379*** -0.142*** -0.590*** 0.031 (0.102) (0.060) (0.019) (0.078) (0.033) Small Entity 0.058* -0.917*** -0.124*** -0.554*** -0.291*** (0.028) (0.013) (0.006) (0.027) (0.013) NBER Category FE Yes Yes Yes Yes Yes Year of Application FE Yes Yes Yes Yes Yes Pending Applications Controls Yes Yes Yes Yes Yes N 1,055,628 1,055,628 1,055,628 1,055,628 1,055,452 R 2 0.054 0.107 0.039 0.024 0.019 Notes: Robust standard errors in parentheses. * p < 0.05, ** p < 0.01, *** p < 0.001. Table 3: Local effect of TRIPS on observed outcomes: OLS Regressions (III) (1) (2) (3) Continuation (CON) Continuation (DIV) Continuation (CIP) After June 1995-0.075*** -0.115*** -0.057*** (0.002) (0.002) (0.002) Small Entity -0.037*** -0.022*** 0.039*** (0.001) (0.001) (0.001) NBER Category FE Yes Yes Yes Year of Application FE Yes Yes Yes Pending Applications Controls Yes Yes Yes N 961,582 961,582 961,582 R 2 0.029 0.032 0.025 Notes: Robust standard errors in parentheses. * p < 0.05, ** p < 0.01, *** p < 0.001. 15

Figure 1: Effect of TRIPS on observed outcomes: OLS Regressions -.5 0.5 1 1.5 0 1 2 3 4 a) Citations by year 5 b) Citations by year 10 -.02 0.02.04.06 -.02 0.02.04.06.08 c) Year 4 renewal rate d) Year 8 renewal rate -.02 0.02.04.06.08 0 2 4 6 d) Year 12 renewal rate e) Number of figures 0 1 2 3 4 f) Number of words (in thousands) 16

Figure 2: Effect of TRIPS on observed outcomes: OLS Regressions 0.1.2.3.4.5 0 1 2 3 4 a) Number of independent claims b) Number of dependent claims -.5 0.5 1 -.1 -.05 0.05.1 c) Ratio of dependent to independent claims d) Whether application is a continuation -.05 0.05.1.15 -.02 0.02.04.06.08 d) Whether application is a divisional continuation e) Whether application is a continuation in part 25 30 35 40 45 f) Time between first USPTO notice and applicant response 17

Figure 3: Number of applications and share of issued applications Number of applications 10000 15000 20000 25000 30000 Time a) Number of applications Share of successful applications.8.85.9.95 Time b) Share of applications that resulted in issued patents 18