Patent Pool

JLEO, V32 N1 1
Patent Pools, Competition, and Innovation—Evidence

from 20 US Industries under the New Deal
Ryan Lampe*
California State University, East Bay
Petra Moser**
Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

New York University and NBER
Patent pools have become a prominent mechanism to reduce litigation risks and
facilitate the commercialization of new technologies. This article takes advan-
tage of a window of regulatory tolerance under the New Deal to investigate the
effects of pools that would form in the absence of effective antitrust. Difference-
in-differences regressions of patents and patent citations across 20 industries
imply a 14% decline in patenting for each additional patent that is included in a
pool. An analysis of the mechanism by which pools discourage innovation indi-
cates that this decline is driven by technologies for which the creation of a pool
weakened competition in R&D. (JEL K21, L24, L41, O31)
1. Introduction
“A patent pool is an arrangement by which two or more patent owners put
their patents together and receive in return a license to use them”
(Vaughan 1956: 39–40). Many pools license their patents as a bundle
to third-party nonmember firms, while other pools only provide for
*California State University, East Bay. Email: ryan.lampe@csueastbay.edu

**New York University and NBER. Email: pmoser@stern.nyu.edu
We thank Tim Bresnahan, Wesley Cohen, Christian Helmers, Eric Hilt, Rick Hornbeck,
Pete Klenow, Annika Lorenz, Mark Lemley, Josh Lerner, Aprajit Mahajan, Shaun McRae,
David Mowery, Tom Nicholas, Roger Noll, Paul Rhode, Tim Simcoe, John Wallis, Gavin
Wright, Peter Zeitz, and seminar participants at the All-U.C. Economic History Conference
on Science and Innovation, U.C. Berkeley, the Clio Meetings in Arizona, DePaul, Duke
University Fuqua Strategy Conference, Loyola, the Melbourne Institute of Applied
Economic and Social Research, the NBER Conference on Patents, Standards, and
Innovation, the Summer Institute of the NBER’s Development of the American Economy
Program, Northwestern Economics, Northwestern’s Searle Center, Stanford University, the
U.B.C. Sauder School of Business, U.C. Irvine, the University of Chicago, and the Western
Economic Association International Conference for helpful comments. We also thank Josh
Lerner, Marcin Strojwas, and Jean Tirole for sharing their data on independent licensing.
Librarians at the National Archives in Chicago, Kansas City, New York City, Riverside, and
San Francisco were instrumental in facilitating access to primary documents.
The Journal of Law, Economics, and Organization, Vol. 32, No. 1

doi:10.1093/jleo/ewv014
Advance Access published August 3, 2015
ß The Author 2015. Published by Oxford University Press on behalf of Yale University.
All rights reserved. For Permissions, please email: journals.permissions@oup.com
2 The Journal of Law, Economics, & Organization, V32 N1
cross-licensing among their members.1 Pools have emerged as a prominent

mechanism to resolve patent wars over technologies, for which competing
firms own mutually infringing patents, such as smart phones and tablet
computers. Pools are also expected to facilitate commercialization and
increase overall welfare by lowering the transaction costs of licensing
(Merges 2001), and by preventing double-marginalization (or “royalty
stacking”), which occurs when individual firms charge excessive license
fees for complementary parts of the same technology (Merges 2001;

Shapiro 2001; Lerner and Tirole 2004). In 2008, the World Health
Organization recommended pools as a policy tool to encourage innov-
ation and improve access to life-saving tests and medication.2 Since then,
pools have formed around diagnostic tests for breast cancer and other
genetic diseases, as well as drugs to treat malaria and HIV.3
Whether or not pools encourage the creation of new medicines and
technologies, however, is difficult to predict. By raising the expected prof-
its from R&D, reductions in litigation risks, licensing fees, and in the
transaction costs of licensing may encourage pool members and outside
firms to innovate more. But the creation of a pool may also discourage
innovation in industries in which a large number of small outside firms
account for the majority of innovations. For example, a pool in the sewing
machine industry (1856–77) discouraged innovation by increasing litiga-
tion risks for outside firms, even as it reduced such risks for members
(Lampe and Moser 2010). By introducing differential license fees, the
sewing machine pool also diverted innovation away from the pool tech-
nology toward inferior substitutes, even though it reduced license fees for
both its members and outside firms (Lampe and Moser 2013).
Contemporary evidence is limited to case studies of pools for technol-
ogy standards; such “standards” pools define protocols that allow firms to
manufacture products such as DVD players and cellphones, for which
compatibility is important (Merges 2001; Shapiro 2001; Rysman and
Simcoe 2008; Simcoe 2012). Standards pools, which have been allowed
to form under antitrust rules that forbid anticompetitive pools, appear to
have mixed effects on innovation. For the CD industry, information on
improvements in the speed of data transfer indicates an increase in
1. Vaughan (1956: 41) includes cross-licensing agreements as a subgroup of patent pools,

while Shapiro (2001: 127) separates patent pools and cross-licensing agreements. We use
Vaughan’s (1956) definition to construct the sample and present separate estimates for
cross-licensing agreements in Section 5.2.
2. WHO Global Strategy and Plan of Action on Public Health, Innovation and
Intellectual Property, 2008 (Available at: http://apps.who.int/gb/ebwha/pdf_files/A61/
A61_R21-en.pdf, accessed March 4, 2013).
3. The HIV Medicines Patent Pool “. . . ‘pools’ multiple patents related to HIV medicines
in one place, which are then licensed out by the same one entity, in order to cut down on
transaction costs for all parties involved. This ‘pooling’ allows more affordable and more
adapted versions of patented drugs to be more easily produced as generics, long before their
20-year patent terms run out. Generic competition both brings down prices and helps spur
innovation” (http://www.medicinespatentpool.org, accessed March 4, 2013).
Patent Pools, Competition, and Innovation 3
innovation after the formation of a pool (Flamm 2013). For DVDs, com-
parable data on transfer speeds and product launches (Flamm 2013), as
well as patents (Joshi and Nerkar 2011) suggest a decline in innovation.
For the MPEG-2 standard pool, patenting declines after the creation of a
pool, as firms shift away from inventing toward implementing new tech-
nologies (Vakili 2012). For the IBM Patent Commons, which made pa-
tents freely available in 2005, Ceccagnoli et al. (2012) document a modest
increase in entry for open-source software segments, in which IBM con-

tributed patents.
This article extends the existing empirical evidence by investigating
changes in innovation for a diverse set of 20 industries that formed
patent pools in the absence of effective antitrust. All of these industries
created traditional (nonstandards) pools for product patents. Our article
extends existing analyzes of such pools (e.g. Gilbert 2004; Lampe and
Moser 2010; 2013) through a direct empirical test for the effects of com-
petition on innovation. Specifically, we examine the effects of competition
in research and development, by testing whether pool members and out-
side firms reduced their patenting activity in technology fields in which
multiple pool members had patented before the pool formed.4
The empirical analysis exploits a period of regulatory tolerance under
the New Deal, when economic policies such as the National Industrial
Recovery Act (NIRA 1933–35), exempted the majority of industries from
antitrust (Haley 2001: 8) to examine the potential effects of pools that
would form in the absence of effective regulation. In this regime of
weak enforcement, regulators allowed pools in a broad range of indus-
tries; this wave of pool formation remained unparalleled until the 1990s.
Arguments for patent pools in the 1930s anticipate arguments today. For
example, the US Supreme Court argued in favor of upholding the
Standard Oil pool for gasoline cracking in 1931 because “An interchange
of patent rights and a division of royalties . . . are frequently necessary if
technical advancement is not to be blocked by threatened litigation.”5
Enforcement of antitrust resumed after March 11, 1938, when President
Roosevelt appointed Thurman Arnold to reorganize the Department of
Justice’s Antitrust Division.6 In 1942, a District Court struck down the
Hartford Empire pool for glassware; in 1945, the US Supreme Court
upheld the Hartford Empire decision arguing that “the history of this
4. Critically, competition in research and development can have different effects on in-
novation than competition in product markets. For example, patent race models (such as
Reinganum 1982; Fudenberg et al. 1983) show that competition in research introduces an
additional risk for innovators who may not be first to develop a new technology (and there-
fore lose the market to a competitor). See Gilbert (2006) for a succinct discussion of these
arguments. Competition in R&D may also be inefficient if competitors’ research efforts are
exactly identical (e.g. Loury 1979); we examine this issue in subsection 6.1.
5. Standard Oil Co. of New Jersey v. United States, 283 U.S. 163, 167–168 (1931).
6. From 1940 to 1949, Justice brought 38 criminal antitrust cases per year, compared with
8.7 per year between 1930 and 1939 (Posner 1970: 376).
country has perhaps never witnessed a more completely successful eco-

nomic tyranny over any field of industry . . . .”7 After this decision, courts
began to shut down pools across industries (Vaughan 1956), creating a
paper trail of court records that we can analyze. Few pools formed until
the Department of Justice (DoJ) revised its antitrust guidelines in 1995 and
approved the MPEG and DVD standards pools between 1997 and 1999.8
A central empirical challenge lies in the ability to distinguish changes in
innovation that occur in response to the creation of a pool from changes

that occur in response to other, unobservable factors. During the New
Deal, government spending on infrastructure and work relief programs
(e.g. Wright 1974; Wallis 1984) may have encouraged innovation by
increasing demand. More generally, rates of innovation may vary across
the life cycle of an industry (Klepper 1996) or as a result of scientific
breakthroughs and experimentation (Rosenberg 1982).
Our analysis addresses this issue by comparing changes in patenting for
technologies that were included in a pool with a control group of closely
related technologies within the same industry that were not included in a
pool. Specifically, we exploit the fact that United States Patent Office
(USPTO) examiners assign patents to primary and cross-reference (second-
ary) subclasses to define the technology space that is covered by a patent.9
Baseline estimates compare changes in patents in the primary subclass of a
pool patent (the pool technology) with changes in cross-reference subclasses
(the control). As closely related technologies, cross-reference subclasses are
influenced by many of the same unobservable factors that influence patent-
ing in pool technologies. Empirically, pre-trends in patents are nearly iden-
tical for pool technologies and cross-reference subclasses; this helps to
address a common concern with difference-in-differences analyses.
The main data set consists of 75,396 patents between 1921 and 1948
across 433 pool subclasses and 828 cross-reference subclasses. Summary
statistics indicate that outside firms that were not pool members,
accounted for 97% of patents, even though pool members produced
more than 60% of output in some industries.10 If outside firms, however,
7. Justice Hugo Black arguing for the majority decision in Hartford-Empire Co. v. United
States, 323 U.S. 386, 436–37 (1945).
8. The revised 1995 guidelines treat licensing agreements as pro-competitive unless they
can be shown to reduce product market competition, and allow pools if they combine com-
plementary patents that are necessary to build a technology (Gallini 2011: 14–15).
9. Although standard data sets, such as the NBER US Patent Citations Data File (Hall et
al. 2001), do not include cross-reference subclasses, data on cross-reference subclasses can be
collected from the USPTO Patent Full-Text Database. Benner and Waldfogel (2008) have
shown that cross-reference subclasses are a useful measure for the technology space that is
covered by a patent.
10. For example, variable condensers to select radio stations (United States v. General
Instrument Corp., 87 F. Supp. 157 (D.N.J. 1949)) and furniture slip covers (United States v.
Krasnov, 143 F. Supp 184 (E.D. Pa. 1956)).
account for the majority of innovations, it is particularly important to

understand how the creation of a pool may affect these firms.
To further mitigate the impact of unobservable policy changes that may
have influenced patenting over time, changes in patenting are measured
relative to a pool-specific year of pool creation (controlling for calendar
year fixed effects). Regressions also include subclass and year fixed effects
to control for variation in the share of innovations that are patented
across technologies and over time (e.g. Moser 2005; 2012).

