When Do Auctions Ensure the Welfare-Maximizing

Allocation of Scarce Inputs?

by

John W. Mayo and David E. M. Sappington

Abstract
We determine when an unfettered auction will ensure the welfare-maximizing allocation of
a scarce input that enhances product quality and may reduce production costs. A supplier
values the input both for this “use value” and for its “foreclosure value,” because once the
input is acquired, it is unavailable to rivals. An unfettered auction often ensures the welfare-
maximizing allocation of an input increment. However, it can fail to do so when the input
would increase relatively rapidly the competitive position of a rival with a moderate compet-
itive disadvantage. Bidder handicapping that ensures auctions generate welfare-maximizing
input allocations di¤er from standard handicapping policies.

Georgetown University (mayoj@georgetown.edu).

University of Florida (sapping@u‡.edu).

We are grateful for very helpful comments and suggestions to the co-editor, anonymous
referees, Evan Kwerel, Preston McAfee, Timothy Tardi¤, and seminar participants at the
Federal Communications Commission, the Center for Research in Regulated Industries 2014
Eastern Conference, the Southern Economics Association 2014 Annual Meeting, and the
Allied Social Sciences Association 2015 Annual Meeting.
1 Introduction
It is well known that a vertically integrated …rm might seek to deny access to an upstream
input in order to foreclose downstream rivals from operating in lucrative retail markets. It is
also well known that a monopolist typically is willing to pay more than a potential entrant
for an essential input because, by foreclosing entry, the monopolist can secure its monopoly
pro…t whereas an entrant can gain at most its share of a smaller duopoly pro…t.1
Recently, foreclosure concerns have expanded to the domain of auctions. For instance,
some have questioned whether leading suppliers of wireless communications services might
outbid smaller rivals in auctions of scarce radio spectrum primarily to limit the ability of
the smaller rivals to develop into e¤ective competitors (U.S. Department of Justice, 2013).
Similarly, the Supreme Court has considered the possibility of “predatory bidding,”whereby
a …rm intentionally bids particularly aggressively for a scarce input in order to limit down-
stream competition from other potential input purchasers.2 Because auctions are commonly
employed to allocate scarce inputs in practice,3 these considerations raise important public
policy concerns.
Numerous authors have recognized that auctions do not always ensure the welfare-
maximizing allocation of scarce inputs.4 However, to our knowledge, the literature does
not provide a clear delineation of the industry conditions under which unfettered input auc-
tions –auctions that allocate inputs to the bidders that value them most highly5 –will, and
1
Corresponding considerations explain why a monopolist may engage in preemptive patenting to exclude
rivals (e.g., Gilbert and Newbery, 1982). Rey and Tirole (2007) provide a comprehensive review of the
literature on foreclosure.
2
Weyerhauser Co. v. Ross-Simmons Hardwood Lumber Co. 127 S. Ct. 1069, 1078 (2007). Blair and Lopatka
(2008) provide an informative discussion of this issue.
3
To illustrate, auctions have been employed to allocate billions of dollars of spectrum among suppliers
of wireless communications services since the mid-1990s. See McAfee and McMillan (1996), Kwerel and
Rosston (2000), Hazlett and Munoz (2009), and Cramton et al. (2011), for example. Timber harvesting
and oil drilling rights also are typically allocated to suppliers of wood and oil products via auction. See
Hendricks et al. (1994), Haile (2001), and Athey et al. (2013), for example.
4
To illustrate, Jehiel and Moldovanu (2003, p. 271) observe that “when the assets for sale . . . are inputs that
will subsequently be used by the successful bidders in imperfect competition with each other . . . auctions
can behave in surprisingly problematic ways.” Eso et al. (2010, p. 542) note that “allocating input(s)
through e¢ cient auctions may be misguided when bidders are competing …rms.”
5
For example, …rst-price and second-price auctions with no bidder subsidies generally have this feature. The
1
will not, ensure the welfare-maximizing allocation of inputs. The purpose of this research is
to provide such a delineation in the context of a common model of industry competition.
The bidders in our model engage in Hotelling price competition after the input auction
concludes. The input being auctioned enhances customer valuation of a …rm’s product and
can reduce the …rm’s production cost. To illustrate, the input might be spectrum that
enables a supplier of wireless communications service to increase the speed and reliability of
its service, which can reduce customer acquisition and retention costs. Each of the …rms in
our model has an initial endowment of the input, and the incremental amount of the input
that is being auctioned is relatively small. Consequently, no …rm can preclude the operation
of its rivals even if it were to acquire all of the available input.
Each …rm in this setting derives both a “use value” and a “foreclosure value” from the
input. The use value arises because the …rm that acquires the input increment can employ
it to enhance its competitive position.6 The foreclosure value arises because, by acquiring
an increment of a scarce input, a …rm precludes its rivals from acquiring the increment.7
Such preclusion does not foreclose the rival in the traditional sense of driving the rival from
the market, but rather in the sense of preventing the rival from employing the increment to
improve its competitive position.
A …rm (“…rm 1”) will value highly, and therefore bid aggressively for, an input increment
that will substantially enhance its competitive position. However, the …rm’s rival will also
bid aggressively for the input in this case in an attempt to prevent …rm 1 from acquiring the

unfettered input auctions that we analyze are isomorphic to the e¢ cient capacity auctions that Eso et al.
(2010, p. 525) consider. The authors de…ne an e¢ cient capacity auction to be one that allocates “each unit
of capacity to the …rm that values it the most.”
6
Formally, a …rm’s “competitive position” is the di¤erence between the valuation that customers place on
the …rm’s product and the …rm’s unit cost of production.
7
Cramton et al. (2011, p. S168) note that auctions may fail to promote economic e¢ ciency because “an
incumbent will include in its private value not only its use value of the [scarce input] but also the value of
keeping [it] from a competitor.” Our terminology parallels that of the U.S. Department of Justice (2013,
p. 10) which, in the context of spectrum auctions, observes, “... the private value [of spectrum] for incum-
bents ... includes not only the revenue from use of the spectrum but also any bene…ts gained by preventing
rivals from improving their services and thereby eroding the incumbents’ existing businesses. The latter
might be called ‘foreclosure value’as distinct from ‘use value’.”

2
input that will substantially enhance its competitive position. These o¤setting valuations of
the rate at which the input increases the competitive positions of the duopolists ensures that
the input allocation is determined at auction by the relative levels of the …rms’competitive
positions. The …rm with the strongest competitive position – and thus the largest market
share –will win an unfettered auction for the input increment.
The resulting allocation of the input maximizes welfare whenever the input increases the
competitive position of the larger supplier at least as rapidly as it increases the competi-
tive position of the smaller supplier. However, an auction can fail to generate the welfare-
maximizing allocation of an input increment when the input would increase relatively slowly
the competitive position of a …rm with a moderately large market share. In this case, the
larger …rm may acquire the input increment in an unfettered auction even though welfare
would be higher if the smaller …rm secured the increment.
In principle, a bid credit for the smaller …rm could be implemented to ensure the welfare-
maximizing allocation of the input increment. (A bid credit re‡ects the fraction of its bid
that a …rm is not required to pay if it wins the auction for the input increment.) However, in
contrast to its typical design in practice, the appropriate bid credit does not re‡ect di¤erences
in pro…tabilities or market shares of the bidding …rms. Instead, in a duopoly setting it re‡ects
the extent to which the input would enhance the competitive position of the smaller …rm
more than it would enhance the competitive position of the larger …rm.
The input auctions we analyze are a special case of auctions with externalities that have
received considerable attention in the literature. Input auctions entail externalities because
the assignment of the input a¤ects both the recipient of the input and competitors that do not
receive the input.8 Studies in this literature often assume that agents are privately informed
about their exogenous valuations of the object being auctioned.9 Our study di¤ers in part
8
Katz and Shapiro (1986) provide an early analysis of the allocation of inputs (technology licenses) in the
presence of externalities. Brocas (2013a) demonstrates that externality considerations can sometimes lead a
supplier to favor the sale of a valuable resource to a non-rivalrous supplier rather than to a direct competitor.
9
Jehiel et al. (1996, 1999), for example, consider settings where the relevant externalities are not systemati-
cally linked to the agents’valuations of the object. Carillo (1998) and Brocas (2013b, 2014), for example,
analyze settings where this linkage is present. Although most studies in this literature consider the alloca-
3
by allowing valuations and externalities to re‡ect the equilibrium outcomes of competition
among sellers of di¤erentiated products.10 Our explicit modeling of industry competition
allows us to identify industry and supplier characteristics that enhance or limit the propensity
of unfettered auctions to generate the welfare-maximizing allocation of inputs.
Our study also di¤ers from the typical analysis in the literature on auctions with external-
ities by focusing on settings with complete information. This focus allows us to demonstrate
most clearly when and why an unfettered auction will fail to ensure the welfare-maximizing
allocation of an input.11 We also show, though, that the key considerations that arise in
settings with complete information remain relevant in settings with incomplete information.
Some related studies assume that bidders engage in Cournot competition after bidding for
capacity at auction. McAfee (1998) …nds that small, capacity-constrained …rms often outbid
larger, unconstrained …rms in part because each unconstrained bidder cannot capture the
full increase in industry pro…t that arises when small producers are precluded from acquiring
additional capacity.12;13 In their analysis of auctions that allocate capacity increments to the
…rms that value them most highly, Eso et al. (2010) …nd that even when all suppliers are

tion of a single object, Ase¤ and Chade (2008) analyze the allocation of multiple objects in the presence of
externalities.
10
Burguet and McAfee (2009, n. 8) observe that “di¤erentiated product models are notoriously challenging
to analyze. However, the analysis of such models represents the natural next step.”
11
Related studies that consider settings with complete information include: (i) Krishna (1993), who con-
siders sequential auctions of capacity; (ii) Jehiel and Moldovanu (1996), who analyze agents’decisions to
participate in auctions with externalities; and (iii) Pagnozzi and Rosato (2014), who demonstrate that, due
to externalities, welfare can be higher when a new supplier enters an industry by acquiring an incumbent
supplier through private bilateral negotiation rather than by auction.
12
McAfee (1998) also analyzes a model in which …rms engage in Cournot competition after bidding to
acquire an input that reduces a …rm’s total and marginal cost of production. He …nds that the …rm with
the smallest initial endowment of the input will win the auction for the input increment. Hendricks and
McAfee (2010) extend models with Cournot competition among users of an input to include competition
among suppliers of the homogeneous input. The authors allow …rms to be both buyers and sellers of
the input. Equilibrium allocations are determined by (strategic, endogenous) supply and demand in their
model, rather than by auctions.
13
Borenstein (1988) observes that the pro…t a …rm secures from a license to operate upstream can di¤er
systematically from the total surplus it generates downstream, so auctions of licenses can fail to generate
welfare-maximizing outcomes. Burguet and McAfee (2009) …nd that auctions of operating licenses max-
imize consumers’ surplus when suppliers face binding …nancing constraints if consumer demand for the
homogeneous retail product is su¢ ciently elastic.

