Bundling of RAND-committed Patents
Anne Layne-Farrar
Charles River Associates, One South Wacker Drive, Chicago, IL 60606
alayne-farrar@crai.com
Michael A. Salinger (corresponding author)
Boston University, Questrom School of Business, 595 Commonwealth Ave., Boston, MA 02215
salinger@bu.edu
November 2015
Abstract: We extend a simplified version of the Gilbert-Katz [Richard Gilbert and Michael Katz (2006)]
(GK) model of patent bundling to incorporate RAND commitments and then use the model to consider
whether a patent holder violates a RAND commitment if it ties a license to its RAND-encumbered
patents to licenses for patents on which it has not made a RAND commitment. In the GK model, the
ability to engage in patent tying makes a patent-owner willing to engage in long term contracting that
prevents it from charging “hold-up” royalties. But a RAND commitment accomplishes the same
objective; and the tying of licenses to patents without RAND obligations to RAND-encumbered patents
creates a risk of reneging on the RAND commitment. Mixed bundling, where the licensor offers licensees
the option of taking a license to RAND-committed patents only or taking a license to the full portfolio
honors the patent-holder’s RAND commitment provided that the royalty for the RAND-encumbered
patents is RAND (regardless of the royalty for the larger portfolio of patent rights). (Pure) p atent
bundling/tying is, however, a common practice that often has sound efficiency justifications. A patentholder can engage in pure bundling/tying of licenses to RAND-encumbered and non-RAND encumbered
patents and still honor its RAND commitments provided that it charges a royalty that would be RAND for
the RAND-encumbered patents alone. The patent owner cannot deduct the value of non-RAND
committed patents from the license fee for the bundle and argue that it has honored its RAND
commitment as long as the difference is RAND for the RAND-committed patents.
Key words: patents, licensing, bundling, tying, RAND, FRAND, intellectual property
JEL classifications: O3, K21
I. Introduction
In this paper, we address the inter-relationship between two common practices in the licensing
of intellectual property: RAND commitments and patent bundling. Patent bundling (or, more precisely,
pure patent bundling) is the practice of licensing patents only in bundles rather than offering licenses to
individual patents on an à la carte basis. A RAND commitment is a commitment to license technology on
“Reasonable and Non-Discriminatory” terms. 1 The question we address is whether, once a patent owner
makes a general commitment to license its patents to all licensees on RAND terms, can it offer those
patents solely in a bundle with other patents? If it can, what are the implications of the RAND
commitment for the royalties it can charge for the bundle? Alternatively, does a RAND commitment on a
patent necessarily entail an obligation to offer it on an unbundled basis?
To address the general issue of bundling RAND-committed patents, one must first understand
why patent bundling is such a common phenomenon, what effect patent bundling has on licensing
terms, and why patent holders make RAND commitments. While there is a substantial and growing
economics literature on bundling in general and some of it specifically focuses on the bundling of
intellectual property (like computer software, music, and video entertainment), the formal literature on
patent bundling is remarkably thin.2 A notable exception is Gilbert and Katz (2006) (henceforth “GK”),
who discuss why the intrinsic features of patents differ from the assumptions of the existing economics
literature on bundling. They present a formal model of bundling based on assumptions designed to
capture some key features of patent licensing. As we explain in more detail below, their results seem to
suggest that patent bundling does not pose a public policy concern and that the case for forcing
companies to license patents on an à la carte basis is weak. Gilbert and Katz did not, however,
incorporate RAND commitments into their model. To assess whether tying RAND-committed and nonRAND-committed patents violates the RAND commit, we incorporate a RAND commitment into a
simplified version of the GK model.
Many standards development organizations (SDOs) ask their members to first disclose any
patents that might be considered essential for compliance with a standard under development and
second to then promise to license any such essential patents on RAND terms and conditions to anyone
requesting a license. Thus, patent holders cannot refuse to license RAND-committed patents, nor can
they license them solely on an exclusive basis. In addition, the “reasonable” requirement places a cap
on the royalty. We discuss the precise nature of the cap in more detail below, but it generally prevents
the patent owner from setting opportunistic royalties that expropriate the value of a licensee’s sunk
investments.
1
As we understand the nomenclature, RAND is synonymous with “FRAND,” which is an acronym for “Fair,
Reasonable, And Non-Discriminatory.” Licensing terms that are both “reasonable” and “non-discriminatory” are
necessarily “fair.”
2 Patent bundling is related to but different from patent pooling, which Shapiro (2001) and Lerner and
Tirole (2004) have analyzed. Patent pooling entails licensing patents of different patent owners in a single package.
Patent bundling refers to licensing multiple patents of a si ngle patent-owner as a package.
2
Patent pledges are not limited to patents essential for cooperative SDO standards in large part
because SDOs are not the only way that standards emerge. In game theoretic terminology, standard
setting is an example of a “coordination game.” Multiple Nash equilibria exist (ex ante) for most industry
technology paths, and there are both private and social benefits from having the industry agree on one.
In this regard, the term “equilibrium” is misleading, as there is no reason to believe that the market will
naturally arrive at one of the so-called equilibria if all firms independently select which of the
alternatives to conform to. 3 SDOs are a forum for explicit coordination, but standards can emerge
organically in the market as well. And coordination can sometimes occur without any explicit
coordination – either in SDOs or via organic evolution in the marketplace. A company might propose a
standard based on its proprietary technology hoping that the rest of the industry will coalesce around its
proposal. A well-known recent example is the standards competition with respect to high definition DVD
technology, with Sony pressing its Blu-Ray laser technology and Toshiba promoting its HD-DVD
technology. A RAND (or other patent) commitment in such settings can have the same economic effect
as in an SDO setting. The patent owner makes the commitment in order to increase the likelihood that
the industry will coalesce around the technology rendering it an industry standard, and other industry
participants then rely on the patent owner’s commitment not to behave opportunistically before
investing sunk costs to develop products based on the standard. As a result, a commitment not to
engage in patent hold-up can be just as necessary for securing investment that results in licensed use of
the patent outside of cooperative standards as it is within SDOs for declared SEPs. Hence, firms often
voluntarily commit publicly to license certain of their patents that are not related to a standard
established by an SDO on RAND terms and conditions. 4
The remainder of this paper is organized as follows. Section II discusses the economics of
bundling in general. Section III then turns to patent bundling with a focus on the GK model. In Section IV,
we present a model of the licensing of a single patent in a setting that otherwise reflects a simplified
version of the GK assumptions. We interpret the RAND commitment as limiting the royalty to what the
licensee would reasonably agree to before it makes any sunk investments. We model a RAND rate as a
“No Opportunism Game” in which the patent owner sets a license fee before the licensee makes any
investments. The No Opportunism Game brings out a key distinction between what we term “value based royalties” and “cost-based royalties.” The former captures the value created by the technology
assuming that inventing around the technology is not feasible. A cost-based royalty is one that is low
enough to eliminate a patent-user’s incentive to try to invent around the patent. As we show, the
conditions when a patent owner would choose a value-based as opposed to a cost-based royalty are
more complicated than one might initially expect. In Section V, we next extend that model to
incorporate the licensing of two patents where the patent holder has made a RAND commitment on just
3
A notable example in which the industry did not coordinate around a standard was the early generations
of cellular telephones in the United States.
4 The Program for Information Justice and Intellectual Property, at Washington College of Law, maintains
a database of more than 150 public such non-SDO patent pledges: http://www.pijip.org/non-sdo-patentcommitments/. Contreras (2014) and Elhauge (2015) discuss the legal basis for enforcing RAND commitments
made outside an SDO setting, see. For other discussions of non-SDO patent pledges, see Layne-Farrar (2014) and
Harkrider (2014).
3
one. We do so with a “Some Opportunism Game” in which the patent owner commits to a royalty for
one of its patents but not for the other. Section VI discusses policy implication s.
