Vertical Integration, Raising Rivals’ Cost and Upstream Collusion∗

Hans-Theo Normann†
Royal Holloway College
University of London
July 23, 2007

Abstract
This paper analyzes the impact vertical integration has on upstream firms’ ability to collude when
downstream firms pay a linear price for the input. As the downstream unit of the integrated firms is
delivered internally at marginal cost, it benefits from a raising-rivals’-cost effect when the input market
is collusive. The main result is that vertical integration facilitates upstream collusion. The paper
discusses the effects underlying this result (punishment, outlet and reaction effect), and continues to
derive conditions for foreclosure, that is, the integrated firm’s (at least partial) withdrawal from the
input market.

Keywords: collusion, foreclosure, vertical integration

JEL classification numbers: D43, L13, L23, L40

∗ I am grateful to Dirk Engelmann, conference participants at the EARIE 2005 meetings in Porto, and seminar audiences

at TU Berlin, Zurich University and the University of East Anglia for helpful comments.
† Department of Economics, Egham, Surrey TW20 0EX, UK, fax: +44 1784 439534, email: hans.normann@rhul.ac.uk
1 Introduction

The anticompetitive effects of vertical integration continue to be an active and controversial topic of
research in industrial organization. Antitrust decisions hostile towards vertical mergers in the US in the

1950s and 1960s were based on the idea that vertical integration can harm competition by removing
resources from the input market, thereby leveraging monopoly power from one market to another. These
arguments have been labeled as naı̈ve (Rey and Tirole, 2005) because they lacked a rigorous formal

basis. The more recent theories of vertical mergers (for example Salinger, 1988; Hart and Tirole, 1990;
O’Brien and Schaffer, 1992; Bolton and Whinston, 1993) formally derived many of the conclusions of
the older theories. In game-theoretic models a connection is established between vertical integration and
potentially anticompetitive outcomes.
These post-Chicago theories of anticompetitive vertical mergers differ in various details, for example,
assumptions about the integrated firm’s market power and the contractual arrangements between the
parties involved. There are several dominating approaches though (Riordan, 2005), including the “raising
rivals’ cost” and the “facilitating collusion” theory.1
The “raising rivals’ costs” theory was first put forward by Ordover, Saloner and Salop (1990), hence-
forth OSS (1990). They argue that vertical mergers might change the incentive to compete in the input
market. More specifically, when a vertically integrated withdraws from the input market, upstream price
competition becomes weaker. This reduction in upstream competition implies a higher price for the in-
put which means higher cost for the non-integrated downstream firms. Since the downstream unit of the
integrated firm benefits when its rivals’ costs are raised, the integrated firm is better off pursuing such a
foreclosure strategy compared to the case where it competes in the input market. In other words, it pays
for the integrated firm to forgo business with non-integrated downstream firms and instead gain from
its downstream rivals becoming less competitive. Only a vertically integrated firm can profitably pursue
such a strategy. It would not make sense for a non-integrated upstream firm to refrain from competign in
the input market (as it would simply lose money) nor would this strategy be feasible for a non-integrated

downstream firm (because it cannot affect the price of the input).

The “facilitating collusion” theory argues that vertical integration might make price agreements among
upstream firms easier. This concern has been expressed in the 1984 Non-horizontal Merger Guidelines and
in several cartel cases (see Riordan, 2005). The idea has recently been formalized in a dynamic model by
1 Riordan (2005) lists the “restoring monopoly power” as a third major theory. See Hart and Tirole (1990), Martin,

Normann and Snyder (2001) and Rey and Tirole (2005).

2
Nocke and White (2006).2 They analyze collusion among upstream firms who use two-part tariffs. Nocke
and White then compare the minimum discount factor required for collusion to be a subgame-perfect

Nash equilibrium with integration to the standard case of vertical separation. It turns out that vertical
integration unambiguously facilitates collusion.
This paper argues that the two approaches are actually closely related. It combines OSS’ (1990)

idea that raising rivals’ cost effects change the incentives of vertically integrated firms to compete in the
input market with the presumption that vertical integration facilitates collusion. In terms of modeling
strategy, the paper merges these two strands of the literature. This is possible because these two lines

of the literature make similar assumptions. Importantly, the integrated firm faces upstream competition
in both groups of models. The stage game is modeled as in OSS (1990) and allows for a raising rivals’
cost effect but, in contrast to the static model of OSS (1990), this paper studies repeated interaction in
a dynamic model. The analysis of the impact of vertical integration on collusion is similar to Nocke and
White (2006). However, departing from Nocke and White (2006) and following OSS (1990), downstream
firms pay simple linear prices for the input here.
The first motivation to study such a setup is that, hitherto, the raising rivals’ cost argument is not
particularly robust. Hart and Tirole (1990) and Reiffen (1992) pointed out that, even though foreclosure
is a profitable strategy for the integrated firm ex ante, it has an incentive to compete in the input
market ex post. Therefore vertical integration may not make any difference at all. To understand this
argument, suppose that the integrated firm does withdraw from the input-good market and that the
price of the input increases as a result. Given the high price in the input market, the integrated firm
has an incentive to deviate. Rather than withdraw from the input market, it will re-enter and undercut
upstream competitors’ price in order to gain the business with the non-integrated downstream firms
which will be delivered either way. The integrated firm’s upstream rivals will anticipate such a deviation
and will expect the re-entry in the market. In that case, upstream competition is the same as without
vertical integration. This implies that, even though foreclosure is profitable for the integrated firm ex
ante, this outcome is not credible and therefore not a subgame perfect Nash equilibrium.
The subsequent literature has attempted to show that the raising-rivals’-cost strategy can be made
credible in more specific models. OSS (1992) analyze a descending-price auction. When non-integrated
downstream firms procure the input in a descending-price auction, the integrated firm will drop out early
from such an auction. This will result in the input market being monopolized by the non-integrated
2 Riordan and Salop (1995) and Chen (2001) are also related to the facilitating collusion theory.

3
upstream firm, leading to the outcome proposed by OSS (1990). The incentive to deviate does not

exist for an integrated firm here. When dropping out of the auction, the integrated firm cannot re-enter
by the rules of the auction and therefore ensure that price competition comes to a halt. Choi and Yi
(2000) assume that the upstream firm can produce a specialized input for its downstream division. The

specialized input serves as a technological commitment not to supply the input market. By contrast,
a non-integrated firm would always produce a generalized input. Finally, Church and Gandal (2000)
analyze a market where the final good consists of a hardware component and complementary software.

Similar to Choi and Yi (2000), when a hardware and a software firm integrate, they will foreclose by
making the software incompatible with rival technologies, whereas a separated firm would always choose
the compatible software.3
However, these papers are only partially successful in re-establishing the OSS (1990) theory. The
reason is that they circumvent the core problem the original analysis posed. When the integrated firm
can make a credible ex-ante commitment not to compete in the intermediate-good market, it goes without
saying that the raising rivals’ cost argument goes through. Without commitment, vertically integrated
firms will compete just like non-integrated upstream firms, as emphasized by Hart and Tirole (1990) and
Reiffen (1992). The crucial question, then, appears to be whether such pre-commitment is feasible. OSS
(1992), Church and Gandal (2000) and Choi and Yi (2000) offer little as an answer to this question.
These papers assume rather specific modeling frameworks where commitment is indeed available for the
integrated firm. Few markets are organized as descending-price auctions and technology commitment or
input specificity may not be widely available strategies either. This would suggest that the rasing rivals’
cost theory has only limited applicability and may not be a sound basis for a broad policy argument
against vertical integration.
This paper does not impose any assumptions on the extensive form which may turn out to be re-
strictive. It attempts to re-establish the rasing rivals’ cost theory of OSS (1990) in a repeated-game
setting. The assumption of repeated interaction is simple, and should plausible for many industries. The
analysis should therefore have relevance for competition policy. The possibility that vertical integration
may be anticompetitive in the repeated game has been suggested by Riordan and Salop (1995) and the
intuition is straightforward. Repeated interaction (Macauley, 1963) can serve as a commitment device for
businesses. It may help the integrated firm in establishing a reputation for foreclosing the input market.
Since both the integrated and non-integrated upstream firms benefit from foreclosing the input market
3 Broadly related to this literature are Chen (2001) and Chen and Riordan (2006) .