4
symmetric ex ante, capacity increments often are allocated asymmetrically. Consequently,
the equilibrium downstream industry con…guration entails one large …rm with no capacity
constraint facing smaller, capacity-constrained rivals.14
Many analyses of auctions with externalities focus on characterizing the properties of
auctions that are optimally designed to achieve a speci…ed objective, such as the maximiza-
tion of the seller’s payo¤.15;16 Although our primary focus is not on auction design, we do
consider how bidding credits can be structured to ensure the welfare-maximizing allocation
of inputs when private and social valuations of inputs diverge. Our …nding that substantial
information is required to ensure this outcome (and so the appropriate design of bidding
credits can be highly problematic in practice) is consistent with the literature’s message
regarding the complexity of auction design in the presence of externalities.

2 Input Allocations
We consider a setting where two …rms engage in price competition after acquiring a
key input (e.g., spectrum) at auction. Let vi denote the value that each consumer derives
from purchasing one unit of …rm i 2 f1; 2g’s product. This value is an increasing function
of the amount of the input (ki ) that …rm i employs (i.e., vi0 (ki ) > 0). The …rm’s unit
cost of production, ci (ki ), also may decline as it acquires more of the input (i.e., c0i (ki )
0). To illustrate, the product in question might be a subscription to the …rm’s wireless
communications service and the functionality admitted by this subscription. Consumers
value this subscription more highly when additional spectrum enables the …rm to increase

14
Eso et al. (2010) also analyze a model of price competition between capacity-constrained suppliers of
di¤erentiated products. Their focus in this analysis, too, is on potential asymmetries in the post-auction
size distributions of industry suppliers.
15
Jehiel and Moldovanu (2000) is an exception because the authors focus on second-price auctions and
consider the optimal design of reserve prices and entry fees. Das Varma (2002) examines how the relative
performance of open and sealed-bid auctions varies according to whether the externalities among bidders
are reciprocal or non-reciprocal.
16
The properties of the optimal auction depend in part on the seller’s powers to compel potential buyers to
participate in the auction. See, for example, Jehiel at el. (1999), Brocas (2003), and Figueroa and Skreta
(2009).

5
the speed and reliability of its wireless communications service.17
All consumers value symmetrically the product enhancement that higher levels of the in-
put provide (e.g., faster download speeds and/or fewer dropped calls). However, consumers
di¤er in their valuations of other elements of the …rms’products (e.g., the color and design
of telephone handsets or the geographic locations of a …rm’s showrooms and service centers).
To capture these di¤erent valuations, we employ the standard Hotelling model of competi-
tion and assume that potential consumers are distributed uniformly on the [ 0; 1 ] interval.
Consumers travel either to point 0 to purchase the product from …rm 1 or to point 1 to
purchase the product from …rm 2. Each consumer experiences unit transportation cost t
as she travels to purchase the product. Therefore, a consumer who travels distance d to
purchase one unit of the product from …rm i 2 f1; 2g and pays price pi for the good receives
net utility vi ( ) pi t d.18
We will refer to the di¤erence between consumer valuation of …rm i’s product and …rm i’s
production cost as …rm i’s value margin, mi (ki ) vi (ki ) ci (ki ). To ensure that both …rms
serve consumers in equilibrium, we assume the …rms’value margins are not too disparate.19

Assumption 1. 3 t < m1 (k1 ) m2 (k2 ) < 3 t for all relevant k1 and k2 .

Given the input allocation (k1 ; k2 ) and the resulting production costs and value margins,
equilibrium outcomes in this model are readily calculated using standard techniques.

Lemma 1. Equilibrium prices, outputs, pro…ts, and consumers’surplus are, for i; j 2 f1; 2g
(j 6= i): pi = ci + 13 [ 3 t + mi mj ]; xi = 1
6t
[ 3 t + mi mj ]; i = 1
18 t
[ 3 t + mi m j ]2 ;

17
A corresponding cost saving can arise when more abundant spectrum allows a …rm to reduce its use of
alternative, less e¢ cient inputs. Increased customer valuation of a …rm’s product also can reduce the …rm’s
customer acquisition and retention costs.
18
Brocas (2008) considers a related model with externalities in which two agents are located at the opposite
ends of a Hotelling line. In her model, a principal must decide where on the line to locate an indivisible
good. Brocas demonstrates how the optimal policy varies according to whether the agents are privately
informed about their valuation of the good or their transportation costs.
19
We also assume that v1 (0) and v2 (0) are su¢ ciently large relative to t that all consumers purchase one
unit of the product in equilibrium.

6
 m1 m2 5t 5 [ m1 m2 ]2 20
and CS = 6t
[ 3 t + m1 m2 ] + 6t
[ 3 t + m2 m1 ] 4 36 t
.

When a …rm secures more of the input at auction, it both enhances its own value margin
and prevents its rival from acquiring the input to increase its value margin. Therefore, the
rate at which a …rm’s equilibrium pro…t increases as it acquires more of the input at auction
is the sum of the marginal “use value”( @@kii ) and the marginal “foreclosure value”( @ i
@kj
) of
the input.21 From Lemma 1, for i; j 2 f1; 2g (j 6= i):

@ i @ i 1
Bi = [ 3 t + mi mj ] m0i (ki ) + m0j (kj ) . (1)
@ki @kj 9t

The …rm that will win an unfettered auction for a marginal input increment is the …rm
that anticipates the largest increase in its equilibrium pro…t from securing the increment,
accounting for relevant use and foreclosure values.22 Therefore, equation (1) provides:

Proposition 1. The …rm with the highest value margin will win an unfettered auction for
the input increment.

Proof. Equation (1) implies that B1 > B2 if and only if:

1 1
[ 3 t + m1 m2 ] [ m01 (k1 ) + m02 (k2 ) ] > [ 3 t + m2 m1 ] [ m02 (k2 ) + m01 (k1 ) ]
9t 9t

, m1 m 2 > m2 m1 , m 1 > m2 . (2)

Proposition 1 implies that the identity of the winning bidder in the input auction is not
a¤ected by di¤erences in the rates at which the input increases the value margins of the two
…rms. One might suspect that …rm 2, say, often would outbid …rm 1 for the input if the

20
The proof of Lemma 1 is presented in the Appendix. The Appendix also presents the proofs of all other
formal conclusions that are not proved in the text.
21
See McAfee (1998) for corresponding discussion.
22
We consider marginal increments here and throughout the ensuing analysis. Doing so allows us to analyze
the rate at which the input increases key variables (e.g., pro…t and welfare) rather than the amount by
which the input increment increases these variables. This focus streamlines the formal analysis without
a¤ecting the qualitative conclusions that would arise from an analysis of discrete input changes that are
su¢ ciently small. The concluding discussion in section 5 notes some of the additional considerations that
can arise in the presence of large, discrete input changes.

7
input increases …rm 2’s value margin substantially more rapidly than it increases …rm 1’s
value margin. However, like …rm 2, …rm 1 values the input highly when it increases …rm 2’s
value margin rapidly. This is the case because …rm 1 recognizes that …rm 2 will become a
substantially more formidable competitor if …rm 2 secures the input. Consequently, …rm 1
is willing to pay relatively handsomely to prevent its rival from becoming considerably more
formidable, just as …rm 2 is willing to pay relatively handsomely to become more formidable.
These o¤setting e¤ects ensure that the equilibrium input allocation is determined solely by
di¤erences in the value margins of the two competitors. The …rm with the largest value
margin serves the most customers and enjoys the largest pro…t margin (recall Lemma 1),
and so will pro…t most from increasing its competitive position by acquiring the marginal
input increment.
In contrast, as Proposition 2 reports, the allocation of the input increment that maximizes
welfare (which is the sum of consumers’surplus and industry pro…t) will depend upon both
di¤erences in the relative value margins of the two competitors and di¤erences in the rates
at which the input increases these margins.

