Turn--Excessive pharma patent protection crushes innovation

By empirically examining how antitrust regulation of patent-enabled monopolies affects follow-on innovations, this study advances the broad policy discussion on interactions between antitrust and patent laws in shaping innovation. As intensively discussed by scholars (Katz and Shelanski, 2005; Sidak and Teece, 2009), the necessity of incorporating “technological innovation” into modern antitrust law by accounting for the Schumpeterian perspective has received increasing recognition from the antitrust law community. The Schumpeterian view posits that market competition and technological innovation are mutually and dynamically influenced in the so-called “innovation cycle.” Firms that are not dominant market players may enjoy monopolistic market profits through future innovation. In contrast, a dominant market player may lose its monopolistic market position due to a competitor's innovation. The dynamics therein create incentives for firms to innovate. This scholarship leads to the important question of how antitrust law jointly shapes market competition and innovation. Our study contributes to this discussion by empirically showing that “patent consolidation” by firms is a venue where enforcement of antitrust law promotes market competition and technological innovation.

The remainder of this paper is structured as follows. Section 2 reviews the literature on patent pools and cumulative innovation to derive our main hypothesis. Section 3 describes the data and methods for empirical analysis. Section 4 presents the findings, and Section 5 discusses the implications.

2. Literature review and hypothesis development

2.1. Antitrust law, patent consolidation, and market competition

Patent transfer can be subject to antitrust regulation because it may enable patent consolidation that allows a few firms to control access to key patented technology for products and thereby effectively monopolize the downstream market (Figueroa and Serrano, 2019; Hahn, 1984). In the United States, two legislative acts have enabled the regulation of patent transfers: the Sherman Act and the Clayton Act. The Sherman Antitrust Act (26 Stat. 209, 15 U.S.C. §§1–7) was enacted in 1890 to regulate unlawful market monopolies. This act enables the regulation of business practices that may discourage market competition. Meanwhile, the Clayton Act was enacted in 1914 (38 Stat. 730, 15 U.S.C. §§12–27, 29 U.S.C. §§52–53) to specify which business practices violate the Sherman Act, elaborating on its antitrust law enforcement scheme and legal remedies. Section 7 of the Clayton Act is considered the foundation for the antitrust regulation of patent transfers. The section states that the federal authorities may regulate business practices that may harm market competition or lead to the formation of a market monopoly. Based on this section, patent transfers that could harm market competition or result in a monopoly are thus subject to regulation (Gotts and Sher, 2012; Klitzke, 1980).

Then, under what conditions patent consolidations by a firm may negatively affect the market competition? The scholarly discussion on patent pools and their impact on market competition provides a hint for the answer. Studies have argued that amassing “complementary” patents through patent pooling by a few firms can spur downstream market competition by addressing the so-called patent holdup or patent thicket problem (Ebersole et al., 2005; Merges, 1999; Shapiro, 2000), which enhances market efficiency (Gilbert, 2004). For instance, when each of the two competing firms owns mutually complementary patents for a product, they may experience significant difficulty in bringing the product to the market because one's patent will hold up another's market operation (i.e., patent holdup). In this case, the problem can be solved by pooling the two firms’ patents and agreeing on the free use of pooled patents.

In contrast, firms aggregating “substitute” patents for essential technology for a product may deter market competition if members of the patent pool are not allowed to license their patents to third parties independently (Lerner et al., 2007; Lerner and Tirole, 2004, 2007). In this case, the patent pool induces collusion by members, thereby resulting in an excessive concentration of market power in the patent pool members. However, scholars have also argued that this anticompetitive effect may not be substantial when the independent licensing by patent pool members with third parties is allowed. This is because independent licensing will lead the pool members to compete with one another on “licensing out” of owned patents to third parties; thus, third parties can access the pooled patents at a competitive price.

A firm's patent consolidation via patent transfers can be seen as the special case of patent pooling by a “single” firm without the competition effect of independent licensing. In this case, the discussion briefed above suggests that a firm's patent consolidation by patent transfer brings an anticompetitive effect when the consolidated patents are on substitute technology.

2.2. Patent consolidation and innovation

For innovation, firms’ patent transfers may bring both positive and negative impacts. On the one hand, a patent transfer facilitates the efficient reallocation of patents to those who can use the patents more effectively (Galasso et al., 2013) while promoting innovator's specialization in developing ideas (Lamoreaux and Sokoloff, 1999). For firms, transferring patents can be a useful instrument in terms of managing their intellectual property (IP) assets. For instance, firms liquidate their patents when they reconfigure their business lines (Serrano and Ziedonis, 2019) and finance operations (Hall, 2019; Hochberg et al., 2018) or when they want to acquire patents from outside to capitalize on emerging technological opportunities (Kelley, 2011). Meanwhile, as Kitch (1977) hinted, an upstream firm's consolidation of patents could enable the firm to mitigate downstream patent races, thereby enhancing the upstream firm's innovation appropriability.

On the other hand, patent transfer enables firms to aggregate patents on important technology for product development, which may negatively impact the innovation activities of competitors by equipping the patent-aggregating firm to impose ex-post patent holdup risk to the competitors’ market operations. Indeed, Kwon (2020) showed firms’ incentives to purchase patents to leverage market competitor's ex-post patent holdup risk. Doing so can effectively increase the rivals' operating cost while making them less active in developing the relevant technologies around the transferred patents.

