Wholesale Bargaining: Models

and Antitrust Implications

Joshua Gans
Melbourne Business School
University of Melbourne
 Background Papers
• Joint work with Catherine de Fontenay
– RAND Journal of Economics, 2005
– Review of Network Economics, 2005
– IJIO, 2004
– “Bilateral Bargaining with Externalities”
• Applications
– “Concentration Measures and Vertical Market Structure”
(JL&E forthcoming)
– “Markets for Competitive Advantage” (w/ Michael Ryall)
– “Network Bargaining” (Martin Byford)
 Wholesale Markets
• Posted prices
– Spengler (JPE): double
marginalisation
– Salinger (QJE): successive
Cournot oligopoly
• Take it or leave it offers
– Hart and Tirole (1990)
– O’Brien and Shaffer (1992)
– McAfee and Schwartz (1994)
– Segal (1999)
– Marx and Shaffer (2004)
• Bargaining
– Inderst and Wey (2003)
– O’Brien and Shaffer (2004)
– Segal and Whinston (2001)
– Stole and Zwiebel (1996)
– Grossman-Hart-Moore
– MacDonald and Ryall (2004)
– Brandenberger and Stuart
(2006)
Antitrust Issues: Traditional Views
How do we analyse How do we analyse vertical
competition between restrictions?
sellers into a wholesale
market?
• Same as any horizontal • Perfect efficient contracting:
market vertical practices only
• Versus countervailing chosen for efficiency
power from buyers reasons
• Versus firms with market
power who can use practices
to extract rents
 New Results
• Changes in competition (e.g., concentration) in
upstream markets have a different impact on
final consumers than changes in downstream
markets
• Vertical practices can have anti-competitive
effects and result in a redistribution of rents
• Can use quantitative bargaining models to
analyse trade-offs
 Outline
1. Our Bargaining Model and Result
2. Treatment of Upstream Competition
3. Analysis of Vertical Integration
4. Future Directions
 Cooperative Bargaining Theory
The Benefits The Problems
• Relates environmental • Presumption that
characteristics to surplus coalitions operate to
division maximise surplus
• Easy to compute – Requires observable and
– E.g., Myerson-Shapley verifiable actions
value is weighted sum of • Coalitional externalities
coalitional values are usually assumed
away
– If considered, impact on
division only (Myerson)
 Non-Cooperative Bargaining Theory
• Benefit: Robust
predictions in the D1
bilateral case
– Nash bargaining
– Rubinstein and Binmore-
Rubinstein-Wolinsky
• Problem: Bilateral case
in isolation cannot deal
with
– externalities U1
– coalitional formation
 We need a theory that can deal with
this …
• Competitive Externalities
– Ds and Us may be competing D1 D2
firms
– Can’t negotiate
• Bilateral Contracts:
– Ds and Us cannot necessarily
observe supply terms of
others
– Connectedness does not
necessarily imply surplus UA UB
maximisation

… while being tractable and intuitive.