Proposition 2. Welfare is highest when the input increment is allocated: (i) to …rm 1 if
5
m01 (k1 ) > m02 (k2 ) + 9t
[ m2 m1 ] [ m01 (k1 ) + m02 (k2 ) ]; and (ii) to …rm 2 if this inequality is
reversed.23

Proof. Lemma 1 implies that the rate at which equilibrium welfare (W CS + 1 + 2)

increases as more of the scarce input is allocated to …rm i (and so is not allocated to …rm
j 6= i) is:
@W @W 1 5 [ mi mj ]
= m0i (ki ) m0j (kj ) + m0i (ki ) + m0j (kj ) > 0 (3)
@ki @kj 2 18 t

5
, m0i (ki ) m0j (kj ) > [ mj mi ] m0i (ki ) + m0j (kj ) . (4)
9t

23
The increase in welfare is the same whether the input increment is allocated to …rm 1 or to …rm 2 when
m01 (k1 ) = m02 (k2 ) + 95t [ m2 m1 ] [ m01 (k1 ) + m02 (k2 ) ].

8
 Proposition 2 reports that when the input increases, say, …rm 1’s value margin more
rapidly than it increases …rm 2’s value margin (so m01 (k1 ) > m02 (k2 ) ), welfare tends to be
highest when a marginal increment of the input is awarded to …rm 1. This allocation tends
to increase consumers’surplus and/or …rm 1’s pro…t relatively rapidly, except when …rm 2’s
value margin substantially exceeds …rm 1’s value margin. In this event, …rm 2 serves many
more customers than …rm 1 serves in equilibrium. (Recall Lemma 1.) Consequently, welfare
can increase most rapidly when …rm 2 employs the marginal input increment to enhance the
value that its substantial customer base derives from its product and/or to reduce the unit
cost of serving this large customer base, even though the increment increases …rm 1’s value
margin more rapidly than it increases …rm 2’s value margin.

3 Comparing Equilibrium and Welfare-Maximizing Allocations

Proposition 3 identi…es the conditions under which an unfettered auction will, and will
not, generate the welfare-maximizing allocation of a marginal input increment.

Proposition 3. An unfettered auction will secure the welfare-maximizing allocation of an

input increment if the increment: (i) increases the value margins of the two …rms at the
same rate (i.e., if m01 (k1 ) = m02 (k2 )); (ii) increases the largest value margin most rapidly
(i.e., if m0i (ki ) > m0j (kj ) and mi > mj ); or (iii) increases most rapidly the value margin
9t 9t
that is the smallest by at least 5
j D j (i.e., if m0i (ki ) > m0j (kj ) and mi < mj 5
j D j for
some i; j 2 f1; 2g, j 6= i), where D [ m02 (k2 ) m01 (k1 ) ] = [ m01 (k1 ) + m02 (k2 ) ] .

If m01 (k1 ) > m02 (k2 ) and m1 m2 2 ( 95t D; 0 ), then …rm 1 will not win an auction
for the input increment even though welfare would be higher if it did win the auction. In
contrast, if m02 (k2 ) > m01 (k1 ) and m1 m2 2 ( 0; 95t D ), then …rm 1 will win an auction for
the input increment even though welfare would be higher if it did not win the auction.

To understand the conclusions in Proposition 3, recall from Proposition 2 that when

the input increment increases the value margins of the two …rms at the same rate, welfare

9
is highest when the input is allocated to the …rm with the highest value margin, because
this …rm serves the most customers in equilibrium. Therefore, because an auction awards
the increment to the …rm with the highest value margin (recall Proposition 1), the auction
ensures the welfare-maximizing allocation of the input increment in this case.
Also recall from Proposition 2 that welfare is maximized by allocating the input increment
to the …rm whose value margin increases most rapidly with the increment, provided the
…rm’s value margin is not too much smaller than its rival’s value margin. Consequently,
if m01 (k1 ) > m02 (k2 ), say, then the increment should be awarded to …rm 1 unless m1 is
su¢ ciently far below m2 . However, from Proposition 1, …rm 2 will win the auction for the
increment whenever m2 exceeds m1 . Consequently, when m01 (k1 ) > m02 (k2 ) and m2 m1
is strictly positive but relatively small, the auction will not secure the welfare-maximizing
allocation of the input increment.
Proposition 2 also reports that when the value margin of …rm 1, say, is su¢ ciently small,
welfare is maximized by allocating the input increment to …rm 2 even when m02 (k2 ) < m01 (k1 ),
so the input increases …rm 2’s value margin relatively slowly. Therefore, an auction secures
the welfare-maximizing input allocation by delivering the input increment to …rm 2 when
m01 (k1 ) > m02 (k2 ) if m1 is su¢ ciently far below m2 .
Figure 1 illustrates the conclusions drawn in Proposition 3. The Figure identi…es the
values of m1 m2 and D [ m02 (k2 ) m01 (k1 ) ] = [ m01 (k1 ) + m02 (k2 ) ] for which an
unfettered auction will, and will not, generate the welfare-maximizing allocation of the input
increment. Firm 1 wins the auction in the right-hand portion of the Figure where > 0, so
m1 > m2 . The resulting input allocation maximizes welfare when m01 (k1 ) > m02 (k2 ) (in the
southeast quadrant of Figure 1) and when m02 (k2 ) m01 (k1 ) > 0 is not too pronounced (in the
lower portion of the northeast quadrant). Similarly, …rm 2 wins the auction in the left-hand
portion of the Figure where m2 > m1 . The resulting input allocation maximizes welfare when
m02 (k2 ) > m01 (k1 ) (in the northwest quadrant of Figure 1) and when m01 (k1 ) m02 (k2 ) > 0
is not too pronounced (in the upper portion of the southwest quadrant).

10
 As consumers become less concerned with the horizontal dimensions of product quality
(i.e., as t declines),24 the solid line in Figure 1 rotates counterclockwise.25 Consequently, the
regions in the northeast and southwest quadrants of the …gure where an unfettered auction
does not secure the welfare-maximizing allocation of the input shrink.

Corollary 1. For a given D, the range of di¤erences between m1 and m2 (i.e., ) for
which an unfettered auction fails to implement the welfare-maximizing allocation of an input
increment declines as the products’horizontal di¤erentiation (i.e., t ) declines.

Proof. From Proposition 3, the set of values for which an unfettered auction fails to
implement the welfare-maximizing allocation of an input increment is: (i) 2 ( 95t D; 0 )
9t
when D < 0; and (ii) 2 ( 0; 5
D ) when D > 0. Both sets of values contract as t declines
because D does not vary with t (since m0i (ki ) is not a function of t ).26

To interpret Corollary 1, recall from Proposition 2 that the welfare-maximizing allocation

of an input is determined both by the di¤erence in the rates at which the input increases
the …rms’value margins (i.e., by j D j) and by the di¤erence in the levels of the …rms’value
margins (i.e., by j j). Technically, Corollary 1 holds because as t declines, the determination
of the welfare-maximizing allocation of the input is in‡uenced relatively more heavily by
the latter di¤erence (the di¤erence in value margins), which is the same di¤erence that
determines which …rm wins the auction for the input increment.27 Consequently, for a given
level of D, the welfare-maximizing allocation coincides with the allocation generated by the
auction for a broader range of values. Intuitively, Corollary 1 holds because a reduction in
t reduces the aggregate transportation costs that consumers incur as the asymmetry in the
24
To illustrate, as t declines, consumers of a wireless communications service would value relatively more
highly vertical dimensions of product quality like call clarity while a¤ording less value to horizontal di-
mensions of product quality such as handset color.
25 5
The equation of the solid line in Figure 1 is D = 9t , so the slope of the line increases as t declines.
26
The changes in t considered in Corollary 1 are those for which Assumption 1 always holds. The variations do
not include those that would cause the equilibrium market structure to change from duopoly to monopoly.
27
Formally, Proposition 2 indicates that the welfare-maximizing allocation of the input increment is de-
termined by the sign of 95t D . As t declines, the weight on D declines relative to the weight on
.

11
…rms’ market shares become more pronounced. Therefore, as t declines, welfare increases
more rapidly as the input is allocated to the …rm that serves the most customers, ceteris
paribus.
A …rm’s equilibrium pro…t increases as it secures more of the input, ceteris paribus. In
contrast, the pro…t of the rival duopolist declines. Consequently, an increased supply of
the input can reduce welfare if the …rm that acquires the input serves substantially fewer
customers than its rival. In this case, the reduction in the pro…t of the large competitor
can exceed the sum of the increase in the pro…t of the small competitor and the increase
in consumers’ surplus. Recall from Proposition 1, though, that the duopolist with the
smallest equilibrium market share never wins an unfettered auction for an input increment.
Consequently, unfettered auctions will avoid allocations of an input increment that reduce
welfare below the level that arises in the absence of the increment. This conclusion is stated
formally in Proposition 4, which refers to W0 , the level of welfare that prevails in the absence
of any input increment.

Proposition 4. Welfare declines below W0 as …rm i 2 f1; 2g acquires more of the input if
and only if …rm i’s value margin is su¢ ciently far below …rm j ( 6= i)’s value margin, i.e.,
9
mi < m j 5
t. Consequently, an unfettered auction will never allocate an input increment in
a manner that reduces welfare below the level that prevails in the absence of the increment.

Before considering extensions of our analysis, we brie‡y consider one possible policy re-
sponse to the fact that unfettered auctions do not necessarily ensure welfare-maximizing
allocations of inputs. In practice, policymakers sometimes grant bid credits to certain po-
tential bidders in order to encourage them to bid more aggressively for scarce inputs. When
…rm i is awarded a bid credit of bi 2 [ 0; 1 ) in a …rst-price auction, the …rm is only required to
pay Bi [ 1 bi ] for an input increment that it wins with a bid of Bi .28 Proposition 5 charac-
terizes the bid credit that ensures a …rst-price auction will generate the welfare-maximizing
28
See Ayres and Cramton (1996), Cramton et al. (2011), and Athey et al. (2013), for example, for additional
discussions and analyses of bid credits in auctions.

