Patent Trading Flows of Small and Large Firms1

Nicolás Figueroa Pontificia Universidad Católica de Chile Carlos J. Serrano Universitat Pompeu Fabra and Barcelona GSE

This draft: October, 2015

1 We thank Patrycja Grzelonska, Victor Aguirregabiria, Anna Lejpras, Matt Mitchell and seminar participants for feedback and suggestions. Nicolás Figueroa is an Assistant Professor of Economics at the Pontificia Universidad Católica de Chile. Carlos J. Serrano is an Assistant Professor of Economics and Business at the Universitat Pompeu Fabra. Corresponding author: Carlos J. Serrano, [email protected].

Abstract This article investigates the patent trading flows of small and large firms and the determinants of these firms’ patent sale and acquisition decisions. We document the role that firm size, technological fit, the economic value of a technology, and patent technology sector plays in patent sales and acquisitions. To do so, we develop a new, comprehensive data set that matches information on patent grant and citations, patent transfers, and the firm size of patent owners over a patent’s lifetime. JEL Codes: L22, L24, O32, O34 Keywords: Transfer of patents; market for patents; innovation; division of innovative labor.

1

Introduction

For centuries, the market for patents —the secondary market for the transfer of the ownership of patent rights— has been important for inventors. For instance, an American inventor and businessman Thomas Edison financed the early stages of his career by selling patents.1 In another example, IBM transferred fifty nine patents covering "target-tracking" technology to Lockheed Martin Corp., a supplier of electro-optical/infrared targeting systems for the U.S. Marine Corps.2 More recently, in 2011, Novell’s patent portfolio -of 880 patents- was sold for $450 million to a consortium of companies led by Microsoft; and later that year, Apple and Microsoft teamed up to buy 6,000 Nortel patents for a record $4.5 billion.3 These examples reflect the pronounced role the market for patents has played in the economy: in helping to raise funds, facilitating technology transfer, and reallocating patent rights to entities that value them the most. Despite its importance, there is little systematic empirical work examining the transfer of patents. An exception is the work by Serrano (2006, 2010), that identifies patent sales through changes in patent ownership recorded at the United States Patent and Trademark Office (USPTO). Serrano documents the effect of firm size, patent age, technology class and the economic value of the technology on the sale of patents. It remains unanswered whether the transferred patents are acquired by either small or large firms and the determinants of these transactions. In contrast, the literature on the licensing of patents has produced a set of facts about the dynamics of cross-firm technology licensing, with a frequent focus on understanding the interactions between small and large firms (see e.g., Teece, 1986; Arora and Gambardella, 1990; Arora, Fosfuri and Gambardella, 2001; Gambardella, Giuri, and Luzzi, 2007; Palomeras, 2007). Patent sales need not follow the same patterns as technology licensing. They not only convey the right over the use of a patented technology (as licensing does), but also the right to let other firms use the technology, to enforce the patent against infringers, and the option to transfer all these rights, including the patent ownership, to other firms. This paper investigates the patent trading flows of small and large firms and examines factors that affect these transactions. Drawing on the literature of economics and management, we examine whether the decision to sell a patent, and whether the patented innovation is acquired 1

Lamoreaux, Sokoloff, and Sutthiphisal (2013) report that Edison sold rights in at least 20 of his first 25 patents. See patent assignment with reel/frame 8430-312, that was executed on August 29, 1996. 3 Both transactions were investigated by the U.S. DOJ. The large-scale of the patent deals and the high prices companies paid to keep patents away from rivals raised competition concerns (Scott Morton and Shapiro, 2014). 2

1

by either small or large firms, depends on the technological fit of an innovation with the current patentee and potential buyers, the economic value of the technology, and the patent technology sector (Teece, 1986; Cassiman and Ueda, 2006). These patterns can provide new evidence on the reallocation of patent rights and guidance in the assessment of models of patent trading and technology transfer. The key challenge in looking at the patent trading across small and large firm is the incomplete or non existing data on the identity of patent owners after a patent sale. The patent assignment records from the USPTO provide detailed information on the timing of change in patent ownership, but they lack standardized information on the identity of the patent buyers (Serrano, 2010). One possibility to identify the patent buyers may be to standardize the reported names that appear in patent assignment records, but many of the buyers create shell companies to hide information about the real party in the transactions, raising questions about the validity of following such an empirical strategy (see e.g., USPTO proposed new rules in 37 CFR § 1.271). Moreover, commonly used strategic alliances databases such as SDC Platinum of Thomson Reuters, which rely on information reported by publicly traded firms to the U.S. Security and Exchange Commission (SEC), are not likely a viable strategy either, in part because many patent deals do not involve publicly traded firms. In this paper, we take a different approach and identify whether a patent owner after a patent sale is either a small or large firm by tracking the patent maintenance fee records. Under section 41 of the U.S. Patent Act, small entities enjoy a 50% reduction over what large entities pay for application and renewal fees. Renewal fees are due no later than the end of the fourth, eight, and twelfth year following the patent’s grant date. Among patents owned by businesses, small entity status is assigned to firms with less than 500 employees (including all subsidiaries); the rest are classified as large firms. Based on this empirical strategy, we can identify the size of the original owner and all subsequent patent owners (small or large) as of each renewal date.4 Patent maintenance fee records are linked at the patent level with patent application, grant, and sales data. The merged data set is a panel of all patents applied for and granted to businesses in the U.S. during the period 1989-1996. For these patents, we constructed the history of patent sales and renewal decisions, and whether the patent owner is a small or large entity both initially and at each renewal date, until 2001. The final data set has 590,873 patents issued to 61,239 4

An obvious caveat of using patent maintenance fee records is that we cannot identify the actual identity of patent owners and the patents in their patent portfolio, except for the initial patentee (patentee as the patent’s grant date).

2

patentees. The main empirical findings of the paper are as follows. First, small firms are more likely to sell their patents, and acquire disproportionately more patents than large firms, especially patents initially owned by other small firms.5 Second, we show that the patents sold have lower technological fit with the initial patentee than the unsold patents. And among patents that were not sold, initial patentees let expire the patents with lower technological fit. Third, we also show that the traded patents are acquired by firms (small or large) that have on average a better technological fit with the patented innovation. Finally, we find that the patents acquired by large firms have on average the highest economic value. This paper contributes to a nascent literature studying the sale of patents using patent assignment data. Lamoureaux and Sokoloff (1997, 1999, 2001) use a sample of sales of individualinventor patents to provide the first historical account of whether organized markets for technology existed in the late 19th and early 20th centuries. Khan and Sokoloff (2004) show that the majority of the technologies patented by the 19th century "great inventors" eventually changed their ownership. Serrano (2006, 2010) creates a dataset on the transfer of patents and presents patterns describing the dynamics of patent sales and renewals.6 Serrano (2011) uses data on patent sales and renewals to quantify the gains from trade in the market for patents. More recently, patent assignment data has been used to measure the extent of the market for technology in an industry (Mani and Nandkumar, 2015), to study patent auctions (Odasso, Scellato and Ughetto, 2015), and to examine the interplay of trading and enforcing patent rights (see e.g., Chien (2011), Galasso, Schankerman, and Serrano (2013), and Haber and Werfel (2015)). Lastly, GonzalezUribe (2014) shows that the probability that a patent is sold increases after the patentee obtains venture capital financing; and Hochberg, Serrano and Ziedonis (2015) find that intensified trading in the secondary market for patents increases the annual rate of startups’ lending. This article is organized as follows. Section 2 describes the new data set and key variables. In Section 3, we provide some theoretical guidance to examine and interpret patent transactions. Section 4 presents the patterns of the patent trading flows across small and large firms. In Section 5, we provide evidence on factors affecting patent sales and acquisitions that may account for these patterns. Finally, concluding remarks close the paper. 5 Disproportionately means that small firms acquire a higher proportion of patents than the proportion of patents that is typically granted to them in a given year. 6 See Marco et al. (2015) for a recent description of the patent assignment raw data.