 Our Approach
• Bilaterality
– Assumes that there are no actions that can be
observed beyond a negotiating pair
– Potential for inefficient outcomes
• Non-cooperative bargaining
– Does not presume surplus maximisation
– Looks for an equilibrium set of agreements
 Our Results
In a non-cooperative model of a sequence of bilateral
negotiations …
• There exists a Perfect Bayesian Equilibrium whereby
– Coalitional surplus is generated by a Nash equilibrium
outcome in pairwise surplus maximisation
– Division is based on the weighted sum of coalitional
surpluses
• We produce a cooperative division of a non-
cooperative surplus
– Strict generalisation of cooperative bargaining solutions
– Collapses to known values as externalities are removed
– Non-cooperative justification for cooperative outcomes
 Some Notation
• Actions
– qij is the input quantity
purchased by Di from Uj D1 D2
– tij is the transfer from Di to Uj
– (A1) Can only observe
actions and transfers you are a
party to (e.g., UB and D2
cannot observe q11 or t11)
• Primitive Payoffs
– Di: π (qiA+qiB, q-iA+q-iB)–ti1–ti2
– Uj: t1j + t2j – c(q1j+q2j)
– Usual concavity assumptions
UA UB
on π (.) and c(.)
 Network State
• Network
– Bilateral links form a graph of
relationships denoted by K
• Initial state: K =
(1A,1B,2A,2B)
– If a pair suffer a breakdown
(e.g., D1 and UA), the new
network is created
• New state: K = (1B,2A,2B)
– (A2) The network state (K) is
publicly observed
 Possible Contracts
• Bilaterality
– As terms of other pairs are unobserved by at least
one member of a pair, supply terms cannot be
made contingent upon other supply contract terms
• Network Observability
– As the network state is publicly observed supply
terms can be made contingent on the network state
– Example:
• q11(1A,1B,2A,2B) = 3 and t11(1A,1B,2A,2B) = 2 and
q11(1A, 2A,2B) = 4 and t11(1A, 2A,2B) = 5 and so on.
 Extensive Form
• Fix an order of pairs (in this case 4)
– Precise order will not matter for equilibrium we focus on
• Each pair negotiates in turn
– Randomly select Di or Uj
– That agent, say Di, makes an offer {qij(K), tij(K)} for all possible K
including Di and Uj.
• If Uj accepts, the offer is fixed and move to next pair
• If Uj rejects,
– With probability, 1-σ , negotiations end and bargaining recommences over the
new network K –ij.
– Otherwise negotiations continue with Uj making an offer to Di.
• Binmore-Rubinstein-Wolinsky bilateral game embedded in a
sequence of interrelated negotiations
– Examine outcomes as σ goes to 1.
 Beliefs
• Game of incomplete information
– Need to impose some structure on out of equilibrium beliefs
– Issue in vertical contracting (McAfee and Schwartz; Segal) in that
one party knows what contracts have been signed with others and
offer/acceptance choices may signal those outcomes
• Simple approach: impose passive beliefs
– Let be the set of equilibrium agreements
– When i receives an offer from j of or
– i does not revise its beliefs about any other outcome of the game
{qˆij ( K ), tˆij ( K )}∀ij , K
qij ( K ) ≠ qˆij ( K ) tij ( K ) ≠ tˆij ( K )
 Equilibrium Outcomes: Actions
• Bilateral Efficiency
– A set of actions satisfied bilateral efficiency if for all ij in
K,
qˆij ( K ) ∈ arg max xij π (q iA + q iB , qˆ − iA( K ) + qˆ − iB ( K )) − c(q1 j + qˆ 2 j( K ))

• Suppose that all agents hold passive beliefs. Then, as

σ approaches 1, in any Perfect Bayesian
equilibrium, each qij(K) is bilaterally efficient (given
K).
 Equilibrium Outcomes: Actions
• Suppose that all agents hold passive beliefs. Then, as σ approaches 1, in any Perfect
Bayesian equilibrium, each qij (K) is bilaterally efficient (given K).
• Intuition
– Negotiation order: 1A,1B,2A,2B and suppose that 1A and 1B have agreed to the equilibrium
actions
– If 2A agree to the equilibrium action, 2B negotiate and as this is the last negotiation, it is
equivalent to a BRW case – so they choose the bilaterally efficient outcome
– If 2A agree to something else, D2 will know this but UB wont
• UB will base offers and acceptances on assumption that 2A have agreed to the equilibrium outcome
(given passive beliefs)
• D2 will base offers and acceptances on the actual 2A agreement. Indeed, D2 will be able to offer (and
have accepted) something different to the equilibrium outcome
– Given this, will 2A agree to something else?
• D2 will anticipate the changed outcome in negotiations with UB
• Under passive beliefs, UA will not anticipate this changed outcome (so its offers don’t change)
• D2 will make an offer based on:

q2 A ∈ arg max q2 A π (q 2 A + q 2 B (q 2 A ), qˆ1 A + qˆ1B ) − c (qˆ1 A + q 2 A) − c (qˆ1B + q 2 B( x 2 A))