12
allocation of an input increment.

Proposition 5. Suppose m0i (ki ) > m0j (kj ) and mj mi 2 ( 0; 95t jD j ) for i; j 2 f1; 2g
(j 6= i). Then a …rst-price auction will ensure the welfare-maximizing allocation of the input
bi 3
increment if …rm i is awarded a bid credit, bi 2 ( 0; 1 ), that satis…es 2 bi
= 5
jD j.

Proof. Equation (1) implies that when …rm i receives bid credit bi 2 ( 0; 1 ) and …rm j (6= i)
receives no bid credit in a …rst-price auction, …rm i will outbid …rm j for the input increment
if:
1 1
[ 3 t + mi mj ] m0i (ki ) + m0j (kj ) > [ 3 t + mj mi ] m0i (ki ) + m0j (kj )
9 t [ 1 bi ] 9t
bi
, 3 t + mi mj > [ 1 bi ] [ 3 t + m j mi ] , mj mi < 3 t .
2 bi
bi 3
Therefore, when 2 bi
= 5
jDj, …rm i will secure the input if:
3 9t
mj mi < 3 t jD j = jD j . (5)
5 5
Expressions (4) and (5) imply that a …rst-price auction with the identi…ed bid credit will
ensure the welfare-maximizing allocation of the input increment.

Proposition 5 reports that in order to ensure a …rst-price auction generates the welfare-
maximizing allocation of an input increment, a bid credit can be awarded to the …rm with
a moderate value margin disadvantage when the input increment would increase its value
margin more rapidly than it would increase the rival’s value margin. The magnitude of
the bid credit should increase with the extent to which the input increases the …rm’s value
margin more rapidly than it enhances the rival’s value margin, ceteris paribus.29
In practice, bid credits often are awarded to competitors that serve relatively few retail
customers.30 Proposition 5 identi…es two ways in which such a policy can fail to ensure
29
Because 2 bibi is increasing in bi , the bid credit that ensures the auction will generate the welfare-
maximizing allocation of the input increment increases with jDj. Recall that jDj increases linearly with
m0i (ki ) m0j (kj ) , holding m0i (ki ) + m0j (kj ) constant, for i; j 2 f1; 2g (j 6= i).
30
Cramton et al. (2011, p. S171) observe that “The most common use of bidding credits has been in U.S.
spectrum auctions, where they are granted to small businesses.”

13
the welfare-maximizing allocation of an input increment. First, …rms that serve the fewest
retail customers may not be the …rms whose value margins increase most rapidly as they
acquire more of the input.31 Second, even when an input increment would increase the
value margin of a small competitor more than it would increase the value margin of a large
competitor, welfare can be highest when the increment is awarded to the large competitor
if its value margin (and thus its market share) su¢ ciently exceeds the value margin of the
small competitor.
It should also be noted that substantial information about prevailing industry conditions
is required to design the bid credits identi…ed in Proposition 5. One must know both the
rates at which the input increases the equilibrium value margins of the industry competitors
and the di¤erence between their equilibrium value margins. In practice, this information can
be di¢ cult, if not impossible, to obtain.32

4 Extensions
Before concluding, we brie‡y discuss two extensions of our model.

A. The Setting with Asymmetric Information

The …rst extension allows for incomplete information about a rival’s ultimate competi-
tive position at the time the input is auctioned. Speci…cally, we assume customer product
valuations (vi ) are common knowledge, but each …rm is privately informed about its pro-
duction cost (ci ) when the auction takes place.33 Each …rm learns its rival’s cost realization
after the auction has concluded, just before the two …rms set their prices simultaneously and
31
To illustrate, suppliers of a wireless communications service that serve many customers may face partic-
ularly severe spectrum shortages. Consequently, the value margins of these …rms may increase relatively
rapidly when they acquire additional spectrum.
32
Policies other than bid credits might be considered to enhance auction performance. Cramton et al. (2011)
discuss the potential roles of restrictions on auction participation and limits on the amount of the input
that particular competitors can acquire in the auction. Milgrom (2004) and Bhattacharya et al. (2014),
among others, note the potential bene…ts of limiting the number of bidders in settings where potential
bidders must incur a cost in order to learn their valuations of the objects being auctioned. Of course,
considerable information typically is required to ensure that these policies generate welfare-maximizing
input allocations.
33
For expositional simplicity, we generally suppress the dependence of vi , ci , and mi on ki in the ensuing
discussion.
14
non-cooperatively.34 For simplicity, we assume ci 2 fciL ; ciH g where ciL < ciH for i 2 f1; 2 g.
To illustrate most readily the central new consideration that arises in this setting, suppose
the input increases each …rm’s value margin at the same rate, regardless of the …rm’s realized
cost.35 It can be shown that the …rm that wins the auction for the input increment in this
case is the …rm with the largest expected value margin di¤erential. Formally, …rm i 2 f1; 2 g
will win the auction for the input increment when its cost is cis (and so its value margin is
mis = vi cis ) and …rm j’s cost is cjz (and so its value margin is mjz = vj cjz ) if:

sL sH Lz Hz
mis mjL + mjH > mjz miL + miH , (6)
s s z z

where sz 2 (0; 1) is the probability that ci = cis and cj = cjz , s sL + sH , and

z Lz + Hz for s; z 2 fL; H g.36

Suppose …rm i has a cost (and margin) advantage in the sense that miL > mjL and
miH > mjH , and thus L miL + H miH > L mjL + H mjH . Then expected welfare is
highest when the input increment is awarded to …rm i. Firm i will win the auction for the
increment if …rm i knows it has a higher margin than …rm j and …rm j knows it has a lower
margin than …rm i. This will be the case, for instance, if: (i) mi = miL and mj = mjH (so
…rm i knows it has its highest margin and …rm j knows it has its smallest margin); or (ii)
miH mjL (so …rm i’s smallest margin is at least as large as …rm j’s largest margin).
More generally, though, …rm j may win the auction for the input increment when ex-
pected welfare (and even ex post welfare) would be higher if …rm i received the increment.
This outcome will arise when the inequality in expression (6) is reversed, so …rm j’s expected
value margin advantage exceeds …rm i’s expected value margin advantage even though …rm
i’s expected value margin (and its actual value margin) exceeds …rm j’s corresponding mar-
34
This information structure implies that the …rms will not alter their bidding strategies in an attempt
to signal (or conceal) their private cost information. See Brocas (2013a) for a related analysis in which
signaling considerations arise.
35
Recall from Proposition 3 that, in the presence of complete information, the auction ensures the welfare-
maximizing allocation of the input in this case.
36
See the Appendix for a proof of this conclusion and proofs of the other conclusions drawn in the ensuing
discussion.

15
gin(s). For example, suppose mjH = 1, miH = 2, mjL = 3, miL = 4, LL = :03, LH = :11,

HL = :60, and HH = :26. Because HL is large relative to LH in this setting, …rm j

believes it has a value margin advantage over …rm i when mj = mjL < mi = miL , and so
…rm j will outbid …rm i for the input increment.37
This example illustrates two more general conclusions. First, the same considerations that
determine when an auction will ensure the welfare-maximizing allocation of an input in the
presence of complete information continue to be relevant in its absence. Second, incomplete
information introduces additional considerations that can promote some variation in the
qualitative conclusions drawn above.

B. The Triopoly Setting

The second extension admits competition among three …rms that are located on the
circumference of a circle (e.g., Salop, 1979). Consumers are distributed uniformly on this
circumference and incur unit transportation cost t when traveling to purchase the product
from any supplier. It can be shown that when the input increases the value margins of the
three competitors at the same rate (so m01 (k1 ) = m02 (k2 ) = m03 (k3 )) in this triopoly setting
with complete information, a …rm will acquire the input increment in an unfettered auction
precisely when welfare is highest if it does so.38 Thus, the natural counterpart to conclusion
(i) in Proposition 3 holds in the triopoly setting.
However, the triopoly setting introduces an important asymmetry that does not arise in
the duopoly setting.39 When …rm 1 acquires the input increment in the duopoly setting, it
37
An auction may fail to generate the allocation of the input increment that maximizes expected welfare
even when the …rms’conditional beliefs are symmetric. To illustrate, suppose " 2 (0; 1) is the conditional
probability that mj = mjs when mi = mis for i; j 2 f1; 2g (j 6= i) and s 2 fL; Hg. Then inequality
(6) implies that …rm j will win the auction for the input increment when mj = mjL > miH = mi if
[2 "] [ mjL miH ] > "[ miL mjH ]. This inequality will hold even when …rm i’s expected value margin
exceeds …rm j’s expected value margin (and so expected welfare is highest when the input increment is
awarded to …rm i, not …rm j) if, for example, mjH = 1, miH = 2, mjL = 3, miL = 4, LL = HH = :2,
and LH = HL = :3 (so " = :4).
38
See Mayo and Sappington (2015) for detailed proofs of the conclusions drawn in the ensuing discussion.
39
The triopoly setting also introduces the possibility of free-riding by individual suppliers. McAfee (1998)
demonstrates that when multiple industry suppliers do not face capacity constraints, each supplier may be
unable to capture the full increase in the pro…t of the unconstrained suppliers that arises when additional
capacity is withheld from capacity-constrained suppliers. The relatively small foreclosure value that each
16
necessarily precludes its only rival, …rm 2, from acquiring the increment. Firm 1 thereby
derives: (i) a marginal use value that is proportional to its equilibrium price-cost margin
(p1 c1 ) and m01 (k1 ); and (ii) a marginal foreclosure value that is proportional to p1 c1 and
m02 (k2 ), for a combined marginal value that is proportional to [ p1 c1 ] [ m01 (k1 ) + m02 (k2 ) ].
Similarly, …rm 2 derives a marginal value from the input increment in the duopoly setting
that is proportional to [ p2 c2 ] [ m02 (k2 ) + m01 (k1 ) ]. The symmetry in each rival’s perceived
gain from enhancing its own value margin and precluding its only rival from enhancing its
value margin implies that the …rm that secures the input increment at auction is the …rm
with the highest equilibrium price-cost margin, which is the …rm with the highest value
margin.
When …rm i secures the input increment at auction in the triopoly setting, though, the
…rm derives no foreclosure value with regard to the rival that would not have obtained the
input even if …rm i had not secured the increment. The resulting asymmetry in the sum of
marginal use and foreclosure values implies that the allocation of the input in the triopoly
setting typically will depend upon both the relative magnitudes of the …rms’value margins
and the relative rates at which these value margins increase as a …rm acquires more of the
critical input.
To illustrate this more general conclusion,40 suppose the …rms are equally spaced around
the circle, …rm 1’s value margin exceeds the value margins of its symmetric rivals (i.e.,
m1 > m2 = m3 ), and the input increases …rm 1’s value margin more slowly than it increases
the value margins of …rm 1’s rivals (i.e., m01 (k1 ) < m02 (k2 ) = m03 (k3 )). When m1 m2 is
large relative to m02 (k2 ) m01 (k1 ), …rm 1 will value the input increment most highly, and so
will be the …rm whose access to the increment is foreclosed should …rm 2 or …rm 3 submit
the highest bid for the input increment. It can be shown that …rm 1 will win the auction