Difference-in-differences estimates of these data imply that pool tech-
nologies produced 14% fewer patents after the creation of a pool for each
additional patent that was included in a pool. Analogous estimates with a
binary treatment variable indicate that pool technologies with any number
of pool patents produced 16% fewer patents after the creation of a pool.
In interpreting these results, it is important to keep in mind that they are
informative only for industries that tend to form patent pools (rather than
randomly selected industries) because the formation of a pool is unlikely
to be exogenous. In the 1930s and today, industries that form pools tend to
be more concentrated, and pool members are often dominant firms. In the
1930s, the market shares of pool members ranged from 41% for railway
springs to 100% for high-tension cables (Lampe and Moser 2012). In the
1990s, members of a pool for laser equipment for eye surgery were the only
firms whose equipment had been approved by the FDA (Federal Trade
Commission [FTC] Docket No. 9286, March 24, 1998). More recently,
members of two pools for DVDs, including Samsung, Hitachi, and Sony,
jointly accounted for 49% of sales in 2001 (Flamm 2013). Moreover, the
timing of a pool’s formation may be endogenous if dominant firms form a
pool to soften competition (in product markets or research and develop-
ment) once a new technology has matured. To help address this issue, we
control for variation in patenting across the life cycle of an industry
through interactions between year and industry dummies. These estimates
confirm the main results: subclasses with 1 additional pool patent pro-
duced 13% fewer patents per year after the creation of a pool, which is
only slightly smaller than the baseline estimate.
To investigate whether the observed decline in patenting may be driven
by a decline in lower quality or “strategic” patents (e.g. Hall and Ziedonis
2001), we extend existing data on patent citations to construct a new data
set of 322,998 citations to 61,694 unique patents between 1921 and 2002
patents and calculate citation-weighted patents as a control for
quality. Analyses of citation-weighted patents imply an 11% decline in
citation-weighted patents for each additional pool patent, which indicates
a moderate decline in strategic patenting.
We also examine whether the observed decline in innovation may be
driven by a decline in the intensity of competition among pool members to
improve the pool technology. To perform this test, we exploit the fact that
the formation of a pool only reduces competition in research and devel-
opment for fields in which multiple pool members were active inventors
before the pool had formed. By comparison, it does not change the inten-
sity of competition in research fields in which a single pool member pa-
tented new technologies before the pool had formed.
This analysis reveals a significant differential decline in patenting for
subclasses in which the creation of a pool combined patents by multiple
firms. The USPTO classification system assigns technologies to subclasses
based on the function that they perform, which implies that pool patents
in the same subclass are more likely to be substitute patents. The DoJ–

FTC 2007 Antitrust Guidelines caution that pools, which combine patents
for substitute technologies, could serve as a mechanism to eliminate prod-
uct market competition, raise prices, and discourage innovation. These
concerns are corroborated by our findings that patenting declined most
significantly when historical pools reduced competition in research and
development by combining patents by multiple pool members in the same
subclass. Historical data also suggest that the USPTO’s system of clas-
sifying inventions by their function offers a straightforward indicator for
pool patents that cover substitutes.
2. Theoretical Predictions
This section presents existing theoretical predictions on patent pools and
suggests ways to extend existing theories of competition and innovation.
Most notably, we propose to extend existing theoretical models to con-
sider the effects of a patent pool on outside (non-member) firms, which
may be an important source of innovation, and in fact account for the
large majority of patents across industries.
A rich theoretical literature has examined the pricing strategies of
pool members and their effects on overall welfare. For example,
Lerner and Tirole (2004) show that the creation of a pool is more
likely to be welfare-enhancing if it combines patents for complementary
technologies, rather than substitutes. Brenner’s (2009) model implies
that predictions in Lerner and Tirole (2004) depend on members’ ability
to prevent entry into the pool. Llanes and Trento (2012) show that
innovators are less likely to join a pool with more members because
revenue would be shared with more firms. Jeitschko and Zhang (2012)
demonstrate that pools of complementary patents may weaken the in-
centive of downstream firms to invest in R&D if they create knowledge
spillovers that lead to a decline in product differentiation and thereby
reduce firms’ profits.
2.1 Pools May Encourage Innovation by Increasing the Returns to R&D

Pools may encourage pool members and outside firms to invest more in
R&D by lowering the transaction costs of licensing (Merges 2001), and by
preventing double-marginalization (or “royalty stacking”), which occurs
when individual firms charge excessive license fees for complementary
parts of the same technology (Shapiro 2001). The impact of license fees
is particularly important for “standards” pools, which combine (essential)

patents, for which no substitutes exist (Lerner and Tirole 2013; Quint
2014). If a subset of firms is vertically integrated, the creation of a pool
that combines complementary patents may also improve welfare by allow-
ing upstream firms to coordinate input prices for the benefit of down-
stream firms (Kim 2004), which in turn may encourage innovation.
Lerner and Tirole (2004) also show that, if pool patents are comple-
mentary and pools increase the profits of pool members by eliminating

double marginalization, then the prospect of a pool may encourage in-
novation (Lerner and Tirole 2004). If the pool combines a fixed number of
standard essential patents then potential pool members will increase their
R&D expenditures as the pool nears formation (Dequiedt and Versaevel
2013). In a two-stage patent race in which successive innovations are sub-
stitutable, innovators in the second stage invest more in R&D, if they are
able to soften product market competition with first-stage innovators by
forming a pool (Denicolò 2002).
The creation of a pool that combines (essential) patents for complemen-
tary technologies, which are necessary to produce a new technology, may
also stimulate innovation. Without a pool, the owners of individual pa-
tents cannot produce the technology without infringing on patents that
are owned by another firm, which enables the other firm to block produc-
tion. A pool, which resolves such blocking patents, may encourage innov-
ation by lowering litigation risks and increasing the expected profits from
investments in R&D (Merges 2001; Shapiro 2001).
2.2 Pools May Discourage Innovation by Reducing Competition

Models of product market competition and innovation, however, imply
that pools, which soften competition among pool members, may discour-
age innovation. Arrow (1962) shows that incentives to innovate are
weaker for monopolists than firms in competitive markets due to a “re-
placement effect,” because the monopolist’s rents from new technologies
replace existing streams of rents. For industries with strong patent pro-
tection, Lee and Wilde (1980) and Reinganum (1982) show that reductions
in the number of rivals reduce equilibrium levels of R&D. Models of
monopolistic competition, however, imply that mechanisms, which limit
product market competition, may encourage R&D and entry by increas-
ing the expected rents of successful innovators (Schumpeter 1942; Dixit
and Stiglitz 1977; Salop 1977). Gilbert and Newbery (1982) identify a
“preemption effect,” which gives incumbents a stronger incentive to in-
novate compared with their competitors because they have more to lose if
a competitor replaces them. Aghion et al. (2001; 2005) extend this litera-
ture to predict an inverted U-shape relationship between product market
competition and innovation, in which—starting from low levels of pre-
existing competition—a shift toward increased competition encourages
incumbents to innovate more by lowering their pre-innovation rents.
For the case of patent pools that form in concentrated industries, these
models imply a decline in innovation, if the creation of a pool reduces

product market competition.
2.3 Grant-Back Rules and Licensing Provisions

The effects of a pool on innovation may also be influenced by the
characteristics of a patent pool, such as grant-back rules and licensing
provisions (e.g. Lerner and Tirole 2004; Lerner et al. 2007; Lerner and
Layne-Farrar 2011). Grant-back rules require pool members to contribute

all patents for related technologies to the pool and make them available
for licensing to other members. Such provisions may encourage innov-
ation by limiting the potential for hold-up (Klein, Crawford, and Alchian
1978; Holmström and Roberts 1998), when opportunistic pool members
threaten to withhold patents for new technologies that other members
need to complement the pool technology (Shapiro 2001; Carlton and
Shampine 2013). The DoJ and FTC (1995, §5.6), however, caution that
Grantbacks may adversely affect competition, however, if
they substantially reduce the licensee’s incentives to engage in
research and development and thereby limit rivalry in
innovation markets.
As noted by the DoJ and FTC, grant-back provisions may also discour-
age innovation by reducing members’ private returns to R&D and by
encouraging members to free-ride on the R&D investments of other mem-
bers. For example, Vaughan (1956: 67) observes that the 1917 aircraft
pool discouraged innovation because “pooling all patents of members
and giving each the right to use the inventions of the other took away
each member’s incentive for basic inventions.” For standards pools that
combine essential patents, Aoki and Nagaoka (2004) predict that firms,
which own essential patents will refuse to join pools with grant-back rules,
especially if pool members share royalties evenly. Theoretical models also
predict that pools, which improve overall welfare, are more likely to allow
members to license their patents independently of the pool (Lerner and
Tirole 2004).11 In Section 5.2, we examine these predictions as an exten-
sion of the main tests.
3. Data
To examine changes in patenting after the creation of a pool, we collected
a new data set of 75,396 issued patents with information on application
years. These data cover patents filed between 1921 and 1948 for 20 indus-
tries that were affected by the creation of a pool. These data cover 10 years
11. Independent licensing may constrain prices only for pools that reduce welfare (but not
pools that increase welfare, Lerner and Tirole 2004). Lerner et al. (2007) find that 28 pools
between 1895 and 2001, which allowed independent licensing, were on average less likely to be
litigated compared with 35 other pools that formed between 1895 and 2001.
before the first pool formed and 10 years after the last pool formed. They
include 433 pool subclasses and 828 cross-reference subclasses without
pool patents in the same industry. To construct measures for the quality
of patented inventions, we also collected 322,998 citations to these patents
after 1921.
These data extend existing data sets on patents by including informa-
tion on the year in which inventors applied for a patent, in addition to the
year in which the patent was issued. This allows us to measure the timing

of invention more accurately because patents may be issued several years
after application (e.g. Popp et al. 2004; Gans et al. 2008). We extract
application years between 1921 and 1948 through a key word search,
which yields application years for 97.68% of 1,093,191 patents issued be-
tween 1921 and 1948.12 The mean lag between application and grant is
2.55 years with a standard deviation of 1.87 years.13
In addition, our data include information on cross-reference subclasses
(Section 3.2), while standard data sets, such as the NBER US Patent
Citations Data File (Hall et al. 2001) report only primary subclasses.
The data also include information on the number of pool members per
subclass (Section 3.2), and extend existing data to include citations from
1921 to 1975 (Section 3.3).
3.1 Pool Patents in 20 Industries, 1930–8

To construct data on patents that were included in a pool, we first col-
lected all mentions of patent pools during the 1930s from Vaughan (1956),
Gilbert (2004), and Lerner et al. (2007); this list includes 41 pools. We then
searched the records of the National Archives in Chicago, Kansas City,
New York, and Riverside for lists of patents and other characteristics of
these pools, including grant-back provisions, and exposure to litigation.
Data on patents that were included in original pooling agreements (pool
patentsi) are available for 20 pools.14 In comparison with the sample in
Lerner et al. (2007), which focuses on characteristics of patent pools, our
sample adds 8 pools and omits 12.15 Both samples are biased toward pools
12. For example, we search the full text of patent grants for the words “iling” (for
“Filing”) and “Ser.” (for “Serial Number”) to recover the year associated with this block
of text. In a random sample of 300 patents, the algorithm correctly records application years
for 296 patents.
13. In comparison, Popp et al. (2004) find that the average US patent between 1976 and
1996 was granted 28 months after the application (with a standard deviation of 20 months).
See the Supplementary Appendix for data on patent counts per year of application and grant.
14. We define pool patentsi to include only the original set of patents that were included in
a pool because patents that were entered after the pool had formed are likely to be influenced
by the agreement. Among 1053 patents by pool members filed after the original agreement,
680 patents (64.58%) covered inventions that were neither a pool technology or a cross-ref-
erence subclasses; 265 (25.17%) belonged to pool subclasses and 108 to cross-reference sub-
classes (10.26%).
15. We also exclude a pool for television and radio apparatus because it included only
foreign members, a pool for male hormones (1937–41) because it was short-lived, a pool for
that became subject to antitrust investigations, which creates a traceable

record of pool patents, and these pools may have been especially prom-
inent or aggressive. To address this issue, we have searched Vaughan
(1956) along with internet archives and court records for mentions of
other pools; these searches have turned up no additional pools beyond
the 20 pools in our sample, which suggests that our sample is reasonably
complete for the 1930s. For 13 pools, pool patents are available from
consent decrees at the National Archives; for 5 pools, pool patents are

included in licensing agreements; for 3 pools, patents are listed in written
complaints, and for another 3 pools, patents are included in the final
judgments. For four pools in railroad springs, Phillips screws, film, and
eyeglasses, patents are listed in more than one source.16
Similar to today, pools in the 1930s formed in relatively concentrated
industries. Members of pools for variable condensers and ophthalmic
frames accounted for 75% and 95% of their markets, respectively; mem-
bers of pools for railway springs, Phillips screws, dry ice, and fuse cutouts
produced between 41% and 75% of total output, and members of pools
for slip covers and high-tension cables accounted for 62% and 100% of
output, respectively.17 Pooling agreements for a total of five pools (fuel
injection equipment, variable condensers, stamped metal wheels, wrinkle
finishes, and slip covers), resolved patent litigation among pool members.
Most of the 20 pools appear to have formed as a mechanism to weaken
competition in product markets and potentially in R&D. Eleven pools did
not license their patents to outside firms, and instead divided the market
for the pool technology into exclusive geographic (seven pools) and prod-
uct segments (four pools) (Lampe and Moser 2012: 73–76). Among nine
pools that licensed their patents to outside firms, six appear to have used
licensing as a mechanism to establish price-fixing schemes (Lampe and
Moser 2012: 77–81). The remaining three pools licensed their patents
to less than four firms each to resolve litigation (Lampe and Moser
2012: 76–77).
Although industries that form pools may not be randomly selected, the
data for the current analysis cover a representative range of industries.
Industries with pools ranged from hydraulic oil pumps (1933–52), ma-
chine tools (1933–55), Philips screws (1933–49), variable condensers
used in radios (1934–53), wrinkle finishes, enamels, and paints (1937–
55), fuse cutouts (1938–48), to stamped metal wheels for automobiles,
buses, and trucks (1937–55) (Table 1). Years of pool formation are
grinding hobs (1931–43) because it combined two patents by the Barber-Colman Company,
and a pool for rail joint bars (1928–44) because it formed in 1928.
16. Seven of these pools transferred patents to a holding company, 9 pools licensed to
third parties, and 4 did both; 8 remaining “pools” were cross-licensing agreements according
to Vaughn’s (1956) definition.
17. Data on market shares are available for eight pools from court records that were
created after the 1942 decision in the Hartford Empire Case. See the Supplementary
Appendix for sources.
Table 1. 20 Patent Pools Formed between 1930 and 1938
Industry Year Members Pool Grant- Independent Prior

formed–year patents backs licensing litigation
dissolved
High tension cables 1930–48 2 73 1 – 0