The detrimental impact of a firm's patent transfers on innovation could be acute when innovation is developed in a highly cumulative manner (Rai, 2001; Scotchmer, 1991) and a firm's patent enforcement is effective in deterring its competitors’ R&D activities (Lampe and Moser, 2010). This is because patenting the technological inputs for a product could prevent the diffusion of that technology for follow-on innovation (Galasso and Schankerman, 2015). This is advantageous for firms seeking the monopolistic market profit (Figueroa and Serrano, 2019; Hahn, 1984). Studies attempted to find empirical evidence (Galasso and Schankerman, 2015; Murray et al., 2016; Murray and Stern, 2007; Sampat and Williams, 2019; Walsh et al., 2003, 2007). For instance, by analyzing the data on public Human Genome projects and Celera's gene-level intellectual property rights (IPRs), Williams (2013) showed that the firm's claim on gene-level IPRs significantly hampered subsequent scientific research and product developments for a long time even though the firm owned the IPRs for a short term. By using the patent–paper pair approach, Murray and Stern (2007) showed that granting patent rights on scientific discovery decreases the diffusion of scientific discovery, indicating that the subsequent scientific research is hampered by patenting. Murray et al. (2016) reached a similar conclusion by showing that the event of the National Institute of Health's agreements on reducing costs to access genetically engineered mice in the late 1990s diversified the pathways of subsequent research and innovation. Based on this finding, they argued that limited access to existing knowledge by patents might have limited the subsequent research diversity. Galasso and Schankerman (2015) showed that the invalidation of patents induced more subsequent inventions, and this effect was particular in the complex technology field where innovation is developed in a highly cumulative manner.

The software sector is one of the “complex technology” areas where innovation is created cumulatively and patent enforcement is effective for market competition. A computer program (i.e., software) is often built cumulatively (Bessen and Maskin, 2009; Cohen and Lemley, 2001; Hall and MacGarvie, 2010; Noel and Schankerman, 2013) on previously developed software—the extent to which one can access the source code of the prior version of the software is crucial for the new software development (Ghidini and Arezzo, 2013a). Meanwhile, software patents are important complementary assets for appropriating software innovation (Fosfuri et al., 2008). Accordingly, in the software sector, owning software patents is crucial for market operations and for seeking strategic benefits in market competition. Indeed, Wen et al. (2013) empirically showed the deterrence of the overall OSS development activities due to software patent enforcement by proprietary companies against the Linux community (OSS). Accordingly, in the software sector, a firm's consolidation of software patents may result in deterring the software innovation activities of competitors.

Our discussion above suggests that when the cumulative innovation is crucial for firms' market operation and patent enforcement is effective, a firm's consolidation of patents on key technology for the cumulative innovation process could discourage the follow-on innovation of its competitors.

In sum, our literature review hypothesizes that a firm's consolidation of the substituting patents on key technology for the cumulative innovation process may bring the anticompetitive effect, which then becomes the subject of the antitrust regulation. In this case, antitrust regulation of patent consolidation could positively affect the follow-on innovation of the patent-consolidating firms’ competitors. More formally:

When a firm consolidates patents on substitute technology for what it already owned, the antitrust regulation of such patent consolidation positively affects its market competitors’ follow-on innovation.1

In the next section, we describe our empirical strategy to test the formulated hypothesis.

3. Empirical setting

3.1. Regulation of Novell's software patents transfers

The empirical work of this study is based on the case of the DOJ's intervention into Novell's software patents transfer to CPTN in 2011. Novell's primary business domain was the Linux-based OSS market. It was also a founder of the Open Invention Network (OIN), the largest nonaggressive defensive patent pool for the Linux community. By joining the OIN licensing program, firms could freely use the patents that are shared through OIN. In return, the OIN licensees and member companies (Novell, Red Hat, IBM, Philips, Sony, and NEC) were obliged to share their patents with other OIN licensees.

In 2010, Attachmate Inc. (hereafter, Attachmate) announced its decision to acquire Novell Inc. Before the acquisition, Novell tried selling about 800 Linux-related software patents to CPTN, a consortium of the four large proprietary software companies (MS, Oracle, EMC, and Apple).2 A two-stage patent transfer process was planned. In the first step, CPTN was to acquire all the Novell patents. In the second step, CPTN was to redistribute the acquired patents to its member companies. For this deal, the four companies of CPTN paid $450 million in cash.

Before December 2010, the members of CPTN have not been known except MS.3 In early December 2010, the German Federal Cartel Office (Bundeskartellamt) fully disclosed the members of CPTN. After this information disclosure, the Open Software Initiative (OSI) requested the German authority to investigate if Novell's patent transfer to CPTN members has violated German antitrust law. In late December 2010,4 two OSS advocates, namely, OSI and the Free Software Foundation (FSF), sent a joint letter to the DOJ to express their concern that Novell's patent transfer would destroy the Linux community and consequently deter its innovation. They argued the following:

•

Novell was dedicated to promoting the Free/Libre and Open Software while using their patents primarily for defensive purposes.

•

Linux software is the major competitor of the products of CPTN member companies.