4
in the long run, this may be an outcome of an equilibrium in the repeated game.
The second motivation arises from the fact that the facilitating-collusion theory has so far been

restricted to non-linear two-part tariffs. While two-part tariffs are relevant in many industries, so will be
linear contracts. Extending the results in Nocke and White (2006) for the case of linear tariffs significantly
strengthens the facilitating-collusion story and adds to its policy relevance.

The next section introduces the market model and an analysis of the static game for both the separated
and vertically integrated industries. Section 3 introduces the repeated game model. Section 4 provides
the analysis of collusion with vertical separation, followed by the key analysis of the paper: a section
on collusion with vertical integration. Section 6 compares vertical integration and vertical separation.
The analysis is extended by an investigation of the foreclosure issue in Section 7, and by allowing for
downstream collusion. Section 9 is the Conclusion.

2 Model and static Nash equilibrium

Apart from minor differences, the stage-game market model is as in OSS (1990). There are n = 2
upstream firms and m = 2 downstream firms. (The main results of the paper are generalized below for
the n, m > 2 case). Call the two upstream firms U 1 and U 2, and the two downstream firms D1 and D2.
The integrated firm will be called U 1-D1. The upstream firms produce a homogenous input. D1 and D2
transform the input on a one-to-one basis into a differentiated final good.
The downstream level is modeled as follows. Downstream firms pay linear prices for the input which
constitute their only cost.4 Define ci as the price per unit firm Di pays. There is differentiated price
competition at the downstream level and Qi (pi , pj ), i, j=1, 2, i 6= j, denotes the demand function of Di.

Accordingly, Di’s profits are

πDi = (pi − ci )Qi (pi , pj ), i, j = 1, 2, i 6= j. (1)

We impose the following assumptions on demand. Demand functions Qi (pi , pj ) are twice continuously
differentiable with ∂Qi /∂pi < 0, ∂Qi /∂pj > 0, and ∂Qi /∂pi + ∂Qi /∂pj < 0, i, j = 1, 2, i 6= j. These
assumptions ensure downward sloping demand with substitutes goods where the effect of a change in
a firm’s own prices dominates the effect resulting from a change of the rival firm’s price. Further, we

assume that goods are strategic complements, that is, ∂ 2 πDi /∂pi ∂pj > 0. A final assumption is that
∂ 2 πDi /∂p2i + ∂ 2 πDi /∂pi ∂pj < 0. This assumptions implies that own effects dominate cross effects also
4 This can be generalized to more complex downstream cost functions. See OSS (1990).

5
in terms of the slope of the demand function. Together with the other assumptions, this is sufficient to
ensure the existence of a unique Nash equilibrium of the stage game.5

Let p∗i (ci , cj ), i, j=1, 2, i 6= j, denote the static Nash equilibrium prices at the D level. In the static
Nash equilibrium, the input prices (ci , cj ) sufficiently describe downstream competition, and we will use
Q∗i (ci , cj ) as a reduced form for Qi (p∗i (ci , cj ), p∗j (cj , ci )), and πDi
∗
(ci , cj ) for πDi = (p∗i (ci , cj )−ci )Q∗i (ci , cj ).

We impose Q∗i (ci , cj ) > 0 throughout.

Given the above assumptions, OSS (1990) show that raising the cost of a downstream rival is profitable,
that is,
∗
∂πDi (ci , cj ) ∂πDi ∂p∗j
= > 0, i, j = 1, 2, i 6= j, (2)
∂cj ∂pj ∂cj
where ∂πDi /∂pj > 0 follows from ∂Qi /∂pj > 0, and ∂p∗j /∂cj > 0 follows from comparative statics of the
first order condition ∂πDj /∂pj = 0.
We now turn to the upstream level. U 1 and U 2 have constant marginal cost which we set equal to
zero.6 The upstream firms compete in prices. More specifically, we assume perfect Bertrand competition
between firms. This implies that the lower of the two upstream prices constitutes the input price non-
integrated downstream firms pay and that upstream firms must meet demand at the announced price.
Given the price of the input, non-integrated downstream firms purchase the number of units they require,
Q∗1 (c1 , c2 ) and Q∗2 (c2 , c1 ). When both firms buy the input from the input market, we get c1 = c2 but
input cost may differ when there is vertical integration.7
Without vertical integration, competition à la Bertrand implies Nash equilibrium prices equal to (zero)
marginal cost on the input market. This static Nash equilibrium is unique. Both downstream firms
purchase the input on the input market and they pay the same price for it. So, we have c1 = c2 = 0.
When U 1 and D1 are integrated, the downstream segment of U 1-D1 is delivered internally at c1 = 0.
This avoids double marginalization and therefore maximizes U 1-D1’s profits. Accordingly, when there is
integration, U 1-D1 and U 2 compete only for the D2 business in the input market.
The novel insight of OSS (1990) was to show that the integrated firm has an incentive to withdraw
from the input market. Suppose U 1-D1 can credibly foreclose the input market. If U 1-D1 completely
5 The assumptions to ensure a unique static Nash equilibrium are merely made to simplify the analysis. Under weaker

assumptions, the stage game has multiple equilibria and one would need to distinguish between stable and unstable equilibria
(see OSS, 1990). Note also that somewhat weaker conditions might be sufficient to guarantee existence and uniqueness (see
Vives, 1999).
6 The assumption of zero marginal cost is made to simplify the analysis but it is not necessarily innocuous as it precludes

certain types of punishment where firms price below cost (e.g., Abreu, 1988). See also footnote 8.
7 We ignore the case where the upstream firms can charge different linear tariffs to different downstream firms. In the

static game, this is without loss of generality. When colluding on the input market, upstream firms will generally want to
charge discriminating prices as we do not impose symmetry of Q∗1 and Q∗2 . However, we are interested only in the likelihood
of collusion here which is unaffected by this assumption. See Bernheim and Whinston’s (1990) irrelevance result.

6
withdraws from the input market, U 2 will became the sole supplier of D2, U 2 will charge some positive
∗
price for the input, so c2 > 0, and U 1-D1 earns πD1 (0, c2 ). Suppose instead that U 1-D1 competes on the
∗
input market. In that case, Bertrand competition implies c2 = 0 and U 1-D1 earns πD1 (0, 0). Comparing
the two cases, we find that U 1-D1 makes no profit upstream in either case, but, when foreclosing, it
∗ ∗
makes a higher profit downstream, πD1 (0, c2 ) > πD1 (0, 0), due to the raising rivals’ cost effect (2). Hence,

U 1-D1 prefers to foreclose ex ante.

However, as emphasized by Hart and Tirole (1990) and Reiffen (1992), U 1-D1 has an incentive to
deviate ex post from the foreclosure outcome. It will re-enter the input market (contrary to its claim
to withdraw) by undercutting U 2’s price because D2 is delivered anyway. As U 2 will anticipate this
deviation, the static Nash equilibrium has U 1-D1 and U 2 charging a price equal to marginal cost. Thus
we have c1 = c2 = 0, just as in the case without integration. U 2 earns zero profits and U 1-D1 earns
∗
πD1 (0, 0) in this unique static Nash equilibrium. The following proposition summarizes these findings.

