Nathan Grognet LLSMS2035 2015 - 2016

Supply Chain Coordination with Contracts

Introduction, scope of coordination
Optimal performance is achievable if the firms coordinate by contracting on a set of transfer
payments such that each firm’s objective becomes aligned with the supply chain’s objective.

Three important questions on the newsvendor model:

• Which contracts coordinate the supply chain? A contract is said to coordinate the
supply chain if the set of supply chain optimal actions is a Nash equilibrium (no firm
has a profitable unilateral deviation from the set of supply chain optimal actions). Here,
the action to coordinate is the retailer’s order quantity.

• Which contracts have sufficient flexibility to allow for any division of the supply chain’s
profit among the firms? There always exists a contract that Pareto dominates a non-
coordinating contract (each firm’s profit is no worse off and at least one firm is strictly
better off with the coordinating contract)

• Which contracts are worth adopting? Those tend to be costly to administer.

The same analysis recipe is usually followed for each variation of the newsvendor model:

1) Identify the type of contracts that can coordinate the supply chain
2) Determine for each contract type the set of parameters that achieves coordination
3) Evaluate for each coordinating contract type the possible range of profit allocations

The “Triple-A Supply Chain” aims on:

• Agility to respond to short-term changes in demand or supply quickly

• Adaptability to adjust the design to meet structural changes in markets, strategies,
products and technologies
• Alignment to establish and monitor common incentive structures with infra-firm
functions and among inter-firm supply chain partners

Analysis of incentive problems

The root of those problems is often when incentives offered lead to actions that increase
variability and reduce total supply chain profits (misalignment of total supply chain objectives
and individual objectives).

Badly-specified incentives and features that complicate incentives design are the main causes.
It happens when each parties are concentrate on their own interests, distributors purchase
quantities to take advantage on price promotion, etc.

Contract components
Every contract needs: specification of decision rights, pricing, minimum purchase
commitments, quantity flexibility, buyback or returns policies, allocation rules, lead times, and
quality.

1
Nathan Grognet LLSMS2035 2015 - 2016

Solutions to incentive problems

• Contracting-based solution: changing the incentives (via contracts)
• Information-based solution: gain information to improve monitoring
• Trust-based solution: use of the relationship to induce coordination
• Structure-based solution: organize the supply chain, the process or the product to align
incentive

Roles of SC contracts
There is a role of coordination for products/services and resource/process allocation, but also
a role of rent allocation concerning business risks and profit allocation.

An action or outcome is contractible if it is observable, verifiable, and enforceable.

Challenges of coordinating contracts

Moral hazard
When one firm can take actions that affect the entire chain, but those actions are not
contractible.

Idiosyncratic investments
When one firm invests in an asset that has value only with respect to its relationship with a
specific supply chain partner.

Adverse selection
When two firms have different levels of information about a key contract parameter prior to
entering into the contract.

Contract-based solutions
Principle: to solve moral hazard (define contractible outcomes that correlate with the right
action) and adverse selection (define after-sales provisions to signal quality (warranties, return-
rights, etc.))

2
Nathan Grognet LLSMS2035 2015 - 2016

Theory: principal-agency theory (no contract is complete) and game theory (even incomplete
contracts may work if they are sound and actors are repeating interaction)

Examples: profit sharing or group incentives (moral hazard), intermediaries for labor law
compliance (adverse selection)

Information-based solutions
Principle: collective more information to bridge asymmetric information

Theory: investments in observability do reduce the profitability of moral hazard or adverse

selection

Examples: mystery shoppers (moral hazard), industry associations (adverse selection)

In practice:

• Collection of demand data and characteristics: CPFR, QR

• Collection of supplier performance indicators: balanced scorecard, Sun
• Collection of inventory data: VMI, CRM
• Collection of customer satisfaction data: mystery shopper
• Integration of supplier data
• Use of commercial databases

Trust-based solutions
Principle: transform episodic relationships to repeat interaction

Theory: win-win interaction (cooperative game theory) is NE in repeated games

Examples: The Box (moral hazard)