Water conditioning 1930–51 3 4 1 0 0
Fuel injection 1931–42 4 22 1 – 1

Pharmaceuticals 1932–45 2 5 1 – 0
Railroad springs 1932–47 2 8 1 0 0
Textile machines 1932–50 2 40 0 – 0
Hydraulic oil pumps 1933–52 2 3 1 0 0
Machine tools 1933–55 5 3 1 0 0
Phillips screws 1933–49 2 2 1 0 0
Color cinematography 1934–50 2 143 1 – 0
Dry ice 1934–52 4 37 1 – 0
Electric generators 1934–53 2 30 1 0 0
Lecithin 1934–47 4 36 1 0 0
Variable condensers 1934–53 3 60 1 0 1
Aircraft instruments 1935–46 2 94 1 – 0
Stamped metal wheels 1937–55 3 90 1 – 1
Wrinkle paint finishes 1937–55 2 20 1 0 1
Fuse cutouts 1938–48 2 3 0 1 0
Ophthalmic frames 1938–48 4 23 0 0 0
Furniture slip covers 1938–49 2 2 0 1 1
Notes: Data from license agreements, written complaints, and court opinions from regional depositories of the
National Archives in Chicago (railroad springs, machine tools, Phillips screws, lecithin, stamped metal wheels,
wrinkle finishes, and fuse cutouts), Kansas City (ophthalmic frames), New York City (high tension cables, water
conditioning, fuel injection, pharmaceuticals, textile machinery, dry ice, electric equipment, variable condensers,
and aircraft instruments), and Riverside (color film). Member firms and pool patents are measured at the time of the
initial pooling agreement. Data on independent licensing are available for 12 pools in our sample from Lerner et al.
(2007).
spread relatively evenly between 1931 and 1938, with 9 pools before 1934
and 11 pools between 1934 and 1938. The average pool was active for
15.80 years and included pool patents that were 4.23 years old when the
pool formed, counting from the year of the patent application.
3.2 Patents per Year in Pool and Cross-Reference Subclasses

The main specifications compare changes in the number of patents per
year in 433 pool subclasses with changes in 828 cross-reference subclasses.
For example, US patent 1,896,594 (issued February 7, 1933) for “Wrinkle
Coating” was included in the 1937 pool for wrinkle paint finishes; this pool
patent belongs to the primary subclass 428/142, which we define as a pool
subclass (Figure 1): “Structurally defined web or sheet—Continuous and
nonuniform or irregular surface on layer or component (e.g., roofing,
etc.)—With transparent or protective coating.”
US patent 1,896,594 is also assigned to five cross-reference subclasses,
such as 428/161 (“Structurally defined web or sheet—Including variation
in thickness—With component conforming to contour of nonplanar
Pool subclasses
Pool Member: Kay & Ess Pool Member: Chadeloid
427/257
U.S. patent 2,077,112 U.S. patent 1,689,892
428/142
Type: Subclass: 524/313 Subclass: Type:
Original 427/257 427/257 Original
Cross-reference 427/288
Cross-reference
subclasses
U.S. patent 1,896,594
U.S. patent 2,069,252 427/288 Subclass: Type:

428/161 428/142 Original
Type: Subclass:
428/499 428/161 Cross-reference
Original 524/313
524/501 428/499 Cross-reference
Cross-reference 524/501
Patent holding company:

New Wrinkle, Inc.
1,689,892
1,896,594
2,069,252
2,077,112
Licensee 1 Licensee 2
Figure 1. Patent Pools, Pool Subclasses, and Cross-Reference Subclasses. The pool for
wrinkles finishes (1937–55) combined 20 patents from Kay & Ess Chemical Corporation
and Chadeloid Chemical Company. The two companies formed a patent holding com-
pany, New Wrinkle, to issue licenses to 185 outside firms. Pool subclasses are subclasses
listed as original subclasses on the pool’s patents. Cross-reference subclasses are sub-
classes listed as cross-reference subclasses on the pool’s patents. US patents 2,077,112
and 1,689,892 both list subclass 427/257 as their original subclass; this is a subclass with
multiple firms and multiple pool patents.
surface”) and 428/499 (“Composite (nonstructural laminate)—Of natural

gum, rosin, natural oil or lac—Next to cellulosic—Natural oil”); cross-
reference subclasses form the control group for the main specifications
(Figure 1).18
To create a searchable database of patents, USPTO examiners assign
patents to primary (“original”) and secondary (“cross-reference”) sub-
classes according to the function that the invention performs. Different
aspects of the technology space are covered by “claims,” which define the
technology space that is covered by each patent.
Examiners assign each patent to a single primary subclass to specify the
primary function that the invention performs. In practical terms, the pri-
mary subclass is the subclass that “receives the most intensive claimed
disclosure” (USPTO 1915: 21), or the largest number of claims. We exploit
this characteristic of the USPTO’s classification system to identify pool
18. Fifteen digest subclasses, which combine patents for all inventions that belong “to a
class but not to any particular subclass” (http://www.uspto.gov/web/offices/ac/ido/oeip/taf/
c_index/explan.htm, accessed May 10, 2014) are dropped from the sample.
patents for substitute technologies: If patents are assigned to primary

subclasses based on the function that they perform, patents that are as-
signed to the same primary subclasses perform the same function, and are
therefore more likely to serve as substitutes for each other.
In addition to the primary subclass, examiners can also assign patents to
one or more cross-reference subclasses. These cross-reference subclasses
include information about the invention that complements the informa-
tion in the primary subclass and help examiners to identify a broader set of

relevant technologies when they search for “prior art.” For example,
cross-reference subclasses may cover technologies that the patentee did
not claim in their application but that the examiner considers to be rele-
vant for the patented technology. For example, if “a patent discloses an
internal combustion engine associated with a specific form of carburetor
[and] the claims relate to the engine parts only [then] the class of Internal-
Combustion Engines should receive the patent, and a cross-reference
should be placed in Carburetors” (USPTO 1915: 32).
An analysis of 118,350 patents by 64 firms in the photographic industry
between 1980 and 2002 suggests that incorporating information on cross-
reference subclasses improves the accuracy of measuring the technology
space that is covered by patented inventions (Benner and Waldfogel 2008).
On average, the 698 pool patents in our sample are assigned to 1.99 cross-
reference subclasses. We exploit this information to define a control group
of technologies that are subject to the same unobservable shocks as the
pool technology but less affected by the creation of a pool.19
Of 433 pool subclasses, 327 include one pool patent, and 106 include
multiple pool patents (Table 2). Thirty-eight subclasses with multiple pool
patents include patents by multiple firms. For example, the wrinkle
finishes pool combined Kay and Ess’ US patent 2,077,112 for “imitation
leather paper” with the Chadeloid Chemical Company’s patent 1,689,892
for “wrinkle finishes.” Both patents are assigned to USPTO subclass
427/257, which covers inventions to produce an “irregular surface . . . by
intentionally employing coating materials which dry to a wrinkled appear-
ance or which crack on drying to produce a ‘crackled’ finish” (Figure 1).
As the USPTO uses the “fundamental, direct, or necessary function [of
an invention] as the principal basis for classification,” two patents as-
signed to the same subclass are more likely to be substitutes than
19. Alternative specifications limit the control to cross-reference subclasses within the
same main class, and expand the control to include all subclasses in the main class (e.g.
class 428 “stock material or miscellaneous articles”). Seven subclasses include patents by
more than one pool; for them, we define the year of pool formation using the earliest pool.
One subclass (352/225) is listed as a pool subclass for two pools. Four subclasses (340/524, 62/
056, 524/594, and 174/152R) are pool and cross-reference subclasses; two subclasses (417/426
and 200/56R) are cross-reference subclasses for two pools. We assign them to the pool that
formed first. For five pools (fuel injection, pharmaceuticals, railroad springs, lecithin, and
aircraft instruments), the pool years include a small number of years after the pool had
dissolved. To be conservative we include these years as pool years.
Table 2. Patents per Subclass and Application Year
Pre-pool formation Post-pool formation All years Pre-pool–post-pool
Raw patents
Pool subclasses (N ¼ 433) 2.535 (3.854) 2.400 (3.049) 2.468 (3.478) 0.210** (0.119)
1 pool patent (N ¼ 327) 2.267 (3.276) 2.293 (2.926) 2.280 (3.107) 0.044 (0.117)
>1 pool patent & 1 pool firm (N ¼ 68) 2.925 (4.585) 2.805 (3.599) 2.867 (4.134) 0.215 (0.367)
>1 pool patent & >1 pool firm (N ¼ 38) 4.196 (6.032) 2.604 (2.978) 3.376 (4.777) 1.631** (0.591)
Cross-reference subclasses (N ¼ 828) 2.695 (3.663) 2.935 (3.880) 2.815 (3.774) 0.211 (0.076)
Citation-weighted patents
Pool subclasses (N ¼ 433) 9.886 (16.466) 15.123 (24.211) 12.487 (20.843) 4.960** (0.628)
1 pool patent (N ¼ 327) 9.123 (15.156) 14.914 (23.818) 11.999 (20.142) 5.507* (0.666)
>1 pool patent & 1 pool firm (N ¼ 68) 11.189 (18.471) 15.810 (25.535) 13.439 (22.307) 4.746 (1.893)
>1 pool patent & >1 pool firm (N ¼ 38) 14.233 (21.969) 15.693 (25.140) 14.985 (23.656) 0.633** (2.585)
Cross-reference subclasses (N ¼ 828) 11.609 (18.792) 19.402 (29.521) 15.496 (25.037) 7.812 (0.561)
Notes: Pool subclasses include at least one pool patent that lists this subclass as the primary subclass. Cross-reference subclasses are subclasses without pool patents that patent examiners have
identified as related technologies. Citation-weighted patents are constructed as 1+ # of citations by later patents (Trajtenberg 1990). We collect citations by searching the full text of patent grants 1921–74
for all patent numbers in our data, adding citations from patent grants 1975–2002 from Hall et al. (2001). Post-pool – pre-pool reports average pre- and post-pool difference across subclasses. Pre-pool
formation includes all years between 1921 and the formation of a pool; post-pool formation includes all years between the formation of a pool and 1948. **, * denotes significant difference at 1%, 5%
compared to cross-reference subclasses. Standard deviations are given in parentheses.

inventions assigned to different subclasses.20 For example, two producers

of equipment for photorefractive keratectomy (PRK) laser eye surgery,
VISX and Summit, formed a pool in 1992.21 In 1998, three years after
developing guidelines for antitrust enforcement of intellectual property
licensing, the FTC challenged and dissolved their agreement, arguing
that, in the absence of a pool “VISX and Summit could have and would
have competed with one another in the sale or lease of PRK equipment by
using their respective patents, licensing them, or both.” The laser eye sur-

gery pool covered 25 patents; 21 of these patents (7 from Summit and 14
from VISX) were assigned to the same subclass.22
3.3 Citations by Patents after 1921 as a Control for Patent Quality