• By the envelope theorem on q2B , this involves a bilaterally efficient choice of q2A .
 Equilibrium Outcomes: Payoffs
• Result: As σ approaches 1, there exists a perfect
Bayesian outcome where agents receive:
vD1 = 1
6 ( 3 ( πˆ (1 A,1B, 2 A, 2 B) − cˆ(1 A,1B, 2 A, 2 B)) + 2 πˆ (1 j, 2 j) − cˆ (1 j, 2 j ) − πˆ (iA, iB) +2cˆ (iA, iB))

vUA = 1
6 ( 3 ( πˆ (1A,1B, 2 A, 2 B) − cˆ(1A,1B, 2 A, 2 B) ) − 2πˆ (1 j, 2 j ) + cˆ(1 j, 2 j ) + πˆ (iA, iB) − 2cˆ(iA, iB) )

• This is each agent’s Myerson-Shapley value over

the bilaterally efficient surplus in each network.
 Remarks
• Stole and Zwiebel adopt a similar approach in proving their
non-cooperative game yields a Shapley value
– Make mistake: do not specify belief structure
• Our most general statement shows that the solution concept is
a graph-restricted Myerson value in partition function space.
– The symmetry in the buyer-seller network case masks some additional
difficulties in the general case
– There is some indeterminacy in the complete graph case
– The cooperative game solution concept has never been stated before
– Nor has it been related to component balance and fair allocation
– So our proof does cooperative game theory before getting to the steps
before
 Ultimate Solution

 
 1 1  vˆ (T , LP )
ϒi ( N , L) = ∑ ∑ (−1) ( p − 1)!
p −1
− ∑
P∈P N T ∈P
 N i∉T ′∈P ( p − 1)( N − T ′ ) 
 T ′≠T 

where:
• N is the set of agents
• P is a partition over the set of agents with cardinality p
• PN is the set of all partitions of N
• L is the initial network (i.e., initial set of bilateral links)
• LP is the initial network with links severed between partitions defined
by P.
 Additional Results
• (No component externalities) Suppose that primitive
payoffs are independent of actions taken by agents
not linked the agent
– Obtain the Myerson value over a bilaterally efficient
surplus.
• (No non-pecuniary externalities) Suppose that the
primitive payoffs are independent of the actions the
agent cannot observe
– Obtain the Myerson value.
• If agreements are non-binding and subject to
renegotiation, the results hold.
 Computability
 s −i  s  

m buyers m
vS1 = ∑   ∑

 m x
( − 1)  
 i   vˆ( m − s, 2)

s =0  s  i = 0 m − i + 2

 
 
 m − x − h − i +1  m − s − h  
 ( −2)   
m m− s m
   m − s   m− s− h  i  (−1)m − s − h vˆ ( s | h )
+ ∑∑     ∑ +
s =0 h =0  s   h  i = 0 m − i + 2 m − h +1 
 
 

Bilaterally efficient surplus

vˆ(m − s, 2) with m-s buyers supplied by
both suppliers
S1 S2
vˆ ( s | h ) Bilaterally efficient surplus if s
buyers are supplied only by S1
and h are supplied only by S2
 Upstream Competition

Why upstream competition should be

treated differently when there is
wholesale bargaining?
2 x 2 Structure (NI)

UA UB

D1 D2
 Model Structure & Notation
• 2 upstream & 2 downstream assets each with an associated
manager (necessary for the asset to be productive); integration
changes ownership but not need to use manager at same level
• Uj can produce input quantities, q1j & q2j to D1 and D2 at cost,
cj(q1j, q2j); quasi-convex
• D1 earns (gross) profits of π 1(q1A,q1B;q2A,q2B); concave in (q1A,q1B)
and non-increasing in (q2A,q2B).
• Industry profit outcomes:

Π ( D1 D2U AU B ) ≡ max q1 A ,q1 B π 1 (q1A , q1B , q 2 A , q 2B ) +π 2 (q 2 A , q 2B , q1A , q1B ) −c A (q1A , q 2 A ) −c B (q1B , q 2B )

q2 A , q2 B

Π ( D1U AU B ) ≡ max q1 A , q1B π1 (q1 A , q1B , 0, 0) − cA (q1 A , 0) − cB (q1B , 0)

Π ( D1U A ) ≡ max q1 A π 1 (q1 A , 0, 0, 0) − cA ( q1 A , 0)

Π ( D1U A , D2U B ) ≡ max q1 A π1 (q1 A , 0, 0, q2 B ) − c A ( q1 A , 0)