3

2

Data

The new facts we present in this paper are derived by merging three datasets. The first dataset contains information about the firm size of the patent owners as of the patent’s application date and renewal dates. We obtained this information using patent level application and maintenance fee records from the USPTO. The second dataset provides the transfer records decisions over the lifetime of patents. The third dataset includes information on patent citations and patent and initial patentee characteristics from patent grant records. The merged dataset is a panel of patents applied for and granted to corporations in the U.S. with their corresponding records on renewal and patent sales for the period 1989-2001.

2.1

Sample construction

Documenting the firm size of patent owners over a patent’s lifetime has been challenging. The work-horse data set of empirical patent research, the NBER Patent Citations data set, provides information on patent characteristics such as issue date, technology class, patent citations, and patent ownership as of the patent’s issue date, but does not keep track of changes in patent ownership post- patent issuance (Hall, Jaffe, and Tratjenberg, 2001). In Serrano (2010), patent assignment data has been used to identify changes in patent ownership, but not the names of patent buyers. An issue is that neither the names of sellers nor buyers are standardized by the USPTO. More critical is that the names reported are not necessarily the real parties in the transactions, making it unfeasible to systematically link these transactions to existing datasets on business characteristics.7 To overcome this challenge, this paper exploits a provision in U.S. patent law that regulates the assertion of small and large entity status and the notification of changes of entity status over the lifetime of patents. Under Section 41 of the U.S. Patent Act, the standard application and maintenance fees are subject to a 50% reduction for small entities.8 Among businesses, patents are assigned a small entity status if the patent holder is a business with less than 500 employees (including all subsidiaries); otherwise it is identified as large. Small entity status is established by 7

Some patent buyers have been found to create shell companies to hide their real name. Under 35 U.S.C. 41(h)(1), fees "charged under 35 U.S.C. 41(a), (b), and (d)(1) shall be reduced by 50 percent with respect to their application to any small business concern as defined under section 3 of the Small Business Act, and to any independent inventor or nonprofit organization as defined in regulations issued by the Director." This is a unique feature of the USPTO, all patentee pay the same fees at the European Patent Office and the Japanese Patent Office. 8

4

a written assertion of entitlement to such status or it is claimed with the payment of filing fees and maintenance fees over the life cycle of patents. After an assertion of small entity status has been established, a second subsequent assertion is not required when the assignment of rights or an obligation to assign rights is to other parties that are small entities too. On the contrary, the notification of loss of entitlement to small entity status, such as upon the sale of a patent from a small entity to a large entity, is required when issue and maintenance fees are due. In short, the regulation implies that the patent’s entity status reflects the actual firm size of patent owners. The first dataset we create combines information extracted from patent application and maintenance fee records to obtain the firm size for patent owners over the lifetime of patents. To identify whether patents were initially owned by small or large businesses, we use the patent application records of issued patents. We have access to these records from January 1, 1980 to December 31, 2000. To identify the firm size of patent owners over a patent’s lifetime, we make use of the patent maintenance fee records from January 1, 1992, to December 31, 2001.9 Maintenance fees are due no later than the end of the 4th, 8th and 12th year after the grant date of the patent.10 By linking both data sources, we can identify the firm size for the patent owners by the grant date and at each of the renewal events for all patents granted in 1989 and until 1996. The second dataset uses registrations of changes in patent ownership recorded at the USPTO in order to discern the sale of a patent. The source of the data is the Patent Assignment Database. When a U.S. patent is transferred, an assignment may be recorded at the patent office to acknowledge a change of ownership. Under Section 261 of the U.S. Patent Act, recording the assignment protects the patent owner against previous unrecorded interests and subsequent assignments. If the patentee does not record the assignment, subsequent recorded assignments will take priority. For these reasons, patent owners have strong incentives to record assignments and patent attorneys strongly recommend this practice (Dykeman and Kopko, 2004). A typical reassignment entry indicates the patent number involved, the name of the buyer (i.e. assignee), the name of the seller (i.e. assignor), the date at which the reassignment was recorded at the patent office and the date at which the private agreement between the parties was signed. We 9 An author obtained access from the USPTO to the maintenance fee records in exchange for advice on their cleaning, processing, and use. A task force at the patent office used the information on the patent entity status in order to better forecast the patent office revenues from maintenance fees. At the time that an author was involved with the patent office, the data in bulk was not publicly available. The processed data can be freely accessed nowadays at: http://www.google.com/googlebooks/uspto.html. 10 The failure to pay the fees by the due date will result in the patent’s immediate expiration.

5

have obtained this information for the period 1989-2001.11 A challenge in using reassignment data is to distinguish changes in patent ownership from other events recorded at the USPTO assignment data. We use an algorithm developed in Serrano (2010) that conservatively drops all the assignments that appear not to be associated with an actual patent trade between distinct entities. The algorithm drops assignments that refer to a "name change" of the patentee, to the patent being used as a collateral ("security interest"), to corrections ("corrections"), changes of address ("change of address"), assignments from an inventor to the employer ("first assignments"), and transactions between entities with the same name. In addition we drop assignments in which the buyer is the assignee as of the grant date of the patent, assignments recorded at the patent application date, assignments to financial institutions, duplicate assignments, etc.12 The details of the procedures we used to deal with the assignment data are explained in Serrano (2010). The remaining, post-cleaning, assignment records represent the transfer of patents and have information about patent numbers. The patent numbers make possible to merge the two previous datasets with the third one: the patent grant and citation data. The source of the patent grant and citation data is the NBER Patent Citations dataset (Hall, Jaffe, and Tratjenberg, 2001). We use patent characteristics such as technology class, citations received, and citations made. We also exploit the unique identifier of each initial patentee, which allows us to construct their patent portfolio as of the patent’s grant date. Matching data on the characteristics of the initial patentee to patents is meaningful as long as the patent is owned by the original patentee at the time of the transaction. To ensure this, we focus our analysis on the first transfer of a patent. As we explained above, we have no detailed information of the patent portfolios of subsequent owners. The majority of patent sales occur in the very first years post-issuance and prior to the first maintenance date (four years after patent issuance). Similarly, we also focus our analysis on patent sales up to the first maintenance date since patents may be let to expire thereafter. For these patents, we can observe whether the initial owner and the owner as of the first renewal date is either a small or large firm. In our empirical analysis we focus on corporate patents, patents that were applied for and 11 A summary of the patent assignment data as well as an detailed discussion of its advantages and disadvantages can be found in Serrano (2006, 2010). 12 By dropping transactions to financial institution we eliminate transactions in which a patent may have been used as collateral in a loan, or that the possibility that patent was transferred to the financial institution in case the loan was not paid back.

6

granted to businesses. Corporate patents represent about 75 percent of all issued utility patents. For these patents we have patentee’s unique identification numbers, which allows us to construct the patentee’s patent portfolios as of the grant date of the patents. The rest of the patents were applied for and issued to federal agencies, universities, individual inventors, and individually owned. In addition, our analysis is restricted to patents granted in between the years 1989 and 1996, for which we have sale and renewal records until 2001. This allows us to observe for at least five years, for each patent cohort, the trading and renewal decisions. The final dataset has 590 873 patents, issued to 61 239 patentees. The dataset contains information on the history of patent sales and renewals, patent and original patentee characteristics, and the firm size of patent holders (both as of the application date and renewal dates). Table 1 presents summary statistics of the patents in our sample. Panel A shows that 11% of these patents were initially owned by small firms, and the rest 89% were owned by large firms. The same panel reports these rates across the grant year of the patents in the sample. The proportion of small firm patents has been growing over the period of study, ranging from 9 percent in 1989-1990 to 12 percent in the years 1995-1996. In Panel B, we report the proportion of patents initially owned by small and large firms across six patent technology classes (Chemical, Computer & Communications, Drugs & Medical, Electric & Electronics, Mechanical, and Other). Small firms patenting is relatively more important in the technology classes of Drugs and Medical (152%) and Other (206%). The lowest share of patenting for small firms is in the Computer and Communications (59%) and Chemical (65%) patent classes.13

3

Factors affecting patent sales and acquisitions

In this section we will consider several hypotheses about the determinants of patent transactions.