•

CPTN member companies have major market shares in the operating systems, virtualization software, and middleware markets.5

The OSI and FSF were concerned that CPTN members might use Novell patents to prevent further development of OSS. They feared this because the CPTN members were not obliged to share their patents with the Linux community and they have a strong motivation to deter the development of the Linux-based software for market competition.

The DOJ responded to this concern by starting an investigation into whether Section 7 of the Clayton Act was violated by Novell's patent transfer. According to the disclosed information, the DOJ requested detailed information about the patent transfer deal from CPTN in late January 2011.6 The 8-K document filed by Novell reports an additional request by the DOJ for further information about the deal in March 2011.7 On April 20, 2011, the DOJ announced its decision to intervene in the deal to address the immediate antitrust concern. The intervention was a partial prohibition of the patent transfer.

First, MS was not allowed to retain any Novell patents. The regulation required the transfer of patents that MS is planning to acquire to CPTN and then transferred back to Novell. Consequently, neither MS nor the other three CPTN members could acquire titles to these patents. Second, EMC agreed not to acquire 33 virtualization software-related patents (i.e., virtual machine technology) because EMC was considered the dominant player in the virtualization software market.8 Third, all the patents retained by Novell became subject to GPLv2 and OIN licensing. Thus, even though Novell disjoined the OIN, these patents were being shared with the Linux community through the OIN license program.

In testing our hypothesis, the DOJ's partial regulation of Novell's patent transfer is a useful case. The CPTN members were the dominant software market players who already owned many software patents that substitute for the Novell patents. Meanwhile, the member companies of the OIN were the primary entities that led the Linux-based OSS development to compete with the CPTN member companies’ software products. The software patent is the crucial complementary asset for OSS appropriation (Fosfuri et al., 2008), whereas IPR protection on software could be damaging to follow-on software developments by those who have no accessibility to the patents (Ghidini and Arezzo, 2013a). Therefore, the CPTN members’ acquisition of the OSS patents from Novell could hamper the software market competition and the OSS innovation. Accordingly, the case of the DOJ intervention into Novell patent transfer meets the conditions of which our hypothesis is likely to support.

Moreover, the quasi-experimental features of the case allow us to conveniently conduct causal analysis. The DOJ's intervention was a result of the German authorities’ disclosure of the list of CPTN members. Hence, the DOJ's regulation was an exogenous event to the involved parties in the patent transfer deal. More importantly, the DOJ's intervention was not to regulate the transfer of a specific set of patents that were particularly important for Linux software development. Instead, the DOJ prohibited the transfer of patents that MS tried to acquire.

3.2. Overview of empirical analysis

Our econometric strategy consists of two analyses. The first analysis is the patent citation analysis. By considering the patent citations accrued to a patent of interest (i.e., patent forward citation) as the proxy for the number of generated follow-on innovations (e.g., Drivas et al., 2017; Galasso and Schankerman, 2015), we estimate the difference in the forward citation counts accrued to the transfer-regulated patents (hereafter, regulated patents) and the transfer-allowed patents (hereafter, transferred patents) after the DOJ's intervention. To this end, we employ the following three complementary econometric methods: cross-sectional data analysis with selected control variables, panel difference-in-differences (DID) analysis, and the synthetic control (SC) approach. If the DOJ's partial regulation of patent transfer was effective to prevent the damaging impact on the follow-on innovation activities of the OIN member companies (i.e., the competitors of the patent-consolidating firms), we expect that the forward citation count of the regulated patents originated from the member companies of OIN should be greater than the forward citations accrued to the transferred patents, after the DOJ's intervention. We additionally conducted several supplementary analyses to rule out rival explanations.

The second analysis is to complement patent citation analysis. Although the patent citation is useful information for measuring the follow-on innovation, not all innovations are patented nor do patents represent innovation. Furthermore, the patent citation rate is often low; thus, measuring the causal impact solely by using patent citation analysis may suffer from the measurement error. To overcome this limitation, we analyze software trademark and copyright-filing activities of companies as alternative measurements of their software innovation activities. Our analysis examines whether the member companies of the OIN experienced a decline in software trademark and copyright-filing activities after the partial transfer of the Novell patents to the CPTN member companies.

3.3. Data

We start by identifying the US patents that were subject to the patent transfer deal between Novell and CPTN by using USPTO patent assignment data (Graham et al., 2018). Our initial search identifies 866 patents. Among them, 760 are the granted patents, which we include in the sample.

In the empirical analyses, we use a variable that counts the number of patent citations accrued to the patent in question beginning 4 years before the DOJ's intervention date (i.e., April 20, 2011) as a control variable. To avoid the truncation problem in the use of this variable, we exclude the patents that were filed after April 21, 2007, resulting in 432 patents in our final sample.

For each patent in the data, we identify which member company of the CPTN became the new patent owner by searching for the patent ownership change history made by CPTN as the patent assignor after April 2011.9 Among the 432 patents, 97 (22.5%) patents were transferred to Oracle (September 2011), 100 patents (23.2%) were reassigned to Apple (June 2012), and 109 patents (25.2%) were transferred to EMC (September 2011). Meanwhile, the rest of the patents (126 patents, 29.2%) were transferred back to Novell (September 2011). The DOJ ruled that the patents that could be transferred to MS must be retained by Novell; thus, the 126 patents are the regulated patents, and the rest (306 patents) are the transferred patents. Fig. 1 depicts the transfer of Novell patents.