Trajtenberg (1990) shows that citation-weighted patent counts—calcu-
lated by adding the number of citations that a patent receives to the
count for each patent (i.e. each patent is weighted as 1 plus the number
of citations it receives)—are correlated with the estimated surplus of im-
provements in computed tomography (CT) scanners. Hall et al. (2005)
establish a positive correlation between the ratio of citations to a firm’s
patents and that firm’s stock market value. Similarly, an analysis of patent
data and stock market returns between 1926 and 2007 (Kogan et al. 2012)
indicates that excess returns on the day of a patent application and the day
of a patent issue are highly correlated with the number of citations to the
patent. An analysis of field trial data for patented inventions in hybrid
corn establishes that citations are positively correlated with the size of
patented improvements in plants, measured as improvements in yields
(Moser et al. 2012).23
To construct historical citations data, we searched the full text of patent
grants between January 4, 1921 and December 31, 1974 for mentions of
75,396 unique patent numbers in our data. Until February 4, 1947,
USPTO patent grants recorded citations anywhere in the text of the
patent document; to extract these citations, we search the full text of
patent grants. After February 4, 1947, the USPTO began to organize
citations in separate sections at the beginning or at the end of patent
documents; we extract citations directly from these sections.24 This
20. USPTO Examiner’s Handbook, Chapter 1, available at http://www.uspto.gov/pa-

tents/resources/classification/handbook/one.jsp (accessed May 13, 2014).
21. The FTC’s written complaint as well as a list of patents included in the pool is available
at: http://www.ftc.gov/os/1998/03/summit.cmp.htm (accessed November 2, 2013).
22. Subclass 606/5 for Surgery-Instrument-Light application-Ophthalmic-Recurving or
reshaping of the eye.
23. Patent citations are checked by professional examiners who remove false citations and
add relevant citations that inventors may have missed. For US patent grants between January
2001 and December 2002, patent examiners added between 21% and 32% of missing citations
(Lampe 2012).
24. To evaluate the quality of these data, we examined page scans for 150 randomly
chosen patents between 1947 and 1974 on Google Patents (www.google.com/patents). This
check indicates that the algorithm correctly identifies 636 of 741 (85.83%) of citations; 5 of
search yields a total of 238,874 citations between January 4, 1921 and

December 31, 1974, to which we added 84,124 citations between
January 7, 1975 and December 31, 2002 from the NBER Data File.
3.4 Nonpatent Measures of Innovation

Despite many benefits, patent data and citations may systematically miss
certain types of innovation that are more likely to be protected by secrecy

and alternative mechanisms (Levin et al. 1987; Cohen et al. 2000; Moser 2012).
This issue may be particularly severe if—after the creation of a pool—R&D
investments shifts toward implementations of the pool technology, which are
not patented (e.g. Vakili 2012). To address this issue, we examine alternative
measures of innovation for the movie industry (Section 6.4); movies account
for the largest number of pool patents in our sample (Table 1).
4. Results
For pool technologies, patenting declined after the creation of a pool, both
in absolute terms and relative to alternative definitions of the control. In
pool subclasses, patents declined from 2.54 per subclass and year before
the creation of a pool to 2.40 afterwards (Table 2), and from 2.80 to 2.48
within a 10-year window before and after the creation of a pool (Figure 2).
By comparison, patents in cross-reference subclasses increased from 2.70
before the creation of a pool to 2.94 afterwards (Table 2).
Data on patents per year, however, indicate that patenting in cross-
reference subclasses declined after a small initial increase (Figure 2),
which is suggestive of negative spillovers from pool to cross-reference
technologies. If these spillovers are economically important, they lead
the difference-in-differences analysis to underestimate the true decline in
patenting as a result of the pool. The data also indicate that patents for
pool subclasses and cross-reference subclasses followed very similar trends
in patenting before the formation of a pool.
4.1 Baseline Estimates

Baseline estimates compare changes in patents per subclass and year for
subclasses with an additional pool patent with changes for cross-reference
subclasses, controlling for subclass and year fixed effects:
Patentsct ¼ a+b1 poolct # pool patentsc +dt +fc +"ct ; ð1Þ
where the variable # pool patentsc counts the number of pool patents in
subclass c, and poolct equals 1 if subclass c includes at least one pool patent
and year t is after the creation of a pool. Year fixed effects t control for
unobservable variation in patenting over time that is common across
technologies, and subclass-fixed effects fc control for unobservable
105 citations that the algorithm missed were misread numbers (i.e. false positives) as a result
of errors in the optical character recognition (OCR).
3.5
Pool Start Year
2.5

2
1.5
-9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
Pool subclass Cross-reference subclass
Figure 2. Patents per Subclass and Year: Pool versus Cross-reference Subclasses. t ¼ 0
denotes the year when a pool forms; the timing of invention is measured at the year of the
patent application. Data include patent counts for 433 pool subclasses that include at
least one pool patent and 828 cross-reference subclasses that patent examiners have
identified as related technologies.
variation in patenting across technologies that is constant over time.

Robust standard errors are clustered at the subclass level.
Under the assumption that changes in patents per year would be com-
parable in pool and cross-reference subclasses in the absence of a pool, the
coefficient for the difference-in-differences estimator poolct * # pool pa-
tentsc measures the causal effect of the creation of a pool. Comparisons of
pre-trends support this assumption, and we will investigate it in more
detail below. In interpreting our results it is, however, important to
keep in mind that they are generalizable only to industries that would
form pools, rather than a randomly selected sample of industries.
Baseline OLS estimates indicate that subclasses with one additional
pool patent produce 0.36 fewer patents per year after the creation of a
pool (significant at 1%, Table 3, column 1).25 Compared with a mean of
2.47 patents per year in pool subclasses (Table 2), this represents a 14.38%
decline in patenting.
Alternative specifications, which define exposure to a pool more flexibly to
allow for nonlinear effects, confirm a significant decline in patenting. In these
specifications, the binary variable 5 1 pool patentsc equals 1 for any subclass
that includes at least one pool patent. Estimates indicate that subclasses with
any number of pool patents produce 0.40 fewer patents after the creation of
a pool (significant at 1%, Table 3, column 2), implying a 16.33% decline in
patenting. These estimates are confirmed by conditional quasi-maximum
25. Estimates remain statistically significant at the 1% level with clustering at the level of
industries with a standard error for poolct * pool patentsc of 0.11 compared with a standard
error of 0.10 with clustering at the subclass level.
Table 3. OLS and QMLE Poisson—Dependent Variable is Patents per Subclass and
Year
OLS OLS QMLE OLS OLS OLS: Excl.

Poisson Pool patents
(1) (2) (3) (4) (5) (6)
Pool * # pool patents 0.355** 0.317** 0.385** 0.337**

(0.096) (0.105) (0.117) (0.094)
Pool * 1 pool patents 0.403** 0.160**

(0.137) (0.051)
Constant 1.975** 1.975** 1.201** 1.975** 1.925**
(0.073) (0.072) (0.442) (0.072) (0.071)
Subclass fixed effects Yes Yes Yes Yes Yes Yes
Year-fixed effects Yes Yes Yes Yes Yes Yes
Industry*year-fixed effects – – – Yes – –
Linear and quadratic trends – – – – Yes –
Standard errors clustered at the subclass level;
** p < 0.01, * p < 0.05
n (# subclasses * 28 years) 35,308 35,308 35,308 35,308 35,308 35,308
R2/Log-likelihood 0.554 0.551 62,824 0.580 0.554 0.557
Notes: The dependent variable counts patents per subclass and year. The timing of invention is measured by the
application year for granted patents. The variable poolct equals 1 after a pool forms. # pool patentsc counts patents
that were included in the initial pooling agreement and list subclass c as their primary subclass. 1 pool patentsc
equals 1 if subclass c contains at least one pool patent. In all, 433 pool subclasses include one or more pool
patents; 828 cross-reference subclasses, which patent examiners have identified as related technologies, form the
control. Column 3 reports the average marginal effect. Column 4 includes year-industry fixed-effects for 18 industries
that contribute more than 5 subclasses to the sample. Phillips screws and slip covers, for which industry fixed
effects are not calculated, contribute 5 and 2 subclasses, respectively.
likelihood fixed-effects Poisson estimates (QMLE Poisson), which address

the count data characteristics of patents.26 QMLE Poisson estimates imply
that pool subclasses produce 14.79% fewer raw patents after the creation of
a pool (significant at 1%, Table 3, column 3).27
4.2 Controlling for Variation in Patenting across the Industry Life Cycle
Part of this decline in patenting may be a result of changes in patenting that
would have occurred even without the creation of a pool. Most import-
antly, firms may be more likely to form a patent pool after their technology
26. See the Supplementary Appendix for details on the distribution of patents across
subclasses. 29.11% of subclass-year observations had zero patentsct, which suggests that
count data models are appropriate but not essential. Our preferred specification for the
(linear) difference-in-differences estimator is OLS because OLS is the best linear unbiased
estimator and because the consistency properties of the OLS are more transparent. For a
binary treatment variable, such as poolct * pool patentsc, Poisson quasi-maximum likelihood
regressions yield consistent estimates for nonlinear difference-in-differences models (Imbens
and Wooldridge 2007). Wooldridge (1999) develops a QMLE for the fixed effects Poisson
model that is robust to heteroskedasticity and serial correlation; Rysman and Simcoe (2008)
implement the estimator for a binary treatment variable. Bertanha and Moser (2014) show
that, in addition to robustness to serial correlation, the QMLE Poisson with conditional fixed
effects is also robust to cross-sectional variation, as long as the structure of cross-sectional
variation is time invariant.
27. The percentage change is calculated as (exp(0.160) 1) 100 ¼ 14.79.
has matured, and patenting in mature industries may decline independently

of the creation of a pool. Consistent with this idea, patenting in the
variable condensers industry, for example, declines prior to the pool, sug-
gesting that the technology may have already matured (see the
Supplementary Appendix for information on patents per year across indi-
vidual industries).
To help control for variation in patenting that occurs over the life cycle
of an industry, we estimate the main specifications with interactions be-

tween year and industry fixed effects to flexibly control for differential
changes in patenting across the industry life cycle. These regressions indi-
cate that subclasses with one additional pool patent produce 0.32 fewer
patents per year after the creation of a pool (significant at 1%, Table 3,
column 4), implying a decline of 12.84%.28 This estimated decline is also
robust to alternative specifications, which control for a separate linear or
quadratic trend for pool technologies. Estimates with quadratic trends
indicate that pool subclasses with one additional pool patent produce
0.39 fewer patents per year after the creation of a pool (significant at
1%, Table 3, column 5), implying a decline of 15.60%.
Consistent with the fact that outside firms produce 97.27% of all pa-
tents in the average industry, excluding all 2058 patents by pool members
leaves the estimates substantially unchanged. Subclasses with an add-
itional pool patent produce 0.34 fewer patents per year after the creation
of a pool (significant at 1%, Table 3, column 6), implying a 14.46% decline
compared with a mean of 2.33 patents per year.
4.3 Time Varying Estimates for the Pre- and Post-Pool Period
To investigate the timing of effects and test for differential pre-trends,
which would violate the identifying assumption of the difference-in-dif-
ferences estimator, we estimate coefficients separately for each year,
allowing the estimated “effect” of a pool to begin before the pool has
formed:
Patentsct ¼ a+bk # pool patentsc +dt +fc +"ct ð2Þ
where # pool patentsc, as well as year and time fixed effects are defined as
above, and k ¼ 17, 16 . . . 17, 18 denotes years before and after the
creation of a pool forms; k ¼ 0 is the excluded period.
Annual coefficients for the pre-period are not statistically significant in
any year except t 1, when estimates imply a 9.12% increase in patenting,
28. Specifications in Table 3, column 3, include interactions between fixed effects for 18
industries (excluding Phillips screws and slip covers) in which pool patents covered more than
5 subclasses. Due to the small number of subclasses for Phillips screws (5) and slip covers (2),
there is little within-cluster variation, and we cannot calculate standard errors with a full set of
year-industry fixed effects. Results are robust to including a full set of year-industry fixed
effects and calculating bootstrapped standard errors, or including a full set of year-industry
fixed effects without clustering standard errors on subclasses.
0.8
Pool Start Year
0.6
0.4
0.2