Structural solutions
Principle: internalize outcomes to overcome incentive problems, vertical integration creates
legal and economic means to align

Theory: solves trivially the problem in theory, but incentive provision in large bureaucracies is
costly and management is less effective

Examples: spearhead factories (adverse selection)

In practice: vertical integration (both backward and forward), assignment of roles, delegation
of procurement/tendering/control, rotation of roles, and product design (module interfaces)

Summary: aligning incentives in supply chain

1) Recognizing goal incongruence: frequently obvious to managers in the supply chain,
can be expected during reengineering efforts

2) Pinpointing the cause of goal incongruence: look at economics of each decision from
decision-maker’s and supply chain’s perspective, identify decisions that would have
been made differently, and trace these differences to moral hazard and pre- and post-
contract private information

3
Nathan Grognet LLSMS2035 2015 - 2016

3) Overcoming goal incongruence: select one or several categories of solution types, and
design a solution to address the specific incongruence

Conclusions
Absolute alignment of incentives in a de-centralized supply chain is not possible except in the
simplest of decision making environments.

In the real-world, we need to identify the decisions that most critically require coordination,
and give priority to aligning the incentives for these critical decisions in the design of supply
chain structures, design of contracts, the use of trust and the collection and analysis of
information.

The effectiveness of any specific solutions depends on its potential power, but also on the
current level of opportunism of the partners involved in a transaction or relation.

Coordinating the newsvendor

The compliance regime (force or voluntary) determines the consequences for failing to adhere
to a contract and therefore matters because it influences the kinds of contracts that coordinate
the supply chain.

Model and analysis

The retailer must choose an order quantity before the start of a single selling season. Let 𝐹 be
the distribution function of demand and 𝑓 its density function. For each demand the retailer
doesn’t satisfy the retailer incurs a goodwill penalty cost 𝑔𝑟 and the analogous cost for the
supplier is 𝑔𝑠 . The retailer earns a salvage price 𝑣 < 𝑐 per unit unsold at the end of the season.

The following sequence of events occurs in this game:

1) The supplier offers the retailer a contract

2) The retailer accepts or rejects the contract
3) Assuming the retailer accepts the contract, he submits an order quantity 𝑞 to the
supplier
4) The supplier produces and delivers to the retailer before the selling season
5) Season demand occurs
6) Transfer payments are made between the firms based upon the agreed contract

The retailer could assume the supplier operates under voluntary compliance, which means the
supplier delivers the amount that maximizes his profit given the terms of the contract.
Alternatively, the retailer could believe the supplier never chooses to deliver less than the
retailer’s order because the consequences for doing so are sufficiently great, that is called force
compliance. Any contract that coordinates the supply chain with voluntary compliance surely
coordinates with forced compliance, but the reverse is not true.

Let 𝑆(𝑞) be the expected sales, min⁡(𝑞, 𝐷),

𝑞 𝑞
𝑆(𝑞) = 𝑞(1 − 𝐹(𝑞)) + ∫ 𝑦𝑓(𝑦)𝑑𝑦 = 𝑞 − ∫ 𝐹(𝑦)𝑑𝑦
0 0

4
Nathan Grognet LLSMS2035 2015 - 2016

Let 𝐼(𝑞) be the expected left over inventory,

𝐼(𝑞) = (𝑞 − 𝐷)+ = 𝑞 − 𝑆(𝑞)

Let 𝐿(𝑞) be the lost sales function,

𝐿(𝑞) = (𝐷 − 𝑞)+ = 𝜇 − 𝑆(𝑞)

And let 𝑇 be the expected transfer payment from the retailer to the supplier.