unconstrained supplier anticipates from securing additional capacity at auction implies that the capacity-
constrained …rms (e.g., new industry entrants) may secure auctioned capacity increments.
40
See Jehiel and Moldovanu (2000), for example, for further discussion of the complications that can arise
in designing auctions in the presence of externalities with more than two buyers.

17
 5t 12[ m1 m2 ]
for the input increment under these conditions if m01 (k1 )=m02 (k2 ) > 5t+15[ m1 m2 ]
rf .
Therefore, the identity of the …rm that wins the auction depends upon both relative value
margins and relative rates at which value margins vary with the input. It can be veri…ed
that welfare is highest when …rm 1 acquires the increment under the speci…ed conditions
25t 24[ m1 m2 ]
if m01 (k1 )=m02 (k2 ) > 25t+48[ m1 m2 ]
> rf . Therefore, as in the duopoly setting, …rm 1’s
relatively high value margin may lead it to acquire the input increment when m01 (k1 ) is
relatively low even though welfare would be higher if one of the …rm’s rivals acquired the
input increment.

5 Conclusions
We have shown that unfettered auctions tend to ensure the welfare-maximizing allocation
of a scarce input when the input increases the value margins of the competing suppliers
symmetrically. However, auctions can fail to allocate scarce inputs so as to maximize welfare
when the input increases relatively rapidly the value margin of a …rm that serves a moderately
small share of the market.
Our …ndings suggest that the insights from the foreclosure literature may require some
modi…cation when considering settings where inputs have foreclosure value but cannot be
employed to fully exclude competitors. We have found that when two …rms compete in
such a setting, the competitor with the larger market share will win an unfettered auction
for an input increment, as the foreclosure literature might suggest. However, the resulting
allocation of the input increment will increase welfare when the increment increases the value
margin of the large …rm at least as rapidly as it increases the value margin of the smaller …rm.
Therefore, when predicting the welfare implications of input allocations or when designing
policies that a¤ect the allocation of scarce inputs, it is important to assess the levels of
prevailing value margins (and associated market shares), the relative rates at which …rms’
value margins change as they acquire more of the input, and whether the amount of the
input being auctioned is su¢ ciently large to admit complete foreclosure of a rival.
Our …ndings imply that, in principle, policies that favor small …rms in securing access
18
to scarce inputs can enhance welfare when additional units of the input would increase
the value margins of (moderately) small …rms relatively rapidly. However, in practice, the
design and implementation of such policies are problematic for at least two reasons. First,
detailed information about the value of the input to individual industry suppliers is required
to know when favoring selected suppliers will enhance welfare. Such information typically
is di¢ cult, if not possible, for policymakers to obtain. Second, even if policymakers are
somehow able to secure the requisite information, implementing the appropriate favoritism
can be challenging.41
The Pioneer’s Preference program that the U.S. Federal Communications Commission
(FCC) established in 1991 illustrates the practical di¢ culties that can arise in attempting
to identify the value of a scarce input –spectrum –to individual entities. The program pro-
vided “a means of extending preferential treatment in the FCC’s licensing process to parties
that demonstrated their responsibility for developing new spectrum-using communications
services and technologies”(FCC, 2000).42 Between 1991 and 1997, the Commission received
more than 1,500 applications for a pioneer’s preference. The …ve such preferences the Com-
mission awarded ultimately enjoyed “only limited deployment” whereas many technologies
that were not awarded a pioneer’s preference ultimately achieved widespread commercial
application (Fusco, 2000).43
The FCC’s experience with spectrum auctions illustrates the problems that can arise
when attempting to favor particular industry competitors. The Commission’s spectrum
auction in 1996 provided special …nancing (i.e., relatively small down payments and low
interest rates) to small …rms. Although the special …nancing may have helped small …rms
acquire spectrum in the auction, several of the …rms either failed to pay for the spectrum they

41
As Cramton (2002, p. 635) observes, “Gauging the right level of set-asides or bidding credits is extremely
di¢ cult. Also, it is nearly impossible to target the favor to the desired group.”
42
See Singh (1996) for additional discussion of the purpose and implementation of the FCC’s Pioneer’s
Preference program.
43
QUALCOMM successfully sued the FCC for failing to award the company a pioneer’s preference for its
code division multiple access (CDMA) technology (Fusco, 2000).

19
won or subsequently failed to supply services using the spectrum. In both cases, the favored
treatment of small …rms led to substantial, costly delays in utilizing the scarce spectrum
(Congressional Budget O¢ ce, 2005). More recently, two …rms that received bidding credits
due to their small size outbid other …rms for large amounts of spectrum. Due to their
status as “very small …rms,” these entities received a 25 percent ($3.25 billion) discount
on the spectrum they secured at auction. It was later discovered that Dish Network (a
company with nearly $14.5 billion in revenue in 2014) was the majority owner of these
…rms (Solomon, 2015; Ayotte and Pai, 2015). Experiences like these illustrate the practical
di¢ culties associated with designing and implementing policies that can improve upon the
performance of unfettered auctions.
Attempts to favor particular bidders for scarce inputs also can con‡ict with other policy
objectives. In particular, such favoritism can reduce the revenue derived from auctions of
scarce inputs.44 Policies that favor particular suppliers on the basis of endogenous character-
istics also can invite welfare-reducing strategic behavior. For instance, if favorable treatment
is a¤orded to …rms with small market shares and high marginal valuations of the input, then
…rms may …nd it pro…table to reduce their market shares (perhaps by reducing the service
quality they deliver to their customers) and to in‡ate their marginal valuations of the input
(perhaps by installing relatively few substitute inputs). The welfare implications of such
strategic behavior warrant careful study.45
We close by mentioning three extensions of our analysis. First, we have only considered
marginal input increments whereas increments of substantial magnitude often are auctioned
in practice. Standard foreclosure concerns can emerge when large input increments are

44
To illustrate, in evaluating the FCC’s policy of awarding preferences to small bidders in auctions for
spectrum, the Congressional Budget O¢ ce (2005, preface) concludes, “partly because of their potentially
less favorable commercial prospects, small bidders may not pay as much at auction for their licenses as
larger bidders pay. As a result, by o¤ering preferences at auction, the government may forgo auction
receipts otherwise available to it.”
45
Mayo and Sappington (2015) extend the present analysis to settings where …rms can supply costly e¤ort
to enhance their value margins. In such settings, the identity of the …rm that secures an input increment
via auction can vary with both prevailing value margins and the manner in which the input alters the
marginal productivities of the …rms’e¤orts.

20
auctioned. Furthermore, the welfare-maximizing allocation of the increments may vary with
such factors as the magnitudes of the increments and the relative post-auction (as well as
pre-auction) market shares of the industry suppliers.
Second, although our analysis has focused on the role of auctions in ensuring the welfare-
maximizing allocation of scarce inputs, policymakers may pursue other objectives in practice.
It can be shown that the key qualitative conclusions drawn above continue to hold if the
social objective is to maximize consumers’surplus rather than welfare. However, there is a
broader set of conditions under which an auction fails to implement the input allocation that
maximizes consumers’surplus.46 Other social objectives are also possible. For example, as
suggested above, policymakers may value the revenue derived from the sale of inputs and/or
seek to promote industry participation by small businesses and minority business owners.47
The potential for auctions to allocate inputs e¢ ciently and achieve these alternative objec-
tives awaits additional research.
Third, although our analysis has focused on the extent to which auctions ensure the
welfare-maximizing allocation of scarce inputs, the basic forces at play in our model likely
are relevant more generally. To illustrate, the expenditures competitors devote to securing
patents on technologies that enhance product quality or reduce production costs seem likely
to be driven largely by the same relative value margin considerations that are central in
our analysis. Furthermore, welfare-maximizing expenditures in these settings seem likely to
re‡ect both relative value margins and the relative rates at which the patented technologies
would increase the value margins of industry competitors. Explicit investigation of such
related considerations awaits further research.