As we explain below, our findings do not necessarily preclude bundling. Due to the efficiency
gains so common with portfolio licensing, we would expect to find lots of bundled licenses even when all
licensors offer the option of RAND-patent only licenses. Second, if the total royalty for the bundle would
be individually reasonable and non-discriminatory for each applicable RAND commitment and each
standard covered by such commitments, and the other terms of the license are similarly reasonable and
non-discriminatory, then bundling does not violate RAND or impose anticompetitive harm on licensees.
In other words, if the patent holder is willing to “throw in” the other patents – either non-RAND
committed patents or patents that are committed for separate and distinct standards – for free, such
that the licensing terms for the bundle would be RAND with or without the inclusion of the additional
patents, the mere inclusion of additional licenses would not necessarily harm consumers. If that
admittedly strict condition is not met, however, pure bundling (i.e., the absence of mixed bundling) is
the equivalent of supra-RAND pricing and is likely to harm consumers. 5
The above proposal has important policy implications. First, one cannot presume
anticompetitive harm simply from the fact that a licensor has tied its patents. An inquiry is required in
that case to determine whether or not the licensing terms meet the conditions spelled out above: is the
rate RAND for just the RAND-committed patents? Second, in disputes over RAND licensing, the patent
holder cannot deduct from the licensing terms the added value of the non-RAND-committed licenses to
get an implicit license fee for the RAND-committed licenses and then assert that this implicit license fee
meets the RAND obligation. Nor can it require licensees to license unrelated patents at a rate that
aggregates the RAND royalties for all of them. Symmetrically, licensees cannot demand a discount for
RAND-only licenses, as compared to RAND-plus licenses, without first establishing that the broader
portfolio license rates and terms are higher than the RAND-only rates and terms. The complexities of
establishing that a RAND-plus license has RAND terms and conditions may push licensors, as a practical
matter, more toward mixed bundling, but the door to RAND-committed patent tying should nonetheless
remain open.
II. Bundling and Tying in General
While a complete review of the extensive economics literature on bundling and tying is
unnecessary for what follows, a few key points are essential.
5
Of course, if a firm offers a RAND-committed patent separately for a royalty that honors its RAND
commitment, nothing precludes if from also offering a license for a bundle of patents that includes other patents
without RAND commitments. With such a mixed-bundling strategy, the difference between the price of such a
bundle and the royalty for licensing just the RAND committed patents may implicitly reflect a price for the patents
that are not RAND-encumbered, though the complementary nature of most portfolio patents leads to nonlinear
pricing that can break that simple implication. Since the RAND commitment does not apply to the additional
patents, it does not constrain the price of the bundle (even though the bundle includes RAND -committed patents).
4
The principle underlying many of GK’s results is the “single monopoly profit” 6 (or, more
generally, “single rent”7) theorem. As is well understood, if a company has a monopoly over two goods –
call them “A” and “B” – that consumers use together in fixed proportions, there is a combined price for
the two goods that maximizes the company’s profits. Any combination of prices that sums to this
optimal combined price or a bundled price equal to the combined price generates the same level of
profits (and consumer surplus). Indeed, the company does not need both A and B to earn this level of
profits as long as the price of the other good equals marginal cost. The company can capture the “single
rent” entirely through the price of A if the market for B becomes competitive and vice versa.
The single rent principle was one of the key underpinnings of the so-called Chicago critique of a
wide range of antitrust policies, including the antitrust treatment of tying. Probably the dominant
interpretation of the principle is that it casts doubt on the coherence of claims that the incentive to tie
goods together is “leveraging” or “foreclosure.” A possible application with respect to patents is that
when two patents are necessary to implement a technology, tying them together cannot create any
harm because the licensing fee the patent owner would charge for the bundle to maximize its profits
would equal the licensing fee it would charge for just one of the patents if it had only one.
An alternative interpretation, however, is that the assumptions underlying the single rent
principle are strong, so understanding the failure of the assumptions provides an organizing framework
for understanding why a single source of rents may not be sufficient to capture all the rents available.
Farrell and Weiser (2003) and Elhauge (2009) have articulated this perspective at length, as did
Krattenmaker and Salop (1986). Of the many strong assumptions underlying it, the single rent principle
presumes that the monopoly or property rights on at least one of the goods are iron-clad. The threat of
entry into both goods can violate this assumption. A threat of entry into just one of the goods does not.
Suppose entry into A is not possible but entry into B is. According to the si ngle rent principle, the seller
should welcome entry into B by a company offering a lower-cost, higher quality, or differentiated
version since the improvement in B would increase the profits it could earn from A. But, the threat of
entry in A as well can make it impossible for the firm to raise the price of A to take advantage of cost
reductions or improvements in B. With the potential for entry into both A and B, tying can raise entry
barriers by forcing two-stage entry. This exception to the single rent principle underlies the analyses of
Choi and Stefanides (2001) and Nalebuff (2004).
Failure of the single rent principle does not, however, necessarily imply that modularization of
competition results in better outcomes. Because of double marginalization, the profits a company can
earn by controlling the rights to both A and B exceed the total profits that separate owners of A and B
would earn. Again, as is well known, given relatively weak assumptions, merging the rights to A and B
6
As evidenced by the title to Elhauge (2009) the term is widely recognized. However, we have not been
able to document the source of the term. Whinston (1990) attributes the arguments to a Chicago oral tradition.
Bowman (1957) recognizes the strong assumptions underlying the principle and a set of exceptions to it when
those assumptions do not apply.
7
While economists (as well as lawyers and courts) use the phrase “single monopoly profit,” we prefer the
term “single rent” because the argument applies to rents of any kind, including patent royalties.
5
can induce a lower price. Eliminating double marginalization is the reason that patent pools can be in
the public interest. As Lerner and Tirole’s (2004) analysis shows, whether patents are complements or
substitutes is more subtle than is the case with consumer goods. However, to the extent that bundles of
separately-owned complementary patents can be combined in the public interest, one must be cautious
about condemning the tying of commonly-owned patents.8
While the formal economics literature has focused on price discrimination and foreclosure as
possible motives for tying, transactional and organizational efficiency is likely to explain much more of
the tying that occurs in practice. A commonly cited example of when tying generates efficiency is that
shoe companies generally sell shoes in pairs rather than as separates. The example is misleading,
however, because it creates the impression that tying for efficiency is limited to cases when there is no
(or at most a de minimis) demand for the individual items. The shoe example is not representative of the
many instances of tying of goods where many customers want only a subset of the package. The
practice is far too common for the specialized models in the price discrimination or foreclosure
literature to explain them all. Virtually no one wants every section of a newspaper. Evans and Salinger
(2005, 2008) argue that one cannot understand the diverse instances of tying without recognizing the
cost of product offering complexity. Before a firm decides on how much to produce (or try to sell) and
how much to charge, it first has to decide exactly what it sells. Even after deciding on its general line of
business, the products a firm offers are typically a small subset of the products and product
combinations it could conceivably offer. Dell revolutionized the personal computer industry by putting in
place systems to customize orders to consumer specifications to a degree that had previously been
unimaginable. But even Dell’s highly customized offerings did not include every conceivable
configuration and, more importantly, Dell and the personal computer industry are the exception. In
general, companies do not and indeed cannot customize their offerings to the precise desires of every
(or, for that matter, any) customer.
While Evans and Salinger did not address the issue of tying patents, the framework they suggest
provides a plausible – indeed, obvious – explanation for why patent tying is such a common
phenomenon. Whenever someone buys a bundle that contains items he does not want, there is a
temptation to ask to purchase just the wanted components at a discount to the bundle price. A cable
television subscriber who does not watch sports might want to purchase a basi c package that excludes
ESPN and other sports channels and pay a lower rate.