0
-9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
-0.2
-0.4
-0.6
-0.8
Figure 3. Annual Coefficients, OLS, Dependent Variable is Patents per Subclass and
Year. Notes: t ¼ 0 denotes the year when a pool forms; the timing of invention is measured
at the year of the patent application. Estimates for k (with 95% confidence interval) in the
regression Patentsct ¼ + k * # pool patentsc + fc + t + "ct where k ¼ 17,. . .,17, 18,
counts years before and after a pool forms. The variable # pool patentsc counts patents
that were included in the initial pooling agreement and that list subclass c as their primary
subclass.
which is consistent with the idea of a patent race leading up to the creation
of a pool (Figure 3). Excluding data for year t 1, however, leaves the
results substantially unchanged, suggesting that changes in innovation as
a result of a patent race alone cannot adequately explain the observed
decline in innovation. Excluding data in t 1 reduces the size of the esti-
mated decline by a small amount to0.34 (significant at the 1% level, not
reported), which implies a 13.95% decline compared with a mean of 2.44
patents.
Most significantly, however, estimates imply that the decline in patent-
ing for pool technologies begins after the creation of a pool and intensifies
over time. Annual coefficients range from 0.17 to 0.30, with an average
0.23 for the first five years, indicating a decline of 9.32%, and from
0.34 to 0.69, with an average of 0.43 for years six and above, indicat-
ing a decline of 17.30% (significant at 5% in years one, three, four and all
years above five, Figure 3).29
29. See the Supplementary Appendix for regressions with industry-year inter-
actions, which confirm these results. Estimates become statistically significant, with an esti-
mate of 0.31 three years after the creation of a pool, implying a decline of 12.68%, and
remain significant through the sample, with an estimate of 0.39 ten years after the creation
of a pool, implying a decline of 15.84%.
4.4 Controlling for Patent Quality through Citations

To investigate whether the large decline in patenting may be driven by a
decline in lower quality or “strategic” patents, we construct citation-
weighted patents, as a quality-adjusted measure for changes in patenting
(following Trajtenberg 1990):
Citation-weighted patentsct ¼ patents by application year 1921 1948ct
+citations in patent grants 1921 2002 to patent applications 1921 1948ct

ð3Þ
Citation-weighted patents increase for both pool and cross-reference sub-
classes over time, because more recent patents are more likely to be cited
(Hall et al. 2001; Mehta et al. 2010).30 This increase, however, is substan-
tially smaller for pool subclasses. After the creation of a pool, the average
pool subclass produced 15.12 citation-weighted patents, compared with
9.89 before. By comparison, the average cross-reference subclass produced
19.40 citation-weighted patents after the creation of a pool, compared
with 11.61 before (Table 2).
Estimates with citation-weighted patents are slightly smaller than esti-
mates with raw patents, but remain statistically significant and large.
Subclasses with an additional pool patent produce 1.42 fewer citation-
weighted patents after the creation of a pool (significant at 1%, Table 4,
column 1), implying an 11.33% decline compared with a mean of 12.49
citation-weighted patents (Table 2).31 Compared with a decline of 14.38%
for raw patents (Table 3, column 1) this implies a moderate decline in stra-
tegic patenting after the creation of a pool. Poisson estimates imply that pool
subclasses produce 10.51% fewer citation-weighted patents after the creation
of a pool (significant at 5%, Table 4, column 3).32 Analyses that exclude
member patents from the sample indicate that subclasses with an additional
pool patent produce 1.42 fewer citation-weighted patents (significant at 1%,
Table 4, column 6), implying an 11.96% decline in quality-adjusted patents,
compared with a mean of 11.84 citation-weighted patents per year.
4.5 Additional Robustness Checks

Additional robustness checks are provided in the Supplementary
Appendix. Results are robust to an alternative definition of the control
30. Results are robust to excluding patent counts and defining citation-weighted pa-
tentsct ¼ citations in patent grants 1921–2002 to patent applications 1921–1948ct (e.g.
Rysman and Simcoe 2008; Galasso and Simcoe 2011). Estimates indicate that subclasses
with an additional pool patent produce 1.06 fewer citation-weighted patents after the creation
of a pool (significant at 1%, not reported), implying a 10.57% decline compared with a mean
of 10.02 citation-weighted patents.
31. Results are robust to alternative tests that remove patents that were not cited, and
weight citation counts by the average number of citations to all patents issued in the same
year.
32. Percentage changes is calculated as (exp(0.111)1) 100 ¼ 10.51.
Table 4. OLS and QMLE Poisson—Dependent Variable is Citation-weighted Patents
OLS OLS QMLE OLS OLS OLS: Excl.

Poisson Pool patents
(1) (2) (3) (4) (5) (6)
Pool * # pool patents 1.415** 1.125** 1.030** 1.417**

(0.289) (0.300) (0.280) (0.285)
Pool * 1 pool patents 2.625** 0.111*
(0.801) (0.051)

Constant 6.608** 6.608** 3.875 6.608** 6.437**
(0.397) (0.396) (2.993) (0.397) (0.390)
Industry*year-fixed effects – – – Yes – –
Linear and quadratic trends – – – – Yes –
** p < 0.01, * p < 0.05
R2/Log-likelihood 0.474 0.474 247,094 0.495 0.474 0.474
Notes: The dependent variable counts patents per subclass and year. Citation-weighted patents are constructed as
1+ # of citations by later patents (following Trajtenberg 1990). The timing of invention is measured by the application
year for granted patents. The variable poolct equals 1 after a pool forms. # pool patentsc counts patents that were
included in the initial pooling agreement and list subclass c as their primary subclass. 1 pool patentsc equals 1 if
subclass c contains at least one pool patent. In all, 433 pool subclasses include one or more pool patents; 828
cross-reference subclasses, which patent examiners have identified as related technologies, form the control.
Column 3 reports the average marginal effect. Column 4 includes year-industry fixed-effects for 18 industries that
contribute more than 5 subclasses to the sample. Phillips screws and slip covers, for which industry fixed effects are
not calculated, contribute 5 and 2 subclasses, respectively.
group that restricts the control group to 631 cross-reference classes in

the same 108 main classes that include at least one of 433 pool sub-
classes, and to an alternative definition that expands the control group
to include all 69,316 subclasses in 108 main classes with pool subclasses
and 61 main classes with cross-reference subclasses. An additional ro-
bustness check excludes two subclasses with exceptionally many pa-
tents: aircraft instruments (12 pool patents) and stamped metal
wheels (10 pool patents). Estimated effects are also robust to restricting
the sample to pools that formed before May 27, 1935, when the US
Supreme Court ruled that price- and wage-fixing in the poultry indus-
try, which had been sanctioned under the NIRA, were unconstitu-
tional.33 A final robustness check estimates 20 separate regressions,
excluding 1 of the 20 industries in each regression, to check whether
the decline in patenting may be driven by a single industry.
5. Changes in the Intensity of Competition

How may the creation of a pool lead to a decline in innovation?
Regulators are concerned that pools may reduce welfare by weakening
33. A.L.A. Schechter Poultry Corp. v. United States, 295 U.S. 495 (1935). Congressional
hearings began to scrutinize patent pools in 1935 (Pooling of Patents, Hearings before House
Committee on Patents on House Resolution 4523, Parts I-IV, 74 Cong (February 11 to March
7, 1935)).
competition in product markets and R&D (e.g. Department of Justice

and Federal Trade Commission 2007). Theoretical predictions on the
effects of competition on innovation are ambiguous (e.g. Schumpeter
1942; Arrow 1962; Aghion et al. 2005), and, to the best of our knowledge,
there is no systematic evidence on the effects of pools on competition and
innovation.
To address this issue, we exploit the fact that changes in the intensity of
competition vary with the number of pool members that are active in-

ventors for a given technology before a pool forms. For subclasses in
which a single pool member is active, the creation of a pool has no
effect on competition to improve the pool technology. For subclasses
with patents by multiple pool members, the pool reduces competition
between pool members by allowing them to combine their patents,
rather than compete with each other to improve substitute technologies.
This differential change allows us to examine the effects of changes in the
intensity of competition to improve substitute technologies, holding con-
stant alternative mechanisms by which the creation of a pool may influ-
ence invention.
5.1 Differential Changes in Patenting in Fields with Reduced Competition

Summary statistics indicate that the observed decline in patenting for pool
technologies was driven by steep declines for technologies in which a pool
combines patents by competing firms. In subclasses with multiple mem-
bers, patents per year decline from 4.20 patents per year before the cre-
ation of a pool to 2.60 afterwards (Table 2). Restricting the data to a
10-year window before and after the creation of a pool, patents per year
decline from 4.43 patents before the creation of a pool to 2.73 patents
afterwards (Figure 4). By comparison, technologies in which a single pool
member owned all patents experienced a much smaller decline from 2.93
to 2.81 patents per year, and subclasses with a single pool patent (and
one pool member by definition) increased slightly from 2.27 to 2.29
patents (Table 2).
To evaluate these differences more systematically we estimate:
Patentsct ¼ a + b1 poolct 1 pool patentc 1 firmc
+ b2 poolct > 1 pool patentc 1 firmc ð4Þ
+ b3 poolct > 1 pool patentc > 1 firmc + dt + fc + "ct
where 1 pool patentc equals 1 if subclass c includes a single pool patent (327
subclasses), >1 pool patentc * 1 firmc equals 1 if subclass c includes multiple
pool patents by a single firm (68 subclasses), and >1 pool patentc * >1 firmc
equals 1 if subclass c includes pool patents by multiple firms
(38 subclasses).
Estimates indicate that subclasses in which a pool combined patents
by multiple pool members produce 1.89 fewer patents per year after the
6
Pool Start Year

2
0
-9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
Pool subclasses with patents by multiple firms
Pool subclasses with patents by 1 member firm
Cross-reference subclass
Figure 4. Patents per Subclass and Year: Pool Subclasses with 1 versus >1 Pool
Member. Notes: t ¼ 0 denotes the year when a pool forms; the timing of invention is
measured at the year of the patent application. Data include 433 pool subclasses that
include at least one pool patent and 828 cross-reference subclasses that patent exam-
iners have identified as related technologies. For 38 pool subclasses, the creation of a
pool combines patents by multiple members.
creation of a pool (significant at 1%, Table 7, column 1), implying a

55.98% decline relative to a mean of 3.38 patents per year (Table 2).
By comparison, estimates are small and not statistically significant for
subclasses with multiple pool patents by a single member and for sub-
classes with a single pool patent by a single member. A Wald test statistic
of 4.65 with a p-value of 0.03 rejects the hypothesis that estimates are equal
for technologies with multiple pool patents by multiple firms and for tech-
nologies with multiple pool patents by a single firm. Estimates with con-
trols for quadratic trends indicate that pool subclasses with multiple
members produce 1.76 fewer patents per year after the creation of a
pool (significant at 1%, Table 5, column 3), implying a decline of
52.07%.34
Citation-weighted counts confirm the differential decline in patenting
for subclasses in which the pool combined patents by competing firms.
34. Results are robust to including separate linear or quadratic trends for pool subclasses
with one pool patent, with one member and multiple pool patents, and with multiple mem-
bers, as well as to excluding two subclasses with many pool patents (aircraft instruments with
12 patents by 1 firm, and stamped metal wheels, with 10 patents by 3 firms). Estimates for this
restricted sample indicate that subclasses with multiple firms produce 1.88 fewer patents per
year after the creation of a pool (significant at 1%), implying a decline of 54.82% compared
with a mean of 3.43 patents per year in pool subclasses.
Table 5. OLS—Subclasses with 1 versus More Pool Member; Dependent Variable is

Patents per Subclass and Year
Raw patents Citation-weighted
(1) (2) (3) (4) (5) (6)
Pool * 1 pool patent 0.235 0.243 0.098 2.080* 1.918* 0.007

(0.137) (0.142) (0.148) (0.840) (0.880) (0.926)
Pool * > 1 pool patent * 1 firm 0.388 0.275 0.250 2.694 1.619 0.600