Fig. 1

Download : Download high-res image (295KB)Download : Download full-size image

Fig. 1. Transfer of Novell Patents.

We then obtain detailed information about each patent in our data from patentview.org (serviced by USPTO). The distribution of the National Bureau of Economic Research (NBER) subcategories shows that most of the 432 patents are related to software (181 Computer Hardware & Software, 103 Electronic Business Method & Software, and 99 Information Storage).

3.4. Variables

Dependent Variables: We define a follow-on innovation as a new technical idea that was built upon a previously existing innovation. Based on this definition, we operationalize the number of follow-on innovations for a patented innovation by using the number of patent citations accrued to the focal patent in a certain time window (e.g., Drivas et al., 2017; Galasso and Schankerman, 2015) (i.e., patent forward citations). In our analysis, we set the time windows for counting forward citations to the period between the beginning on the date of the DOJ's intervention and up until 4 years later (i.e., 4-year window).

For hypothesis testing, we use the two separately counted forward citation variables by the citation origin: citations from the competitors of the patent-consolidating firms (PostCiteOIN) and citations from the patent-consolidating firms (PostCiteCPTN).

The first dependent variable, PostCiteOIN, is defined as the number of forward citations originated from the patents belonging to the members of OIN (i.e., Sony, NEC, IBM, Red Hat, and Philips), excluding Novell. Given that the OIN members were the competitors of the patent-consolidating firms in software market, this variable is a proxy for the number of follow-on innovations created by the competitors of the patent-consolidating firms (i.e., CPTN members) after the DOJ intervention.

The second dependent variable is PostCiteCPTN, which accounts for the number of forward citations originated from the CPTN members’ patents following the DOJ intervention. This variable is a proxy for the number of follow-on innovations created by the firms that were trying to consolidate the Novell patents.

Independent Variable: The independent variable Regulated is a binary variable that takes the value of 1 for the regulated patents and 0 for the transferred patents. The regulated patents comprise the treatment group, and the transferred patents belong to the control group.10

Control Variables: To rule out probable confounding effects in estimating the impact of the DOJ intervention, we introduce several control variables into our regression analysis. The regulation of the patent transfer by DOJ was “targeted” to the Novell patents that were planned to be acquired by MS. Thus, we select the variables that are likely to simultaneously correlate with the MS’ selection of Novell patents for consolidation and the number of follow-on innovations of the patent.

First, the number of created follow-on innovations for a patent after the DOJ intervention may be highly dependent upon the degree to which the patent has become the basis for the subsequent patented innovation before the intervention. Meanwhile, the DOJ decision might have considered the technological importance of the patent of interest for software development in general. To rule out this confounding effect, we control for the number of created follow-innovations for each Novell's patent based on the number of forward citations that the patent of interest received from 4 years before the DOJ intervention to the date of the DOJ intervention. We count the number of forward citations up to 6 months before the intervention date because the full list of the CPTN member companies was disclosed in December 2010. PreCiteOIN takes the forward citation counts that originated from the CPTN member companies during the pre-intervention period. This variable is introduced as a control variable in regressing PostCiteOIN. Likewise, when regressing the PostCiteCPTN, we control for PreCiteCPTN that takes the patent forward citation counts originated from the CPTN member companies before the DOJ intervention.

Second, we control for the patent's legal scope by using the number of independent patent claims (nClaim). The broader its legal scope, the greater the probability that the patent limits the legal scope of subsequent patents (Marco et al., 2019) and that the patent is involved in patent infringement litigation (Lanjouw and Schankerman, 1997). Patents with a broad legal scope could be useful in strategic patent enforcement. Thus, MS might be interested in acquiring these patents, but their transfer could catch the attention of the DOJ. Controlling for nClaim allows us to partial out this probable confounding effect.

Third, as the DOJ stated, its primary concern was the probable discouragement of further development of Linux-based applications, middleware, virtualization software, and operating system as a result of the patent transfer. Thus, the transfer of patents relevant to these application areas could become the primary target of the DOJ's intervention, while these patents were targeted by MS for strategic purposes. Meanwhile, these applications could be related to emerging technological opportunities in the software market, and thus, the relevant patents could receive more patent citations than those for other applications. To partial out this spurious effect, we introduce a binary variable DoJConcern that takes the value of 1 if the patent of interest contains at least one of the following clauses in its abstract: Linux, Unix, software, middleware, open-source, operating system, and virtual machine. The search was able to identify 98 patents in the sample.

Fourth, the member companies of OIN had major stakes in the GNU/Linux software market, whereas the GNU/Linux software was the primary competitor of MS products. Hence, MS might have the incentive for acquiring patents that closely relate to OIN members’ technologies for monopolistic market profit (see the “Incentive for a firm to preempt patents for monopolistic market profit” (Chang, 1995)). Simultaneously, patents relevant to the technological domain of the OIN members could have been a part of a promising technology area, thus attracting more follow-on innovations than other software patents. To rule out this explanation, we control for whether the patent of interest has a technological link with the patents owned by the OIN members by using the patent citation information. CiteOINPat takes the value of 1 if the patent of interest cites patents owned by the OIN members before the DOJ intervention, and 0 otherwise.