The retailer’s profit function is

𝜋𝑟 (𝑞) = 𝑝𝑆(𝑞) + 𝑣𝐼(𝑞) − 𝑔𝑟 𝐿(𝑞) − 𝑐𝑟 𝑞 − 𝑇

⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡= (𝑝 − 𝑣 + 𝑔𝑟 )𝑆(𝑞) − (𝑐𝑟 − 𝑣)𝑞 − 𝑔𝑟 𝜇 − 𝑇

The supplier’s profit function is

𝜋𝑠 (𝑞) = 𝑔𝑠 𝑆(𝑞) − 𝑐𝑠 𝑞 − 𝑔𝑠 𝜇 + 𝑇

And the supply chain’s profit is

Π(𝑞) = 𝜋𝑟 (𝑞) + 𝜋𝑠 (𝑞) = (𝑝 − 𝑣 + 𝑔)𝑆(𝑞) − (𝑐 − 𝑣)𝑞 − 𝑔𝜇

Let 𝑞° = arg max Π(𝑞) be the supply chain optimal order quantity, it satisfies
𝑐−𝑣
𝑆 ′ (𝑞°) = 𝐹̅ (𝑞°) =
𝑝−𝑣+𝑔

And let 𝑞𝑟∗ be the retailer’s optimal order quantity.

The wholesale price contract

𝑇𝑤 (𝑞, 𝑤) = 𝑤𝑞

The retailer’s unique optimal quantity satisfies

(𝑝 − 𝑣 + 𝑔𝑟 )𝑆 ′ (𝑞𝑟∗ ) − (𝑤 + 𝑐𝑟 − 𝑣) = 0

Since 𝑆′(𝑞) is decreasing, 𝑞𝑟∗ = 𝑞° only when

𝑝 − 𝑣 + 𝑔𝑟
𝑤=( ) (𝑐 − 𝑣) − (𝑐𝑟 − 𝑣)
𝑝−𝑣+𝑔

According to that equation, the supplier should earn a non-positive profit, since 𝑤 ≤ 𝑐𝑠 . As a
result, the wholesale price contract generally does not coordinate the supply chain.

There is a “double marginalization” problem because there are two margins and neither firm
considers the entire supply chain’s margin when making a decision.

Let 𝑤(𝑞) be the unique wholesale price that induces the retailer to order 𝑞𝑟∗ units

𝑤(𝑞) = (𝑝 − 𝑣 + 𝑔𝑟 )𝐹̅ (𝑞) − (𝑐𝑟 − 𝑣)

𝜋𝑠 (𝑞, 𝑤(𝑞)) = 𝑔𝑠 𝑆(𝑞) + (𝑤(𝑞) − 𝑐𝑠 )𝑞 − 𝑔𝑠 𝜇

The compliance regime doesn’t matter with this contract, the supplier surely produces and
delivers whatever quantity the retailer orders.

5
Nathan Grognet LLSMS2035 2015 - 2016

An “Increasing Generalized Failure Rate” (IGFR) demand distribution implies that there is a
unique sales quantity 𝑞𝑠∗ that maximizes the supplier’s profit.

Two performance measures can be used:

• The efficiency of the contract Π(𝑞𝑠∗ )⁄Π(𝑞°)

• The supplier’s profit share 𝜋𝑠 (𝑞𝑠∗ , 𝑤(𝑞𝑠∗ ))⁄Π(𝑞𝑠∗ )

The product of those ratios is the supplier’s share of the supply chain’s optimal profit

𝜋𝑠 (𝑞𝑠∗ , 𝑤(𝑞𝑠∗ )) 𝜋𝑠 (𝑞𝑠∗ , 𝑤(𝑞𝑠∗ )) Π(𝑞𝑠∗ )

=
Π(𝑞°) Π(𝑞𝑠∗ ) Π(𝑞°)

The retailer’s profit represents compensation for bearing risk. There’s no variation in the
supplier’s profit, but the retailer’s profit varies with the realization of demand. As the coefficient
of variation of the demand decreases the retailer faces less demand risk and therefore his
compensation is reduced. If the retailer were risk averse, the supplier would have to provide
for yet more compensation. Instead, the retailer is being compensated for the risk that demand
and supply do not match.

The buy-back contract

The supplier charges the retailer 𝑤 per unit purchased, but pays the retailer 𝑏 per unit
remaining at the end of the season.