The key question to ask with respect to such a request (as well as a legal mandate to honor it) is,
“What is the limiting principle?” Some large innovative companies that license their technology have
thousands of patents. Yet, they might offer them in only a handful of bundles, and they might not offer
any of them individually. For a company with 1,000 patents, the number of possible combinations of
8
To be sure, participants in patent pools often offer to license their patents separately as well as part of
the pool, and whether they do can be a factor in the determination of whether a pool is in the public interest. See
the discussion in Lerner and Tirole (2004). While at least one paper, Quint (2014), develops theory demonstrating
that the inclusion of patents that are not perfectly complementary need not be anticompetitive or harmful, the
general consensus is that pools should be restricted to essential patents only to prevent foreclos ure of alternative
technologies for optional features.
6
patents is on the order of 10301, which is about googol cubed! The notion that the licensor is obligated to
unbundle any arbitrary bundle and offer a discount means that it would have to set 10 301 different
possible prices and then monitor and enforce compliance for all those different configurations. As a
practical matter, the number of different combinations that licensees might demand would obviously be
much smaller, but there is nonetheless a cost of having more complex product offerings. As a result,
patent licensors necessarily offer a small subset of the patent bundles that they could conceivably offer;
and it should come as no surprise (and should not necessarily be a public policy concern) if many
licensees only use a subset of the patents they license.
It also should not be surprising if licensees want bundles to be inclusive. A licensee takes a
license to avoid being sued for patent infringement. A company that licenses just a subset of the patents
that it needs to implement a technology risks a patent-infringement suit even if it pays the royalties due
on the patents it does license. As a result, licensees might demand patents in an inclusive bundle that
completely protects them from the threat of a suit. To the extent that such bundles cover both RANDcommitted patents and non-RAND committed patents, v including all the patents that a licensee might
conceivably use in conjunction with one or more of a licensor’s patents is itself a form of commitment
on the part of a patent owner not to behave opportunistically by suing for patent infringement on a
non-RAND committed patent that the licensor of its RAND-committed patents ends up infringing.
III.The Gilbert-Katz (GK) Model
The Gilbert-Katz (GK) model addresses two enduring issues with respect to patent licensing,
along with the possible interaction between the two: patent bundling and patent hold up. The possibility
of patent hold-up adds a layer of complexity to the implications of the single-rent theorem because
there are two levels of rents to consider: the ex ante rents before licensees engage in sunk investment
(including R&D efforts to invent around the patents) and the ex post rents after they do.
In the formal set up of the model, the owner of two complementary patents (“the IP owner”)
can license to a single licensee (the “manufacturer”). The technologies covered by the patents are
strongly complementary.9 That is, the technologies only create value in combination with each other.
The IP owner does not have manufacturing capability, so it must license its technology to the
manufacturer to capture any value from it. The value that can be realized from the technology is a
function of the level of complementary investment, which only the manufacturer can make. The
potential hold-up problem derives from the fact that if the manufacturer does not obtain a license to
the technology prior to investing in complementary assets, the IP owner might be able to expropriate
the contribution of the manufacturer’s investment. In other words, if the manufacturer does not obtain
an early license, the IP holder can practice hold up.
9
Strictly speaking, they are only strongly complementary if inventing around them is not economically
practical.
7
A simple numerical example illustrates the point. Suppose that without any investment by the
manufacturer, the patented technology yields a value of 20 (unrealizable by the patent holder, by
assumption, as explained above). With efficient investment in complementary assets by the
manufacturer, however, the technology embodied in an end product yields gross (i.e., before taking
account of the cost of the manufacturer’s investments) benefits of 100 (unrealizable by the
manufacturer absent the initial contribution by the patent holder). But the investments needed to
generate the additional value (the end product) have a cost. Suppose that cost is 30, so that the
potential net value of the combined technology (patent plus manufacture) embodied in the end product
is 70.
Given these assumptions, the manufacturer would be foolish to invest in the complementary
assets before obtaining a license for the patented technology. If it did so, the IP owner could insist on a
license fee of 100 (or just below it). After the fact, the manufacturer would rationally accept this offer, in
which case it would lose its sunk investment of 30. In anticipation of this possibility, the manufacturer
would not invest in complementary assets and the most the IP owner could charge as a license fee
would be 20, the value created by the patented technology without any complementary manufacturing
investment. The IP owner has an incentive to commit up front to a license fee of 70 for the bundle,
which allows the manufacturer to recover its investment cost (and also to choose the most efficient
level of complementary investment). 10 GK refer to licenses entered into before the manufacturer invests
as “long term” licenses and licenses signed after the manufacturer invests as “short term” licenses.
Up to this point, whether the technology is based on one patent or two is irrelevant. The
manufacturer needs both. Similarly, whether the IP owner offers the patents separately or in a bundle is
also irrelevant. It can offer the patents as a bundle at a license fee of 70. Alternatively, it can offer them
à la carte with individual prices that add to a cumulative license of 70. Profits and consumer welfare
(and, therefore, total surplus) are all the same under the various options for charging a total of 70. The
irrelevance of bundling under these conditions is an application of the single rent principle. One patent
that is essential for a product gives the patent holder the ability to extract the same total license fee as it
can with two such patents given that the manufacturer needs a license to both patents to create the
end product.
An additional feature of the GK model that makes bundling potentially relevant is that the
manufacturer can invest in R&D to invent around one or both patents. The outcome of the
manufacturer’s R&D is random. Thus, for a given level of investment, the manufacturer might invent
around just “patent A,” just “patent B,” both, or neither. 11 The manufacturer’s incentive to invest in R&D
depends on whether it has licensed the patents on a bundled or an à la carte basis. If it licenses the
10
A simplifying assumption in GK is that the IP owner makes take-it-or-leave-it offers to the manufacturer,
which implies that the IP owner is able to capture the entire available surplus. As they explain, the model would be
more complicated if the IP owner and manufacturer bargained over the available surplus, but the risk would
remain that the IP owner would expropriate part of the manufacturer’s investment if the manufacture invested
prior to obtaining a license.
11 The assumptions and the phenomenon that the GK model examines are similar to those explored by
Choi and Stefanadis (2001), discussed above.
8
bundle, it only earns a return on its R&D if it succeeds in inventing around both patents. If it licenses on
an à la carte basis, then successfully inventing around just one of the patents will lower the license fees
it owes the IP owner. Thus, holding the total license fee constant, the manufacturer has more of an
incentive to invest in R&D under à la carte licensing than under bundled licensing. As a result, the IP
owner can limit the manufacturer’s incentive to invest in R&D to invent around its patents by licensing
its patents solely as a bundle. This effect is often stated as a reason why patent bundling is problematic.
In the GK model, however, the manufacturer’s innovative efforts are assumed to be duplicative, working
only to replace the existing patented technology as opposed to extending it. The benefit R&D potentially
yields for the manufacturer is a private one (if successful, it enables the manufacturer to avoid license
fees), but it provides no social benefit. Furthermore, since the GK model assumes patent license fees are
lump sums, there is no distortion in the product market that could be corrected by invent-around
patents licensed at lower per unit fees.
Suppose the manufacturer waits until it learns the outcome of its R&D before obtaining a license.
That is, within the GK model, suppose the manufacturer decides to rely on short term contracting in
hope of being able to avoid a license for one of the patents. From the perspective of reaping additional
rewards from its R&D (by getting value when it successfully invents around just one of the patents), the
strategy does not provide the manufacturer with any benefits. (This point is yet another implication of
the single rent principle.)