Proposition 1 In the static game, vertical integration does not have any impact as the input market is
competitive both with and without vertical integration.

3 Repeated-game framework

We now analyze collusion in the infinitely repeated game. Time is indexed from t = 0, ..., ∞. Firms
discount future profits with a common factor δ, where δ ∈ (0, 1). When analyzing the repeated game,
denote by πic the profit a firm earns when both firms adhere to collusion. We require πic to be strictly
larger than the profit i makes in the static Nash equilibrium. Let πid denote the profit when firm i defects.
πip is the profit when punishment is triggered. We employ simple trigger strategies with reversion to the
static Nash equilibrium here.8 Nash reversions are credible threats.
We look for collusive equilibria that are subgame perfect. In order the prevent defection in period

t = 0, the one-time gains from deviating today, πid − πic , must be smaller than the loss due to punishment,
πic − πip , made in every future period t = 1, ..., ∞ for both firms. That is,

δ
πid − πic ≤ (πic − πip ) ; i = 1, 2. (3)
1−δ
8 As demonstrated in Nocke and White (2006), using simple penal codes as an optimal punishment scheme (Abreu, 1988)

actually fails in extensive-form games like this. The reason is that the downstream firms should be involved in the most
severe punishment scheme.

7
This condition can also be expressed in the following way

πid − πic
δ≥ := δi ; i = 1, 2, (4)
πid − πip

where δi denotes the minimum discount factor required for firm i to adhere to collusion and i = 1, 2.
In words, whenever the actual discount factor, δ, is larger than both firms’ minimum discount factor,

collusion is feasible.
When comparing minimum discount factors with vertical separation and integration, we will analyze
where a lower discount factor is required. Further, we will say that a collusive Nash equilibrium maximizes

the scope for collusion when it requires the lowest discount factor possible. We follow Bernheim and
Whinston (1990) and Compte, Jenny and Rey (2002) in that we sometimes look for the distributions of
market shares that maximize the scope for collusion.

4 Collusion with vertical separation

Without integration, it is straightforward to solve for the minimum discount factor. There are two
independent upstream firms, U 1 and U 2, and both downstream firms, D1 and D2, purchase in the input
market. Suppose upstream firms collude on some price they charge for the input. Denoting this collusive
input price by c, we obtain c1 = c2 = c.
Let π col denote total upstream industry profit when collusion is successful, that is, π col = c(Q∗1 (c, c) +
Q∗2 (c, c)). Let s ∈ (0, 1) denote the market share U 1 has in the input market. Accordingly, U 1 and U 2
earn π1c = sπ col and π2c = (1 − s)π col , respectively, when colluding. When firm i defects, it will earn
πid = π col by undercutting the collusive price c. Nash reversions yield πip = 0 for both firms with vertical
separation.
Let δU i denote the minimum discount factor required for U i to adhere to collusion. Using (4), the
minimum discount factors required are

δU 1 = (π col − sπ col )/π col = 1 − s (5)

and

δU 2 = (π col − (1 − s)π col )/π col = s. (6)

We have δU 1 + δU 2 = 1, and a symmetric division of the input market, s = 1/2, maximizes the scope for
collusion. We obtain the result that minimum discount factors are

δU 1 = δU 2 = 1/2 (7)

8
under vertical separation.

Proposition 2 Collusion can be supported as a subgame perfect Nash equilibrium in the separated in-
dustry if only if δ ≥ 1/2.

5 Collusion with vertical integration

5.1 Preliminaries

We will now analyze the repeated game when U 1 and D1 are vertically integrated but U 2 and D2 are
not. We will consider collusion at the upstream level, that is, a implicit agreement between U 1-D1 and
U 2.
One implication of the vertical integration of U 1 and D1 is that D1 obtains the input internally at
the price c1 = 0 and D2 only buys on the input market. The reason is that, for an integrated firm, the
effective price of the input is always equal to own marginal production cost (Bonanno and Vickers, 1988).
Henceforth, we have c1 = 0 accordingly. Since D2 only purchases the input externally, let c2 (rather than
c) denote the price of the input for clarity here.
Collusion between U 1-D1 and U 2 involves c2 , p1 and s. Consider these in turn.

– The input price charged to D2 will be collusive, that is, c2 > 0.9 For a collusive equilibrium to be
individually rational for U 2, it must yield a profit strictly better than the static Nash profit. This
requires c2 > 0 strictly.

– U 1-D1’s downstream price, p1 , will also be part of the collusion. This is not say that U 1-D1 colludes
with D2 at the downstream level. However, as we will see in detail below, p1 affects the both the
collusive and the defection profits of U 1-D1 and U 2 at the upstream level. Therefore, U 1-D1 will
take these effects into account when setting p1 , that is, the downstream price p1 will be chosen
to optimize the profitability and feasibility of upstream collusion. Note that this is an important
difference to the case where U 1 and D1 are vertically separated.10

– Third, s will be a part of the collusive agreement. As with vertical separation, market shares will be
set such as to minimize the incentives to deviate. Let s denote U 1-D1’s market share in the input
9 The notation is deliberately sloppy here. Strictly speaking, c2 denotes D2’s input cost and not the price of an upstream
firm. We refrain from introducing extra notation for upstream firms’ actions as the (lowest) posted price on the input
market is always equal to D2’s cost.
10 Below, we will consider a special case where p is not collusive and p = p∗ maximizes D1’s short-run profit.
1 1 1

9
 market and 1 − s is U 2’s market share. Because D2 only buys the input on the market, market
shares refer to the external input market. (By contrast, with vertical separation, s and 1 − s refer

to D1 and D2’s purchases, that is, Q1 + Q2 .)

The punishment will be triggered when either firm deviates from the collusive input price, c2 , or
when U 1-D1 deviates from the collusive upstream price, p1 . Importantly, U 1-D1 can observe whether
U 2 deviates from c2 before setting p1 in the period of the deviation.

In terms of notation, whenever p1 is not equal to p∗1 (c1 , c2 ), we cannot apply to shortcut notation
Q∗i (ci , cj ) for the downstream outputs. In such cases, let Q1 (p1 , p2 (c2 )) and Q2 (p2 (c2 ), p1 ) denote outputs
in the sense that p2 will be the (myopic) best reply to p1 given c2 , and p1 is the collusive downstream
price given c1 = 0. Using this notation, collusive profits are p1 Q1 (p1 , p2 (c2 )) + sc2 Q2 (p2 (c2 ), p1 ) and
(1 − s)c2 Q2 (p2 (c2 ), p1 ), for U 1-D1 and U 2, respectively.

5.2 Benchmarks

The following three benchmarks will be useful in the analysis. First, there is the joint-profit maximum
of U 1-D1 and U 2 in terms of p1 and c2 . Formally

{pjpm
1 , cjpm
2 } := arg max p1 Q1 (p1 , p2 (c2 )) + c2 Q2 (p2 (c2 ), p1 ). (8)
p1 ,c2

Let Qjpm
1 = Q1 (pjpm
1 , p2 (cjpm
2 )) and Qjpm
2 = Q2 (p2 (cjpm
2 ), pjpm
1 ).
Second, we have the downstream price that maximizes the joint profits of U 1-D1 and U 2 for any
given c2 . That is,
pb1 (c2 ) := arg max p1 Q1 (p1 , p2 (c2 )) + c2 Q2 (p2 (c2 ), p1 ) (9)
p1

formally. Of course, pjpm

1 and cjpm
2 relate to pb1 in that pjpm
1 = pb1 (cjpm
2 ).
Another benchmark is the price of the input that maximizes U 2’s profits if it is a monopolist in the
D2 market and provided that U 1-D1 plays its myopic best reply at the downstream level, p∗1 (c). Denote
this price by cmon
2 and define formally

cmon
2 := arg max c2 Q∗2 (c2 , 0). (10)
c2

Accordingly, define π mon := cmon

2 Q∗2 (cmon
2 , 0). The input-good price cmon
2 is also important in the static
analysis of OSS (1990).

10
5.3 Optimal downstream price after an upstream deviation

When a firm deviates from the collusive upstream price, c2 , U 1-D1 can observe the deviation and can
therefore respond to the deviation by charging a downstream price different from the collusive one. In

order to analyze the profitability of deviations at the upstream level, we need to know how U 1-D1 best
responds to them at the downstream level.
No matter which firm deviates, U 1-D1 should always set its downstream price such that it maximizes