(0.379) (0.384) (0.379) (1.994) (1.974) (2.035)
Pool * > 1 pool patent * > 1 firm 1.890** 1.600** 1.758** 7.231** 5.445* 5.187*
(0.589) (0.573) (0.588) (2.521) (2.478) (2.448)
Constant 1.975** 1.055* 1.975** 6.608** 1.983 6.608**
(0.073) (0.442) (0.073) (0.397) (3.303) (0.397)
Industry*year fixed effects – Yes – – Yes –
Linear and quadratic trends – – Yes – – Yes
** p < 0.01, * p < 0.05
R2 0.553 0.579 0.553 0.474 0.495 0.474
Notes: The dependent variable counts patents per subclass and year. Citation-weighted patents are constructed as
1+ # of citations by later patents (following Trajtenberg 1990). The timing of invention is measured by the application
year for granted patents. 1 pool patentc equals 1 if there was exactly 1 pool patent that was included in the initial
pooling agreement and listed subclass c as their primary subclass (327 subclasses). >1 pool patentc *1 firmc
equals 1 if subclass c includes more than 1 pool patent owned by a single member firm (68 subclasses). >1 pool
patentc * >1 firmc equals 1 if subclass c includes patents owned by multiple member firms (38 subclasses). In all,
433 pool subclasses include one or more pool patents; 828 cross-reference subclasses, which patent examiners
have identified as related technologies, form the control. Columns 2 and 5 include year-industry fixed-effects for 18
industries that contribute more than 5 subclasses to the sample. Phillips screws and slip covers, for which industry
fixed effects are not calculated, contribute 5 and 2 subclasses, respectively.
Subclasses with patents by multiple members produce 7.23 fewer citation-

weighted patents after the creation of a pool (significant at 1%, Table 5,
column 4), implying a decline of 48.25% relative to a mean of 14.99 cit-
ation-weighted patents per year (Table 2). By comparison, estimates are
small and not statistically significant for subclasses with multiple patents
by a single firm and for subclasses with a single pool patent by a single
firm.
Alternative specifications compare changes in patenting for subclasses
with multiple pool patents by multiple firms with changes for subclasses
with multiple pool patents by the same firm, using subclasses with a single
pool patent as a control. Estimates (reported in the Supplementary
Appendix) exceed the baseline estimates.
5.2 Pool Characteristics: Grant-Back Rules, Licensing, and Prior Litigation

Predictions on the effects of pool characteristics, including grant-back
rules and licensing provisions, are theoretically ambiguous. Sixteen of
the 20 pools in our sample included grant-back rules (Table 1).
Difference-in-differences estimates for the interaction pool * # pool
patents * no grant-backs indicate that pools without grant-backs
produced 0.43 additional patents per year compared with pools with
grant-backs (significant at 1%, Table 6, column 1). Compared with an
estimate of 0.39 for pool * # pool patents (significant at 1%, Table 6,
column 1), this implies that pool technologies in pools without grant-
backs experienced no statistically significant decline in patenting (with a
p-value of 0.747).35
Although our data do not include detailed data on pool characteristics,
12 of the 20 pools in our sample are also included in Lerner et al. (2007);

two of these pools, for fuse cutouts and furniture slip covers, allowed
independent licensing. Estimates of the interaction pool * # pool patents
* independent licensing are positive, but not statistically significant
(Table 6, column 2). These results are confirmed in archival sources,
which suggest that pools that did not allow independent licensing, may
have used licensing as a tool to weaken product market competition. For
example, the Phillips pool did not allow independent licensing, and instead
used licensing as a mechanism to weaken the development and manufac-
ture of competing cross-recessed head screws. According to a government
complaint, in 1945, it persuaded the National Lock Company, which had
begun to produce its own similar design of cross-recessed screws, to license
the Phillips technology, abide by the pool’s pricing scheme, and refrain
from producing and selling non-Phillips cross-recessed screws. By 1944,
the pool and its licensees sold 75% of all cross-recessed head screws in the
United States.36
An additional test considers a more narrow definition of a patent pool
as a pool that licenses its patents to third parties or transfers all pool
patents to a holding company that makes the patents available for licen-
sing. Vaughan (1956: 41), Carlson (1999: 369), and Shapiro (2001: 127)
refer to pools that only provide for cross-licensing as cross-licensing agree-
ments; Vaughan’s original definition includes these cross-licensing agree-
ments as a subgroup of patent pools, while Shapiro (2001) separates
patent pools and cross-licensing agreements.
Among 20 pools that formed in the 1930s, 7 pools established a patent
holding company, 9 pools licensed to third parties; 4 pools established a
patent holding company and licensed third parties.37 Another 8 pools were
cross-licensing agreements (Vaughan 1956: 41, Shapiro 2001: 127) be-
tween two or more firms that did not license to other firms.
35. Estimates for a restricted sample of patents by pool members indicate that members of
pools without grant-back rules produced 0.08 additional patents per year compared with
members of pools with grant backs (significant at 1%, not reported). Compared with an
estimate of pool * # pool patents of 0.02 (significant at 1%, not reported), this implies
0.06 additional patents per year.
36. United States v. Phillips Screw Co., CCH 1949 Trade Cases ô 62,394 (N.D. Ill. Civil
No. 47-C-147; Complaint, 1947; Consent Judgment, 1949).
37. See the Supplementary Appendix for details on each pool’s licensing practices.
Table 6. Pool Characteristics, OLS—Dependent Variable is Patents per Subclass and

Year
(1) (2) (3) (4)
Pool * # pool patents 0.389** 0.454** 0.464** 0.251

(0.100) (0.147) (0.119) (0.138)
Pool * # pool patents * no grant-backs 0.427**
(0.148)

Pool * # pool patents * independent licensing 0.447
(0.358)
Pool * # pool patents * cross-license 0.207
(0.182)
Pool * # pool patents * prior litigation 0.248
(0.187)
Constant 1.975** 1.757** 1.975** 1.975**
(0.072) (0.128) (0.073) (0.072)
Subclass fixed effects Yes Yes Yes Yes
Year-fixed effects Yes Yes Yes Yes
** p < 0.01, * p < 0.05
n (# subclasses * 28 years) 35,308 10,388 35,308 35,308
R2 0.554 0.537 0.554 0.554
Notes: The dependent variable counts patents per subclass and year. The timing of invention is measured by the
application year for granted patents. The variable poolct equals 1 for years after the pool forms. # pool patents
counts pool patents that list subclass c as their primary subclass. In all, 433 pool subclasses include one or more
pool patents; 828 cross-reference subclasses form the control. The variable no grant-backsc equals 1 for four pools
that did not include grant-back provisions; independent licensingc equals 1 for two pools that allowed members to
grant licenses independently of the pool; cross-licensec equals 1 for eight pools that involve two-firm cross-licensing
agreements; prior litigationc equals 1 for five pools in which court records indicate that members were engaged in
patent litigation prior to the formation of a pool.
Estimates with a narrower definition of a pool imply a larger decline in

patenting (0.46, significant at 1%, Table 6, column 3), compared with
baseline estimates that use the original definition in Vaughan (1956, with
an estimate of 0.36, Table 3, column 1). This difference suggests that
the observed decline in patenting was driven primarily by pools that
satisfy the narrower definition. Estimates for pool * # pool patents * cross-
licensing are positive but small and not statistically significant (0.21, Table 6,
column 3).
Predictions for a fourth characteristic of patent pools, the existence of
prior litigation, are also theoretically ambiguous. Pools, which resolve
litigation as a result of overlapping patents (e.g. Shapiro 2001), may en-
courage innovation if they reduce the risk of future litigation. The exist-
ence of prior litigation may, however, also be an indicator of severe patent
thickets, which may not be resolved completely by the pool, or it may
indicate pools that combine substitute patents, which are predicted to
discourage innovation (Lerner and Tirole 2004; Lerner et al. 2007). Five
of 20 pools in our sample formed to resolve litigation. Estimates for pool *
# pool patents * prior litigation are negative, at 0.25, but not statistically
significant (Table 6, column 4).
6. Welfare Implications
In this section we discuss the welfare implications of the creation of a
patent pool. We summarize alternative hypotheses, under which a decline
in innovation may be associated with an increase (or a reduction) in wel-
fare and discuss evidence in favor of against these hypotheses.
6.1 Elimination of Duplicative Research Efforts

Although patent data indicate an unambiguous decline in patenting after

the creation of a pool, the welfare implications of this decline may have
been positive, if the creation of a pool eliminated duplicative research that
was socially wasteful (e.g. Loury 1979). Alternatively, the creation of a
pool may have triggered an increase in duplicative research if it encour-
aged firms to race to create patents that could be included in the pool
(Baron and Pohlmann 2011; Dequiedt and Versaevel 2013).
Changes in patenting are broadly consistent with the idea of a patent
race leading up to a pool, but these effects are small and cannot explain the
observed decline in innovation after the creation of a pool. Annual coef-
ficients for the pre-period are positive and statistically significant in year
t 1, when estimates imply a 9.12% increase in patenting (Figure 3).
Excluding data for t 1, however, leaves the size of the estimated decline
at 0.34 (significant at the 1% level, not reported), which is only slightly
smaller than the baseline estimate of 0.36 (Table 3, column 1).
Compared with a mean of 2.41 patents, this implies a 13.95% decline in
patenting. Excluding data for both t 1 and t 2 leaves the size of the
estimated decline nearly unchanged at 0.33 (significant at the 1% level,
not reported).
6.2 Strategic Patenting and Litigation

A reduction in patenting may also be welfare-increasing if the creation of a
pool reduced the need for strategic patenting by lowering litigation risks
for member firms. Surveys of research labs indicate that many firms value
patents as a means to defend themselves from litigation (Levin et al. 1987;
Cohen et al. 2000). In a sample of 95 publicly traded semiconductor firms,
firms with large capital investments increased their propensity to patent
between 1979 and 1995 as a strategic response to the threat of litigation
and hold-up (Hall and Ziedonis 2001). If the creation of a pool reduces the
threat of litigation for both pool members and outside firms, it may induce
a welfare-improving reduction in raw patent counts along with a shift
toward higher quality patents.
Consistent with a moderate increase in patent quality, analyses of cit-
ation-weighted patents indicate a smaller decline in patenting after the
creation of a pool. Even controlling for quality, however, the decline in
patenting remains economically large; subclasses with an additional pool
patent produce 11.33% fewer citation-weighted patents (Table 4, column
1), which implies that the creation of a pool also discouraged the creation
of high-quality patents.
For 19th-century sewing machines, data on patents and improvements

in technical performance indicate that innovation declined in response to
the creation of a pool, which increased litigation risks for outside firms
even as it reduced such risks for members (Lampe and Moser 2010). Pool
members may be particularly aggressive in defending their patents if their
pool combines weak and potentially invalid patents for substitute tech-
nologies. If a substitute patent is found invalid, competitors can produce
the pool technology, which undermines the pool’s market power. If one of

a set of complementary pool patents is found invalid, the pool can con-
tinue to prevent competitors based on the remaining patents (Gilbert
2004; Choi 2010). This suggests that litigation risks may be particularly
high if pools combine patents for substitutes.
6.3 License Fees

Finally, pools that reduce innovation may be welfare-enhancing if the
pool offers its technology at a royalty rate that is low enough to make
outside firms better off if they adopt the pool technology, rather than
invest in R&D to improve a competing design. This argument is particu-
larly strong for pools of complementary patents, which may reduce roy-
alty rates and increase profits by eliminating double marginalization.
The experience of the sewing machine pool indicates that the creation of a
pool that combines complementary patents may reduce welfare even if it
leads to a substantial reduction in license fees. When the sewing machine
pool formed in 1856, it reduced license fees from $25 per machine for all
firms (compared with a typical price around $75) to $15 for outside firms
and $5 for member firms. This fee structure strongly favored pool members
and put outside firms at a significant competitive disadvantage. Data on
patents and firm entry indicate that outside firms shifted their R&D efforts
away from the pool technology toward improving a technologically inferior
substitute for the pool technology. Entrants with this technology faced
significantly higher failure rates (Lampe and Moser 2013).
While there is no systematic data on license fees for the 1930s, the
available evidence is not indicative of welfare-enhancing licensing
schemes. After forming a pool in 1934, three producers of variable con-
densers settled an infringement suit with one of their competitors, Federal
Instrument Company, by licensing their technology at a sufficiently high
rate that Federal became insolvent in 1938.38 Six pools licensed their pa-
tents to nine or more outside firms. The licensing strategies of these pools
are suggestive of a “divide and conquer” approach by which pools capture
licensees to close off the market to innovating rivals (Rasmussen et al.
1991; Segal and Whinston 2000). For example, the Phillips screws pool
required licensees to refrain from producing and selling non-Phillips cross-
recessed screws. Another pool for wrinkle paint finishes set a very low
royalty of 5 cents per gallon relative to a sales price of $2.85 per gallon,
38. United States v. General Instrument Corp., 115 F. Supp. 582 (D.N.J. 1953).
licensed to 185 firms, and produced almost all the wrinkle finishes pro-
duced in the United States. This case is consistent with Gallini (1984),
which suggests that pools can reduce welfare if pool members use royalties
to discourage rivals from developing potentially superior technologies.
6.4 Delays in the Adoption of New Technologies