Fifth, the genuine value of a patent is realized through a patent portfolio (Choi and Gerlach, 2017; Parchomovsky and Wagner, 2005). Firms often obtain patents to organize their patent portfolio into a well-interconnected web of IP rights. In this regard, MS probably tried acquiring Novell patents related to MS's already-owned patents. Meanwhile, the patents closely related to the patents of MS might be relevant to emerging technological opportunities in the software market, attracting more follow-on innovation than other patents. To control for this effect, we introduce CiteMSPat as a control variable, which takes the value of 1 if the patent of interest has cited any MS-owned US patents filed before the DOJ's intervention.

Sixth, Novell's patent transfers were the subject of the German authorities’ investigation regarding the violation of the antitrust law. Hence, those patented innovations that were protected in Germany could have caught the attention of the German authorities, which, in turn, triggered the DOJ's intervention later. As shown by many studies, patents with protections in multiple jurisdictions often contain commercially valuable inventions (e.g., Harhoff et al., 2003; Lanjouw et al., 1998). MS may have also planned to purchase Novell's commercially valuable patents in the first place. We control for this effect by introducing EuroFam that takes the value of 1 if the Novell patents have EPO family patents. Finally, we control for the patent application year (AppYr) and technology field fixed effects (TechFE) by using the NBER subcategory (Hall et al., 2001). Table 1 summarizes the descriptions of these key variables.

[TABLE OMITTED]

Our hypothesis expects that the patent forward citations originated from OIN member companies will be greater for a regulated patent than a comparable transferred patent, after the DOJ intervention.

3.5. Descriptive analysis

Table 2 provides the summary statistics for the key variables. A simple comparison of the mean values reveals notable differences between the regulated and transferred patents in the two forward citation variables. On average, the regulated patents received more citations both before and after the DOJ's intervention than the transferred patents. This suggests that even though the DOJ's intervention was triggered by a seemingly exogenous event, the two groups of patents are different in their technological importance, indicating the necessity of controlling for the forward citation count accrued to the patents before the DOJ's intervention.

[TABLE OMITTED]

[TABLE OMITTED]

4. Results

4.1. Regression results

Cross-sectional data analysis. We fit our data to the econometric model employed by Galasso and Schankerman (2015). This model controls for the pre-intervention dependent variable (in our analysis, the PreCiteOIN and PreCiteCPTN). Therefore, the estimated coefficient of the independent variable captures the difference in the number of follow-on innovations for the regulated and transferred patents that had the same number of follow-on innovations before the DOJ's intervention. We used the ordinary least squares (OLS) estimator with robust standard error for regression. For the estimation, we take natural logarithms of the two dependent variables incremented by 1 as these variables have right-skewed distributions and zero as the minimum value. Table 4 presents the regression results.

[TABLE OMITTED]

The first column reports the regression result using ln(PostCiteOIN + 1) as the dependent variable. The regression result shows a positive and statistically significant coefficient of Regulated at the 0.05 significance level. The number of follow-on innovations originating from OIN member companies for a regulated patent has been estimated to be 15.6% greater than that for a comparable transferred patent, following the intervention.

The second column presents the estimation result using ln(PostCiteCPTN + 1) as the dependent variable. The coefficient of Regulated is statistically insignificant at the 0.1 significance level, indicating no evidence of the difference between regulated and transferred patents in the number of CPTN member's follow-on innovation after the DOJ's intervention.

To further check the empirical specificity of our findings, we additionally run the regression for the forward citation counts accrued to the regulated and transferred patents from those who were not the OIN nor CPTN members (ln(PostCiteOthers + 1)). Column 3 shows that the coefficient of Regulated is statistically insignificant at the 0.1 significance level—there is no evidence showing that the DOJ's intervention influenced the creation of follow-on innovation by individuals or organizations that were not involved in Novell's patent transfer deal. Our regression results are robust to the count-variable regression models (negative binomial and Poisson models) and a placebo test. Appendix A reports the robustness check.

Panel DID analysis. One may raise a concern that our cross-sectional data analysis may suffer from omitted variable bias because the control variables we introduced into the regression may not be sufficient to capture all the probable confounding effects of differences in patent-level characteristics. To address this concern, we fit our data to the patent-year fixed effect panel OLS regression model by reformatting our data into the panel data. We then create the time dummy variables corresponding to each patent citation year from 4 years before to 4 years after the DOJ's intervention date. Then, the interaction terms between these time dummy variables and Regulated are created. In the panel DID analysis, we regress yearly counted patent forward citations originated from OIN and CPTN member companies, respectively. Fig. 2 presents the estimated coefficients of the interaction terms with their 95% confidence intervals.

[FIGURE OMITTED]

Note that the estimated coefficients of the interaction terms (TimeDummies x Regulated) for the pre-intervention period are not far from 0 for the regression on the forward citation count originated from both OIN and CPTN member companies. This indicates the parallel time trend of the forward citation counts accrued to the regulated and transferred patents pre-DOJ intervention. Meanwhile, in the post-intervention period, the coefficients of the interaction terms are positive and statistically significant at the 0.05 significance level only for the forward citation count originated from OIN. This result confirms that the OIN members developed more follow-on innovations for the regulated patents than the transferred patents after the DOJ intervention. In contrast, there is no evidence showing that the DOJ's intervention affected the follow-on innovation development by the CPTN member companies.