46
Mayo and Sappington (2015) show that if m01 (k1 ) > m02 (k2 ) and m1 m2 2 ( 9 t D; 0 ), then …rm 1 will
not win an auction for the input increment even though consumers’surplus would be higher if it did win
the auction. In contrast, if m02 (k2 ) > m01 (k1 ) and m1 m2 2 ( 0; 9 t D ), then …rm 1 will win an auction
for the input increment even though consumers’surplus would be higher if it did not win the auction.
47
Cramton et al. (2011, p. S169) report that they “consider the primary goal of the regulator to be economic
e¢ ciency.”However, the authors also note other goals of spectrum auctions, including revenue generation.

21
 Appendix
This Appendix provides the proofs of formal conclusions that were not proved in the
text. This Appendix also proves the conclusions cited in section 4A and outlines the proofs
of the conclusions drawn in section 4B.48

Proof of Lemma 1.
b 2 [ 0; 1 ] ) of the consumer who is indi¤erent between purchasing the
The location ( x
product from the two …rms when consumers place value vi on …rm i 2 f1; 2g’s product and
when …rm i charges price pi for its product is determined by:
1
v1 p 1 t x b = v2 p2 t [ 1 x b] ) x b = [ t + v1 p1 (v2 p2 ) ] . (7)
2t
(7) implies that …rm 1’s pro…t is:
1
1 = [ p1 b =
c1 ] x [ p1 c1 ] [ t + v1 p1 (v2 p2 ) ] . (8)
2t
(8) implies that …rm 1’s pro…t-maximizing price is determined by:
@ 1 1
= 0 , p1 = [ t + c1 + v1 v2 + p2 ] . (9)
@p1 2
Corresponding calculations reveal that …rm 2’s pro…t-maximizing price is determined by:
1
p2 = [ t + c2 + v2 v1 + p1 ] . (10)
2
(9) and (10) imply that the prices the …rms will charge in equilibrium and the …rms’corre-
sponding price-cost margins are, for i; j 2 f1; 2g (j 6= i):
1 1
pi =[ 3 t + vi vj + 2 ci + cj ] ) p i ci = [ 3 t + mi mj ] . (11)
3 3
These prices are readily employed to show that the equilibrium outputs of the …rms are:
1
xi = [ 3 t + mi mj ] for i; j 2 f1; 2g (j 6= i). (12)
6t
(11) and (12) are readily employed to show that the …rms’equilibrium pro…ts (given v1 ,
v2 , c1 , and c2 ) are:
1
i = [ pi ci ] xi = [ 3 t + mi mj ]2 for i; j 2 f1; 2g (j 6= i). (13)
18 t
Let m1 m2 denote the di¤erence between the value margins of …rms 1 and 2.
Also let 1 + 2 denote equilibrium industry pro…t. Then (13) provides:
2
1 1
= [3t + ]2 + [3t ]2 = t + . (14)
18 t 18 t 9t
(11) and (12) imply that equilibrium consumers’surplus is:
48
See Mayo and Sappington (2015) for additional detail.

22
 Zx1 Z1
CS = [ v1 p1 t x ] dx + [ v2 p2 t (1 x) ] dx
0 x1

m1 m2 5t 5 2 1 2
5t
= [3t + ]+ [3t ] = [ m1 + m2 ] + . (15)
6t 6t 4 36 t 2 36 t 4

Proof of Proposition 3.
Propositions 1 and 2 imply that when m01 (k1 ) = m02 (k2 ), welfare increases most rapidly
when the input increment is allocated to the …rm with the highest value margin, which
an unfettered auction ensures. The propositions also imply that: (1) …rm 1 will win the
auction for the input increment if > 0; and (2) welfare is highest when …rm 1 wins the
9t
auction if > 5 D. Therefore, the auction ensures the welfare-maximizing allocation of
the input increment if m01 (k1 ) > m02 (k2 ) and either m1 > m2 (so …rm 1 wins the auction)
or m1 < m2 95t j D j (so …rm 2 has a substantial value margin advantage and wins the
auction). The auction also ensures the welfare-maximizing allocation of the input increment
9t
if m02 (k2 ) > m01 (k1 ) and either m2 > m1 (so …rm 2 wins the auction) or m2 < m1 5
jDj
(so …rm 1 has a substantial value margin advantage and wins the auction).

From Proposition 1, …rm 2 wins the auction if < 0. From Proposition 2, welfare is
9t
highest when …rm 1 wins the auction if > 5 D < 0. Therefore, …rm 1 does not win
the auction when 2 ( 95t D; 0 ), even though welfare would be higher if …rm 1 did win the
auction.
From Proposition 1, …rm 1 wins the auction if > 0. From Proposition 2, welfare is
9t
highest when …rm 2 wins the auction if < 5 D > 0. Therefore, …rm 1 wins the auction
9t
when 2 ( 0; 5 D ), even though welfare would be highest if …rm 2 won the auction.

Proof of Proposition 4.
From (14) and (15), equilibrium welfare is:
1 5 2 t
W = CS + = [ m1 + m2 ] + (16)
2 36 t 4
@W 1 9
) = m01 (k1 ) [ 9 t + 5 ] T 0 , T t.
@k1 18 t 5

The proof that @W

@k2
T 0 , m2 T m1 9
5
t is analogous, and so is omitted.
@W
(16) and Proposition 1 imply that if > 0, then …rm 1 wins the auction and @k1
=
1 5
m01 (k1 ) 2
+ 18 t
> 0. (16) and Proposition 1 imply that if < 0, then …rm 2 wins
23
 @W 1 5 @W
the auction and @k2
= m02 (k2 ) 2 18 t
> 0. If = 0, then (16) implies that @ki
=
1
2
m0i (ki ) > 0 for i = 1; 2.

The Setting with Asymmetric Information

Let i (ci ; cj ) denote …rm i’s pro…t when its cost is ci and …rm j’s cost is cj . Also let
E i (ci ) denote …rm i’s ex ante expected pro…t when its cost is ci . Then Lemma 1 implies
that for i; j 2 f1; 2 g (j 6= i) and for s; z 2 fL; H g (z 6= s):
1
i (ci ; cj ) = [ 3 t + m i m j ]2
18 t
1
) E i (cis ) = ss
[ 3 t + mis mjs ]2 + sz
[ 3 t + mis mjz ]2 . (17)
18 t s s

Di¤erentiating (17) provides:

@E i (cis ) 1
= f ss [ 3 t + mis mjs ] m0is (ki ) + sz [ 3 t + mis mjz ] m0is (ki ) g (18)
@ki 9t s

and
@E i (cis ) 1
= ss [ 3 t + mis mjs ] m0js (kj ) + sz [ 3 t + mis mjz ] m0jz (kj ) .
@kj 9t s
(19)
(18) and (19) imply that the rate at which …rm i’s expected pro…t increases as it secures
more of the input (and so …rm j secures less of the input) when …rm i has cost cis is, for
z 6= s:
@E i (cis ) @E i (cis )
Bi (cis ) =
@ki @kj

1
= ss [ 3 t + mis mjs ] m0is (ki ) + m0js (kj )
9t s

+ sz [ 3 t + mis mjz ] m0is (ki ) + m0jz (kj ) . (20)

(18) and (19) also imply that the rate at which …rm j’s expected pro…t increases as it secures
more of the input when …rm j has cost cjz is, for s 6= z:

@E j (cjz ) @E j (cjz )
Bj (cjz ) =
@kj @ki

1
= zz [ 3 t + mjz miz ] m0jz (kj ) + m0iz (ki )
9t z

+ sz [ 3 t + mjz mis ] m0jz (kj ) + m0is (ki ) . (21)

24
 (20) and (21) imply that …rm i will win the auction for the input increment when …rm
i’s cost is cis and …rm j’s cost is cjz if, for s; y; z; w 2 fL; Hg (y 6= s and w 6= z), Bi (cis ) >
Bj (cjz ) ,
ss sy
[ mis mjs ] m0is (ki ) + m0js (kj ) + [ mis mjy ] m0is (ki ) + m0jy (kj )
s s

zz wz
> [ mjz miz ] m0jz (kj ) + m0iz (ki ) + [ mjz miw ] m0jz (kj ) + m0iw (ki ) . (22)
z z

Equation (6) in the text follows from (22).

Let W (ci ; cj ) denote welfare when …rm i’s cost is ci and …rm j’s cost is cj . Also let E W
denote ex ante expected welfare. Then Lemma 1 implies that for i; j 2 f1; 2 g (j 6= i) and
for s; z 2 fL; H g:
mi mj t [ mi m j ]2
W (ci ; cj ) = [ 3 t + mi mj ] + [ 3 t + mj mi ]
6t 6t 4 36 t
H X
X H nm mjz
is
) EW = sz [ 3 t + mis mjz ] + [ 3 t + mjz mis ]
s=L z=L
6t 6t
)
2
[ mis mjz ] t
. (23)
36 t 4
Di¤erentiating (23) provides:

@E W XH X H
1 mis mjz
= sz [ 3 t + 2 mis 2 mjz ] m0is (ki ) m0is (ki )
@ki s=L z=L
6t 18 t

1 X X
H H
= sz [ 9 t + 5 mis 5 mjz ] m0is (ki ), and (24)
18 t s = L z = L

@E W XH X H
1 mis mjz
= sz [ 3 t + 2 mjz 2 mis ] m0jz (kj ) + m0jz (kj )
@kj s=L z=L
6t 18 t

1 X X
H H
= sz [ 9 t + 5 mjz 5 mis ] m0jz (kj ). (25)
18 t s = L z = L

(24) and (25) imply that the rate at which expected welfare increases as the input increment
is awarded to …rm i (and not to …rm j) is:
1 X X
H H
@E W @E W
= sz f [ 9 t + 5 mis 5 mjz ] m0is (ki )
@ki @kj 18 t s = L z = L
[ 9 t + 5 mjz 5 mis ] m0jz (kj ) g . (26)
25
Assumption A1. m0is (ki ) = m0iz (ki ) > 0 and m0is (ki ) = m0js (kj ) > 0 for i; j 2 f1; 2g (j 6= i)
and s; z 2 fL; Hg (z 6= s).