Our numerical example from above clarifies the options – absent any R&D by the manufacturer –
that the IP owner has. Namely, the IP owner could 1) charge 70 for a bundle of the two patents, 2)
charge 35 per patent for each technology, 3) charge 70 for “patent A” and 0 for “patent B,” or 4) charge
70 for “patent B” and 0 for “patent A.” The options yield the patent holder the ability to earn the same
profit of 70, regardless of whether the manufacturer takes a long term contract or not. If the
manufacturer gets to wait to see which patent it needs (e.g., under short term licensing), then the IP
owner will know which patent the manufacturer requests12 and will choose its licensing terms
accordingly. So, if the manufacturer innovates around “patent A” and requests a license for “patent B,”
then the IP owner can charge 70 for a license to “patent B” and 0 for “patent A.” If, on the other hand,
the manufacturer innovates around “patent B” and requests a license for “pate nt A” only, then the IP
owner can charge 70 for a license to “patent A” and 0 for “patent B.”
In combination, GK interpret their results to suggest that many of the concerns with patent
bundling are misplaced. In their model, there is a benefit to “long term” licenses (or ex ante from the
manufacturer’s perspective) because such licenses prevent the IP owner from expropriating the
manufacturer’s returns to complementary investments (which are socially beneficial). But a ban on
patent bundling acts as a deterrent to long-term license agreements because the IP owner can
accomplish with short term licensing some of what it could accomplish by bundling. Allowing patent
bundling in long-term licenses does allow the IP owner to discourage “invent-around” innovation by the
manufacturer, but such innovative effort is socially wasteful by assumption under the GK model.
12
Within the GK model, the outcome of the manufacturer’s R&D is common knowledge.
9
IV. RAND Commitments
The intent of the GK model was to assess whether there is any reason to be concerned with the
bundled licensing of patents. They conclude that there is not. The question we need to consider is
whether the model justifies a conclusion that there is no reason to be concerned about the tying of
patents that are not RAND-committed to patents that are.
The possibility of RAND commitments is an important consideration in assessing the policy
conclusions GK infer from their results. In their model, bundling facilitates long-term contracts which are
in turn necessary to protect patent users against the risk that the patent-owner will try to expropriate
the value of their sunk investments in complementary assets. RAND commitments are, however, an
alternative mechanism to accomplish the same ultimate objective of preventing hold-up. The premise
behind RAND commitments is that potential patent users will sometimes wait until they have sunk
investments into the use of a particular technology before seeking a license. 13 Within the context of the
GK model, short-term contracting occurs in practice. As a starting point, we interpret RAND
commitments as commitments not to seek short-term contracts that are more favorable to the patentowner than it could have obtained with long-term contracts. As we will see, though, the implications of
this simple hypothesis are more subtle than one might initially expect.
To incorporate a RAND commitment into the GK model of the bundling of two patents, we first
need to formalize what RAND means for the licensing of a single patent. Here, we assume a simplified
version of the GK model. Consider a patent owner where use of the patent yields net benefits of B but
requires expenditure of a sunk cost, S.14 With expenditure R, the licensee can invent around the patent
with probability p.15
Given how we interpret RAND, we need to analyze what terms the licensor and manufacturer
would agree to before the manufacturer incurs sunk costs in complementary assets and decides
whether to invest in R&D. To do so, we analyze a “No Opportunism Game” in which the patent owner
must commit to a license fee before manufacturer chooses whether to invest in R&D or in
complementary inputs. Table 1 lays out the timing of the game. At Time 1, the patent owner sets a
royalty, L. At Time 2, the manufacturer decides whether to invest in R&D to circumvent the patent and
in complementary inputs. If the manufacturer invests in R&D, the outcome is revealed at Time 3. At
Time 4, the manufacturer decides whether to produce.
13
See Swanson and Baumol (2005).
GK allow for different levels of S and for B to be a function of S. The levels of sunk cost that are optimal
for both the licensee and society are then part of the model solution.
15 GK allow for a range of R and let p be an increasing function of R.
14
10
Table 1
No Opportunism Game
Time
Decision
Maker
Decision
1
Patent Owner
2
Manufacturer
3
Random
4
Manufacturer
L
R&D,
complementary
investment
R&D succeeds
or fails
Production
At stage 1, there are three levels of royalties to consider: LE = B, LV = B – S, and LC = R/p. (The subscripts
denote, “expropriation,” “value,” and “cost.”16 ) At stage 2, the options are to invest in both
complementary inputs and R&D, to invest just in complementary inputs, or not to invest at all.
To capture the effects we are interested in, assume that pB < R + S. When pB > R + S, the
manufacturer would invest in both R&D and complementary assets even if the IP holder chooses LE at
stage 1, and the IP holder would, for some parameters, choose LE to charge B with probability (1 – p).
Since our interest in the no opportunism game is to examine cases when the IP holder has an incentive
to commit not to expropriate the value of sunk investments, we need to restrict attention to parameter
values when it would have an incentive to make such a commitment.17 Even with this assumption, it is
possible for p(B – S) > R, which will imply that the royalty under value licensing exceeds the royalty
under cost licensing.
Given this restriction (which allows us to restrict attention to the choice between LV and LC at Time
1), the sub-game perfect solution of the game is as follows. At time 4, the manufacturer produces if it has
invested in R&D and the R&D succeeded. If either it did not invest in R&D or if it invested and the R&D
failed, it produces if L ≤ Lv = B – S. Since the IP holder only gets a positive payoff if the manufacturer
produces, it will choose L ≤ Lv at stage 1. As a result, the manufacturer will produce at stage 4 and invest
in complementary inputs at stage 2.
The manufacturer still has to decide whether to invest in R&D at stage 2. If it does not invest in
R&D, its payoff is B – S – L. If it does invest, its expected payoff is p(B – S – R) + (1 – p)(B – S – R – L). The
value for L that makes the manufacturer indifferent between investing and not investing is LC = R/p.18
16
We explain why R/p is cost-based below.
An alternative modeling approach would be to assume that the manufacturer chooses S only after it
observes the outcome of its R&D. Under that assumption, the IP owner would never successfully charge a license
fee of B. In Section V, however, we show that we can interpret S as the opportunity cost of not pursuing an
alternative standard, and that interpretation is particularly relevant for cases when IP holders make RAND
commitments. Under that interpretation, it would not be reasonable to assume that it is possible to delay the
investment decision.
18 Given that the IP holder is not going to charge above L , the manufacturer is produces regardless of
V
whether its R&D succeeds. Thus, the expected value of the R&D is pL.
17
11
If LC > LV, then the IP owner charges LV. Investing in R&D to invent around the patent is not
profitable in this case, so the IP owner can charge a value-based royalty and not face a risk of having its
patent invented around. If LC < LV , then the IP owner gets an expected payoff of (1 – p)LV if it chooses
value licensing and LC if it chooses cost licensing. It chooses cost-based licensing if:
(1)
(B – S) ≥ R/p ≥ (1 – p)(B – S),
and L = Lv = B – S otherwise. If p* is the probability that makes the patent owner indifferent between the
value and cost-based license fee, then the condition for p* is quadratic with roots:
1 ± 1- 4
(2)
p* =
R
B -S
2
Equation (2) merits some elaboration. When 4R > B – S, the equation has no real roots. Figure 1
illustrates such a case with B – S = 100 and R = 30. The horizontal line is the value-based royalty, the
curve is the cost-based royalty, and the downward-sloping line is the patent owner’s expected royalty if
it charges a value-based royalty. For all values of p, the expected cost of inventing around the patent
exceeds the patent owner’s expected royalties if it charges a value -based royalty. However, when the
expected cost of inventing around the patent exceeds B – S, the most the patent-holder can charge is Lv
= B – S. Given the parameter values underlying Figure 1, the patent owner maximizes its expected
royalties with value licensing when p is between 0 and 0.3 and with cost licensing for values of p above
0.3.
Figure 2, in which B – S = 100 and R = 20, illustrates the case in which equation (2) has two real
roots. As in Figure 1, there is a range of low values of p where the expected cost of inventing around the
patent exceeds the value of the patent. In that range, a value-based royalty maximizes the patentowner’s expected royalties. For values of p above that range, matters are more complicated than in
Figure 1. Within that range, there is a range bounded by the two real roots of equation (5) in which the
patent holder gets higher expected royalties with value licensing than it does with cost-based licensing
even though the manufacturer responds by investing in R&D to try to avoid the need to take the license.