3.1

Technological fit

Firms’ decisions to sell and acquire new innovations can depend on the technological fit of an innovation with the current patentee and potential buyers. The technological fit of an innovation is the degree to which it relates to the research activities of a firm. Cassiman and Ueda (2006) define technological fit as the cost savings from internal commercialization in contrast to external one. The cost savings are derived from the internal technical and organizational capabilities 13

We have followed Hall, Jaffe, and Tratjenberg (2001) to classify patents into technology categories or fields.

7

of a firm (Teece, 1986). These capabilities, which Cohen and Levinthal (1990) associate with firm’s research activities, allow the firm to appropriate value from the innovation. Not all new innovations necessarily perfectly fit the capabilities of the original inventor because firms can find it hard control ex-ante the exact nature and scope of new innovations (Rosenberg ,1996). This brings about possible gains from trade and generates a role for reallocating patented innovations to other firms, who may be a better technological fit. Cassiman and Ueda’s model predicts that innovations with a high fit are more likely to be developed internally and thus less likely to be sold. Moreover, because the theoretical result does not depend on where innovations come from, the same prediction can apply to innovations that originate outside of the firm. Firms should therefore be more likely to acquire patented innovations with high technological fit. These also studies suggest that patent acquisitions by small and large firms can depend on the manner in which their research activities relate to each other. If small firms operate in technology ’niches’, and thus the innovations that they create may be a better technological fit for firms working in related technologies, small firms may disproportionally acquire patented technologies from other small firms. As Cohen and Levinthal (1990) point out, firms working in related technologies may own the technical capabilities to better select external innovations and appropriate value from them (Arora and Gambardella, 1994; Cassiman and Veugelers, 2006). Alternatively, if small and large firms coexist in the technology space, and the former’s research activities are anchored around large firms, we can expect large firms to be more likely to acquire the new innovations of small firms. Large firms may also be morel likely to acquire patents, particularly from small firms, when large firms have a comparative advantage in large-scale development (Arrow, 1983). Another implication of technological fit is that small firms are more likely to sell their technologies. The intuition of this result also originates in the costs savings derived from the internal technical capabilities of firms. Because firms find it hard to control the exact nature of and scope of new innovations, the broad research activities that typically characterize large firms also provides them better opportunities to integrate the new in-house innovations within the firm (Nelson, 1959; Henderson and Cockburn, 1996). As a result, small firms are more likely to sell their new technologies than large firms.

8

3.2

Economic value of the technology

Patent sale and acquisition decisions may also depend on the economic value of the technology. Since prices and market transactions must be paid to transfer a patented innovation, only the technologies that provide enough rents to cover transaction costs will be acquired. Serrano (2006) developed a theoretical model where patents may be transferred because some firms generate higher returns than others using a given patent, but transferring a patent and adopting the technology involves a sunk transaction cost. In this model, the author shows that patents with high returns are more likely to be traded because the higher the returns the lower is the necessary improvement factor of potential buyers over the current owner to amortize the transaction costs. The decision to sell and acquire a patent can also depend on access to capital (see e.g., Kulatilaka and Lin, 2006). In general, financially constrained innovators, who typically do not have the funds to make R&D investments, are more likely to sell their innovations to others, particularly innovations with the highest economic value. The economic value of the technology may also affect whether patents are acquired by small or large firms. Large firms may acquire the technologies of higher economic value for several reasons. Large firms can obtain capital more cheaply in financial markets because their revenuestream is more stable than that of small firms and they typically have more tangible assets that can be used as collateral to secure a loan. There is empirical evidence showing that large firms have better access to financial markets to raise capital (Beck, Demirgüç-Kunt, Maksimovic, 2008). Another theoretical argument that leads to the same prediction is by Figueroa and Serrano (2013), who argue that large firms have superior abilities than small firms to reallocate new inhouse innovations within the firm, which it makes large firms more selective in their technology acquisitions; as a result, they acquire on average the technologies with highest economic value.

3.3

Patent technology classes

Patent trading flows can differ across technology sectors. There are several reasons. The first one is that market transaction costs may be different across technology sectors. There is evidence showing that patent legal rights over a technology, which can arguably reduce market transaction costs, increase the likelihood of licensing the technology (Gans, Hsu and Stern, 2008). A second reason is that the gains from trade in patents, and the corresponding benefits of the division of labor, can also vary across technology sectors (Arora, Fosfuri, and Gambardella, 2001).

9

4

Patent trading flows of small and large firms

We begin the empirical analysis describing patent sales and acquisitions of small and large firms. Table 2 reports information on patent sales and acquisitions by small and large firms. Panel A reveals that patent sales are more common among small firms. As shown in Column (1), 9 percent of the patents granted to small firms are sold within four years of being granted, whereas large firms only transfer 55 percent of their patents.14 As said above, our analysis of patent transactions focuses on patent sales within four years after their issuance. The rest of the columns present similar findings for six different technology classes. Of particular interest is that in the drugs and pharmaceutical sectors, as well as in chemicals, where patent rights are more effective, the probability of firms to sell their technologies is higher than in other sectors. The high propensity to sell patents by small firms is consistent with the technological fit hypothesis previously discussed in the paper. Small firms are more likely to sell their technologies because they may find it harder than large firms to find a technological fit within the firm for their own innovations. We study this possibility in the next section. Panel B shows that small firms acquire disproportionately more patents in the secondary market than large firms. To show this, we compute the proportion of traded patents that had been acquired by small firms and large firms and then adjust for their respective patenting activity. Although small firms applied for and were granted 107 percent of all the patents in our sample, they acquired 16 percent of the traded patents. In other words, the acquisition rate of small firms is about 50% higher than their patenting rate. In contrast, large firms, who applied for and were granted 892 percent of the patents, acquired 84 percent of the traded patents, which is just about 6% less than their patenting activity. The panel also reveals that this phenomenon is consistent across the six technology sectors analyzed. To explore why small firms have a higher propensity to acquire patents than large firms, Panel C examines in more detail the patent trading flows between small and large firms. We look at the patents sold by small and large firms separately. Among patents sold by small firms, we find that the vast amount of them, 67 percent, are acquired by other small firms. This share is about six times times larger than the proportion of patents granted to small firms in the sample (67 percent of small patents are acquired by small firms vs. 107 percent of corporate patents are granted to small firms.) In contrast, the proportion of large patents acquired by other large firms 14

These rates are low because in our analysis trade is truncated by the fourth year since a patent’s grant date.