U 1-D1’s short run profits after a deviation. This is because, in the case of a deviation at the upstream
level in period t, the punishment will be triggered in t + 1 regardless of the downstream price U 1-D1
charges in t. However, U 1-D1’s optimal downstream price following a defection will differ depending on
whether U 1-D1 or U 2 was the defector.
First, if U 1-D1 itself deviates at the upstream level by charging some cd2 < c2 , U 1-D1 will capture the
entire upstream profit in addition to its downstream profit. In that case, U 2 earns nothing and U 1-D1’s
defection profit is equal to the joint profit of U 1-D1 and U 2 which is maximized if and only if p1 =
pb1 (cd2 ) where pb1 is as in (9). Hence, pb1 (cd2 ) is U 1-D1’s optimal downstream price in this case. Second,
if U 2 is the deviator at the upstream level and sets some cd2 < c2 , U 1-D1 does not gain any upstream
profit. U 1-D1’s profit is now equal to D1’s profit only which is maximized if and only if p1 = p∗1 (cd2 ). We
summarize

Proposition 3 If U 1-D1 defects by charging cd2 < c2 , U 1-D1’s optimal downstream price is pb1 (cd2 ). If
U 2 defects by charging cd2 < c2 , U 1-D1’s optimal downstream price is p∗1 (cd2 ).

Finally, note that U 1-D1 will deviate from p1 in period t only if there is a defection at the upstream

level in t. If U 1-D1 defects from the collusive p1 even though both firms charge the equilibrium c2 at the
upstream level, the punishment will be triggered in t + 1. But if there is no more collusion in the next
period, then U 1-D1 can increase its defection profit by deviating from c2 before deviating from p1 .

5.4 Optimal deviation at the upstream level

If U 1-D1 defects at the U level, we know it will charge pb1 (cd2 ) afterwards at the D level. This implies

that the highest defection profit U 1-D1 can obtain results if c2 = cjpm
2 as in (8). Since U 1-D1’s defection
profit is monotonically increasing as long as c2 < cjpm
2 , we conclude that U 1-D1’s optimal defection is
c2 − ε if c2 ≤ cjpm
2 and cjpm
2 if c2 > cjpm
2 , where ε denotes an infinitesimally small margin.

11
 If U 2 defects, we know U 1-D1 will set p∗1 (cd2 ) at the D level. With a downstream price of p∗1 (c2 ),
the highest defection profit U 2 can obtain results when U 2 charges cmon
2 as in (10). Since U 2’s profit
is monotonically increasing in c2 up to cmon
2 , U 2’s optimal defection is c2 − ε if c2 ≤ cmon
2 and cmon
2 if

c2 > cmon
2 . We have

Proposition 4 U 1-D1’s optimal upstream defection is cd2 = c2 − ε if c2 ≤ cjpm

2 and cd2 = cjpm
2 if
c2 > cjpm
2 . U 2’s optimal upstream defection is cd2 = c2 − ε if c2 ≤ cmon
2 and cd2 = cmon
2 if c2 > cmon
2 .

5.5 Defection profits

We can use Propositions 3 and 4 to derive the profits resulting from a deviation.

Proposition 5 Given a collusive upstream price of c2 , optimal defection implies profits of

π jpm if c2 > cjpm


π1d = 2
pb1 Q1 (b
p1 (c2 ), p2 (c2 )) + c2 Q2 (p2 (c2 ), pb1 (c2 )) if c2 ≤ cjpm
2
π2mon if c2 > cmon

π2d = 2
c2 Q∗2 (c2 , 0) if c2 ≤ cmon
2

Note that the deviation profits depend on the collusive upstream price, c2 , but not on the collusive
downstream price, p1 .

5.6 Optimal collusive downstream price

For some given collusive upstream price c2 , it is easy to see that the optimal collusive downstream price
is p1 = pb1 (c2 ). The intuition is that the defection profits stated in Proposition 4 are not affected by p1 .
If neither defection nor punishment profits are affected by p1 , the incentive constraints are relaxed as far
as possible when the collusive profits are maximized. This is the case if and only if p1 = pb1 (c2 ).

To formally prove this claim, note that U 1-D1’s collusive profit is p1 Q1 (p1 , p2 (c2 ))+sc2 Q2 (p2 (c2 ), p1 ).
Plugging the expression for U 1-D1’s collusive profit, its defection profit π1d as in Proposition 5, and the
∗
static Nash profit πD1 (0, 0) into the incentive constraint (3), we obtain

∗ δ
π1d −p1 Q1 (p1 , p2 (c2 ))+sc2 Q2 (p2 (c2 ), p1 ) ≤ (p1 Q1 (p1 , p2 (c2 )) + sc2 Q2 (p2 (c2 ), p1 ) − πD1 (0, 0)) (11)
1−δ

12
for U 1-D1. U 2’s collusive profit is (1 − s)c2 Q2 (p2 (c2 ), pb1 (c2 )), its defection profit is π2d as in Proposition
5, and the static Nash profit is zero. Thus, (3) becomes

δ
π2d − (1 − s)c2 Q2 (p2 (c2 ), p1 ) ≤ (1 − s)c2 Q2 (p2 (c2 ), p1 ) (12)
1−δ

for U 2. Pooling the incentive constraints and rearranging, we get

π1d + π2d − (p1 Q1 (p1 , p2 (c2 )) + c2 Q2 (p2 (c2 ), p1 ))

δ≥ ∗ (0, 0)
(13)
π1d + π2d − πD1

Since the defection profits, πid , do not depend on p1 , choosing p1 to maximize p1 Q1 (c2 , p1 ) + c2 Q2 (c2 , p1 )
relaxes the pooled incentive constraint as far as possible. This is the case if and only if p1 = pb1 (c2 ). Since
pb1 (c2 ) also maximizes collusive profits (for any c2 ), it is optimal also when the incentive constraint is not
binding. Thus we have established

Proposition 6 Given some collusive upstream price c2 , the optimal collusive downstream price is pb1 (c2 ).

The proposition has an interesting implication. Recall that the pb1 (c2 ) maximize p1 Q1 (.) + c2 Q2 (.)
whereas the p∗1 (c2 ) maximize p1 Q1 (.) only. Since ∂Q2 /∂p1 > 0, given some c2 , we have pb1 (c2 ) > p∗1 (c2 ),
and since products are strategic complements at the downstream level, not only p1 but also p2 will be
higher than p∗1 and p∗2 , respectively. Now, pb1 (c2 ) and the according p2 will be the downstream prices
with vertical integration and p∗1 (0, p2 ) and p∗2 (c2 , 0) are the downstream price with vertical separation.
Thus, we conclude that—given the same upstream price c2 —the vertical integration of U 1 and D1 implies
higher prices at the downstream level.
The intuition behind this effect is that U 1-D1 faces a trade-off regarding its downstream price when

it is vertically integrated. Lowering p1 implies higher sales and profits for D1. But if D1 sells more, D2
will sell less and therefore profits at the upstream level will go down. As the optimal collusive upstream
price, pb1 (c2 ), maximizes the profits of D1 and those gained in the upstream input market, prices will be
higher compared to the case where U 1 and D1 are separated. This is also the case when s = 0.
Chen (2001) was the first to discover this anticompetitive effect of vertical integration on downstream
prices. The market model in that paper is similar to the present one although Chen (2001) analyzes a
static game and allows for differences in the cost functions. Specifically, he analyzes the possibility that
downstream firms choose the upstream supplier from which they will purchase the input before prices are
set. In this case, vertical integration has the same anticompetitive effect as above (see Chen’s Lemma 1)

13
even in the static game. When the integrated firm sells some of the input to D2 at higher than marginal
cost, its incentives to compete at the downstream level are reduced. Therefore, D2 finds it worthwhile

to select the integrated firm as its supplier.