Based on Figures 1 and 2 and the above discussion, we can establish:
Proposition 1: At time 1 in the No Opportunism Game, the patent owner charges Lv = B – S when
B – S ≤ R/p. It also charges Lv = B – S when B – S > R/p and p is within the range given by
equation (5). Otherwise, it charges Lc = R/p.
In the introduction, we mentioned a complication with respect to interpreting the
“reasonableness” restriction as what the patent owner would charge (and the manufacturer would
agree to) ex ante. Proposition 1 includes that complication. Intuitively, one might expect a reduction in
the expected cost of inventing around a patent to make it more likely that the patent owner opt for
cost-based licensing. According to Proposition 1 (and as Figure 2 i llustrate), a reduction in the expected
12
cost of inventing around a patent might induce the patent owner to choose a value -based royalty rather
than a cost-based royalty. 19
One can debate what this result implies about RAND royalties. On the one hand, if the RAND
commitment is simply a commitment not to behave opportunistically, then the model captures that
commitment. The patent owner is not behaving opportunistically by setting a RAND rate that leaves an
incentive for the manufacturer to invent around the patent. On the other hand, if the purpose of a
RAND rate is to help establish a standard, then a rate that leaves an incentive to invent around the
patent does not accomplish that objective since the standard would fail if the manufacturer successfully
invents around it. We do not seek to resolve this debate here. Instead, we allow for two possible
interpretations of RAND. The “looser” interpretation is the one implied by the No Opportunism Game.
The stricter “stricter” interpretation is the minimum of the value-based and cost-based rates.20
V. Ex Ante Competition to be the Standard – An Alternative
Interpretation of B and S
The two essential features giving rise to the need for standards are 1) competing technologies
each of which could serve as the standard and 2) network externalities. The GK model does not explicitly
capture those elements. In this section, we explore a “Standards Competition” game that captures the
two essential features that give rise to the need for standards and show how the opportunity cost of not
adopting the alternative standard has the same effect as sunk costs in complementary inputs in the No
Opportunism Game.
Suppose that there are two competing technologies, α and β, and that there are three types of
𝑖
customers, I, II, and III. Let 𝐵𝑗𝑘
be the per person benefit that Group i (i ∈ 𝐼, 𝐼𝐼, 𝐼𝐼𝐼) gets from technology
j (j ∈ α, β) given that the standard is technology k (k ∈ α, β). Group I prefers α to β but is primarily
𝐼 > 𝐵𝐼 > 𝐵𝐼 . Group II is loyal to α and will choose it even if it is not
interested in standardization: 𝐵𝛼𝛼
𝛽𝛽
𝛼𝛽
𝐼𝐼
𝐼𝐼
𝐼𝐼𝐼
𝐼𝐼𝐼 . Let φ and φ be
the standard while group III is similarly loyal to β. That is, 𝐵𝛼𝛽
> 𝐵𝛽𝛽
and 𝐵𝛽𝛼
> 𝐵𝛼𝛼
II
III
the proportion of buyers in groups II and III (with the proportion in group I being (1 - φII - φIII).
To keep matters simple, we assume that there is no additional cost of adopting a standard other
than the opportunity cost of not adopting the alternative standard and that the probability of inventing
around the patent(s) is 0. Table 2 lays out the timing of the game. At Time 1, the owners of both
19
The distinction between cost-based and value-based licensing is reminiscent of the distinction in Lerner
and Tirole between when the competition margin and the demand margin binds. With the cost-based license, the
binding constraint is the competition from the manufacturer’s own R&D. Since the model here does not have the
licensee heterogeneity present in Lerner and Tirole, the demand margin is simply a maximum willingness-to-pay
rather than a distribution across potential licensees in the willingness to pay.
20 However one would resolve this issue with perfect and costless enforcement, one might question
whether the looser interpretation would be possible to implement. It would create situations in which patent
owners would argue for a value-based fee on the grounds that it is relatively easy and inexpensive to invent
around the patent.
13
technologies each offer a royalty (Lα and Lβ, respectively) conditional on being accepted as the standard.
Further assume that only Type I manufacturers participate in the standard setting effort, so that Type II
and Type II manufacturers simply take the standards outcome as given and have no voice in the
cooperative development of the standard. Therefore, at Time 2, the Type I manufacturers choose which
technology to adopt as the standard. At Time 3, the owner of the technology not selected as the
standard can choose a license fee. At Time 4, manufacturers decide whether to produce.
Table 2
Standards Competition Game
Time
Decision
Maker
Decision
1
Patent Owners
Lα, Lβ
2
Type I
Manufacturers
α or β
3
Owner of nonStandard Patent
Lα’ or Lβ ’
4
Manufacturers
Produce or shut
down
As with the No Opportunism Game, we assume a sub-game perfect Nash equilibrium. The
solution to the game is as follows. Suppose the Type I manufacturers choose α at Time 2. Then Type III
𝐼𝐼𝐼
manufacturers produce at Time 4 if Lβ’ ≤ 𝐵𝛽𝛼
and shut down otherwise. Type I and Type II manufacturers
𝐼𝐼𝐼
produce (using the α technology). At Time 3, the owner of the β technology chooses Lβ’ = 𝐵𝛽𝛼
(and earns
𝐼𝐼𝐼
𝐼 - L ≥ 𝐵𝐼 - L , and β otherwise.
φIII 𝐵𝛽𝛼
). At time 2, Type I producers select α if 𝐵𝛼𝛼
α
β
𝛽𝛽
𝐼𝐼𝐼
Now consider Time 1. The owner of the β technology can guarantee itself φIII 𝐵𝛽𝛼
. If it offers a
royalty that induces the producer to accept it as a standard, however, it can increase its share from φIII
to (1 - φII). The lowest royalty it would offer to be the standard would be:
(3)
𝐿̃𝛽 =
𝜙𝐼𝐼𝐼
1− 𝜙𝐼𝐼
𝐼𝐼𝐼
𝐵𝛽𝛼
.
If the owner of the β technology offers 𝐿̃𝛽 , then the owner of technology α would have to offer:
(4)
𝐼 − 𝐵𝐼 = 𝐵𝐼 − (𝐵𝐼 − 𝐿
̃ 𝛽 ).
𝐿𝛼 = 𝐿̃𝛽 + 𝐵𝛼𝛼
𝛼𝛼
𝛽𝛽
𝛽𝛽
As long as:
(5)
𝐼𝐼
𝐼 − (𝐵𝐼 − 𝐿
̃𝛽 ) (1 – φIII) > φII 𝐵𝛼𝛽
𝐵𝛼𝛼
.
𝛽𝛽
If the owner of technology α offered any royalty above that given in equation (4), the owner of
technology β would undercut it in order to induce producers to accept it as a standard. Thus, given (5),
no royalty for α can exceed the value given by equation (4). Since we have already established that the
owner of the β technology will not offer a royalty below that given by equation ( 3) and that the royalty
given by (4) is the best response by the owner of the α technology to the royalty given by equation (3),
we have established:
14
Proposition 2: At Time 1 in the Standards Competition Game, the owner of technology α offers
the royalty given by equation (4) and the owner of technology β offers the royalty given by
equation (3). The Type I manufacturers choose the α technology as the standard, and the owner
𝐼𝐼𝐼
of the β technology chooses a royalty of 𝐵𝛽𝛼
at Time 3.