10

is just 56 percent higher than the proportion of patents granted to large firms (942 percent of large patents are acquired by large firms vs. 892 percent of all corporate patents are granted to large firms.) These findings reveal that small firms have a much higher propensity to acquire patents than large firms, but only patents originating in firms of the same size. This finding may be consistent with technological fit hypothesis. If small firms operate in technology niches, and thus the innovations that they create may be a better fit for firms working in related technologies, small firms may disproportionally acquire patented technologies from other small firms. This is an issue we plan to explore in the next section. Finally, Panel D puts together patent sales and acquisitions and report the percentage of all patents owned by small and large firms that were acquired. In a typical small firm, 93 percent of the patents (as of age five) had been acquired externally. For larger firms, acquired patents account for 56 percent of the patent portfolios. These findings show that external sources of patented innovations account for a significant share of the firms’ patent portfolios, particularly for small firms. The evidence that firms significantly rely on external sources for innovation and patent rights suggests that externally-sourced technologies must increase the acquirer’s performance compared to what it would be if they had no access to the market for innovation. To summarize, we have documented the patent trading flows between small and large. First, small firms are more likely to sell their patents. Second, small firms have a higher propensity to acquire patents than large firms, particularly patents originating from small firms. Finally, putting patent sales and acquisitions together we find that small firms own a larger share of external technologies than large firms in their patent portfolios. The patterns on the sale of patents are consistent with studies documenting intense patent licensing and selling activity by small firms and individual inventors (see e.g., Anand and Khanna, 2000; Gambardella, Giuri, Luzzi, 2007; Serrano, 2010). The findings on the patent acquisitions are opposite to those reported in the literature on technology licensing. This literature has documented an increasing division of labor between small firms who typically specialize in innovation but lack the capacity for large development, and large firms whose comparative advantage lies in the commercialization of these innovations (Arora and Gambardella, 1990; Arora, Fosfuri, and Gambardella, 2001). In the next section, we explore factors affecting patent sales and acquisitions by small and large firms that may account for these patterns.

11

5

Determinants of patent sales and patent acquisitions

The literature discussed in the earlier sections associates patent sales and acquisitions of small and large firms with technological fit, the economic value of the technology and patent technology sectors.

5.1

Technological fit of the patent with the initial patentee

Discerning the technological fit between a patent and the initial patentee is challenging. Ideally, we would like a measure that captures the degree to which a patentee’s new innovation fits with its technical and organizational capabilities to develop and commercialize the innovation (Cassiman and Ueda, 2006). Patented innovations could fit the initial patentee better than other firms if the patentee owns these capabilities (Teece, 1986). Alternatively, the fit of the patented innovation with the current patentee could be weak when access to these capabilities is difficult or very costly. The number of patents embedded in a product and the form in which this product interacts with the rest of the firm could provide some evidence about the technological fit of the patent with the initial patentee (Cohen and Levinthal, 1990), but there are no datasets providing such detailed information. Because patented knowledge of an initial patentee and other firms is considered to be prior art which patent applicants and patent office examiners are required to cite, we use patent citations to construct a variable designed to reflect to what degree an innovation is a good technological fit with the initial patentee relative to the rest of firms. While citations made in patents are not a perfect measure of what firms are good fit for a new patented innovation, they often have been considered a proxy of the relevant technology that the new patented innovations follow-on (and thus related technologies.) We therefore define    as the number of patent citations made in a patent document to the patentee’s patent portfolio relative to the total number of patent citations made to patents of all patentees. In other words, the share of backward citations that the focal patents makes to patents that had been granted to the owner of the focal patent. A patented innovation with higher    is assumed to have a better technological fit with the initial patentee, and thus this patentee has better possibilities than other firms to develop internally and commercialize the innovation. The measure is similar in spirit to the technological-relatedness proxy used in Lanjouw and Schankerman (2001): the authors use the self-citation ratio as a measure to capture the relatedness between firm patents and subsequent technological activity by the firm. Graham (2004) uses a similar citation-based measure to proxy

12

for a firm’s control over the technology trajectory. Table 4 presents descriptive statistics for the fit of new patented innovations with the initial patentee. Panel A reports the mean level of    for the traded and untraded patents. As discussed above, traded patents are those sold within four years of being granted. Consistent with the technological fit hypothesis, traded patents command 30 percent lower fit with the initial patentee (0093) than the patents that were not sold (0136).15 To provide additional supporting evidence that patented innovations with higher    are those with a better technological fit with the initial patentee, we also exploit the sample of patents that were untraded by the end of age four. Among these patents, the renewed patents have a higher fit measure with the patentee (   = 0139) than the patents that were let to expire (0117)16 This is what we should expect: patentees keep and renew those patents from which they can extract more value than other firms, i.e., that are a better fit for them. Finally, again consistent with the technological fit hypothesis, Panel B shows that smaller firms also have granted patents with lower    measures. Table 5 reports the results of several regressions. Regression analysis allows us to control for factors other than technological fit that may affect the decision to sell a patent. The first regression is a linear-probability OLS regression. The dependent variable is  .   indicates whether patent  is sold to another entity within four years of being granted. The explanatory variables are  ,    , and dummy variables   and   A commonly used proxy for the economic value of a technology is the number of patent citations received (Trajtenberg, 1990), which may be a factor in the decision to sell a patent.  is equal the cumulative number of patent citations that a patent receives as of age five. Similarly,     controls for the stock of the patentee’s patents, which is a proxy for the opportunities for technological fit within the firm. The dummies   (36 patent sub-categories) and  (years 1989 to 1996) account for technology sector and year effects that may affect patent trading. Consistent with the technological fit hypothesis, we find that the correlation between    and   is negative and significant. The result implies that traded patents have a lower technological fit with initial patentee, even controlling for numerous factors and allowing for patent and firm-specific observed differences. We also use alternative regression specifications. There is the concern that firms that specialize 15 16

The difference is significant at p-values  01. The difference is significant at p-values  01. The sample mean of    is 0134.

13

in patent trading may also be firms with low and persistent measures of   . To account for this possibility, we use a fixed-effects linear-probability OLS regression. The fixed-effects specification allows us to exploit the within firm variation in the technological fit of new patents with the initial patentee. Finally, we run a Probit regression with the explanatory variable defined above. This allows us to explicitly model the binary nature of the dependent variable. In all alternative regression specifications, we find that traded patents have a lower technological fit with the initial patentee than the untraded patents. Also consistent with the technological fit hypothesis, we find that among the untraded patents, the patents let to expire at the first renewal date have on average lower technological fit with the initial patentee.17 There are at least two possible ways to interpret the negative correlation between    and   The first interpretation of the result is that the lower the technological fit of a new patented innovation with an initial patentee the higher is the likelihood that the patent is sold, which is consistent with the predictions in Cassiman and Ueda (2006). This interpretation is also consistent with Arora and Ceccagnoli (2006), which shows that the propensity to license a patented innovation to others decreases with the availability of complementary assets. Our result together with the previous finding that patents granted to small firms have on average low    can account, in part, for why small firms are more likely to sell their patents than large firms. An alternative interpretation of the result is that patentees intentionally create some patented innovations in order to sell them to firms that have a better technological fit than the original patentee. However, the observed negative correlation of    and   in the fixed-effects specification suggests that patentees that may specialize in the selling of patents are not necessarily those that persistently create patents with low measure of   . Although the nature of the transaction is different depending on the interpretation of the coefficient, technological fit plays a role in the decision to sell a patent in both interpretations.

5.2

Technological fit of the patent with potential buyers

The next variable aims to capture the fit between a patented innovation and patentees that may be potential buyers. Among the patents that are sold, the technological fit of the patent may be highest for those potential buyers working in related technologies, which may have the technical and organizational capabilities to select and effectively adopt the technology (Cohen 17

We report this finding in supplemental analysis, available upon request.