𝑇𝑏 (𝑞, 𝑤, 𝑏) = 𝑤𝑞 − 𝑏𝐼(𝑞) = 𝑏𝑆(𝑞) + (𝑤 − 𝑏)𝑞

A retailer should not profit from left over inventory, so assume 𝑏 ≤ 𝑤. If the retailer’s salvage
value is higher than the supplier’s, the retailer salvages the units and the supplier credits the
retailer for those units, which is sometime referred to as “markdown money”.

𝜋𝑟 (𝑞, 𝑤𝑏 , 𝑏) = (𝑝 − 𝑣 + 𝑔𝑟 − 𝑏)𝑆(𝑞) − (𝑤𝑏 − 𝑏 + 𝑐𝑟 − 𝑣)𝑞 − 𝑔𝑟 𝜇

𝑝 − 𝑣 + 𝑔𝑟 − 𝑏 = 𝜆(𝑝 − 𝑣 + 𝑔)
𝑤𝑏 − 𝑏 + 𝑐𝑟 − 𝑣 = 𝜆(𝑐 − 𝑣)
𝜋𝑟 (𝑞, 𝑤𝑏 , 𝑏) = 𝜆(𝑝 − 𝑣 + 𝑔)𝑆(𝑞) − 𝜆(𝑐 − 𝑣)𝑞 − 𝑔𝑟 𝜇 = 𝜆Π(𝑞) + 𝜇(𝜆𝑔 − 𝑔𝑟 )

It follows that 𝑞𝑟∗ = 𝑞° is optimal for the retailer.

𝜋𝑠 (𝑞, 𝑤𝑏 , 𝑏) = Π(𝑞) − 𝜋𝑟 (𝑞, 𝑤𝑏 , 𝑏) = (1 − 𝜆)Π(𝑞) − 𝜇(𝜆𝑔𝑠 − (1 − 𝜆)𝑔𝑟 )

So the buy-back contract coordinates with voluntary compliance as long as 𝜆 ≤ 1. Voluntary

compliance actually increases the robustness of the supply chain. Suppose the retailer is not
rational and orders 𝑞 > 𝑞°, the supplier will then correct the retailer’s mistake by delivering only
𝑞° units. However, he can’t correct it if the retailer orders less than 𝑞°.

The 𝜆 parameter acts to allocate the supply chain’s profit between two firms. The retailer earns
the entire supply chain profit when
Π(𝑞°) + 𝜇𝑔𝑟
𝜆= ≤1
Π(𝑞°) + 𝜇𝑔

6
Nathan Grognet LLSMS2035 2015 - 2016

And the supplier earns the entire supply chain profit when
𝜇𝑔𝑟
0≤𝜆=
Π(𝑞°) + 𝜇𝑔

That coordination of the supply chain requires the simultaneous adjustment of both the
wholesale price and the buy-back rate. The players should never negotiate these parameters
sequentially, instead negotiations should always allow simultaneous changes. The parameter 𝜆
can be seen as the retailer’s share.
𝑏 + 𝑔𝑠
𝑤𝑏 (𝑏) = 𝑏 + 𝑐𝑠 − (𝑐 − 𝑣) ( )
𝑝−𝑣+𝑔

In general, a contract coordinates the retailer’s and the supplier’s action whenever each firm’s
profit is an affine function of the supply chain’s profit.

The revenue-sharing contract

The supplier charges 𝑤𝑟 per unit purchased plus the retailer gives the supplier a percentage of
his revenue (salvage revenue is also shared). Let 𝜙 be the fraction of supply chain revenue the
retailer keeps, so (1 − 𝜙) is the fraction the supplier earns.