Observation 1. Suppose Assumption A1 holds. Then expected welfare increases most

rapidly when the input is awarded to the …rm with the highest expected value margin.

Proof. (26) implies that when assumption A1 holds, expected welfare increases more rapidly
when the input is awarded to …rm i than when it is awarded to …rm j if:
@E W @E W 5 XH XH
0
= m (ki ) sz [ mis mjz ] > 0
@ki @kj 9 t is s=L z=L

, L miL + H miH > L mjL + H mjH .

Observations 2 –4 consider the case where mis > mjs for i; j 2 f1; 2g (j 6= i) and s 2 fL; Hg.

Observation 2. Suppose Assumption A1 holds, ci = ciL , and cj = cjH . Then …rm i will
win the auction for the input increment.

Proof. From (22), …rm i will win the auction for the input increment in this setting if:
LL LH HH LH
[ miL mjL ] + [ miL mjH ] > [ mjH miH ] + [ mjH miL ]
L L H H

LH LH HH LL
, [ miL mjH ] + > [ mjH miH ] [ miL mjL ]
H L H L

, [ miL mjH ] LH [ LH + LL + LH + HH ]

> HH [ LL + LH ] [ mjH miH ] LL [ HH + LH ] [ miL mjL ]

HH L LL H
, miL mjH > [ mjH miH ] [ miL mjL ] .
LH [ L + H ] LH [ L + H ]

The left-hand side of this inequality is strictly positive and the right-hand side is strictly
negative, so the inequality holds.

Observation 3. Suppose Assumption A1 holds and miH mjL . Then …rm i will win the
auction for the input increment.

Proof. (22) implies that …rm i will win the auction for the input increment when Assumption
A1 holds if, for s; z 2 f L; H g ( z 6= s ):

ss sz zz sz
mis mjs + mjz > mjz miz + mis
s s z z
26
 zz sz ss sz
, mis + miz + mis > mjz + mjs + mjz . (27)
z z s s

The inequality in (27) holds because miL > miH mjL > mjH .

Observation 4. Suppose Assumption A1 holds, ci = ciL , cj = cjL , and mjL > miH .
Then there are conditions under which …rm j will win the auction even though both ex ante
expected welfare and ex post welfare would be greater if the input increment were allocated to
…rm i.

Proof. Suppose mjH = 1, miH = 2, mjL = 3, miL = 4, LL = :03, LH = :11, HL = :60,

and HH = :26. In this setting, ex post welfare is highest when the input increment is
allocated to …rm j because mjL > miH and Assumption A1 holds. Firm i has a higher
expected value margin than …rm j in this setting because:

L miL + H miH = :14 [ 4 ] + :86 [ 2 ] = :56 + 1:72 = 2:28

> 2:26 = 1:89 + :37 = :63 [ 3 ] + :37 [ 1 ] = L mjL + H mjH .

From (22), …rm j will win the auction for the input increment in this setting when mi = miL
and mj = mjL if:

LL LH HL LL
[ miL mjL ] + [ miL mjH ] < [ mjL miH ] + [ mjL miL ]
L L L L

2 LL HL LH
, [ miL mjL ] < [ mjL miH ] [ miL mjH ]
L L L

LH HL
, miL mjL + [ miL mjH ] < [ mjL miH ]
2 LL 2 LL

:11 :60
, 4 3+ [4 1] < [3 2] , 1 + 5:5 < 10 .
:06 :06

The Triopoly Setting

Let xi denote the location of …rm i 2 f1; 2; 3 g on a circle with unit circumference, where
0 = x1 < x2 < x3 < 1. Also let xij denote the location of the consumer who is indi¤erent
between purchasing from …rm i and …rm j. It is readily veri…ed that:
1 1
xij = [ xi + xj ] + [ vi vj + pj pi ] . (28)
2 2t
These market boundaries permit a speci…cation of each …rm’s pro…t as a function of
prevailing prices. When each …rm chooses its price to maximize its pro…t, given the prices
set by its competitors, the resulting equilibrium prices are, for i; j; y 2 f1; 2; 3g (j 6= i and
27
y 6= i):
1
pi = [ 2 v i v j v y + 3 c i + c j + c y + t ( 1 + Li ) ] , (29)
5
where Li denotes the sum of the distances between …rm i and each of its rivals along the
circle circumference. (28) and (29) can be employed to demonstrate that equilibrium pro…ts
are, for i; j; y 2 f1; 2; 3g (j 6= i and y 6= i):
1
i = [ 2 m i m j m y + t ( 1 + Li ) ] 2 . (30)
25 t
@ i @ i @ i
Let Bi @ki ij @kj iy @ky denote the rate at which …rm i’s equilibrium pro…t
increases as it acquires the input increment and thereby precludes …rm j from acquiring
the increment with probability ij and precludes …rm y from acquiring the increment with
probability iy . (30) implies:
2
Bi = [ 2 mi mj my + t ( 1 + Li ) ] 2 m0i (ki ) + ij m0j (kj ) + iy m0y (ky ) . (31)
25 t
Straightforward but tedious calculations reveal that welfare in this setting is:
Z1 Zx12 Zx2
W = [ v1 c1 t ( 1 x ) ] dx + [ v1 c1 t x ] dx + [ v2 c2 t ( x2 x ) ] dx
x31 0 x12
Zx23 Zx3 Zx31
+ [ v2 c2 t(x x2 ) ] dx + [ v3 c3 t ( x3 x ) ] dx + [ v3 c3 t(x x3 ) ] dx
x2 x23 x3

1
= 8 (m1 )2 + 8 (m2 )2 + 8 (m3 )2 8 m1 m2 8 m1 m3 8 m2 m3
25 t
+ t [ 11 m1 + 3 m2 + 11 m3 + 8 m1 x2 8 m1 x3 + 8 m2 x3 8 m3 x2 ]
t 1
+ 56 x2 x3 + 44 (x2 )2 + 48 (x3 )2 + 50 x3 + 56 x3 + 23 t + (x2 )2 + (x3 )2 . (32)
100 2

Assumption A2. m01 (k1 ) = m02 (k2 ) = m03 (k3 ) m0i (ki ), a strictly positive constant.

@W @W @W
Let Gi @ki ij @kj iy @ky denote the rate at which equilibrium welfare increases as
the input increment is awarded to …rm i, which precludes …rm j from acquiring the increment
with probability ij and precludes …rm y from acquiring the increment with probability iy .
Then (31) and (32) imply that when Assumption A2 holds:

12 0 1 t
Bi = m (ki ) mi ( mj + my ) + ( 1 + Li ) , and (33)
25 t i 2 2

24 0 t
Gi = m (ki ) mi ( ij mj + iy my ) + ( Li [ ij Lj + iy Ly ] ) . (34)
25 t i 3
28
Case 1. m1 + 3t L1 > m2 + 3t L2 > m3 + 3t L3 and Assumption A2 holds.
It can be veri…ed that B1 > B2 > B3 in Case 1. Therefore, when …rm 1 secures the input
increment at auction, it precludes …rm 2 (not …rm 3) from acquiring the increment, and so
12 = 1 and 13 = 0. Also, if …rm 2 were to increase its bid to the point where it won the
auction for the input increment, it would preclude …rm 1 (not …rm 3) from acquiring the
increment, and so 21 = 1 and 23 = 0. Similarly, if …rm 3 were to increase its bid to the
point where it won the auction for the input increment, it would preclude …rm 1 (not …rm
2) from acquiring the increment, and so 31 = 1 and 32 = 0.
(33) and (34) then imply that …rm 1 will win an unfettered auction for the input increment
and welfare is highest when …rm 1 wins the auction in Case 1. Analogous arguments reveal
that the same is true when m1 + 3t L1 > m3 + 3t L3 > m2 + 3t L2 . Corresponding conclusions
for settings in which …rm 2 or …rm 3 wins the auction for the input increment imply that
when Assumption A2 holds, a …rm wins the auction for an input increment if and only if
welfare is highest when the …rm wins the auction.
Case 2. x2 = 13 , x3 = 23 , m1 > m2 = m3 , m01 (k1 ) < m02 (k2 ) = m03 (k3 ) .
Suppose that m1 m2 is su¢ ciently large relative to m02 (k2 ) m01 (k1 ) that 21 = 31 =1
and 23 = 32 = 0 in this case. Then it is readily veri…ed that:

2 5
B1 = 2 ( m1 m2 ) + t [ 2 m01 (k1 ) + m02 (k2 ) ] ; (35)
25 t 3

2 5
B2 = B3 = m2 m1 + t [ 2 m02 (k2 ) + m01 (k1 ) ] ; (36)
25 t 3

1 25 25
G1 = 16 ( m1 m2 ) + t m01 (k1 ) 8 ( m2 m1 ) + t m02 (k2 ) ; (37)
25 t 3 3
and
G2 = G3 = G1 . (38)

(35) and (36) imply that …rm 1 will win the auction if:

5 5
2 ( m1 m2 ) + t [ 2 m01 (k1 ) + m02 (k2 ) ] > m2 m1 + t [ 2 m02 (k2 ) + m01 (k1 ) ]
3 3

5 5
, m01 (k1 ) 5 ( m1 m2 ) + t > m02 (k2 ) 4 ( m2 m1 ) + t
3 3
5
m01 (k1 ) 3
t 4 [ m1 m2 ] 5 t 12
, > 5 = rf . (39)
m02 (k2 ) 3
t + 5 [ m1 m2 ] 5 t + 15

(37) and (38) imply that welfare is highest when …rm 1 acquires the input increment if:

G 1 > G2 = G 3 = G1 , G1 > 0 ,
29
 25 25
16 ( m1 m2 ) + t m01 (k1 ) > 8 ( m2 m1 ) + t m02 (k2 )
3 3
25
m01 (k1 ) 3
t 8 [ m1 m2 ] 25 t 24
, > 25 = rw . (40)
m02 (k2 ) 3
t+ 16 [ m1 m2 ] 25 t + 48

Observation. rw > rf .