5.7 Optimal collusive upstream pricing

Proposition 7 The optimal collusive upstream price is c2 = cjpm

2 .

The proof is simple but requires some space and is therefore delegated to the appendix. A rough sketch
of the proof is as follows. Proposition 5 implies that we need to consider the three parameter regions
c2 ≥ cjpm
2 , cmon
2 ≤ c2 < cjpm
2 , and c2 < cmon
2 . When c2 > cjpm
2 , it turns out that δ is increasing in c2 and
joint profits are decreasing in c2 . Hence, it cannot be optimal to charge c2 > cjpm
2 . If cmon
2 ≤ c2 < cjpm
2

and cmon
2 > c2 , δ is decreasing in c2 and joint profits are increasing in c2 . Hence, cjpm
2 must be the optimal
collusive upstream price.

5.8 Summary: optimal collusion

We now summarize the results on collusion when U 1 and D1 are integrated. We note that, from Propo-
sition 6, the optimal downstream price conditional on c2 is pb1 (c2 ) and, from Proposition 7, the optimal
upstream price is c2 = cjpm
2 . Hence, the overall optimal downstream price is pb1 (cjpm
2 ) = pjpm
1 .

Proposition 8 Optimal collusive prices are c2 = cjpm

2 and p1 = pjpm
1 . That is, collusion between U 1-D1
and U 2 is feasible if and only if the joint-profit maximum can be sustained as a subgame perfect Nash
equilibrium. Optimal defection yields profits of π jpm for U 1-D1 and π2mon for U 2.

Coordinating on a different collusive equilibrium is not worthwhile as this requires a higher discount
factor and reduces the joint profits.

6 Comparison of vertical integration and vertical separation

6.1 Vertical integration facilitates collusion

We now turn to a key point of the analysis, the comparison of the minimum discount factor with vertical
integration and vertical separation. Above, we saw that collusion can be supported as a subgame perfect

14
Nash equilibrium in the separated industry if only if δ ≥ 1/2. With vertical integration, optimal collusion

described as in (8) implies the following. For U 1-D1, (3) becomes

  δ
(1 − s)cjpm
2 Qjpm
2 ≤ scjpm jpm
2 Q2 + p jpm jpm
1 Q1 − π ∗
D1 (0, 0) (14)
1−δ
∗
as πD1 (0, 0) is the punishment profit. For U 2, (3) becomes

δ
π2mon − (1 − s)cjpm
2 Qjpm
2 ≤ (1 − s)cjpm
2 Qjpm
2 (15)
1−δ

as U 2’s punishment profit is zero. Pooling these incentive constraints, we obtain that collusion is a
subgame perfect Nash equilibrium only if
  δ
π2mon ≤ cjpm
2 Qjpm
2 + pjpm
1 Qjpm
1
∗
− πD1 (0, 0) (16)
1−δ

Comparing this constraint to the threshold obtained with vertical separation, collusion requires a
discount factor smaller than 1/2 if and only if

π2mon < cjpm

2 Qjpm
2 + pjpm
1 Qjpm
1
∗
− πD1 (0, 0). (17)

∗
This inequality holds since π2mon + πD1 (cmon
2 , 0) < pjpm
1 Qjpm
1 + cjpm
2 Qjpm
2 from the definition of the joint
∗ ∗
profit maximum and since πD1 (0, 0) < πD1 (cmon
2 , 0) due to the raising rival’s cost effect (2). Thus, we
have established

Proposition 9 Vertical integration facilitates collusion.

The proposition states that collusion requires a lower discount factor with vertical integration com-
pared to the δ = 1/2 benchmark which results with vertical separation. The result is based on the
assumption that U 1-D1 and U 2 collude optimally (that is, they collude by charging c2 = cjpm
2 and
p1 = pjpm
1 and that they also defect optimally). There may be other collusive equilibria with vertical
integration which require minimum discount factors higher than 1/2, however, joint profits are lower in
such equilibria.

6.2 Discussion and comparison with Nocke and White (2006)

We now discuss the effects that drive the main result, Proposition 9. We will also compare the results to

Nocke and White (2006).

15
 Beforehand, it is useful to introduce the following notation. Let r denote U 1-D1’s share of the collusive
profit; 1 − r is U 2’s share, accordingly. The market share variable s relates to r in that

pjpm
1 Qjpm
1 + scjpm
2 Qjpm
2
r= . (18)
pjpm
1 Qjpm
1 + cjpm
2 Qjpm
2

In the analysis of vertical separation (Section 4), market shares in the input market and the shares of
collusive profits are identical, that is s = r. When U 1 and D1 are integrated D1’s profit is part of the

collusive profit. Thus upstream market shares and shares of the collusive industry profits differ. What
we are interested in is a comparison of integration and separation when upstream firms have the same
share of the collusive profits (r). It is less meaningful to conduct such a comparison when upstream firms
have the same market share in the input market (s) because the profits in the input market are rather
different with vertical integration.
Using the t notation, (4) becomes for U 1-D1

(1 − r)π jpm
δU 1−D1 = ∗ (0, 0) > 1 − r
π jpm − πD1
(19)

∗
where the inequality is due to πD1 (0, 0) > 0 and where 1 − r is the benchmark for the separated industry
(see (5)). That is, when we compare a separated and an integrated firm which get an identical share
of collusive profits, r, the minimum discount factor for the integrated firm is higher. Nocke and White
(2006) call this the punishment effect. The punishment effect occurs because the integrated firm makes
∗
a positive profit in the static Nash equilibrium (πD1 (0, 0) > 0) whereas a nonintegrated firm makes no
profit in the static Nash equilibrium. Thus, the punishment is less harsh for an integrated firm which,
all else equal, makes collusion more difficult to sustain.
For U 2, (4) becomes
π mon − (1 − r)π jpm
δU 2 = <r (20)
π mon
where the inequality is due to π mon < π jpm and where r is the benchmark for the separated industry
(see (6)). There are two effects here, both were also found in Nocke and White (2006). The outlet effect
arises from the fact that the non-integrated upstream firm can no longer sell to the downstream affiliates
of the integrated rival when deviating. All else equal, this reduces the deviation profits and therefore
reduces the minimum discount factor required for collusion.. The reaction effect arises because U 1-D1
can observe if U 2 deviates at the upstream level and can adapt its downstream price accordingly (see
also the analysis of downstream defection in Proposition 3). This reduces deviation profits further and
therefore implies an additional reduction of the minimum discount factor.

16
 We can disentangle the two effects as follows. Assume hypothetically that U 1-D1 cannot react at
the D level to any deviation of U 2. In that case, U 2 would earn at least cjpm
2 Qjpm
2 but less than π jpm .

Thus, we would still obtain δU 2 < r even if there was no reaction effect. Due to the reaction effect,
U 2 gets even less profit than in the hypothetical scenario, namely π2mon < cjpm
2 Qjpm
2 . This inequality
is established from cjpm
2 Qjpm
2 > cjpm
2 Q∗2 (since pb1 (c2 ) > p∗1 (c2 )) and cjpm
2 Q∗2 > cmon
2 Q∗2 = π2mon (by

revealed preference).
The net effect of the negative punishment effect and the positive outlet and reaction effect is positive,
of course—otherwise, we would not have obtained the result that vertical integration facilitates collusion

(Proposition 9). Pooling the incentive constraints yields (16) as above.