4.2. Synthetic control method

One of the limitations in drawing causal inferences using the prior two analyses is that the estimator could be biased due to the difference between the treatment and control groups in the treatment-free time trend of the outcome variables in either pre- or post-treatment (or both) period. Our panel DID regression result showed that the assumption of the parallel time trend of outcome variables for regulated and transferred patents in the pretreatment period holds. However, this does not guarantee the parallel time trend of outcome variables in the post-treatment period if the DOJ had not regulated the patent transfer deal. We thus mitigate this econometric concern by reanalyzing our data using a synthetic control (SC) approach as a complementary method (Kreif et al., 2016). The SC approach mechanically forces the pretreatment period outcome of the treatment and control group to become virtually identical. Under the assumption that the free-treatment post-treatment outcomes of the treatment and control groups can be expressed into a linear combination of selected covariates and pretreatment outcomes, the SC method yields an unbiased estimation of the treatment effect (Abadie et al., 2010).

The dataset employed in this study has multiple regulated patents (treatment units) and transferred patents (control units). Hence, we use the method for multiple-treated and control unit case as proposed by Kreif et al. (2016). Our SC analysis presented in Appendix B reaches the same conclusion as our DID analysis results.

In sum, our cross-sectional data regression analysis, panel DID estimation, and the SC method found that the DOJ's regulation has positively affected the OIN members’ development of follow-on innovation for Novell's patents.

We conduct supplementary empirical analyses to rule out three rival explanations of our findings in the next sections. The first supplementary test is to specify the origin of our findings. The second analysis is to examine whether our findings were stemmed from the fact that MS was excluded from the patent consolidation rather than the DOJ's intervention effect. The last analysis is to ensure that our findings are about the impact of DOJ's intervention on follow-on “innovation” rather than “patenting” activities of the OIN member companies.

4.3. Origin of the effect

Our literature review explains our findings that the DOJ's regulation mitigated patent consolidation's discouragement effect on follow-on innovations (hereafter, mitigation effect). Yet, another explanation could be that the DOJ's regulation “promoted” follow-on innovation for regulated patents by OIN members (hereafter, promotion effect). We conduct additional empirical analyses to examine which explanation is more credible.

If the mitigation effect worked, the number of OIN members’ follow-on innovation for the transferred patents should have been lower than that of comparable patents that were not involved in the transfer deal (hereafter, counterpart patents). In contrast, if the promotion effect worked, the number of follow-on innovations for the regulated patents created by OIN members should have been greater than that of the comparable counterpart patents.

For the test, we consider the patents that were originally filed by the four CPTN members as the counterpart patents. The CPTN members were in intensive market competition in the software market with the OIN members, while participating in the consortium for acquiring Novell's patents. Hence, we argue that the patents filed by CPTN member companies are a reasonable comparison group. The forward citation counts accrued to the counterpart patents are the baseline in this analysis. To implement this empirical strategy, we choose the US patents that were filed between January 1, 1990, and April 20, 2007, by the four CPTN members.

The regression results in Table 5 present that the number of OIN members’ follow-on innovations for the transferred patents is about 6.4% lower than that of the comparable counterpart patents in the post-DOJ regulation period. Meanwhile, the coefficients of Regulated are statistically insignificant at the 0.1 significance level. Our additional analysis indicates that the CPTN members’ consolidation of the Novell patents harmed the development of OIN members’ follow-on innovations; thus, the DOJ's intervention is seen to have mitigated this effect. The existence of the “promotion effect” by the regulation is not evidenced.

[TABLE OMITTED]

4.4. Buyer effect or regulation effect?

In the main analysis, we used the Novell patents that were transferred to Oracle, EMC, and Apple as a comparison group of the regulated patents. A downside of this empirical strategy is its incapacity to isolate the effect of the antitrust regulation from the probable confounding effect by the fact the patent acquirer was not MS. That is, the effect of regulation could be particular for patent consolidation by three firms of the CPTN, but not necessarily so if MS was allowed to acquire the Novell patents.

To address this concern, we replace the comparison group with the software patents that were transferred to MS from America Online (AOL) in June 2012. In 2012, AOL announced that it sold some of its software patents to MS for patent monetization.11 AOL was also involved in developing OSS just as Novell did, and the patent transfer timing was close enough to Novell's patent transfer (i.e., 1-year difference). Hence, AOL's patents that MS could acquire (hereafter, AOL-MS patents) can be a reasonable alternative comparison group of the regulated patents. If our finding was not solely of the buyer's effect, a similar finding as that of the main regression is likely to be observed. In contrast, if the patent buyer effect was behind our findings, the number of follow-on innovations for the regulated and the AOL-MS patents are expected to be indifferent.

We collect information on the US patents that were transferred from AOL to MS in June 2012,12 which were filed before June 2008 (4-year before the transfer). In the USPTO patent assignment data, we could identify 90 AOL-MS patents. Then, we conduct the same regression analysis as the main regression. Table 6 presents the regression result. In this analysis, the AOL-MS patents have become the comparison group for both regulated and transferred patents.

[TABLE OMITTED]

The statistical significances and signs of the coefficients of Regulated are consistent with those in the main regression result. Meanwhile, the coefficients of Transferred are statistically insignificant at the 0.1 significance level. This result indicates that our finding was not driven by the “buyer's effect,” but more of the DOJ's regulation effect.