Since we have assumed away the possibility of inventing around the standard, the pricing at
Time 1 in the Standards Competition game corresponds to value pricing in the No Opportunism Game
𝐼 corresponding to B and to 𝐵𝐼 − 𝐿
̃ 𝛽 corresponding to S. Note that the opportunity cost of not
with 𝐵𝛼𝛼
𝛽𝛽
adopting the alternative standard is the value net of the royalty (rather than the entire value). Since we
formulated the game with Bertrand competition at Time 1, the royalty that the owner of β is assumed to
command requires that β have some value even if it is not the standard. More generally, though,
particularly if the patent owners compete against each other to be included in multiple standards,
licensing competition in practice might be “softer” than what we have modeled. For example, suppose
each patent has no value when it is not the standard. In that case, under the Standards Competition
Game, it would be willing to offer a royalty-free license at Time 1. In practice, however, competition may
be over technologies with some outside value and hence patent owners are unlikely to set their rates at
royalty free.
VI. Incorporating RAND Commitments into the GK Model
Having formalized the meaning of a RAND commitment on a single patent, we can now analyze
the tying of a RAND-committed patent to a non-RAND committed patent within our simplified version of
the GK model. To keep matters simple, we simply assume a sunk cost of complementary inputs (as in
Section IV) without specifying whether they are out-of-pocket costs or the opportunity cost of not
adopting an alternative technology.
A.
Analysis
Assume that the patent-owner owns patents over two technologies needed to capture value B
with investment S into complementary inputs (as in GK). Let R1 and R2 be the R&D expenses needed to
invent around each patent and p1 and p2 be the respective probabilities of success. Suppose the patentowner makes a RAND commitment on patent 1 but not patent 2.
Table 3 lays out the timing of the “Some Opportunism Game,” where the name indicates that
the incumbent cannot set an opportunistic license fee for patent 1 but it can for patent 2. At Time 1, the
patent owner sets the license fee for patent 1. At Time 2, the manufacturer decides whether to invest in
R&D with respect to each patent and whether to invest in complementary inputs. It can try to invent
around both patents, just patent 1, just patent 2, or neither. At time 3, the results of the R&D become
15
known (by everyone). At time 4, the patent owner then gets to choose a license fee for patent 2.21 At
Time 5, the manufacturer decides whether to produce.
Table 3
Some Opportunism Game
Time
Decision
Maker
Decision
1
Patent
Owner
L1
2
Manufacturer
R&D for each
patent,
complementary
investment
3
Random
R&D
succeeds or
fails
4
Patent
Owner
L2
5
Manufacturer
Produce or
shut down
The manufacturer’s choice at Time 5 limits what the Patent Owner can charge at Time 4, but it
does not prevent it from expropriating any sunk investments. If, at Time 4, the manufacturer needs a
license for patent 2, the patent owner can choose a license fee of B if the manufacturer has invented
around patent 1 and B – L1 if it has not. Under those circumstances, the manufacturer loses all its sunk
costs (in complementary inputs and any R&D it performed). At stage 2, the manufacturer in principle has
eight possible choices. However, given that the patent owner can expropriate any costs that the
manufacturer incurs at Time 2, we can immediately rule out any choice that does not entail investing in
R&D to invent around patent 2. As a result, the manufacturer has only three realistic options to consider
at time 2. It can invest in complementary assets and R&D to invent around both patents, it can invest in
complementary inputs and R&D just to invent around patent 2, or it can not invest at all. With the first
option, there are four possibilities for the resolution of uncertainty. With the second option, there are 2.
Suppose the manufacturer invests just in R&D to circumvent patent 2, which implies that it will
need to pay L1 if it chooses to produce. If the R&D is successful and it produces, its payoff is B – S – L1 –
R2; and it will choose to produce as long as L1 ≤ B – S. The patent owner’s pay-off is L1. If its R&D is not
successful, the patent owner sets L2 = B – L1. The payoff to the manufacturer will be –S – R2.
Now suppose the manufacturer invests in R&D to get around both patents. If both R&D efforts
succeed, the manufacturer does not need either license. Its payoff is B – S – R1 – R2 and the patent owner
gets 0. If it successfully invests around patent 2, it needs a license for patent 1. Its payoff is B – S – R1 – R2
– L1 and the patent owner gets L1 . If its patent 2 R&D fails then, regardless of whether its patent 1 R&D
21
Assuming that the manufacturer can wait to learn the outcome of its R&D before deciding whether to
invest in complementary inputs would not alter the results qualitativel y. Here, we assume that the manufacture
decides whether to invest in the complementary inputs at the same time that it makes its R&D decisions to
accommodate the interpretation of S as the opportunity cost of not having selected an alternative possible
standard.
16
succeeds, the patent owner chooses an opportunistic license fee for patent 2. The patent owner gets total
license fees of B and the payoff to the manufacturer is –S – R1 – R2.
The manufacturer’s expected payoff is p2(B – L1 ) – S – R2 if it invests just in patent 2 R&D and
p2[B – (1 – p1)L1] – S – R1 – R2 if it invests in R&D to circumvent both patents. Its expected payoff is 0
from investing in just patent 2 R&D if 𝐿1 = 𝐿𝑣1 = B – (S + R2)/ p2, which is value-based licensing. Its
expected payoff is equal between the two alternatives if 𝐿1 = 𝐿𝑐1 = R1/p1p2 , which is cost-based licensing.
Note that the probability of unsuccessful R&D affects both the value -based and cost-based license fees,
but in different directions. It lowers the value-based license fee because the risk of losing sunk costs
when the manufacturer’s patent 2 R&D is unsuccessful reduces the value of patent 1. In contrast, the
risk of unsuccessful patent 2 R&D raises the license fee the patent owner can charge without providing
the manufacturer an incentive to invest around patent 1 because it can expropriate any benefits from
successful patent 1 R&D through the patent 2 license fee.
As in the No Opportunism Game, the patent owner cannot charge more than the value -based
license fee. Thus, the cost-based license fee is only a consideration when it is less than the value based
license fee. When it is, the condition for the patent holder to get equal expected payoffs from value based and cost-based licensing is [B – (S + R2 )/p2](1 – p1 ) = R1 /p1p2, which has roots:
𝑅
(6)
𝑝1∗ =
1
1±√1−4
𝑆+𝑅 2
𝐵−
𝑝2
2
Thus, we can state:
Proposition 3: In the Some Opportunism Game, if R1 /p1p2 ≥ B – (S + R2)/p2, the patent owner
charges a license fee of 𝐿𝑣1 = B – (S + R2 )/p2 for patent 1. It also 𝐿𝑣1 if R1/p1 p2 ≥ B – (S + R2)/p2 and
p1 lies within the range defined by equation (9). Otherwise, it charges a license fee of R1/p1p2 .
The reason for exploring the No Opportunism and Some Opportunism Games is to compare
them with the Opportunism Game, in which the patent owner chooses L1 after the manufacturer has
made sunk investments. Table 4 shows the timing of the Opportunism Game. The only difference
between the Some Opportunism Game and the decision that the patent owner had made at what had
been labeled Time 1 is delayed until what had been Time 5 and is now Time 4 (given the elimination of
the first step.
Table 4
Opportunism Game
Time
Decision
Maker
Decision
1
Manufacturer
R&D for each
patent,
complementary
investment
2
Random
R&D
succeeds or
fails
17
3
Patent
Owner
L1, L2
4
Manufacturer
Production
The solution to the game is straightforward. As long a p1 p2B ≥ S + R1 + R2 , the manufacturer
invests in both types of R&D. If both succeed, it gets a payoff of B – S – R1 – R2. If at least one fails, the
patent owner charges a license fee of B,22 leaving the manufacturer with payoff – S – R1 – R2.
In the opportunism game, the manufacturer invests in R&D and production occurs only if B ≥ (S
+ R1 + R2)/ p1p2. In contrast, in the Some Opportunism Game, production occurs if B ≥ (S + R2)/ p2, which
is a weaker condition as long as p2 < 1.
Implications for Patent Tying
B.