So far the results fully confirm the analysis in Nocke and White (2006). In the model with linear
contracts, we find the same effects, and they have the same sign as in the model with two-part tariffs.
There is one difference though.
Nocke and White (2006) also find a lack-of-commitment effect. This effect does not occur in this
model, and it easy to see why. In Nocke and White (2006), downstream prices are such that they
maximize industry profits (comprising U 1-D1, U 2 and D2) when U 1-D1 and U 2 successfully collude. If
the integrated firm deviates, however, it will choose a downstream price that maximizes it own profits and
will this choose a more competitive price at the downstream level. This will be anticipated by D2 and
thus U 1-D1 cannot obtain maximum industry profits when deviating. This is the lack-of-commitment
effect. In the this model, the effect does not occur because upstream firms charge a linear price for the
input, and D2 makes a positive profit which the upstream firms cannot seize as a result. Therefore,
U 1-D1 and U 2 do not collude by maximizing industry profits. Instead, as seen above, they collude by
maximizing the joint profits of U 1-D1 and U 2. When U 1-D1 deviates, U 2’s profit is zero and the joint
profits U 1-D1 and U 2 all go to U 1-D1. Thus U 1-D1 will set the same downstream price when deviating
and there is no lack-of-commitment effect. (See also Proposition 3.)

6.3 Example with linear demand

In order to illustrate some of the results derived above, consider the following parametrized version of
the model, also used in OSS (1990), as an example. Demand is assumed to be linear, symmetric and the
demand intercept is, without loss of generality, normalized to one

Qi (pi , pj ) = 1 − bpi + d(pj − pi ), i, j, = 1, 2; i 6= j, (21)

17
where b, d ≥ 0. It simplifies the analysis to rewrite this as

Qi (pi , pj ) = 1 − kpi + dpj , i, j, = 1, 2; i 6= j, (22)

where k = b + d. Products are entirely heterogenous if d = 0 while d → ∞ (in which case d = k) would

imply perfectly homogenous goods. Di’s profit is

πDi = (1 − kpi + dpj )(pi − ci ), i, j, = 1, 2; i 6= j. (23)

In the appendix, explicit solutions for cjpm

2 , pjpm
1 , the profit expressions involved, and the δi can

be found. The appendix also shows how s is used to relax the incentive constraint as far as possible.
Essentially, the optimal s is implicitly defined by δU 1−D1 (s) = δU 2 (s), using the δi in (19) and (??).
Figure 1 plots the optimal s and the resulting δi as a function of the parameter of product differenti-
ation, d. We note a few observations

• If d = 0, we have s = δ1 = δ2 = 1/2. This is intuitive because, when products are entirely

independent, there is no raising rival’s cost effect and firms face the same incentives as with vertical
separation.

• For d ∈ (0, 2.895), s < 1/2 is optimal, whereas, for d ≥ 2.895, s ≥ 1/2 is optimal.

• δ decreases monotonically in d.

Using the algebra in the appendix, we can also take limd→∞ . In that case, we obtain s = 1 and δ = 0.

6.4 Extensions

It is relatively straightforward to see that the results of the previous section hold with more than two
upstream firms. Suppose there are now n > 2 upstream firms, one of which is vertical integrated, say

U 1-D1. What does change is that the profit made in the D2 market is divided among more firms, so
collusion will generally be more difficult. But collusion will be still more likely under vertical integration.
Under vertical separation, symmetric division of the market maximizes the scope for collusion. In
that case each upstream firm has a minimum discount factor of δU i = (n − 1)/n, i = 1, ..., n.
With vertical integration, assume as above that the integrated firm has a market share of s and the
n − 1 non-integrated firms symmetrically split the rest so that each has a share of (1 − s)/(n − 1). If
firms collude optimally by charging cjpm
2 and pjpm
1 , U 1-D1’s incentive constraint is
  δ
(1 − s)cjpm
2 Qjpm
2 ≤ p jpm jpm
1 Q1 + scjpm jpm
2 Q2 − π ∗
D1 (0, 0) (24)
1−δ

18
as in the duopoly case. For the nonintegrated upstream firms U j, j = 2, ..., n, (3) becomes
   
1−s 1−s δ
π2mon − cjpm
2 Qjpm
2 ≤ cjpm
2 Qjpm
2 . (25)
n−1 n−1 1−δ

Pooling these incentive constraints, and solving for δ, we obtain

  δ
(n − 1)π2mon ≤ pjpm
1 Qjpm
1 + scjpm jpm
2 Q2 − π ∗
D1 (0, 0) . (26)
1−δ

Comparing this constraint to the threshold obtained with vertical separation, collusion requires a discount

factor lower than (n − 1)/n if and only if

π2mon < pjpm

1 Qjpm
1 + cjpm
2 Qjpm
2
∗
− πD1 (0, 0)

which is the same condition as in the duopoly case. Thus, collusion is more likely under vertical integration
also with n > 2 upstream firms.
Finally, the results do not change qualitatively when there are m > 2 downstream firms. To begin
with, note that raising one downstream firm’s cost has the same qualitative effects as raising two or more
downstream firms’ cost. Further, what matters are the market shares s and 1 − s the upstream firms
have in the input market. For m = 2, industry profits in the input market were denoted by c(Q∗1 + Q∗2 )
and c2 Q∗2 with vertical separation and integration, respectively. This can be extended with Q∗3 , ..., Q∗m
for the m > 2 case. Whether it is one or more non-integrated downstream firms purchasing on the input
market is therefore immaterial and, thus, does not change the results qualitatively. As the model does
not impose a symmetry assumption on the D firms, Q∗2 simply can be thought of representing the sales
of all non-integrated downstream firms.

7 Foreclosure

It is the purpose of this paper to put together the raising-rivals’-cost and facilitating-collusion theories of

vertical integration. Proposition 9 shows that vertical integration facilitates collusion, and in the optimal
collusive equilibrium the cost of the nonintegrated downstream firm are indeed raised.
This section attempts to strengthen the relation between the two theories. In what follows, we will
analyze a scenario where the downstream price of the integrated firm, p1 , maximizes the short-run profits
of D1 rather than U 1-D1’s profits. Such behavior is plausible when D1 operates as a profit center, for
example. Under this simplifying assumption downstream pricing will be exactly as in the static model

of OSS (1990). In other words, we can analyze the behavior of the upstream firms in the repeated game

19
given the same downstream Nash prices as in the static game. The main purpose of this exercise is to
check whether the outcome derived in OSS (1990) is an equilibrium in the repeated game. A further

assumption (which is without loss of generality, though) is to assume that U 1-D1 and U 2 charge the
monopoly price as in (10) for the input.

Assumption Throughout section 7, assume c2 = cmon

2 and p1 = p∗1 (0, cmon
2 ).