Proof. From (39) and (40):

rw > rf , [ 25 t 24 ] [ 5 t + 15 ] > [ 25 t + 48 ][5t 12 ]

2 2
, 25 [ 15 ] t 24 [ 5 ] t 24 [ 15 ] > 25 [ 12 ] t + 48 [ 5 ] t 48 [ 12 ]

2 2
, [ 375 120 + 300 240 ] t > [ 360 576 ] , 315 t > 216 .

30
 ∆
𝑚2 > 𝑚1 𝑚1 = 𝑚2 𝑚1 > 𝑚2

Firm 1 wins the auction,

but welfare is higher
𝑚2′ > 𝑚1′ when firm 2 wins 9𝑡
∆= 5
𝐷

The auction maximizes

welfare by allocating The auction maximizes
𝐷 𝑚2′ = 𝑚1′ the input to firm 2 welfare by allocating
the input to firm 1

𝑚2′ < 𝑚1′

Firm 2 wins the auction,
but welfare is higher
when firm 1 wins

Figure 1. Outcomes under Unfettered Input Auctions.

 References
Ase¤, J. and Chade, H. “An Optimal Auction with Identity-Dependent Externalities.”
RAND Journal of Economics, Vol. 39 (2008), pp. 731-746.

Athey, S., Coey, D. and Levin, J. “Set-Asides and Subsidies in Auctions.” American Eco-
nomic Journal: Microeconomics, Vol. 5 (2013), pp. 1-27.

Ayotte, K. and Pai, A. “Ending Welfare for Telecom Giants.”Wall Street Journal, February
4, 2015.

Ayres, I. and Cramton, P. “De…cit Reduction through Diversity: A Case Study of How
A¢ rmative Action at the FCC Increased Auction Competition.”Stanford Law Review, Vol.
48 (1996), pp. 761-816.

Bhattacharya, V., Roberts, J. and Sweeting, A. “Regulating Bidder Participation in Auc-

tions.”RAND Journal of Economics, Vol. 45 (2014), pp. 675-704.

Blair, R. and Lopatka, J. “Predatory Buying and the Antitrust Laws.” Utah Law Review,
Vol. 2 (2008), pp. 415-470.

Borenstein, S. “On the E¢ ciency of Competitive Markets for Operating Licenses.”Quarterly

Journal of Economics, Vol. 103 (1988), pp. 357-385.

Brocas, I. “Endogenous Entry in Auctions with Negative Externalities.” Theory and Deci-
sion, Vol. 54 (2003), pp. 125-149.

. “Optimal Choice of Characteristics for a Nonexcludable Good.”RAND Journal

of Economics, Vol. 39 (2008), pp. 283-304.

. “Selling an Asset to a Competitor.”European Economic Review, Vol. 57 (2013a),

pp. 39-62.

. “Optimal Allocation Mechanisms with Type-Dependent Negative Externalities.”

Theory and Decision, Vol. 75 (2013b), pp. 359-387.

. “Countervailing Incentives in Allocation Mechanisms with Type-Dependent Ex-

ternalities.”Journal of Mathematical Economics, Vol. 50 (2014), pp. 22-33.

Burguet, R. and McAfee, R.P. “License Prices with Financially Constrained Bidders.”Jour-
nal of Regulatory Economics, Vol. 36 (2009), pp. 178-198.

Carrillo, J. “Coordination and Externalities.”Journal of Economic Theory, Vol. 78 (1998),

pp. 103-129.

Congressional Budget O¢ ce. “Small Bidders in License Auctions for Wireless Personal Com-
munications Services.”Congress of the United States, October 2005 (http://www.cbo.gov/
publication/17409).
31
Cramton, P. “Spectrum Auctions.” In M. E. Cave, S.K. Majumdar, and I. Vogelsang, eds.
Handbook of Telecommunications Economics, Vol. 1. Amsterdam: North Holland, 2002.

Cramton, P., Kwerel, E., Rosston, G. and Skrzypacz, A. “Using Spectrum Auctions to
Enhance Wireless Competition.”Journal of Law and Economics, Vol. 54 (2011), pp. S167-
S188.

Das Varma, G. “Standard Auctions with Identity-Dependent Externalities.”RAND Journal

of Economics, Vol. 33 (2002), pp. 689-708.

Eso, P., Nocke, V. and White, L. “Competition for Scarce Resources.” RAND Journal of
Economics, Vol. 41 (2010), pp. 524-548.

Federal Communications Commission. “Pioneer’s Preference Program.” O¢ ce of Engineer-

ing and Technology, Updated January 13, 2000 (http://transition.fcc.gov/oet/faqs/pioneerfa
qs.html).

Figueroa, N. and Skreta, V. “The Role of Optimal Threats in Auction Design.” Journal of
Economic Theory, Vol. 144 (2009), pp. 884-897.

Fusco, P., “FCC Grants QUALCOMM Pioneer Preference.” InternetNews.com. June 12,
2000 (http://www.internetnews.com/xSP/article.php/392251).

Gilbert, R. and Newbery, D. “Preemptive Patenting and the Persistence of Monopoly.”

American Economic Review, Vol. 72 (1982), pp. 514-526.

Haile, P. “Auctions with Resale Markets: An Application to U.S. Forest Service Timber
Sales.”American Economic Review, Vol. 91 (2001), pp. 399-427.

Hazlett, T. and Munoz, R. “A Welfare Analysis of Spectrum Allocation Policies.” RAND

Journal of Economics, Vol. 40 (2009), pp. 424-454.

Hendricks, K. and McAfee, R.P. “A Theory of Bilateral Oligopoly.”Economic Inquiry, Vol.

48 (2010), pp. 391-414.

Hendricks, K., Porter, R. and Wilson, C. “Auctions for Oil and Gas Leases with an Informed
Bidder and a Random Reservation Price.”Econometrica, Vol. 62 (1994), pp. 1415-1444.

Jehiel, P. and Moldovanu, B. “Strategic Nonparticipation.” RAND Journal of Economics,

Vol. 27 (1996), pp. 84-98.

. “Auctions with Downstream Interaction among Buyers.”RAND Journal of Eco-

nomics, Vol. 31 (2000), pp. 768-91.

. “An Economic Perspective on Auctions.” Economic Policy, Vol. 18 (2003), pp.

269-308.

32
 and Stacchetti, E. “How (Not) to Sell Nuclear Weapons.” American Economic
Review, Vol. 86 (1996), pp. 814-829.

. “Multidimensional Mechanism Design for Auctions with Externalities.”Journal

of Economic Theory, Vol. 85 (1999), pp. 258-293.

Katz, M. and Shapiro, C. “How to License Intangible Property.”Quarterly Journal of Eco-

nomics, Vol. 101 (1986), pp. 567-590.

Krishna, K. “Auctions with Endogenous Valuations: The Persistence of Monopoly Revis-

ited.”American Economic Review, Vol. 83 (1993), pp. 147-160.

Kwerel, E. and Rosston, G. “An Insiders’ View of FCC Spectrum Auctions.” Journal of
Regulatory Economics, Vol. 17 (2000), pp. 253-289.

Mayo, J. and Sappington, D. “Technical Appendix to Accompany ‘When Do Auctions Ensure

the Welfare-Maximizing Allocation of Scarce Inputs?’,”2015 (http://people.clas.u‡.edu/sapp
ing/).

McAfee, R. P., “Four Issues in Market Design.” Revista de Análisis Económico, Vol. 13
(1998), pp. 7-24.

and McMillan, J. “Analyzing the Airwaves Auction.” Journal of Economic Per-

spectives, Vol. 10 (1996), pp. 159-175.

Milgrom, P., Putting Auction Theory to Work. Cambridge University Press, 2004.

Pagnozzi, M. and Rosato, A. “Entry by Takeover: Auctions vs. Bilateral Negotiations.”

Università di Napoli Federico II discussion paper, March 2014.

Rey, P. and Tirole, J. “A Primer on Foreclosure.” In M. Armstrong and R. Porter, eds.,

Handbook of Industrial Organization: Volume 3. Amsterdam: North-Holland, 2007.

Salop, S. “Monopolistic Competition with Outside Goods.”Bell Journal of Economics, Vol.

10 (1979), pp. 141-156.

Singh, K. “The FCC’s Pioneer’s Preference Policy: An Innovative Idea Grows Old and
Weary.”Rutgers Computer & Technology Law Journal, Vol. 22 (1996), pp. 143-188.

Solomon, S.D. “How Loopholes Turned Dish Network into a ‘Very Small Business’.” New
York Times, February 24, 2015 (http://www.nytimes.com/2015/02/25/business/dealbook/
how-loopholes-transformed-dish-network-into-a-very-small-business.html?_r=0).

U. S. Department of Justice. “Ex Parte Submission of the United States Department of

Justice.” In the Matter of Policies Regarding Mobile Spectrum Holdings, Docket No. 12-
269, April 11, 2013.

33