4.5. Patent or innovation? Evidence from copyright and trademark data

In our research design, we used patents as the outcome of a firm's innovation activity. The shortcoming of this operationalization is that our findings may have restrictively indicated decreased software “patenting” activities of OIN members by the partial transfer of the Novell patents to the three CPTN members. To analyze the software innovation activities more comprehensively, we conduct two complementary analyses using two alternative measurements of a firm's software innovation activities; software copyrights, and trademarks filing activities.

Software Copyrights: Copyright is a nonpatent IP right that allows a limited monopoly on technical ideas on software (Graham and Mowery, 2003). Thus, examining firms’ software copyright-filing activities can measure their software innovation activities. The subject of software copyright protection is not a conceptual or mathematical algorithm, but program source code or the computer file itself, which makes copyright filing a more direct measure of firms’ software innovation activities.13 If the partial transfer of Novell patents did not affect OIN members’ overall software innovation activities, their copyright-filing intensity on new computer programs (or source code) is expected to remain, following the partial patent transfer. In contrast, if Novell's partial transfer of patents to the CPTN members has negatively affected software innovation activities by OIN members, their software copyright-filing activities should have decreased after the transfer. To operationalize this test, we collect information about copyrights filed by OIN member companies from the US copyright office's database.14 The copyright-filing activity of MS serves as the baseline in this analysis because MS was not in the OIN program and was prevented from acquiring Novell patents by the regulation. We categorize copyright as software copyright if the subject of the copyright is “computer program.” Fig. 3 displays the results.

[FIGURE OMITTED]

On the left side, we present the total number of software copyrights filed by the OIN members and MS, before and after the partial transfer of the Novell patents.15 The comparison shows that OIN members’ software copyright-filing activities decreased after the partial transfer of the Novell patents.

On the right side, we present the change in the between-firm share of software copyrights filed by the OIN members and MS, calculated by using the total number of software copyrights filed by the top 100 software companies as the denominator.16 The OIN members’ average share of software copyrights substantially reduced after the partial transfer of the Novell patents, whereas MS's did not.

Software Trademarks: Firms’ product or service innovation can be measured by their trademark-filing activities because trademark law protects products’ brand names. As such, studies use trademark-filing activities of firms to capture firms’ innovative activities (Flikkema et al., 2014; Mendonça et al., 2004; Srinivasan et al., 2008; Stoneman, 2010). If Novell's partial transfer of its patents to the CPTN members negatively affected OIN members’ software innovation activities, their software trademark-filing activities should have dropped after the transfer. We consider MS's trademark-filing activities on software as a baseline in this analysis, as we did in the copyright data analysis. We retrieve information about the trademarks that were filed by the OIN members and MS from the USPTO trademark data (Graham et al., 2013).17 We mark a trademark as a “trademark on software” if the trademark document contains the clauses of “software for” or “computer program(s) for” in its goods and service description while being categorized into international class “009″ (“Electrical and scientific apparatus”).

Fig. 4 presents the total number of software trademarks (left) and between-firm share of software trademarks filed by the OIN members and MS, before and after the partial transfer of the Novell patents. The data show that the OIN members experienced decreased software trademark-filing activities after Novell's partial transfer of patents.

[FIGURE OMITTED]

In sum, our additional analysis using copyright and trademark data supports the conclusion that the CPTN member's consolidation of a part of Novell patents has negatively affected the OIN members’ software innovations, and, accordingly, the DOJ's partial regulation of patent transfer helped OIN members’ software innovation activities.

5. Discussion and conclusions

This study examined how the antitrsut regulation of firms' patent consolidations affects the development of follow-on innovations. Based on our reconciliation of relevant prior studies, we hypothesized that in a market sector where cumulative innovation is crucial to the competition, a firm's consolidation of patents on substitute upstream technology for what it already possesses discourages the development of follow-on innovations by the patent-consolidating firm's competitors. Accordingly, regulation of such patent consolidation is expected to positively affect the development of follow-on innovations by the patent-consolidating firm's competitors.

We then empirically tested the derived hypothesis using the case of the DOJ's partial regulation of Novell's software patents transfer in 2011.

Our empirical analysis used patent citation, software copyright, and software trademark data. Results indicated that antitrust regulation of the patent transfer mitigated the negative impact of the patent consolidation on the follow-on innovations development of the patent-consolidating firms’ competitors.

'Does our finding imply that regulating patent consolidations is necessary for social welfare? This study does not provide a complete answer. First, the DOJ's regulations primarily aim to encourage downstream market competition. In contrast, this study focused on the impact of regulations on follow-on innovations. The overall impact of regulations on social welfare should not be dictated by our findings because even though innovation and market competition are related to social welfare, they are not the only ones that compose social welfare. A more comprehensive analysis of how regulations of patent transfer affect other dimensions of social welfare is necessary.

Second, the regulation of patent consolidation can bring undesirable consequences to other dimensions of social welfare from a market competition dynamics standpoint. In particular, when failing firms try liquidating their patents, and the only buyer of the patent is the patent-consolidating firm, the authority's restriction of such patent transfer deals may impose an exit cost to the patent-selling firm. The imposed exit cost discounts the value of the patent ex-ante while limiting the patent holder's enforceability. Restricted patent enforceability could attract market entry by less-efficient firms (Gilbert and Shapiro, 1996); hence, regulation of patent transfer deals may even harm consumer welfare.