To fully assess the welfare implications of a patent pool, we would ideally

examine changes in prices and sales for products that were differentially
affected by the creation of a pool. Although such data are not available for
the industries in our sample, we can examine data on the adoption of a new
technology for the movie industry. In this example, the creation of a pool
eliminated competition to develop a simpler, cost-effective process to
shoot movies in colors, which would have replaced a pool member’s exist-
ing technology.
In the early 1930s, Technicolor was the market leader in professional
color cinematography with its cumbersome and expensive but high-
performing three-strip method for filming movies in color. Technicolor’s
three-strip technology produced an exceptionally vivid color scheme by
running three separate strips of film, which covered different parts of the
color spectrum, through a specialized camera. Its technology put the com-
pany in complete control of the market and enabled Technicolor to charge
high rental fees for cameras and film. In 1935, shooting a movie in color,
using Technicolor’s three-strip method, increased production costs by 30%
compared with an average of $300,000 per feature (Kindem 1979: 34).
At the same time, however, Technicolor and its competitor Eastman
Kodak began to pursue independent research to develop the “monopack”
technology, a simpler method for filming movies in color, which runs a
single strip of celluloid through a regular black-and-white camera.39
Substantially less cumbersome and less costly than the three-strip process,
the monopack threatened Technicolor’s monopoly in color movies.
On June 25, 1934, Technicolor and Kodak formed a pool to combine
their patents, and research on monopack slowed. In 1941, Technicolor
introduced a monopack film, but it posed no threat to the three-strip
method, because it was too grainy for studio work.40 The 1934 agreement
also included a covenant that Kodak “refrain from engaging in the com-
mercial processing of wide ‘monopack’ film.” Although Technicolor and
Kodak removed the covenant on December 14, 1945, a government com-
plaint in 1947 observed that Kodak
continued to refrain from the commercial processing of wide
“monopack” film, from licensing others to engage in such
39. United States v. Technicolor, Inc. (S.D. Calif. Civil No. 7507-M; Complaint, 1947).
40. Filmmakers used this technology exclusively for outdoor shots that they could not
reach with Technicolor’s bulky three-strip cameras (Haines 2003: 28; Basten 2005: 127).
processing, and, with minor exceptions, from selling such film

with the right to process to customers other than
Technicolor . . . the development of the art of professional
color cinematography by others than Technicolor has been
retarded, to the detriment of the general public, the motion
picture industry, and the film manufacturing industry.41
To examine whether the pool did in fact retard the “art of professional

color cinematography,” we collected data on the color scheme of 1911
feature-length movies between 1930 and 1960 from the catalogs of the
American Film Institute (www.afi.com). These data confirm that the cre-
ation of a pool did, in fact, delay technical change. While the pool was
active (1934–50) the majority of movies continued to be filmed in black-
and-white (see Figure A6 in the Supplementary Appendix).
On August 18, 1947, the DoJ filed a complaint against Technicolor and
Kodak. After this, Kodak invested more than $3 million in R&D to im-
prove color film in 1948 alone, compared with a total of $15 million be-
tween 1921 and 1948 (Frost and Oppenheim 1960: 124). On November 24,
1948, a consent decree made Kodak’s patents available for compulsory
licensing to outside firms. A second consent decree on February 28, 1950
required Technicolor to license its patents to outside firms. The decree also
terminated arrangements that forced producers to use Technicolor’s three-
strip cameras.42
In the same year, Kodak introduced Eastmancolor, a monopack film
that was fine-grained enough for commercial filmmakers.43 In 1952,
Kodak introduced an improved monopack technology with higher
emulsion speed and better grain structure that allowed the monopack
technology to meet the standards of major Hollywood studios.44
Buoyed by substantial costs savings, the number of US color movies
increased from 121 in 1952 to 174 in 1954. As a share of all US movies,
color films increased from 2.28% in the 1930s to 33.06% in 1952 and
57.62% in 1954. Universal Pictures’ “Foxfire,” released in 1955, was the
41. United States v. Technicolor, Inc. (Civil No. 7507-M, S.D. Calif.; Complaint, 1947,
paras. 18 and 27). See the Supplementary Appendix for estimates using data from the color
cinematography industry.
42. United States v. Technicolor, Inc., CCH 1950–51 Trade Cases ô2,586 (S.D. Calif. Civil
No. 7507-M; Complaint, 1947; Consent Judgments, 1948 and 1950).
43. Eastmancolor commercialized the principle of colored dye couplers (US patents
2,428,054 and 2,449,966; Frost and Oppenheim 1960: 115–6). Monopack film stock contained
silver halides that were exposed to light through thin layers of filters to generate black and
white latent images of red, green, and blue. The exposed film then went through three bleaches
that included tiny color globules (dye couplers), which replicated the latent silver image. After
the silver had washed away, a three color image remained, in which each hue was represented
by a thin layer of dye couplers (Haines 2003: 28).
44. Records of the Eastman Kodak Company accessed on May 25, 2012 at http://motion.
kodak.com/motion/Products/Chronology_Of_Film/1940–1959/index.htm.
last live-action American film that used the three-strip method (Basten
2005: 129).45
7. Conclusions
This article has taken advantage of a unique window of regulatory toler-
ance under the New Deal to examine pools that may form in the absence
of effective antitrust and investigate their effects on innovation.

Difference-in-differences analyses indicate that the creation of a pool led
to a 14% decline in patents for each additional patent that was included in
a pool. Estimates with a binary variable for pool technologies imply a 16%
decline in patenting after the creation of a pool. These results are robust to
flexible controls for unrelated changes in patenting over time (e.g. as a
technology matures), alternative definitions of the control, and a broad
range of additional specifications. Estimates with citation-weighted pa-
tents confirm a substantial decline in quality-adjusted patents, and suggest
a moderate decline in strategic patenting after the formation of the pool.
An analysis of alternative, nonpatent measures of innovation for the
movie industry shows that the creation of a pool reduced the rate of
technical progress in color film.
To investigate the role of competition in determining the effects of a
pool, we exploit the structure of the USPTO classification system, which
assigns inventions to USPTO subclasses according to the function that
they perform, so that inventions that are covered by patents in the same
USPTO subclass can act as substitutes by fulfilling the same function. For
technologies that are covered by these subclasses, the creation of a pool is
more likely to reduce competition to improve substitute technologies.
Analysis of historical patents and citations indicate that the observed de-
cline in patenting after the creation of a pool was driven by steep declines
for technologies in which the creation of a pool combined patents in the
same USPTO subclass by competing firms. These results are robust to
using subclasses with a single pool patent—in which the creation of a
pool did not affect the intensity of competition—as an alternative control
to compare changes in subclasses with multiple pool patents by a single
versus multiple firms.
Are these results generalizable to modern pools? Our historical analysis
suggests that pools, which combine patents in the same subclass, may dis-
courage innovation by reducing competition to improve the pool technol-
ogy. Although systematic data across industries are not available for
modern pools, anecdotal evidence from individual industries is suggestive.
For example, two producers of equipment for photorefractive keratec-
tomy (PRK) laser eye surgery, VISX and Summit, formed a pool in
45. An analysis of color usage by studios that were differentially affected by the 1948
Paramount decree, which forced studios to divest from movie theaters, reveals no direct link
between the decree and the use of color by the major studios (United States v. Paramount
Pictures, Inc. et al., 334 U.S. 131 (1948), Gil and Lampe 2014).
1992. In 1998, three years after developing guidelines for antitrust enforce-
ment of intellectual property licensing, the FTC challenged and dissolved
their agreement, arguing that, in the absence of a pool “VISX and Summit
could have and would have competed with one another in the sale or lease
of PRK equipment by using their respective patents, licensing them, or
both.” Twenty-one of 25 patents in this pool were assigned to the same
subclass.46
More generally, our findings are consistent with a positive link between

competition and innovation. Pools that have been allowed to form after
the DoJ and FTC developed their intellectual property guidelines have
combined standard-essential patents, for which no substitutes exist by
definition. The agencies’ policy to forbid the creation of pools that com-
bine substitute patents prevents a reduction in product market competi-
tion for technological standards. Our results, however, suggest that this
policy may also encourage competition to improve R&D and thereby
encourage innovation.
Funding
Moser thanks the Center for Advanced Studies in the Behavioral Sciences
(CASBS), the National Science Foundation (through CAREER Grant
1151180), the Stanford Institute for Policy Research (SIEPR) and the
Olin Center for Law and Economics at Stanford for financial support.
Lampe thanks CASBS, the Driehaus College of Business and the
Department of Economics at DePaul University for financial support.
Siyeona Chang, Josh Wagner, Aleksandra Goulioutina, and Marina
Kutyavina provided valuable research assistance.
Supplementary material
Supplementary material is available at Journal of Law, Economics, &
Organization online.
References
Aghion, Philippe, Christopher Harris, Peter Howitt, and John Vickers. 2001. “Competition,
Imitation and Growth with Step-by-step Innovation,” 68 Review of Economic Studies
467–92.
Aghion, Philippe, Nick Bloom, Richard Blundell, Rachel Griffith, and Peter Howitt. 2005.
“Competition and Innovation: An Inverted-U Relationship,” 120 Quarterly Journal of
Economics 701–28.
Aoki, Reiko, and Sadao Nagaoka. 2004. “The Consortium Standard and Patent Pools,” Hi-
Stat Discussion Paper No. 32, Hitotsubashi University.
Baron, Justus, and Tim Pohlmann. 2011. “Patent Pools and Patent Inflation,” Cerna
Working Paper Series 2010:13, MINES ParisTech.
46. Subclass 606/5 for Surgery-Instrument-Light application-Ophthalmic-Recurving or

reshaping of the eye.
Basten, Fred E. 2005. Glorious Technicolor: The Movies’ Magic Rainbow. Camarillo, CA:
Technicolor.
Benner, Mary, and Joel Waldfogel. 2008. “Close to you? Bias and Precision in Patent-based
Measures of Technological Proximity,” 37 Research Policy 1556–67.
Bertanha, Marinho, and Petra Moser. 2014. “Spatial Errors in Count Data Regressions,”
Journal of Econometric Methods, forthcoming.
Brenner, Steffen. 2009. “Optimal Formation Rules for Patent Pools,” 40 Economic Theory
373–88.
Carlson, Steven C. 1999. “Patent Pools and the Antitrust Dilemma,” 16 Yale Journal on