14

and Levinthal, 1990; Cassiman and Ueda, 2006). Similarly, firms, especially new firms, can buy patents rights in related technologies, which can provide them "freedom-to-operate" and facilitate cross-licensing negotiations with other firms (Hall and Ziedonis, 2001; Galasso, Schankerman, and Serrano, 2013). However, distinguishing the technological fit between a patent and its potential buyers is very challenging, in part because we know little information about their capabilities and cross-licensing opportunities. To capture the firms that may have a good technological fit with new patented innovations, we also use the citations that these innovations make to the patented technologies of other firms. Since for each patent, and its corresponding cited patents, the initial patentee is linked with the patentees of the cited patents, we can use patent citation data to capture what firms (small or large) would be the most suitable potential buyers. We define    as the share of patent citations made (excluding self-citations) to patentees with the same size than the initial patentee.18 For example, among patents initially owned by small firms, the patents with higher    are assumed to have a better technological fit with potential buyers that are small firms. Building on this idea, if patent trading flows were associated with technological fit, then we would expect that, among patents initially owned by small firms, the patents acquired by other small firms are those with more citations made towards small firms, i.e.,. higher    .19 Table 3 reports the technological fit of new innovations with potential buyers, for small and large firms. If the mean    is higher than the share of patents granted to firms with the same size as the initial patentee, then these firms have a higher propensity to work on the relevant technology of other firms of the same size. For small firms, the mean    (168%) is much higher than the proportion of patents initially owned by small firms (107%) (about 57 percent higher than the rate of a random citation pattern).20 In sharp contrast, the proportion of all citations made by large firms to patents initially owned by other large firms (945%) is marginally higher than the proportion of patents initially owned by large firms (892%) 18 A caveat of using patent maintenance fee records is that we cannot identify the actual identity of patent buyers and the patent in their portfolios. For this reason, we cannot compute share of patent citations made to the patent portfolios of specific patentees that are potential buyers. Instead, we classify potential buyers into two groups: small and large, as we can observe patent citations made to both groups for all patents. 19 In other words, our hypothesis is that if research activities between small firms (or between small and large firms) were interconnected in some non-random fashion, then we would expect that part of the trading flows between small and large firms could be accounted for this variable. We are assuming that the proportion of small and large firm patents that could be potentially cited is equal to the proportion of patents initially owned by small and large firms too. 20 The patent citations we use exclude self-citations; and the rate of random citations is the one where citations made are based on the stock of small and/or large firm patents that may be potentially cited. The ratio 01680107 = 157 indicates that the mean of    is 57 percent higher than random citation.

15

(about 6 percent higher than random citation.) These findings show that small firms have a propensity to work around the relevant technologies of other small firms, which is consistent with claims that small firms operate in technology niches (Bloom, Schankerman, and Van Reenen, 2013). Large firms’ research, instead, is much broader than small firms. As seen in the Table, the results are qualitatively similar across different technology classes. Next, we examine whether    is associated with the direction of the observed patent trading flows of small and large firms. Table 3 reveals that for small firms the ratio of    relative to the proportion of patents granted to firms of the same size is highest in the Computer & Communications and the Chemical patent categories ( 0122 0059 and

0134 0065 

respectively). Interestingly, the ratio of the proportion of small firm traded patents acquired by other small firms relative to the proportion of patents applied for and granted to small firms is also highest in the Computer & Communication and the Chemical patent classes. These findings suggest that the most likely patent buyers are also the firms (small or large) with the highest fit. Regression results are presented in Table 6. The first analysis is a linear-probability OLS model. The dependent variable is      indicating that a patent traded is acquired by a small entity. The explanatory variables are the interactions    ∗  and    ∗   the control variables  and    a dummy variable for firm size  , and patent technology classes   and patent grant year  dummies There are two results we want to highlight. First, the coefficient of    ∗  is positive and significant, indicating that small firm patents with higher proportion of patent citations made to small firms are more likely to be acquired by small firms too. Second, the coefficient of    ∗  is negative and significant, proving that large firm patents with higher proportion of patent citations made to large firms are also more likely to be acquired by large firms. The rest of the regression analysis shows that similar results were obtained using a linear-probability fixed-effect OLS model and a Probit model. This exercise confirms that, independently of the initial patentee and the technology class of the patent, traded patents are more likely to be acquired by the firms (small or large) that on average have the highest technological fit with the patent. Together with our finding that small firms work in related technologies, this result could account for why small firms acquire disproportionately more patents originating from other small firms. This is consistent with the technological fit hypothesis: firms acquire patents that are more likely to be a good technological 16

fit. This finding also relates to recent empirical work on licensing that provides evidence of positive complementarities between own R&D and external technology acquisition (Arora and Gambardella, 1994; Cassiman and Veugelers, 2006). Relatedly, Ceccagnoli et al. (2010) focuses on U.S. Food and Drug Administration (FDA) new drug approvals and considers both technology licensing and mergers and acquisitions. At the same time, we emphasize that our result does not rule out the possibility that patentees that ultimately attempted to sell some patented innovations to other firms may also be more likely to create innovations that intentionally build on the research of potential buyers that are a good technological fit for the patent (and thus the patents cite these firms.) However, similar results were obtained in a firm fixed-effects specification, suggesting that such patentees may not be the type of firms that specialize in the selling of patents and thus persistently create innovations that build on the innovations of potential buyers that happen to have a good technological fit with the new innovation.

5.3

Economic value of the technology

Finally, we turn the focus to the relationship between the economic value of the technology and patent acquisitions by small and large firms. Because we have neither data on acquisition prices nor information on expected patent values, we can only rely on imperfect measures of "value." A commonly used proxy for the economic value of a technology is the number of patent citations received (Trajtenberg, 1990). Following this line of work, the variable  is equal the cumulative number of patent citations that a patent receives as of age five. Our assumption is that the higher the number of patent citations a patent receives the higher is on average the economic value of the technology. Panel A of Table 7 reports the number of patent citations received of acquired and untraded patents, for small and large firms. The table shows that large firms tend to acquire more highly cited patents than small firms, particularly among patents originating from small firms. In contrast, among traded patents originating from large firms, there are no significant differences in the number of patent citations received. Moreover, both small and large firms buy improvements over their patent portfolio, i.e., they acquire patents that have higher number of patent citations received than their granted and untraded patents.21 To examine the robustness of these findings, we perform regression analysis. The dependent 21 Similar results were obtained when examining whether a patent owner pays patent maintenance fees (see Panel B of Table 7). The payment of such fees is another common proxy for the economic value of a technology.

17

variable is     . The explanatory variables of interest are  ∗  and  ∗  . The two interactions allow us to ascertain the potentially distinct role that the economic value of the technology may have in the acquired patents initially owned by small or large firms. A negative and significant coefficient of  ∗  indicates that, among patents initially owned by small firms, patents acquired by large firms have on average higher number of patent citations received. Instead, a negative and significant coefficient of  ∗  shows that, among patents initially owned by large firms, the patents with higher number of citations received are less likely to be acquired by small firms (and thus more likely to be acquired by a large firm). The regression analysis also controls for factors that may affect the decision to sell a patent such as technological fit, Patent portfolio, a firm size dummy, and technology class and patent grant year effects. Table 8 presents the results of the relationship between patent acquisitions and the economic value of the technology. We find that the coefficient of  ∗  is negative and significant, indicating that among patents initially owned by small firms, large firms acquire those that on average have the highest number of patent citations received. As for the patents initially owned by large firms, the coefficient of the interaction  ∗  is positive but not significant, implying that among patents initially owned by large firms, there are no significant differences in the number of patent citations received between the patents acquired by small and large firms. These results were obtained across several specifications, including an OLS with patentee fixed-effects and a Probit model. The regression results confirm that large firms acquire on average the technologies with highest economic value in the market, particularly patented technologies originating from small firms. This finding is consistent with the determinants of patent acquisitions discussed in an earlier section. Large firms have lower costs than small firms to raise capital in order to purchase the highest valued technologies (Beck, Demirgüç-Kunt, Maksimovic, 2008). Alternatively, Figueroa and Serrano (2013) argue that large firms, who may have better opportunities to find a productive use for their new innovations within the firm, can retain more often their new innovations with highest value. This mechanism makes large firms to be on average more selective than small firms in their technology acquisitions, which ultimately can account for why large firms have a higher propensity to acquire technologies of high economic value. To sum up, the empirical analysis in this Section shows that the patterns in the data are broadly consistent with the theories of the determinants of patent sales and acquisitions discussed 18

in an earlier section. We find that the decision to sell a technology is negatively correlated with measures of the technological fit of the innovation with the initial patentee. Also consistent with the technological fit hypothesis, we show that the decision that firms acquire patents is positively correlated with the technological fit of the patent with potential buyers (small or large). Finally, we also find that the technologies with the highest economic value, particularly those originating in small firms, are more likely to be acquired by large firms. These factors, under some assumptions, can account for the observed patent trading flows of small and large firms.