𝑇𝑟 (𝑞, 𝑤𝑟 , 𝜙) = (𝑤𝑟 + (1 − 𝜙)𝑣)𝑞 + (1 − 𝜙)(𝑝 − 𝑣)𝑆(𝑞)

𝜋𝑟 (𝑞, 𝑤𝑟 , 𝜙) = (𝜙(𝑝 − 𝑣) + 𝑔𝑟 )𝑆(𝑞) − (𝑤𝑟 + 𝑐𝑟 − 𝜙𝑣)𝑞 − 𝑔𝑟 𝜇
𝜙(𝑝 − 𝑣) + 𝑔𝑟 = 𝜆(𝑝 − 𝑣 + 𝑔)
𝑤𝑟 + 𝑐𝑟 − 𝜙𝑣 = 𝜆(𝑐 − 𝑣)
𝜋𝑟 (𝑞, 𝑤𝑟 , 𝜙) = 𝜆Π(𝑞) + 𝜇(𝜆𝑔 − 𝑔𝑟 )

Hence, 𝑞° is the retailer’s optimal order quantity.

𝜋𝑠 (𝑞, 𝑤𝑟 , 𝜙) = Π(𝑞) − 𝜋𝑟 (𝑞, 𝑤𝑟 , 𝜙) = (1 − 𝜆)Π(𝑞) − 𝜇(𝜆𝑔 − 𝑔𝑟 )

So 𝑞° is also the supplier’s optimal quantity as long as 𝜆 ≤ 1. The revenue-sharing contract

coordinate the supply chain and arbitrarily allocate its profit.

The revenue sharing and the buy-back contract are equivalent when

𝑤𝑏 − 𝑏 = 𝑤𝑟 + (1 − 𝜙)𝑣
𝑏 = (1 − 𝜙)(𝑝 − 𝑣)

The revenue-sharing contract generates the same profits for the two firms for any realization
of demand as the following buy-back contract

𝑤𝑏 = 𝑤𝑟 + (1 − 𝜙)𝑝
𝑏 = (1 − 𝜙)(𝑝 − 𝑣)

The quantity-flexibility contract

The supplier charges 𝑤𝑞 per unit purchased but then compensates the retailer for his losses on
unsold units. The retailer receives a credit from the supplier at the end of the season equal to
(𝑤𝑞 + 𝑐𝑟 − 𝑣) min(𝐼, 𝛿𝑞), where 𝐼 is the left over inventory and 𝛿 a contract parameter. This
contract fully protects the retailer on a portion of the retailer’s order whereas the buy-back

7
Nathan Grognet LLSMS2035 2015 - 2016

contract gives partial protection on the retailer’s entire order. If the supplier didn’t compensate
the retailer for the 𝑐𝑟 cost per unit then the retailer would receive only partial compensation
on a limited number of units, which is called a backup agreement.
𝑞
𝑇𝑞 (𝑞, 𝑤𝑞 , 𝛿) = 𝑤𝑞 𝑞 − (𝑤 + 𝑐𝑟 − 𝑣) ∫ 𝐹(𝑦)𝑑𝑦
(1−𝛿)𝑞

Where the last term is the retailer’s compensation for unsold units, up to the limit of 𝛿𝑞 units.

𝜋𝑟 (𝑞, 𝑤𝑞 , 𝛿) = (𝑝 − 𝑣 + 𝑔𝑟 )𝑆(𝑞) − (𝑐𝑟 − 𝑣)𝑞 − 𝑇𝑞 (𝑞, 𝑤𝑞 , 𝛿) − 𝜇𝑔𝑟

𝑞
⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡= (𝑝 − 𝑣 + 𝑔𝑟 )𝑆(𝑞) − (𝑤𝑞 + 𝑐𝑟 − 𝑣)𝑞 + (𝑤𝑞 + 𝑐𝑟 − 𝑣) ∫ 𝐹(𝑦)𝑑𝑦 − 𝜇𝑔𝑟
(1−𝛿)𝑞

To achieve coordination it’s necessary (but not sufficient) that the retailer’s first order condition
holds at 𝑞°. The wholesale price that satisfies this is

(𝑝 − 𝑣 + 𝑔𝑟 )(1 − 𝐹(𝑞°))
𝑤𝑞 (𝛿) = − 𝑐𝑟 + 𝑣
1 − 𝐹(𝑞°) + (1 − 𝛿)𝐹((1 − 𝛿)𝑞°)