Regulation 359–99.
Carlton, Dennis, and Allan L. Shampine. 2013. “An Economic Interpretation of FRAND,”
9 Journal of Competition Law & Economics 531–52.
Ceccagnoli, Marco, Christopher Forman, and Wen Wen. 2012. “Patent Pools, Thickets, and
the Open Source Software Entry by Start-up Firms,” Working Paper, Georgia Institute of
Technology.
Choi, Jay Pil. 2010. “Patent Pools and Cross-Licensing in the Shadow of Patent Litigation,”
51 International Economic Review 441–60.
Cohen, Wesley M., Richard R. Nelson, and John P. Walsh. 2000. “Protecting Their
Intellectual Assets: Appropriability Conditions and Why U.S. Manufacturing Firms
Patent (or Not),” NBER Working Paper No. w7552.
Denicolò, Vincenzo. 2002. “Sequential Innovation and the Patent-Antitrust Conflict,” 54
Oxford Economic Papers 649–68.
Department of Justice and Federal Trade Commission. 2007. Antitrust Enforcement and
Intellectual Property Rights: Promoting Innovation and Competition. Washington, DC:
GPO. http://www.justice.gov/atr/public/hearings/ip/222655.htm (accessed February 8, 2013).
Dequiedt, Vianney, and Bruno Versaevel. 2013. “Patent Pools and Dynamic R&D
Incentives,” 36 International Review of Law and Economics 59–69.
Flamm, Kenneth. 2013. “A Tale of Two Standards: Patent Pools and Innovation in the
Optical Disk Drive Industry,” NBER Working Paper No. w18931.
Frost, George E., and S. Chesterfield Oppenheim. 1960. “A Study of the Professional Color
Motion Picture Antitrust Decrees and Their Effects,” 4 Patent, Trademark, and Copyright
Journal of Research and Education 108–49.
Fudenberg, Drew, Richard Gilbert, Joseph Stiglitz, and Jean Tirole. 1983. “Preemption,
Leapfrogging and Competition in Patent Races,” 22 European Economic Review 3–31.
Galasso, Alberto, and Timothy S. Simcoe. 2011. “CEO Overconfidence and Innovation,” 57
Management Science 1469–84.
Gallini, Nancy T. 1984. “Deterrence by Market Sharing: A Strategic Incentive for
Licensing,” 74 American Economic Review 931–41.
———. 2011. “Private Agreements for Coordinating Patent Rights. The Case of Patent
Pools,” 38 Journal of Industrial and Business Economics 5–29.
Gans, Joshua S., David H. Hsu, and Scott Stern. 2008. “The Impact of Uncertain Intellectual
Property Rights on the Market for Ideas: Evidence from Patent Grant Delays,” 54
Management Science 982–97.
Gil, Ricard, and Ryan Lampe. 2014. “The Adoption of New Technologies: Understanding
Hollywood’s (Slow and Uneven) Conversion to Color, 1940–70,” 74 The Journal of
Economic History 987–1014.
Gilbert, Richard J. 2004. “Antitrust for Patent Pools: A Century of Policy Evaluation,” 3
Stanford Technology Law Journal. http://stlr.stanford.edu/STLR/Articles/04_STLR_3
(accessed February 8, 2013).
———. 2006. “Competition and Innovation,” 1 Journal of Industrial Organization
Education, Article 8. Available at: http://www.bepress.com/jioe/vol1/iss1/8.
Gilbert, Richard J., and David M. G. Newberry. 1982. “Preemptive Patenting and the
Persistence of Monopoly,” 72 American Economic Review 514–26.
Haines, Richard W. 2003. Technicolor Movies: The History of Dye Transfer Printing.
Reprint. Jefferson, NC: McFarland & Company.
Haley, John O. 2001. Antitrust in Germany and Japan: The First Fifty Years, 1947–1998.
Seattle, WA: University of Washington Press.
Hall, Bronwyn H., Adam B. Jaffe, and Manuel Trajtenberg. 2001. “The NBER Patent
Citation Data File: Lessons, Insights and Methodological Tools,” NBER Working
Paper No. w8498.
———. 2005. “Market Value and Patent Citations,” 36 RAND Journal of Economics 16–38.
Hall, Bronwyn H., and Rosemarie H. Ziedonis. 2001. “The Patent Paradox Revisited: An
Empirical Study of Patenting in the U.S. Semiconductor Industry, 1979–1995,” 32 RAND
Journal of Economics 101–28.

Holmström, Bengt, and John Roberts. 1998. “The Boundaries of the Firm Revisited,” 12
Journal of Economic Perspectives 73–94.
Imbens, Guido, and Jeffrey M. Wooldridge. 2007. “What’s New in Econometrics?,”
Prepared for the National Bureau of Economic Research Summer Institute, Cambridge
MA: July/August 2007.
Jeitschko, Thomas D., and Nanyun Zhang. 2012. “Adverse Effects of Patent Pooling on
Product Development and Commercialization,” Working Paper EAG 12–5, Department
of Justice.
Joshi, Amol M., and Atul Nerkar. 2011. “When Do Strategic Alliances Inhibit Innovation by
Firms? Evidence from Patent Pools in the Global Disc Industry,” 32 Strategic
Management Journal 1139–60.
Kim, Sung-Hwan. 2004. “Vertical Structure and Patent Pools,” 25 Review of Industrial
Organization 231–50.
Kindem, Gorham A. 1979. “Hollywood’s Conversion to Color: The Technological,
Economics and Aesthetic Factors,” 31 Journal of the University Film Association
29–36.
Klein, Benjamin, Robert G. Crawford, and Armen Alchian. 1978. “Vertical Integration,
Appropriable Rents, and the Competitive Contracting Process,” 21 Journal of Law and
Economics 297–326.
Klepper, Steven. 1996. “Entry, Exit, Growth, and Innovation over the Product Life Cycle,”
86 American Economic Review 562–83.
Kogan, Leonid, Dimitris Papanikolaou, Amit Seru, and Noah Stoffman. 2012.
“Technological Innovation, Resource Allocation, and Growth,” NBER Working Paper
No. w17769.
Lampe, Ryan. 2012. “Strategic Citation,” 94 Review of Economics and Statistics 320–33.
Lampe, Ryan, and Petra Moser. 2010. “Do Patent Pools Encourage Innovation?
Evidence from the 19th-Century Sewing Machine Industry,” 70 Journal of Economic
History 871–97.
———. 2012. “Patent Pools: Licensing Strategies in the Absence of Regulation,” 29
Advances in Strategic Management 69–86.
———. 2013. “Patent Pools and Innovation in Substitutes – Evidence from the 19th-century
Sewing Machine Industry,” 44 RAND Journal of Economics 757–78.
Lee, Tom, and Louis L. Wilde. 1980. “Market Structure and Innovation: A Reformulation,”
94 Quarterly Journal of Economics 429–36.
Lerner, Josh, Marcin Strojwas, and Jean Tirole. 2007. “The Design of Patent Pools: The
Determinants of Licensing Rules,” 38 RAND Journal of Economics 610–25.
Lerner, Josh, and Jean Tirole. 2004. “Efficient Patent Pools,” 94 American Economic Review
691–711.
———. 2013. “Standard-Essential Patents,” NBER Working Paper No. w19664.
Levin, Richard C., Alvin K. Klevorick, Richard R. Nelson, and Sidney G. Winter. 1987.
“Appropriating the Returns from Industrial Research and Development,” 3 Brookings
Papers on Economic Activity 783–831.
Llanes, Gastón, and Stefano Trento. 2012. “Patent Policy, Patent Pools, and the
Accumulation of Claims in Sequential Innovation,” 50 Economic Theory 703–25.
Loury, Glenn C. 1979. “Market Structure and Innovation,” 93 Quarterly Journal of
Economics 395–410.
Mehta, Aditi, Marc Rysman, and Tim Simcoe. 2010. “Identifying the Age Profile of Patent
Citations: New Estimates of Knowledge Diffusion,” 25 Journal of Applied Econometrics
1179–204.
Merges, Robert. 2001. “Institutions for Intellectual Property Transactions: The Case of
Patent Pools,” in R. Dreyfuss, D. Zimmerman, and H. First, eds., Expanding the
Boundaries of Intellectual Property. Oxford, UK: Oxford University Press.
Moser, Petra. 2005. “How Do Patent Laws Influence Innovation? Evidence from
Nineteenth-Century World Fairs,” 95 American Economic Review 1214–36.
———. 2012. “Innovation without Patents – Evidence from World’s Fairs,” 55 Journal of

Law & Economics 43–74.
Moser, Petra, Joerg Ohmstedt, and Paul W. Rhode. 2012. “Patents Citations and the Size of
the Inventive Step - Evidence from Hybrid Corn,” SSRN Working Paper No. 1888191.
Popp, David, Ted Juhl and Daniel K. N. Johnson. 2004. “Time in Purgatory: Examining the
Grant Lag for U.S. Patent Applications,” 4 Topics in Economic Analysis & Policy. doi:
10.2202/1538-0653.1329 (accessed February 8, 2013).
Posner, Richard A. 1970. “A Statistical Study of Antitrust Enforcement,” 13 Journal of Law
and Economics 365–419.
Quint, Daniel. 2014. “Pooling with Essential and Nonessential Patents,” 6 American
Economic Journal: Microeconomics 23–57.
Rasmussen, Eric B., Mark Ramseyer, and John S. Wiley, Jr. 1991. “Naked Exclusion,” 81
American Economic Review 1137–45.
Reinganum, Jennifer F. 1982. “A Dynamic Game of R and D: Patent Protection and
Competitive Behavior,” 50 Econometrica 671–88.
Rosenberg, Nathan. 1982. Inside the Black Box: Technology and Economics. New York, NY:
Cambridge University Press.
Rysman, Marc, and Tim Simcoe. 2008. “Patents and the Performance of Voluntary Standard
Setting Organizations,” 54 Management Science 1920–34.
Schumpeter, Joseph A. 1942. Capitalism, Socialism, and Democracy. London: Routledge.
Segal, Ilya R., and Michael D. Whinston. 2000. “Naked Exclusion: Comment,” 90 American
Economic Review 296–309.
Shapiro, Carl. 2001. “Navigating the Patent Thicket: Cross Licenses, Patent Pools, and
Standard Setting,” in A. B. Jaffe , J. Lerner, and S. Stern, eds., Vol. 1 of Innovation
Policy and the Economy. Cambridge, MA: MIT Press.
Simcoe, Timothy. 2012. “Standard Setting Committees: Consensus Governance for Shared
Technology Platforms,” 102 American Economic Review 305–36.
Trajtenberg, Manuel. 1990. “A Penny for Your Quotes: Patent Citations and the Value of
Innovations,” 21 RAND Journal of Economics 172–87.
United States Patent Office (USPTO). 1915. The Classification of Patents. Washington, DC:
GPO.
Vakili, Keyvan. 2012. “Competitive Effects of Collaborative Arrangements: Evidence from
the Effect of the MPEG-2 Pool on Outsiders’ Innovative Performance,” Working Paper,
University of Toronto.
Vaughan, Floyd L. 1956. The United States Patent System: Legal and Economic Conflicts in
American Patent History. Norman, OK: University of Oklahoma Press.
Wallis, John J. 1984. “The Birth of the Old Federalism: Financing the New Deal,” 44 Journal
of Economic History 139–59.
Wooldridge, Jeffrey M. 1999. “Distribution-free Estimation of Some Nonlinear Panel Data
Models,” 90 Journal of Econometrics 77–97.
Wright, Gavin. 1974. “The Political Economy of New Deal Spending: An Econometric
Analysis,” 56 Review of Economics and Statistics 30–38.

Patent Pool

Transféré par

Informations du document

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Patent Pool

Transféré par

Droits d'auteur :

Formats disponibles

JLEO, V32 N1 1

Patent Pools, Competition, and Innovation—Evidence

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

*California State University, East Bay. Email: ryan.lampe@csueastbay.edu

The Journal of Law, Economics, and Organization, Vol. 32, No. 1

cross-licensing among their members.1 Pools have emerged as a prominent

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

1. Vaughan (1956: 41) includes cross-licensing agreements as a subgroup of patent pools,

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

country has perhaps never witnessed a more completely successful eco-

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

account for the majority of innovations, it is particularly important to

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

2.1 Pools May Encourage Innovation by Increasing the Returns to R&D

is particularly important for “standards” pools, which combine (essential)

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

2.2 Pools May Discourage Innovation by Reducing Competition

models imply a decline in innovation, if the creation of a pool reduces

2.3 Grant-Back Rules and Licensing Provisions

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

3.1 Pool Patents in 20 Industries, 1930–8

that became subject to antitrust investigations, which creates a traceable

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

Table 1. 20 Patent Pools Formed between 1930 and 1938

Industry Year Members Pool Grant- Independent Prior

High tension cables 1930–48 2 73 1 – 0

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

3.2 Patents per Year in Pool and Cross-Reference Subclasses

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

Patent holding company:

surface”) and 428/499 (“Composite (nonstructural laminate)—Of natural

patents for substitute technologies: If patents are assigned to primary

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

Pre-pool formation Post-pool formation All years Pre-pool–post-pool

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

inventions assigned to different subclasses.20 For example, two producers

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

3.3 Citations by Patents after 1921 as a Control for Patent Quality

20. USPTO Examiner’s Handbook, Chapter 1, available at http://www.uspto.gov/pa-

search yields a total of 238,874 citations between January 4, 1921 and

3.4 Nonpatent Measures of Innovation

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

4.1 Baseline Estimates

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

variation in patenting across technologies that is constant over time.

OLS OLS QMLE OLS OLS OLS: Excl.

Pool * # pool patents 0.355** 0.317** 0.385** 0.337**

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

likelihood ﬁxed-effects Poisson estimates (QMLE Poisson), which address

has matured, and patenting in mature industries may decline independently

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

4.4 Controlling for Patent Quality through Citations

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

4.5 Additional Robustness Checks

Table 4. OLS and QMLE Poisson—Dependent Variable is Citation-weighted Patents

OLS OLS QMLE OLS OLS OLS: Excl.

Pool * # pool patents 1.415** 1.125** 1.030** 1.417**

Downloaded from https://academic.oup.com/jleo/article-abstract/32/1/1/2579509 by guest on 10 March 2019

group that restricts the control group to 631 cross-reference classes in

5. Changes in the Intensity of Competition

competition in product markets and R&D (e.g. Department of Justice

Pool * # pool patents 0.355 0.317 0.385 0.337

Pool * # pool patents 1.415 1.125 1.030 1.417

Pool * # pool patents 0.389 0.454 0.464** 0.251