Informed by the above results, we can now consider the implications of offering a RANDcommitted patent only in a bundle with a patent for which the owner has not made a RAND
commitment.
Presumably, the RAND commitment on patent 1 does not extend to patent 2 even if the
manufacturer might want a license for patent 2 to use in conjunction with patent 1. If the patent owner
licenses the two patents separately and the license fee for patent 1 honors the RAND commitment, the
patent owner can charge whatever it chooses – including an opportunistic fee of B – for patent 2.
Moreover, if the patent owner offers patent 1 separately at a rate that honors its RAND commitment,
that commitment places no constraint on what it would charge for a license that covers both patents.
But, what if it only offers patent 1 in conjunction with patent 2? What implications does the
RAND commitment have on allowed royalty?
While patent 2 is not RAND-encumbered, it is worth considering the implications of the RAND
commitment if it had. Suppose R1/p1 and R2/p2 are both greater than B – S so that any RAND royalty will
be value-based. In that case, the single rent principle applies. The combined royalties that the patent
owner would charge add to B – S. Selling the two together and charging B – S is one feasible option, and
any transactions cost savings would make it the preferred option.
Now suppose that R1/p1 and R2/p2 are both low compared to B – S so that a RAND royalty would
be cost-based. Under separate licensing, the royalties would be R1/p1 and R2/p2 . If the patent holder ties
the two together, it can charge (R1 + R2)/p1 p2 ≥ R1/p1 + R2/p2 with strict inequality as long as at least one
of the probabilities is less than 1. Here, the effect of tying is similar to the effect in Choi and Stefanides
(2001) and Nalebuff (2004). Just as tying eliminates the value of single stage entry in those models, it
eliminates the value of inventing around just one patent in the GK framework. 23
22
Because of the single rent principle, the patent owner can set L1 = B if the manufacturer’s R&D on
patent 2 succeeds, L2 = B if the manufacturer’s R&D on patent 1 succeeds, or a bundled price of B in either case.
23 GK finesse this issue. They restrict attention to the case where (R + R )/p p > B – S, so that no one
1
2
1 2
would try to enter by inventing around both patents. Under those conditions, they invoke the single-rent theorem
to argue that the patent holder would need only one patent to behave opportunistically in short term contracting.
Further, they argue that any R&D to circumvent the patents is socially wasteful. (Given their assumption of
18
Now suppose that R1/p1 > B – S but R2 /p2 is small. If so, a RAND commitment on patent 1 alone
implies a royalty of B – S – R2/p2 whereas a RAND commitment for all the patent-owner’s patents that
are relevant for the standard would warrant a royalty for the bundle of B – S. In this case, the patent
owner might try to justify a higher license fee by virtue of licensing the two patents in a bundle rather
than licensing patent 1 on a stand-alone basis.
Even accepting the argument that the RAND royalty for the bundle can exceed the RAND royalty
for just patent 1, RAND for the bundle rests critically on the assumption that the patent holder makes a
RAND commitment on patent 2. If it did not, then the manufacturer would have no assurances that the
patent holder would not try to charge an opportunistic license fee for patent 2; it therefore would not
agree to a royalty of B – S prior to making any investments. If so, tying the patents together and arguing
that the proposed royalty have honored a RAND commitment on the two patents does not imply that it
honors a RAND commitment just on patent 1.
Whether one adopts the stricter or looser meaning of a RAND commitment, the patent owner
only honors the commitment if the manufacturer can license patent 1 for what we have analyzed to be
the RAND rate. The requirement does not necessarily preclude offering patent 1 only in conjunction
with patent 2, and the patent holder may choose to do so. However, the extension of the license to
include patent 2 does not increase the price that satisfies the RAND obligation. In assessing whether a
license fee on a bundle of patents that includes patents that are not RAND-encumbered with patents
that are, the question to ask is whether the license fee would satisfy the RAND commitment even if the
bundle included only the RAND-committed patents.
VII.
Conclusions and Policy Implications
The question we address in this paper is whether a patent holder that has made a RAND
commitment on a patent violates that commitment by offering a license to that patent only in a bundle
of patents that include patents that are not RAND-committed.
The starting point of our formal modeling is the GK model of patent bundling. That model
captures a key element of the problem we are addressing – the desirability of patent licenses that
protect patent users against hold up that effectively expropriates the value of the patent user’s sunk
investments in assets needed to capture the value of the technology. In the GK model, patent tying is
socially desirable because it is necessary to induce patent holders to enter into what they call long -term
licenses, i.e., licenses that the patent users can enter into prior to incurring sunk costs.
The desirability of bundling within the GK model does not, however, fully resolve the policy issue
that we address. One cannot use an economic model to analyze the implications of tying patents
without RAND-commitments to patents with RAND commitments unless RAND commitments are
explicitly addressed within the model, and the GK model does not take RAND commitments into
perfectly efficient licenses, license fees do not create any product market distortions.) If one rela xes those
assumptions, however, tying of RAND-committed patents may be harmful (at least before one takes account of
transaction cost issues).
19
account. Incorporating RAND commitments into the GK model turns out to be important for two
reasons. First, the purpose of RAND commitments is to solve precisely the same problem that patent
bundling solves in the GK model: they are a commitment not to expropriate manufacturers’ sunk costs
(not to practice hold-up). The possibility of making a RAND commitment at least weakens and
potentially eviscerates the argument that the primary effect of patent bundling is to prevent hold -up.
Second, one of the reasons that patent bundling is benign in the GK model is that the single rent
principle applies for long run contracting. The patents in the GK model are perfect complements, so the
patent owner only needs one patent to extract the full rent. But a RAND commitment necessarily
implies a limit on the royalty for the RAND-committed patent. Just as the single rent principle does not
apply to a good that is subject to price regulation, it does not apply to the combination of RANDcommitted and non-RAND-committed patents. 24
In the concern with licensing RAND-committed patents only in bundles with non-RAND
committed patents, the word “only” is key. As long as a patent owner offers a RAND-committed patent
(or a set of RAND-committed patents that are all necessary to implement a given standard) and charges
a RAND royalty, nothing precludes it from also offering a bundle that includes patents that are not
RAND-committed with those that are and the RAND commitment places no limit on the royalty for the
bundle. Using standard economic terminology, if the patent holder engages in mixed bundling, it honors
its RAND commitment by charging a RAND rate for the RAND-committed patent.25
The licensing practice that potentially gives rise to a concern is when the patent owner engages
in tying (not mixed bundling).26 While the GK results are not sufficient to dismiss concerns about tying
licenses for patents without RAND commitments to licenses of RAND-committed patents, the licensing
of patents only in bundles is a common practice with sound efficiency justifications. Thus, while there is
a legitimate concern that this type of tying might be a way to evade a RAND commitment, a ban on the
practice would ignore the significant transaction-cost justifications for the bundling of patents.
Importantly, the problem with offering RAND-committed patents only in bundles is not the bundling per
se but, rather, the terms on which the bundle is offered. A RAND commitment is a commitment to
make the RAND-committed patent(s) available for a reasonable rate. While that rate might not be
known when the patent-holder makes the RAND commitment, there is a process for determining a
24
Farrell and Weiser (2003) refer to this exception to the single rent principle as the “Baxter exception.”
Strictly speaking, “mixed bundling” entails offering all the components of a bundle separately in
addition to the bundle. In the GK model and our simplified version of it, mixed bundling would entail offering not
only a license for the bundle and a stand-alone license for patent 1, but also a stand-alone license for patent 2.
Moving away from these simple models, however, a practical application of mixed bundling would entail offering a
license to the portfolio of RAND-patents alone, offering a second license to the non-RAND patents (either as a
portfolio or in rational subsets), and offering a third license with a bundle of the RAND and non -RAND patents
together, rather than having to offer individual licenses of each RAND-patent and each non-RAND patent, for the
product offering cost reasons we explain above.