6

Conclusion

This study has presented novel evidence on the market for patents -the secondary market for the transfer of the ownership of patent rights-, an important yet under-explored source of R&D incentives for firms. We present a set of patterns on the patent trading flows of small and large firms and examine some determinants of these transactions. The findings provide evidence on the role that firm size, technological fit, the economic value of the technology and patent technology class plays in patent sales and acquisitions. To do so, we exploited patent owners’ maintenance fee records at the USPTO and constructed a new data set that matches information on patent trades and firms’ size for both sides of patent transactions. There are four key empirical findings in the paper. First, small firms are more likely to sell their patented innovations, and have a higher propensity than large firms to acquire external innovations, especially patented innovations initially owned by other small firms. Second, the decision to sell and renew a patented innovation depends on the technological fit of the innovation with the initial patentee. We show that patents that are sold are those with a lower technological fit with the initial patentee; and among patents that were not sold, the ones with the lowest technological fit are more likely to be discontinued, i.e., expired. Third, patent acquisitions by small and large firms also depend on the technological fit of the innovation with potential buyers. We find that the higher the technological fit between the patent and potential buyers (small or large), the higher is the probability that the focal patent will be acquired by a buyer (small or large) with highest technological fit. Finally, we find that the patented innovations acquired by large firms have on average the highest economic value, especially for patented innovations sold by small firms. These findings are broadly consistent with the theories discussed in the paper. Four conclusions can be drawn. First, the fact that patented innovations with a lower tech-

19

nological fit are those with a higher predicted likelihood of changing ownership, and that patents that change ownership are more likely to be renewed, suggests that the market for innovation generates efficiency gains by reallocating patented innovations to firms with a better technological fit (and higher value.) Second, this also indicates that the market for innovation will likely be an important source of incentives to invest in R&D, especially for small firms being typically at a disadvantage to develop internally their newly created innovations. This disadvantage can potentially discourage firm’s entry and growth opportunities, especially when market transaction costs are significant. Third, the result showing that traded patents are more likely to be acquired by firms with the highest technological fit can inform market practitioners about who the potential buyers for these patents might be. This evidence can be useful for lenders when assessing the salvage value of patent assets used as collateral to secure debt financing. The information can also be useful for researchers when designing economic models that estimate gains from trade in the market for patents. Finally, the finding that the small firms’ technologies with the highest economic value are on average acquired by large firms suggests that, at least for these innovations, the direction of the patent trading flows takes place in ways consistent with the classical theory of the division of innovative labor. At the same time, the paper also has several limitations. Our data has no information on the business characteristics of firms acquiring patents other than whether the new owner is a small or large firm. Moreover, the transaction data has no sale prices or contractual terms of patent sales. We plan to investigate these issues in further research.

References Anand, B., and T. Khanna (2000): “The Structure of Licensing Contracts,” Journal of Industrial Economics, 48, 103—135. Arora, A., and M. Ceccagnoli (2006): “Patent Protection, Complementary Assets and Firms’Incentives for Technology Licensing,” Management Science, 52, 293—308. Arora, A., A. Fosfuri, and A. Gambardella (2001): Market for Technology: The Economics of Innovation and Corporate Strategy. The MIT Press. Arora, A., and A. Gambardella (1990): “Complementarity and External Linkages: The Strategies of the Large Firms in Biotechnology,” Journal of Industrial Economics, 38 (4), 361—379. (1994): “Evaluating technological information and utilizing it: Scientific knowledge, technological capability and external linkages in biotechnology,” Journal of Economic Behavior and Organisation, 24 (1), 91—114.

20

Arrow, K. (1983): “Innovation in Large and Small Firms,” in Entrepreneurship, ed. by J. Ronen. Lexington Books, Lexington, MA. Beck, T., A. Demirgüç-Kunt, and V. Maksimovic (2008): “Financing patterns around the world: Are small firms different?,” Journal of Financial Economics, 89(3), 467—487. Bloom, N., M. Schankerman, and J. V. Reenen (2013): “Identifying Technology Spillovers and Product Market Rivalry,” Econometrica, 81(4), 1347—1393. Cassiman, B., and M. Ueda (2006): “Optimal project rejection and new firm start-ups,” Management Science, 52, 262—275. Cassiman, B., and R. Veugelers (2006): “In Search of Complementarity in the Innovation Strategy: Internal R&D and External Knowledge Acquisition,” Management Science, 52 (1), 68—82. Ceccagnoli, M., S. J. Graham, M. J. Higgins, and J. Lee (2010): “Productivity and the Role of Complementarity Assets in Firms’ Demand for Technology Innovations,” Industrial and Corporate Change, 19 (3), 839—869. Chien, C. (2011): “Predicting Patent Litigation,” Texas Law Review, 90, 283—295. Cohen, W. M., and D. A. Levinthal (1990): “Absorptive Capacity: A New Perspective on Learning and Innovation,” Administrative Science Quarterly, 35 (1), 128—152. Dykeman, D., and D. Kopko (2004): “Recording Patent License Agreements in the USPTO,” Intellectual Property Today, August, 18—19. Figueroa, N., and C. J. Serrano (2013): “Patent Trading Flows of Small and Large Firms,” NBER Working Paper No. 18982. FTC (2012): “Patent Assertion Entity Activities Workshop,” Washington DC. Galasso, A., M. Schankerman, and C. J. Serrano (2013): “Trading and Enforcing Patent Rights,” The Rand Journal of Economics, 44(2), 275—312. Gambardella, A., P. Giuri, and A. Luzzi (2007): “The Market for Patents in Europe,” Research Policy, 36(8), 1163—1183. Gans, J. S., D. Hsu, and S. Stern (2008): “The Impact of Uncertain Intellectual Property Rights on the Market for Ideas: Evidence from Patent Grant Delays,” Management Science, 54, 982—997. Gonzalez-Uribe, J. (2014): “Venture Capital and the Appropriation of Innovation Externalities,” LSE Working Paper. Graham, S. J. (2004): “Hiding in the Patent’s Shadow: Firm’s Used of Secrecy to Capture Value from New Discoveries,” GaTech Ti:Ger Working Paper Series. Haber, S., and S. H. Werfel (2015): “Why Do Inventors Sell to Patent Trolls? Experimental Evidence for the Asymmetry Hypothesis,” Stanford University Working Paper. Hall, B. H., A. B. Jaffe, and M. Trajtenberg (2001): “The NBER Patent Citation Data File: Lessons, Insights and Methodological Tools,” NBER Working Paper 8498. Hall, B. H., and R. H. Ziedonis (2001): “The patent paradox revisited: an empirical study of patenting in the US semiconductor industry, 1979-1995,” RAND Journal of Economics, 32 (1), 101—128. Henderson, R., and I. Cockburn (1996): “Scale, Scope, and Spillovers: The Determinants of Research Productivity in Drug Discovery,” RAND Journal of Economics, 27 (1), 32—59. 21