 Upstream Merger

UA UB UA UB

D1 D2 D1 D2
 Upstream Merger

UA UB

D1 D2
 Impact on Efficiency
• Bilateral negotiations for upstream supply under
upstream competition
max q1 A π 1 (q1 A , q1B ; q2 A , q2 B ) − cA (q1 A , q2 A )

• Bilateral negotiations for upstream supply under

upstream monopoly
max q1 A π 1 (q1 A , q1B ; q2 A , q2 B ) − cA (q1 A , q2 A ) − cB (q1B , q2 B )

No difference in outcomes so no impact on efficiency

 Illustrative Example (Distribution)
• Assumptions
– Demand: P = 1 – (q1 + q2)
– Costs: none
– Symmetry
– Monopoly profits = ¼
Merger means that if negotiations
breaks down with one downstream
firm, they split monopoly profits
with remaining one.

Upstream Upstream
Competition Monopoly
Output 2/3 2/3
Total Profits 2/9 2/9
Di payoff 0.051 0.0324
Uj payoff 0.0601 0.1574
 Incentives to Merge
• Upstream firms jointly gain:
– One third of the profits from a UB Monopoly
– Intuition: the possibility that a breakdown could
generate this was used by downstream firms as
leverage on the other upstream firm
• Downstream firms jointly lose this
– Face higher transfers
 Upstream Competition
• Changes to upstream competition have a different
impact to changes in downstream competition
• Fragmentation amongst downstream firms drives
impact on consumers, input and output choices. It
constrains upstream market power.
• Extreme: permit upstream mergers when there is no
vertical integration
– Leads to additional upstream investment (maybe over-
investment)
– May lead to reduced downstream entry
 Vertical Integration

What is the competitive impact of

vertical integration?
 Effect of Integration
UA UB
BI
UA UB

D1 D2

D1 D2 UA UB

FI
D1 D2
 Will D1 and UA profit from VI?

UC UM
1 ˆ (D D U U )
∆UC Π ˆ (D D U U )
2 1 2 A B 1
2∆UM Π 1 2 A B
+1∆ Π ˆ (D D U )
FI 6 UC 1 2 A + 1 ∆ Πˆ ( D D U )
UM 1 2 A

( )
6
ˆ (D D U ) − Π (D U )
+ 16 Π 1 2 B 2 B

1 ˆ (D D U U )
∆UC Π 1 ˆ (D D U U )
∆UM Π
2 1 2 A B 2 1 2 A B

+1∆ Π ˆ (D D U ) +1∆ Π ˆ (D D U )
BI 6 UC 1 2 A 6 UM 1 2 A

(
+ 16 Π ( D1U AU B ) − Π ( D2U B ) ) (
+ 16 Π ( D2U AU B ) − Πˆ ( D1 D2U B ) )
 Comparisons
• FI versus BI
Πˆ ( D D U ) ≥ Π( D U U )
1 2 B 2 A B
• UC versus UM
1
3 ( Πˆ ( D D U
1 2 B )
) − Π( D2U B )
ˆˆ ˆˆ
≥ ΠUM ( D1 D2U AU B ) − ΠUC ( D1 D2U AU B )
As downstream products become more differentiated, strategic VI is
more likely under upstream competition than upstream monopoly
 Impact on Efficiency
• Bilateral negotiations for upstream supply under NI
max q1 A π 1 (q1 A , q1B ; q2 A , q2 B ) − c A (q1 A , q2 A )
max q2 A π 2 (q2 A , q2 B ; q1 A , q1 B ) − c A (q1 A , q 2 A )

• Bilateral negotiations for upstream supply under VI

max q1 A π 1 (q1 A , q1B ; q2 A , q2 B ) − cA (q1 A , q2 A )
max q2 A π 1 (q1 A , q1 B ; q2 A , q2 B ) + π 2 (q 2 A, q 2 B; q1 A, q1 B) − c A(q 1 A, q 2 A)

Incentive to ‘raise rivals’ costs’

 VI and Foreclosure
• Upstream competition
– With homogenous inputs (and some symmetry), VI does not change
efficiency
• Upstream monopoly
– VI leads to industry profit maximisation (with symmetry and
substitutability); D2 is not supplied any inputs
– Under FI, D2 still receives a payoff of:

(
vD2 ( FI ) = 121 Π ( D1D2U AU B ) − Π ( DU
i j) )
Technical foreclosure but downstream firm still valuable in
disciplining internal negotiations
Quantitative Evaluation

How can mergers impacting on

vertical market structure be
evaluated?
 Wholesale Bargaining
• N firms indexed by i
– Each may operate in an upstream and/or
downstream segment
– si: downstream share
σ i: upstream share
• Perfect substitutes (downstream)
– Market demand: P(Q)
– Costs: upstream (Ci(.)), downstream (ci(.))
 Lerner Index for Vertical Chain
• Negotiations between i and j:

( )
max qij P (Q ) qij + ∑ k ≠ j q ik − c i (q ij ,.) − C i (.) + P (Q ) ∑ k q jk −c j(.) −C j(q ij,.)

P (Q) + P (Q )∑ k ( qik + q jk ) − ∂qij − ∂qijj = 0

∂ci ∂C
′

P′(Q)∑ k ( qik + q jk )
∂ci ∂C j
P− − 1
= ( si + s j )
∂qij ∂qij
⇒ =−
P P ε
 Vertical HHI
• The average Lerner index is:

∑ ∑
N N
1
ε s ( si + σ i − qii / Q) = HHI +
i =1 i
1
ε
1
ε s ( σi −qii / Q)
i =1 i

HHI ≡ ∑
N 2
s
i =1 i

• If there is a preference for internal supply,

∑
N
VHHI = 1
ε s max{ si ,σ i }
i =1 i
 Properties
• Ranges between 0 (perfect competition) and 10,000
(downstream monopoly)
– Collapses to HHI (Downstream) when all downstream
firms are net buyers of inputs or non-integrated
– If there is integration then VHHI > HHI
• Upstream concentration not relevant
– Non-integrated upstream mergers do not change VHHI
– Only look upstream if merger involves a net supplier
 Some Examples
• Example 1:
– 4 equal sized upstream firms and 10 equal sized
downstream ones
– Up HHI = 2500; Down-HHI = 1000 = VHHI
– Vertical merger leaves HHI’s unchanged (no concern) but
raises VHHI to 1150 (potential concern)
• Example 2:
– 8 downstream firms with 10% share and a 9th with 20%
share
– If vertical merger involves large firm then HHI does not
change but VHHI goes from 1300 to 1400 (no concern)
despite higher concentration.
 Approach #2: Successive Oligopoly
• Firm’s post unit prices in wholesale market
• With linear demand and costs (and homogenous
inputs):

( )
N N N N 2

∑s ∑σ (σ j − s j ) + ∑ ∑
( si − min[ si ,σ i ])(σ j −min[ s j ,σ j ])
VHHI = 1
max[ s j ,σ j ] + 1
ε
j =1
j ε
j =1
j
j =1 i =1
1− ∑ i min[ si ,σ i ]

• Like bilateral bargaining but with additional

distortions (that can be removed by vertical
integration)
 Application: Exxon-Mobil in California
Company Upstream (Refining) Market Share Downstream (Retailing) Market Share (%)
(%)
Chevron 26.4 19.2
Tosco 21.5 17.8
Equilon 16.6 16
Arco 13.8 20.4
Mobil 7 9.7
Exxon 7 8.9
Ultramar 5.4 6.8
Paramount 2.3 0
Kern 0 0.3
Koch 0 0.2
Vitol 0 0.2
Tasoro 0 0.2
PetroDiamond 0 0.1
Time 0 0.1
Glencoe 0 0.1
 Concentration Measures
Concentration Pre-Merger Post-Merger Post-Merger Post-Merger
Measure with Exxon with Exxon
Refinery Retail
Divestiture Divestiture
Upstream HHI 1800 1900 1800 1900

Downstream HHI 1600 1800 1800 1600

VHHI Contracting 2100 2400 2400 2100

VHHI Cournot 2200 2500 2700 2100

 Future Directions
• Other Vertical Practices
– Exclusive dealing
– Negotiations over linear prices
• Quantitative Analysis
– Construct simulation model of bargaining
– Empirical tests of vertical market structure
concentration measures and pricing