This is a coordinating wholesale price if the retailer’s profit function is concave which holds
when 𝑣 − 𝑐𝑟 ≤ 𝑤𝑞 (𝛿) ≤ 𝑝 + 𝑔𝑟 − 𝑐𝑟 and that range is satisfied with 𝛿 ∈ [0,1]. The wholesale
price 𝑤𝑞 (𝛿) is increasing in 𝛿.
𝑞
𝜋𝑠 (𝑞, 𝑤𝑞 (𝛿), 𝛿) = 𝑔𝑠 𝑆(𝑞) + (𝑤𝑞 (𝛿) − 𝑐𝑠 )𝑞 − (𝑤𝑞 (𝛿) + 𝑐𝑟 − 𝑣) ∫ 𝐹(𝑦)𝑑𝑦 − 𝜇𝑔𝑠
(1−𝛿)𝑞

The supply chain coordination under voluntary compliance is not assured with a quantity-
flexibility contract even if the wholesale price is 𝑤𝑞 (𝛿). Channel coordination is achieved with
forced compliance since then the supplier’s action is not relevant.

When 𝛿 = 0, the retailer earns at least the supply chain optimal profit. When 𝛿 = 1, it is the
supplier.

The sales-rebate contract

The supplier charges 𝑤𝑠 per unit purchased but then gives the retailer an 𝑟 rebate per unit sold
above a threshold 𝑡.
𝑤𝑠 𝑞 𝑞<𝑡
𝑞
𝑇𝑠 (𝑞, 𝑤𝑠 , 𝑟, 𝑡) = {
(𝑤𝑠 − 𝑟)𝑞 + 𝑟 (𝑡 + ∫ 𝐹(𝑦)𝑑𝑦) 𝑞 ≥ 𝑡
𝑡

When 𝑞 ≥ 𝑡 the retailer pays 𝑤𝑠 − 𝑟 for every unit purchased, an additional 𝑟 per unit for the
first 𝑡 units purchased and an additional 𝑟 per unit for the units not sold above the 𝑡 threshold.

𝜋𝑟 (𝑞, 𝑤𝑠 , 𝑟, 𝑡) = (𝑝 − 𝑣 + 𝑔𝑟 )𝑆(𝑞) − (𝑐𝑟 − 𝑣)𝑞 − 𝑔𝑟 𝜇 − 𝑇𝑠 (𝑞, 𝑤𝑠 , 𝑟, 𝑡)

But this contract is interesting only if it achieves supply chain coordination for 𝑡 < 𝑞°.

𝑤𝑠 (𝑟) = (𝑝 − 𝑣 + 𝑔𝑟 + 𝑟)𝐹̅ (𝑞°) − 𝑐𝑟 + 𝑣

Concerning the allocation of profit, the retailer earns more than Π(𝑞°) with 𝑡 = 0. There must
be a 𝑡 such that 𝜋𝑟 (𝑞°, 𝑤𝑠 (𝑟), 𝑟, 𝑡) = Π(𝑞°), the retailer earns the supply chain’s profit.

8
Nathan Grognet LLSMS2035 2015 - 2016

Let 𝑞̅ = arg max 𝜋𝑟 (𝑞, 𝑤𝑠 (𝑟), 𝑟, 𝑡), when 𝑡 = 𝑞°, the retailer earns 𝜋𝑟 (𝑞̅ , 𝑤𝑠 (𝑟), 𝑟, 𝑡), and with a
q≤t
sufficiently large 𝑟 such that profit is zero; the supplier earns the supply chain’s profit.

The sales-rebate contract is parameter rich: these three parameters are more than sufficient to
coordinate one action and to redistribute rents.

𝜋𝑠 (𝑞, 𝑤𝑠 (𝑟), 𝑟, 𝑡) = −𝑔𝑠 (𝜇 − 𝑆(𝑞)) − 𝑐𝑠 𝑞 + 𝑇𝑠 (𝑞, 𝑤𝑠 (𝑟), 𝑟, 𝑡)

The sales-rebate contract doesn’t coordinate the supply chain with voluntary compliance.