Hochberg, Y. V., C. J. Serrano, and R. H. Ziedonis (2014): “Patent Collateral, Investor Commitment, and the Market for Venture Lending,” NBER Working Paper No. 20587. Khan, Z. B., and K. L. Sokoloff (2004): “Institutions and Democratic Invention in 19th Century America,” American Economic Review, 94, 395—401. Kulatilaka, N., and L. Lin (2006): “Impact of Licensing on Investment and Financing of Technology Development,” Management Science, 52(12), 1824—1837. Lamoreaux, N., and K. Sokoloff (1997): “Inventors, Firms, and the Market for Technology: U.S. Manufacturing in the Late Nineteenth and Early Twentieth Centuries,” NBER Working Paper H0098. (1999): “Inventive Activity and the Market for Technology in the United States, 18401920,” NBER Working Paper 7107. Lamoreaux, N. R., and K. L. Sokoloff (2001): “Market Trade in Patents and the Rise of a Class of Specialized Inventors in the 19th-Century United States,” American Economic Review P&P, 91 (2), 39—44. Lamoreaux, N. R., K. L. Sokoloff, and D. Sutthiphisal (2013): “Patent Alchemy: The Market for Technology in US History,” Business History Review, 87, 3—38. Lanjouw, J. O., and M. Schankerman (2001): “Characteristics of Patent Litigation: A Window on Competition,” The Rand Journal of Economics, 32, 129—151. Mani, D., and A. Nandkumar (2015): “The Differential Impacts of Markets for Technology on the Value of Technological Resources: An Application of Group-based Trajectory Models,” Strategic Management Journal, forthcoming. Marco, A. C., A. F. Myers, S. J. Graham, P. A. D’Agostino, and K. Apple (2015): “The USPTO Patent Assignment Dataset: Descriptions and Analysis,” USPTO Working Paper No. 2015-2. Nelson, R. R. (1959): “The Simple Economics of Basic Scientific Research,” Journal of Political Economy, 67, 297—306. Odasso, C., G. Scellato, and E. Ughetto (2015): “Selling patents at auction: an empirical analysis of patent value,” Industrial and Corporate Change, 24(2), 417—438. Palomeras, N. (2007): “An Analysis of Pure-Revenue Licensing,” Journal of Economics and Management Strategy, 16(4), 971—994. Serrano, C. J. (2006): “The Market for Intellectual Property: Evidence from the Transfer of Patents,” Ph.D. thesis, University of Minnesota. (2010): “The Dynamics of the Transfer and Renewal of Patents,” RAND Journal of Economics, 41, 686—708. (2011): “Estimating the Gains from Trade in the Market for Innovation: Evidence from the Tranfer of Patents,” NBER Working Paper No. 17304. Shapiro, C., and F. M. Scott-Morton (2014): “Strategic Patent Acquisitions,” Antitrust Law Journal, 2, 463—499. Teece, D. J. (1986): “Profiting from technological innovation: Implications for integration, collaboration, licensing and public policy,” Research Policy, 15, 285—305. Trajtenberg, M. (1990): “A Penny for Your Quotes: Patent Citations and the Value of Innovations,” The Rand Journal of Economics, 21 (1), 172—187.

22

Table 1: Summary Statistics A. Proportions of Patents Initially Owned by Small and Large Firms by Grant Year Patent Grant Year (in groups of two years)

Small firms Large firms Number of Patents

All patents

1989-1990

1991-1992

1993-1994

1995-1996

0.107 (.000) 0.892 (.000) 590,873

0.089 (.001) 0.910 (.001) 135,507

0.098 (.001) 0.902 (.001) 144,224

0.116 (.001) 0.884 (.001) 151,839

0.123 (.001) 0.877 (.001) 159,303

B. Proportions of Patents Initially Owned by Small and Large Firms by Patent Technology Class

Small firms Large firms Number of Patents

All patents

Chemical

0.107 (.000) 0.892 (.000) 590,873

0.065 (.001) 0.934 (.001) 126,865

Patent technology classes Computer Drugs & Elec. & & comm medical Electro 0.059 (.001) 0.941 (.001) 85,924

Note: Standard errors in parenthesis.

23

0.152 (.002) 0.848 (.002) 49,857

0.080 (.001) 0.920 (.001) 116,232

Mechani

Other

0.121 (.001) 0.878 (.001) 122,642

0.206 (.001) 0.793 (.001) 89,352

Table 2: Patent Sales and Acquisitions by Small and Large Firms

All patents (1) A. Proportion of patents sold Small firm 0.090 Large firm All firms

(.001) 0.055 (.000) 0.059 (.000)

Chemical (2) 0.091 (.003) 0.058 (.001) 0.061 (.001)

Patent Technology Classes Computer Drugs & Elec. & & comm medical Electro Mechanical (3) (4) (5) (6) 0.130 (.005) 0.050 (.001) 0.055 (.001)

Other (7)

0.112 (.004) 0.073 (.001) 0.079 (.001)

0.096 (.003) 0.044 (.001) 0.049 (.001)

0.074 (.002) 0.049 (.001) 0.052 (.001)

0.078 (.002) 0.068 (.001) 0.070 (.001)

0.214 (.006)

0.143 (.005)

0.180 (.005)

0.219 (.005)

0.611 (.017)

0.649 (.016)

0.745 (.013)

0.738 (.012)

0.103 (.005)

0.048 (.003)

0.061 (.003)

0.064 (.003)

B. Proportion of patent acquisitions by small firms Prop. of patents acquired by small firms 0.160 (.002)

0.116 (.004)

0.100 (.004)

C. Patent trading flows between small and large firms Prop. of small firm patents acquired by small firms 0.673 (.006)

0.669 (.017)

0.531 (.019)

Prop. of large firm patents acquired by small firms 0.058 (.001)

0.056 (.003)

0.030 (.003)

D. The importance of external patented technologies for small and large firms Prop. of patents acquired in the patent portfolios of small and large firms as of age 5 Small firms 0.093 0.107 0.110 0.114 0.093 0.081 Large firms Number of patents

(.001) 0.056 (.000)

(.004) 0.060 (.001)

(.005) 0.052 (.001)

(.004) 0.080 (.001)

(.003) 0.045 (.001)

(.002) 0.049 (.001)

0.081 (.002) 0.071 (.001)

509,047

106,876

78,685

42,074

102,808

104,797

73,807

Note: A patent is traded if it was sold within four year of its grant date. Stand. errors in parenthesis

24

Table 3: Patent Trading Flows and Technological Fit

All patents (1)

Chemical (2)

Patent technology classes Computer Drugs & Elec. & & comm medical Electro Mechani (3) (4) (5) (6)

A. Proportion of patents initially owned by small and large firms Small firms 0.107 0.065 0.059 0.152 0.080 Large firms

(.000) 0.892 (.000)

(.001) 0.934 (.001)

(.001) 0.941 (.001)

(.002) 0.848 (.002)

(.001) 0.920 (.001)

0.121 (.001) 0.879 (.001)

B. Average proportion of citations made to firms of same size (excluding self-cites) Small firms 0.168 0.134 0.122 0.195 0.119 0.172 Large firms

(.001) 0.945 (.000)

(.003) 0.966 (.000)

(.003) 0.951 (.000)

(.004) 0.915 (.001)

(.003) 0.952 (.000)

(.003) 0.944 (.001)

Other (7) 0.206 (.001) 0.793 (.001)

0.213 (.003) 0.908 (.001)

Note: standard errors in parenthesis.

Table 4: Technological Fit of the Innovation for Kept, Traded, and Expired Patents Patentee fit A. Patents Kept, Traded, and Expired Kept patents 0.136

At least one self-citations

(.000) 0.139 (.000) 0.117 (.001)

0.370 (.001) 0.381 (.001) 0.305 (.002)

Traded patents

0.093 (.001)

0.304 (.002)

All patents

0.134 (.000)

0.366 (.001)

Kept and renewed Kept and expired

B. Patents initially owned by small and large firms Small firms 0.054 0.194 (.001) 0.143 (.000)

Large firms

Note: Standard errors in parenthesis.