Given this setup, we analyze the incentives for collusion again. Consider collusive profits, πic , first.
At the upstream level, U 1-D1 makes a profit of sπ mon from delivering D2 on the input market. At the
∗ ∗
downstream level, U 1-D1 makes a profit of πD1 (0, cmon
2 ). Therefore, π1c = sπ mon + πD1 (0, cmon
2 ). U 2
makes a collusive profit of π2c = (1 − s)π mon . Then consider a defection. When defecting, either firm
will undercut cmon
2 by an infinitesimally small margin and will obtain π mon in the period of defection. It
∗
follows that π1d = π mon + πD1 (0, c2 ) and π2d = π mon . Punishment (Nash) profits are π1p = πD1
∗
(0, 0) and
π2p = 0.
Using these expressions and plugging them into (4), the minimum discount factors under vertical
integration are
(1 − s)π mon
δ1 = ∗ (0, cmon ) − π ∗ (0, 0) < 1 − s,
π mon + πD1
(27)
2 D1
∗ ∗
where the inequality is due to πD1 (0, cmon
2 ) − πD1 (0, 0) > 0, and

π mon − (1 − s)π mon

δ2 = = s. (28)
π mon

We obtain δ1 +δ2 < 1. Therefore, suitable values of s push both minimum discount factors below 1/2. We
conclude that collusion with vertical integration requires a lower discount factor than upstream collusion
with vertical separation even when we impose the above assumptions.
It is instructive here to take a closer look at the foreclosure effect of vertical integration. We can
interpret the market share of the integrated firm, s, as an indicator of foreclosure in that s = 0 would
correspond to the case where U 1-D1 completely withdraws from the input market. Similarly, market
outcomes with s < 1/2 can be interpreted as a partial withdrawal because U 1-D1’s market share is less

than a symmetric market division would suggest.

Bearing this notion of foreclosure in mind, it is easy to see that s < 1/2 is actually crucial under
vertical integration. From (19) and (??), we have δ1 = (1 − s)z, for some z < 1, and δ2 = s. It follows
that, when firms maximize the scope for collusion, s < 1/2 is a necessary condition. Similarly, Proposition

20
?? only holds if s < 1/2. If not, we would get δ2 > 1/2 and the minimum discount factor required under

vertical separation and vertical integration would be the same.11

We summarize

Proposition 10 Assuming c2 = cmon

2 and p1 = p∗1 (0, cmon
2 ), we obtain the following results. (i) Vertical
integration facilitates collusion. (ii) With vertical integration, foreclosure (s < 1/2) maximizes the scope

for collusion. (iii) Moreover, s < 1/2 is a necessary condition for vertical integration to require a
minimum discount factor of less than 1/2.

The proposition predicts a foreclosure effect in a probabilistic sense. That is, for some discount
factors, an (at least partial) withdrawal of the integrated firm from the input market will occur. We
saw above that a symmetric division of the market, s = 1/2, does maximize the scope for collusion in
the separated industry. Here, with vertical integration, a market share of less than 50% is necessarily
required to maximize the scope for collusion.

8 Conclusions

This paper analyzes the impact vertical integration has on upstream firms’ ability to collude when down-
stream firms pay a linear price for the input. As the downstream unit of the integrated firms is delivered
internally at marginal cost, it benefits from a raising-rivals’-cost effect when the input market is collusive.
The main result is that vertical integration facilitates upstream collusion. The paper discusses the effects
underlying this result (punishment, outlet and reaction effect), and continues to derive conditions for
foreclosure, that is, the integrated firm’s (at least partial) withdrawal from the input market.

Appendix
Proof of Proposition 7

The Proposition states that the optimal collusive upstream price is c2 = cjpm
2 . From Proposition 5, we
need to distinguish three parameter regions: (i) c2 ≥ cjpm
2 , (ii) cmon
2 ≤ c2 < cjpm
2 , and (iii) c2 < cmon
2 .
(i) If c2 ≥ cjpm
2 , Proposition 6 implies a collusive profit of pb1 Q1 (b
p1 (c2 ), p2 (c2 )) + sc2 Q2 (p2 (c2 ), pb1 (c2 ))

for U 1-D1 and Proposition 5 implies a defection profit of pjpm

1 Qjpm
1 + cjpm
2 Qjpm
2 . For U 2, we obtain
11 This condition is not restricted to the c = cmon case; it turns out to be a general requirement. It is straightforward to
2 2
see that c2 = cmon
2 minimizes δ2 for any s. We also know δ2 = s if c2 = cmon
2 , so, we have δ2 ≥ δ2 |c=cmon
2
= s. Therefore
δ2 < 1/2 only if s < 1/2.

21
a collusive profit of (1 − s)c2 Q2 (p2 (c2 ), pb1 (c2 )) and a defection profit of π2mon . Punishment profits are
∗
πD1 (0, 0) and zero for U 1-D1 and U 2, respectively. Pooling the incentive constraints, (3) yields
δ
pjpm
1 Qjpm
1 + cjpm
2 Qjpm
2 + π2mon − (b
p1 Q1 (.) + c2 Q2 (.)) ≤ (b ∗
p1 Q1 (.) + c2 Q2 (.) − πD1 (0, 0)) . (29)
1−δ
When c2 ≥ cjpm
2 , the term pb1 Q1 (.) + c2 Q2 (.) is decreasing in c2 . This implies that the minimum discount
factor required is increasing whereas collusive profits are decreasing in c2 when c2 ≥ cjpm
2 .

(ii) If cmon
2 ≤ c2 < cjpm
2 , U 1-D1 has a collusive profit of pb1 Q1 (.) + sc2 Q2 (.) and a defection profit of
pb1 Q1 () + c2 Q2 (.). U 2, has a collusive profit of (1 − s)c2 Q2 (p2 (c2 ), pb1 (c2 )) and a defection profit of π2mon .
Punishment profits are as above. Pooling the incentive constraints, we get

∗ δ
π2mon ≤ (b
p1 Q1 (.) + c2 Q2 (.) − πD1 (0, 0)) . (30)
1−δ
Since the term pb1 Q1 (.) + c2 Q2 (.) is increasing in c2 , increasing c2 in region (ii) weakens the incentive
constraint and yields higher collusive profits.
(iii) If c2 < cmon
2 , U 1-D1 has a collusive profit of pb1 Q1 (.) + sc2 Q2 (.) and a defection profit of pb1 Q1 () +
c2 Q2 (.). U 2, has a collusive profit of (1 − s)c2 Q2 (p2 (c2 ), pb1 (c2 )) and a defection profit of c2 Q∗2 . Pooling
the incentive constraints, we get
δ
c2 Q∗2 ≤ (b ∗
p1 Q1 (.) + c2 Q2 (.) − πD1 (0, 0)) (31)
1−δ
and solving for δ
c2 Q∗2
δ≥ ∗ (0, 0) .
pb1 Q1 (.) + c2 Q2 (.) + c2 Q∗2 − πD1
Both the numerator and the denominator are increasing in c2 but since ∂(b
p1 Q1 (.) + c2 Q2 (.))/∂c2 > 0
when c2 < cmon
2 , the numerator increases more quickly in c2 . Therefore, the minimum discount factor is
decreasing in c2 whereas collusive profits are increasing in c2 .
Taking the parameter regions (i) to (iii) together, the minimum discount factor is decreasing in c2 as

long as c2 < cjpm

2 and it is increasing in c2 if c2 > cjpm
2 . Thus, choosing c2 = cjpm
2 maximizes the scope
for collusion. As c2 = cjpm
2 also maximizes the collusive profits (by definition), the optimal upstream
price is c2 = cjpm
2 .