Third, whether promoting follow-on innovations is even desirable from a social welfare standpoint must be considered. In some sectors, follow-on innovations often result in marginal improvements to existing innovations. The encouragement of the follow-on innovation development of existing ideas could result in the duplication of R&D in some fields; therefore, restricting patent transfers to encourage “follow-on” innovation could result in an inefficient allocation of resources for innovation (Gilbert and Newbery, 1982; Wright, 1983).

Our empirical analysis was based on a case in the software sector; hence, one may question to what extent our findings are generalizable across sectors. The rationale derived from our literature review suggests that similar findings can be expected in sectors characterized by two features—(1) cumulative innovation that is crucial for firms’ market operations and (2) patent enforcement that is effective for market competition. For instance, telecommunications technology, like software, is often developed by inheriting key functions in prior generational technology, whereas many technological components therein are patentable and effectively enforceable. In contrast, in the chemical sector, individual innovation is highly distinct from existing innovation; hence, the cumulativeness in the innovation process may be less prominent than the software or telecommunication device sector. Accordingly, our findings may not hold in this sector. Whether the sectoral heterogeneity exists indeed and what factors further explain the probable sector-level heterogeneity are intriguing questions for future studies.

Our cross-sectional data analysis showed that the DOJ's regulation increased follow-on innovations of OIN member companies by 15.6%. Is this impact economically significant? Although our study alone cannot answer this question, prior studies that estimated the change in patent forward citation rate induced by other types of exogenous events provide benchmarks. For instance, Galasso and Schankerman (2015) estimated a change in the patent forward citation rate after the invalidation of patents by patent opposition. Their estimation showed that, on average, the invalidation of a patent increased the rate of five-year patent forward citations by 50%. A similar study by Gaessler et al. (2017) found that the invalidation of patents increased the five-year-window patent forward citation by 23%. A 15.6% change is below these benchmarks in magnitude. However, we argue that this does not necessarily discount the economic significance of the impact of the DOJ regulation because it was an inherently different event from the patent invalidation. Moreover, our measurement of follow-on innovations was based on four year-window patent forward citation count instead of five-year-window citation count.

Firms may seek monopolistic profit, in addition to the patent transfer, by consolidating patents through preemptive patenting (Gilbert and Newbery, 1982)—patenting before their market competitors obtain them while not using the patented technology nor licensing. However, how preemptive patenting may affect “innovation” and its welfare implications remains ambiguous. From the Schumpeterian viewpoint, firms’ incentive for preemptive patenting could function as the driver for firms to engage more in R&D. Meanwhile, the presence of the preemptive patenting strategy may trigger the competitors to explore the new pathway of innovation. In contrast, the preemptive patenting on marginally improved technology or with overly broad/uncertain patent scope may discourage the competitor's market operation and innovation in the short run. This could be particular in the sector where innovation is created cumulatively, as we have discussed above. This theoretical ambiguity calls for empirical studies on whether and how a firm's patents consolidation through preemptive patenting affects innovation.

In the present study, we elucidated that the formation of a market monopoly can be prevented by antitrust intervention into a firm's patent consolidation while encouraging the development of follow-on innovation. Our conclusion suggests further elaboration on the modern understanding of the relationship between antitrust and patent laws is neccessary, at least when it comes to innovation. The two laws become complementary when the right to access the substituting patents on an upstream technology is distributed across innovators; when the patents are owned by a single firm, complementarity between antitrust and patent law may not work. This implication echoes the ongoing policy and scholarly discussion on how the modern antitrust law should accommodate technological innovation (Katz and Shelanski, 2005; Khan, 2016; Sidak and Teece, 2009). Thus far, the discussion concerning the necessity of accommodating the Schumpeterian view on the relationship between market competition and innovation has been largely theoretical. Moreover, how the antitrust law plays a role in promoting technological innovation remains an open question. Our study suggests that firms’ patent consolidation through patent transfer is one of the venues where antitrust law enforcement can shape market competition and technological innovation. This implication also indicates the importance of extending the scope of antitrust intervention into “patent-only” transfers in cooperation with patent policymakers to promote market competition and technological innovation.

This study has several limitations that we wish future studies to address. First, our empirical analysis was based on a single case of antitrust regulation of patent transfer. Although the case was useful for identifying the causal impact of the antitrust regulation, a comprehensive conclusion and understanding need analyses on more cases.

Second, we focused on examining the impact of antitrust regulation of a firm's consolidation of patents on “substitute” technology. Yet, firms may acquire external patents on technology that is “complementary” to what they own. Here, the patent transfer may result in the generation of more innovation by improving the efficiency of the innovation process. In contrast, the aggregation of patents on complementary technology may concentrate more market power in the patent-aggregating firm by enabling the birth of a dominant “platform” (e.g., Haucap and Heimeshoff, 2014; Khan, 2016; Moore and Tambini, 2018), which has become a new issue in the antitrust policy domain. How the antitrust regulation of a firm's consolidation of patents on “complementary technology” affects innovation and market competition is an interesting question worthy of future study.