25

(.002) 0.387 (.001)

Table 5: Marginal Effects of Patentee Fit on the Probability that a Patent is Traded Estimation method Dependent variable

Patentee fit Citations

(1)

(2)

(3)

(4)

(5)

(6)

OLS Fixed Effects

OLS

OLS

Probit

Probit

Probit Random Effects

Traded

Traded

Traded

Traded

Traded

Traded

Coef.×102

Coef.×102

Coef.×102

Mar.Eff.×102

Mar.Eff.×102

Coef.×102

-1.35*** (0.12) 0.03*** (0.006)

-3.13*** (0.37) 0.09*** (0.02) -0.0002*** (0.0000)

-3.13*** (0.41) 0.10*** (0.02) -0.0003*** (0.0000)

-3.56*** (0.50) 0.07*** (0.01) -0.0003** (0.0000)

-3.61*** (0.56) 0.09*** (0.02) -0.0003** (0.0000)

-22.15*** (1.80) 0.65*** (0.07) 0.002*** (0.001)

Yes Yes Yes

Yes Yes No

No No No

Yes Yes No

No No No

Yes Yes No

All patents

All patents

All patents

All patents

All patents

All patents

61,239 590,873

61,239 590,873

61,239 590,873

61,239 590,873

61,239 590,873

61,239 590,873

Patent Portfolio Controls Grant Year Technology Firm fixed effects Sample Firms Observations

Note: Standard errors are clustered at the firm in columns 1-5. Statistical significance: *** 1 percent, ** 5 percent, and * 10 percent. Traded=1 if the patent is sold. Patent citations: number of forward cites by age 5 Technology Dummies are generated using the 36 technology subcategories defined in Hall et al. (2001). Patent Portfolio: total number of patents granted to the patentee from 1975 or the year the patentee first patented.

26

Table 6: Marginal Effects of Patent Acquirer Fit on the Probability that a Traded Patent is Sold to Small Firms (1)

(2)

(3)

(4)

(5)

(6)

OLS Fixed Effects

OLS

OLS

Probit

Probit

Probit Random Effects

Traded to Small

Traded to Small

Traded to Small

Traded to Small

Traded to Small

Traded to Small

Coef.×102

Coef.×102

Coef.×102

Mar.Eff.×102

Mar.Eff.×102

Coef.×102

7.47*** (1.90) -7.71*** (0.66) 0.02 (0.02)

6.09** (2.73) -6.73*** (1.19) -0.09** (0.04) -0.0004*** (0.0000) 51.55*** (1.58)

6.10** (2.75) -7.16*** (1.18) -0.18*** (0.04) -0.0004*** (0.0000) 51.47*** (1.59)

2.64** (1.29) -7.49*** (1.11) -0.07* (0.04) -0.001*** (0.000) 39.99*** (2.93)

2.83** (1.39) -8.13*** (1.14) -0.16*** (0.05) -0.001*** (0.000) 39.43*** (2.93)

24.29** (12.24) -51.92*** (9.68) -0.62** (0.32) -0.01*** (0.003) 227.5*** (9.06)

Yes Yes Yes

Yes Yes No

No No No

Yes Yes No

No No No

Yes Yes No

Sample

Traded Patents

Traded Patents

Traded Patents

Traded Patents

Traded Patents

Traded Patents

Firms Observations

9,179 34,599

9,179 34,599

9,179 34,599

9,179 34,599

9,179 34,599

9,179 34,599

Estimation method

Dependent variable

Patent Acquirer Fit*Small Patent Acquirer Fit*Large Citations Patent Portfolio Small Controls Grant year Technology Firm fixed effects

Note: Standard errors are clustered at the firm level. Statistical significance: *** 1 percent, ** 5 percent, and* 10 percent. Traded to Small=1 if a traded patent was sold to a small firm. Small=1 if patent was initially owned by a small patentee; it is zero othewise. Large=1 if patent owned by a large patentee; it is zero othewise. Citations: number of forward cites by age 5. Technology Dummies are generated using 36 technology subcategories as defined in Hall et al. (2001). Patent Portfolio: number of patents granted to the patentee since 1975 or the year the patentee first patented. Grant year is the calendar year a patent was issued.

27

Table 7: Economic Value of the Technology A. Mean number of patent citations received Initially owned

All patents Small firms

Small firms Large firms

3.44 (.021) 3.48 (.007)

4.12 (.100) 3.59 (.129)

Acquired by Large firms 5.85 (.181) 3.70 (.033)

Not Traded All firms 4.68 (.090) 3.69 (.032)

3.31 (.021) 3.47 (.007)

B. Patent expiration rates at age five Initially owned

All patents Small firms

Small firms Large firms

0.192 (.002) 0.132 (.000)

0.186 (.006) 0.057 (.006)

Note: standard errors in parenthesis

28

Acquired by Large firms 0.017 (.003) 0.119 (.002)

Not Traded All firms 0.131 (.004) 0.115 (.002)

0.198 (.002) 0.133 (.000)

Table 8: Marginal Effects of Patent Citations on the Probability that a Traded Patent is Acquired by Small Firms (1)

(2)

(3)

(4)

(5)

(6)

OLS Fixed Effects

OLS

OLS

Probit

Probit

Probit Random Effects

Traded to Small

Traded to Small

Traded to Small

Traded to Small

Traded to Small

Traded to Small

Coef.×102

Coef.×102

Coef.×102

Mar.Eff.×102

Mar.Eff.×102

Coef.×102

6.57*** (1.92) -7.12*** (0.68) -0.27*** (0.07) 0.01 (0.03)

6.09** (2.71) -6.61*** (1.19) -0.67*** (0.13) 0.09** (0.04) -0.0004*** (0.0000) 55.28*** (1.67)

6.10** (2.73) -7.02*** (1.18) -0.74*** (0.14) -0.002 (0.04) -0.0005*** (0.0000) 55.14*** (1.67)

2.70** (1.30) -7.46*** (1.12) -0.23*** (0.06) 0.08* (0.05) -0.001*** (0.000) 43.31*** (2.97)

2.87** (1.33) -8.07*** (1.15) -0.33*** (0.07) -0.01 (0.06) -0.001*** (0.000) 42.64*** (3.01)

25.14** (12.25) -51.66*** (9.69) -2.01*** (0.45) 0.58 (0.40) -0.01*** (0.003) 239.1*** (11.13)

Yes Yes

Yes Yes

No No

Yes Yes

No No

Yes Yes

Sample

Traded Patents

Traded Patents

Traded Patents

Traded Patents

Traded Patents

Traded Patents

Firms Observations

9,179 34,599

9,179 34,599

9,179 34,599

9,179 34,599

9,179 34,599

9,179 34,599

Estimation method

Dependent variable

Patent Acquirer Fit*Small Patent Acquirer Fit*Large Citations*Small Citations*Large Patent Portfolio Small Controls Grant year Technology

Note: Standard errors are clustered at the firm level. Statistical significance: *** 1 percent, ** 5 percent, and * 10 percent. Traded to Small=1 if a traded patent was sold to a small firm. Small=1 if patent was initially owned by a small patentee; it is zero othewise. Large=1 if patent was initially owned by a large patentee; it is zero othewise. Citations: number of forward cites by age 5. Technology Dummies are generated using 36 technology subcategories as defined in Hall et al. (2001). Patent Portfolio: number of patents granted to the patentee since 1975 or the year the patentee first patented. Grant year is the calendar year a patent was issued.

29