8.1 Optimal collusion with linear demand

Step 1: Derivation of the joint-profit maximum. The general analysis in the main part of the paper shows
that optimal collusion involves the joint-profit maximizing prices. Therefore, we need to find

arg max p1 Q1 (p1 , p2 ) + c2 Q2 (p2 , p1 ). (32)

c2 ,p1

22
Assuming the linear demand function in (22), the joint profits are

p1 (1 − kp1 + dp2 ) + c2 (1 − kp2 + dp1 ) . (33)

We first solve for the downstream prices. U 1-D1’s first-order condition with respect to p1 is 1 −

2kp1 + dp2 + c2 d = 0. D2 plays the myopic best reply to p1 given the price of the input, c2 . D2’s profit
is p2 (1 − kp2 + dp1 ), the first-order condition is 1 − 2kp2 + dp1 + kc = 0. We use the two first-order
condition to find the explicit solution for the downstream prices for a given c2

2k + d + 3kc2 d
p1 = (34)
4k 2 − d2
2k + d + c2 (2k 2 + d2 )
p2 = . (35)
4k 2 − d2

We now derive the joint-profit maximizing upstream price. Plugging the downstream prices (34) and
(35) into the profit function (33) and maximizing with respect to c2 , we get

(2k + d) 4k 2 − 2dk + d2
cjpm
2 = (36)
2k(k − d)(8k 2 + d2 )

which implies the following downstream prices

8k 2 + 2dk − d2
pjpm
1 = (37)
2 (k − d) (8k 2 + d2 )
12k 3 − 4dk 2 + 2d2 k − d3
p2 (cjpm
2 ) = .
2k (k − d) (8k 2 + d2 )

Given these prices, the joint maximum implies profits of

(2k + d) 4k 2 + d2 − dk 8k 2 + 2dk − d2
pjpm
1 Qjpm
1 = 2 (38)
4k (k − d) (8k 2 + d2 )
at the downstream level, and



2k 2 + d2 4k 2 − 2dk + d2 (2k + d)
cjpm
2 Qjpm
2 = 2 (39)
2k (8k 2 + d2 ) (k − d)

at the upstream level.

Step 2: Defection and punishment profits. U 1-D1’s defection profit is simply pjpm
1 Qjpm
1 + cjpm
2 Qjpm
2
∗
as just derived and its punishment profit is πD1 (0, 0). U 2’s defection profit is π2mon and its punishment
∗
profit is zero. Thus, we still need to derive π2mon and πD1 (0, 0). Both expressions depend on downstream
prices p∗i (ci , cj ), that is, the prices that maximize the short-run profits of Di. With the linear demand

23
specification, Di’s profit is πDi = (1 − kpi + dpj )(pi − ci ), i, j, = 1, 2; i 6= j and myopic maximization at

the downstream level yields Nash equilibrium prices of

2k + d + k 2 ci + kdcj
p∗i (ci , cj ) = (40)
4k 2 − d2

and equilibrium outputs

2k + d − (2k 2 − d2 )ci + kdcj
Q∗i (ci , cj ) = k (41)
4k 2 − d2
∗
(see also OSS, 1990). Downstream profits are πDi (ci , cj ) = (Q∗i )2 /k. Thus we obtain

∗ k
πD1 (0, 0) = 2. (42)
(2k − d)

The monopoly price, cmon

2 , maxmizes c2 Q∗2 (c2 , 0) and is easily derived as

2k + d
cmon
2 = (43)
2(2k 2 − d2 )

and the monopoly profit is

(2k + d) k
π mon = cmon
2 Q∗2 (cmon
2 , 0) = (44)
4 (d − 2k) (−2k 2 + d2 )

Step 3: Minimum discount factors. Plugging these expressions in to (4), we obtain the following
minimum discount factors

(1 − s)cjpm
2 Qjpm
2
δU 1−D1 = (45)
pjpm
1 Qjpm
1 + cjpm
2 Qjpm
2
∗ (0, 0)
− πD1
2

2 (2k − d) (2k + d) 2k 2 + d2
= (1 − s) (46)
(k + d) (4k 2 − 2dk + d2 ) (8k 2 + d2 )

and

π mon − (1 − s)cjpm
2 Qjpm
2
δU 2 = mon
(47)
π


2 4k 2 − 2dk + d2 4k 4 − d4 (2k − d)
= 1 − (1 − s) 2 .
k 2 (8k 2 + d2 ) (k − d)

Step 4. Optimal markets shares. Optimal market shares should be fixed such that they minimize
max{δU 1−D1 , δU 2 }. Since δU 1−D1 is decreasing in s whereas δU 2 is increasing in s, the optimal s solves

δU 1−D1 = δU 2 . Explicit solutions for the optimal s and the resulting minimum disocvunt factor can be
obtained but they are not particularly infrmative. These explicit solutions are plotted in Figure 1.

24
References
[1] Abreu, D., 1988. On the Theory of Infinitely Repeated Games with Discounting. Econometrica 56,
383-396.
[2] Bernheim, D.C., Whinston, M.D., 1990. Multimarket Contact and Collusive Behavior. RAND Jour-
nal of Economics 39, 191-225.
[3] Bolton, P., Whinston, M.D., 1993. Incomplete Contracts, Vertical Integration and Supply Assurance.
Review of Economic Studies 60, 121-148.
[4] Chen, Y., 2001. On Vertical Mergers and Their Competitive Effects. RAND Journal of Economics
32, 466-496.
[5] Chen, Y., Riordan, M., 2006. Vertical Integration, Exclusive Dealing and Ex Post Cartelization.
Forthcoming: RAND Journal of Economics.
[6] Choi, J.P., Yi, S.-S., 2000. Vertical Foreclosure and the Choice of Input Specifications. RAND Journal
of Economics 31, 717-743.
[7] Church, J., Gandal, N., 2000. Systems Competition, Vertical Merger, and Foreclosure. Journal of
Economics and Management Strategy 9, 25-51.
[8] Compte, O., Jenny, F., Rey, P., 2002. Capacity Constraints, Mergers and Collusion. European
Economic Review 46, 1-29.
[9] Cooper, J.C., Froeb, L.M., O’Brien, D., Vita, M.G., 2005. Vertical antitrust policy as a problem of
inference. International Journal of Industrial Organization 23, 639-664.
[10] Hart, O., Tirole, J., 1990. Vertical Integration and Market Foreclosure. Brookings Papers on Eco-
nomic Activity, Microeconomics, 205-276.
[11] Macauley, S., 1963. Non-Contractual Relations in Business. American Sociological Review 28, 55-70.
[12] Martin, S., Normann, H.T., Snyder, C.M., 2001. Vertical Foreclosure in Experimental Markets.
RAND Journal of Economics 32, 466-96.
[13] Nocke, V., White, L., 2006. Do Vertical Mergers Facilitate Upstream Collusion?, Forthcoming: Amer-
ican Economic Review.
[14] Normann, H.T. (2004). Equilibrium Vertical Foreclosure in the Repeated Game. SSRN Working
Paper #708441.
[15] O’Brien, D.P., Shaffer, G., 1992. Vertical Control with Bilateral Contracts. RAND Journal of Eco-
nomics 23, 299-308.
[16] Ordover, J.A., Saloner, G., Salop, S.C., 1990. Equilibrium Vertical Foreclosure. American Economic
Review 80, 127-42.
[17] Ordover, J.A., Saloner, G., Salop, S.C., 1992. Equilibrium Vertical Foreclosure: Reply. American
Economic Review 82, 698-703.
[18] Reiffen, D., 1992. Equilibrium Vertical Foreclosure: Comment. American Economic Review 82, 694-
97.
[19] Rey, P., Tirole, J., 2005. Foreclosure. In: Handbook of Industrial Organization, vol. III. M. Arm-
strong, Porter, R. (Eds.). North Holland.
[20] Riordan, M., 1998. Anticompetitive Vertical Integration by a Dominant firm. American Economic
Review 88, 1232-48.
[21] Riordan, M., 2005. Competitive Effects of Vertical Integration. Forthcoming in: Handbook of An-
titrust Economics, Paolo Buccirossi (ed.), MIT Press.
[22] Riordan, M., Salop, G., 1995. Evaluating Vertical Mergers - A Post Chicago Approach. Antitrust
Law Journal 63, 513-568.

25
[23] Salinger, M., 1988. Vertical Mergers and Market Foreclosure, The Quarterly Journal of Economics
103, 345-356.
[24] Vives, X., 1999. Oligopoly Pricing - Old Concepts and New Ideas. MIT Press.

26