26 If the patent owner offers only the bundle, it engages in “pure bundling.” If, in addition to offering a
license to the bundle, it also offers a license to patent 2, it no longer engages in pure bu ndling but it is still tying
patent 2 to patent 1.
25
20
RAND range and for assessing rates as RAND if the patent(s) were offered separately. 27 As long as the
royalty rate for the bundle of patents would be considered RAND for the RAND-committed patent alone,
and the non-royalty terms of the license are similarly RAND, then the patent owner is honoring the
commitment even though it is including other patents in the bundle.
Determining a RAND rate is not a simple matter. Part of the compli cation stems from the
difficulty of measuring the factors that matter. As our model clarifies, part of the complication is
determining whether RAND implies value-based or cost-based royalties. But these complications are
inherent in any RAND determination. They do not alter and should not obscure the analysis of whether
offering a RAND-committed patent only in a bundle with other patents necessarily violates the RAND
commitment. It does not, but it only does not if the total license fee would be RAND for th e RANDcommitted patents alone.
This principle limits the arguments that licensees and licensors can make. The limit on a licensee
is that it cannot argue that it necessarily deserves a discount from the bundled price if it only wants the
bundle in order to get access to the RAND-committed patents. As long as the licensor can show that its
royalty for the bundle would be RAND just for the RAND-committed patents, it is honoring its RAND
commitment. The limit on the licensor that engages in tying is that it cannot compute an implicit royalty
for licenses to the RAND-committed patents by subtracting the value of the licenses to the additional
patents in the bundle from the royalty for the bundle and then argue that this implicit royalty is RAND.
Allowing such an argument as a defense would create an obvious way for patent owners to avoid their
RAND commitments.
27
While this process is still evolving, several court rulings offer frameworks for determining RAND rates .
See, e.g., Microsoft v. Motorola, 2013 WL 2111217 at *12 (W.D. Wash. Apr. 25, 2013); In re Innov atio IP Ventures,
LLC Patent Litig., 2013 WL 5593609 at *8-10 (N.D. Ill. Oct. 3, 2013); Ericsson Inc. v. D-Link Systems, Inc., 773 F.3d
1201 (Fed. Cir. 2014); and In the Matter of Certain Wireless Devices with 3G and/or 4G Capabilities and
Components Thereof, ITC Inv. No. 337-TA-868 at 123-24 (June 13, 2014), available at
http://www.essentialpatentblog.com/wp-content/uploads/sites/234/2014/07/2014.06.26-Initial-Determinationon-Violation-PUBLIC-337-TA-868smMRC.pdf.
21
Acknowledgments
We thank Bernhard Ganglmair, Rich Gilbert, John Harkrider, Elizabeth Wang and two anonymous
referees for helpful suggestions. We are also grateful to Google for financial support for early versions of
this research. The opinions represent the views of the authors and not those of any organization.
22
REFERENCES
Adams, William J. and Janet L. Yellen (1976) “Commodity Bundling and the Burden of Monopoly,” The
Quarterly Journal of Economics, vol. 90, pp. 475-498.
Bowman, Ward S. (1957) “Tying Arrangements and the Leverage Problem, Yale Law Journal, vol. 67 19 .
Carlton, Dennis W. and Michael Waldman (2002) “The Strategic Use of Tying to Preserve and Create
Market Power in Evolving Industries,” The Rand Journal of Economics, vol. 33, pp. 194-220.
Choi, Jay Pil and Christodoulos Stefanadis (2001) “Tying, Investment, and the Dynamic Leveraging
Theory,” Rand Journal of Economics, vol. 32, pp. 52-71.
Contreras, Jorge (2014) “Market Reliance and Patent Pledges,” Utah Law Review (forthcoming).
Cooper, Robin and Robert S. Kaplan (1988) “Measure Costs Right: Make the Right Decisions,” Harvard
Business Review, vol. 66, pp. 96-103.
Elhauge, Einer (2009) “Tying, Bundled Discounts and the Death of the Single Monopoly Profit Theory”
Harvard Law Review, vol. 123, pp. 397 - 481.
Elhauge, Einer (2015), “Treating RAND Commitments Neutrally” Journal of Competition Law &
Economics (forthcoming).
Evans, David S. and Michael A. Salinger (2005), “Why do Firms Bundle and Tie? Evidence from
Competitive Markets and Implications for Tying Law,” Yale Journal on Regulation, vol. 22, pp. 37-89.
Evans, David S. and Michael A. Salinger (2008), “The Role of Cost in Determining when Firms Offer
Bundles and Ties,” Journal of Industrial Economics, vol. 56, pp. 143-168.
Farrell, Joseph and Philip J. Weiser (2003), “Modularity, Vertical Integration, and Open Access Policies:
Towards a Convergence of Antitrust and Regulation in the Internet Age,” Harvard Journal of Law &
Techology, vol. 17, pp. 85-134.
Gilbert, Richard J. and Katz, Michael L. (2006), "Should good patents come in small packages? A welfare
analysis of intellectual property bundling," International Journal of Industrial Organization, vol.
24(5), pp. 931-952.
Gabaix, Xavier and David Laibson (2006) “Shrouded Attributes, Consumer Myopia, and Information
Suppression in Competitive Markets,” Quarterly Journal of Economics, vol. 121, pp. 505-540.
Harkrider, John (2014) “REPs Not SEPs: A Reasonable and Non-Discriminatory Approach to Licensing
Commitments,” Competition Policy International.
23
Layne-Farrar, Anne (2014) “Moving Past the SEP RAND Obsession: Some Thoughts on the Economic
Implications of Unilateral Commitments and the Complexities of Patent Licensing”, George Mason
Law Review, Vol. 21, No.4.
Lerner, Josh and Jean Tirole (2004) “Efficient Patent Pools,” American Economic Review, vol. 94, pp. 691711.
Krattenmaker, Thomas G. and Steven C. Salop (1986) “Anticompetitive Exclusion: Raising Rivals’ Costs to
Achieve Power over Price,” The Yale Law Journal, vol. 96, pp. 209-293.
McAfee, R. Preston, John McMillan, and Michael D. Whinston (1989) “Multiproduct Monopoly,
Commodity Bundling, and Correlation of Values,” The Quarterly Journal of Economics, Vol. 104, No.
2, pp. 371-383.
Nalebuff, Barry (2004) “Bundling as a Barrier to Entry,” Quarterly Journal of Economics, vol. 119, pp. 159187.
Quint, Daniel (2014) “Economics 0f Patent Pools When Some (But Not All ) Patents Are Essential”,
American Economic Journal: Microeconomics, 6(1), pp. 23 – 57.
Schmalensee, Richard (1984), “Gaussian Demand and Commodity Bundling,” Journal of Business, 57 (1),
S211-S230.
Shapiro, Carl (2001) “Navigating the Patent Thicket: Cross Licenses, Patent Pools, and Standard Setting,”
in Adam Jaffe, Josh Lerner, and Scott Stern (eds.), Innovation Policy and the Economy, Vol. 1,
(Cambridge, MA: MIT Press).
Stigler, George J. (1963) “United States v. Loew’s Inc.: A note on block-booking,” in P. Kurland, Supreme
Court Review : 1963 (Chicago: University of Chicago Press) pp. 152-57.
Swanson, Daniel G. and William J. Baumol (2005) “Reasonable and nondiscriminatory (RAND) royalties,
standards selection, and control of market power,” Antitrust Law Journal, vol. 73, pp. 1-58.
Whinston, Michael (1990) “Tying, Foreclosure, and Exclusion,” American Economic Review, vol. 80, pp.
837-59.
24
Figure 1
Cost-Based vs. Value-Based Royalties
(No real roots to Equation 5)
Figure 2
Cost-Based vs. Value-Based Royalties
(Real roots to Equation 5)
25