The quantity-discount contract

This contract considers an “all unit” quantity discount. The wholesale price 𝑤𝑑 (𝑞) is decreasing
in 𝑞.

𝑇𝑑 (𝑞) = 𝑤𝑑 (𝑞)𝑞

𝜋𝑟 (𝑞, 𝑤𝑑 (𝑞)) = (𝑝 − 𝑣 + 𝑔𝑟 )𝑆(𝑞) − (𝑤𝑑 (𝑞) + 𝑐𝑟 − 𝑣)𝑞 − 𝑔𝑟 𝜇

One technique to obtain coordination is to choose the payment schedule such that the
retailer’s profit equals a constant fraction of the supply chain’s profit.

𝑆(𝑞)
𝑤𝑑 (𝑞) = ((1 − 𝜆)(𝑝 − 𝑣 + 𝑔) − 𝑔𝑠 ) ( ) + 𝜆(𝑐 − 𝑣) − 𝑐𝑟 + 𝑣
𝑞

The above is decreasing in 𝑞 if 𝜆 ≤ 𝜆̅, where

𝑝 − 𝑣 + 𝑔𝑟
𝜆̅ =
𝑝−𝑣+𝑔

𝜋𝑟 (𝑞, 𝑤𝑑 (𝑞)) = 𝜆(𝑝 − 𝑣 + 𝑔)𝑆(𝑞) − 𝜆(𝑐 − 𝑣)𝑞 − 𝑔𝑟 𝜇

= 𝜆(Π(𝑞) + 𝑔𝜇) − 𝑔𝑟 𝜇

Hence, 𝑞° is optimal for the retailer and the supplier. The parameter 𝜆 acts to allocate the supply
chain’s profit. However, the upper bound on 𝜆 prevents too much profit from being allocated
to the retailer with a quantity discount. The 𝑤𝑑 (𝑞) coordinates even if 𝜆 > 𝜆̅, but then 𝑤𝑑 (𝑞) is
increasing in 𝑞. In that case the retailer pays a quantity premium.

Discussion
The revenue sharing and buy-back contracts are equivalent. Each of the five contracts
coordinates by inducing the retailer to order more than he would with a wholesale price
contract. Revenue sharing and quantity-flexibility contracts do this by giving the retailer some
downside protection: if demand is lower than 𝑞, the retailer gets some refund. The sales rebate-
contract does this by giving the retailer upside incentive: if demand is greater than 𝑡, the retailer
effectively purchases the units sold above 𝑡 for less than their cost of production. The quantity
discount coordinates by adjusting the retailer’s marginal cost curve so that the supplier earns
progressively less on each unit.

Except for the standard wholesale price and quantity-discount contract that only require a
single transaction, the mentioned contracts are costly to administer. The supplier must monitor

9
Nathan Grognet LLSMS2035 2015 - 2016

the number of units the retailer has left at the end of the season, or the remaining units must
be transported back to the supplier, depending on where the units are salvaged.

With the exception of the quantity-discount contract, the contracts shift risk between the two
firms: as the retailer’s share of profit decreases, his risk decreases and the supplier’s risk
increases.

The supplier’s exposure to demand uncertainty with some of the coordinating contracts could
matter to the supplier if the retailer chose an order quantity other than 𝑞°. Under voluntary
compliance the supplier can avoid an excessive ordering error by shipping only 𝑞°, but with
forced compliance the supplier bears the full risk of an irrational retailer.

Revenue sharing and quantity discounts always coordinate the supplier’s action with voluntary
compliance, quantity-flexibility contracts generally, but not always, coordinate the supplier’s
action and sales-rebate contracts never do.

No consider the application of these contracts in a setting with heterogeneous retailers that
do not compete. If only one contract is offered, then it coordinates all of the retailers as long
as the set of coordinating contracts doesn’t depend on something that varies across the
retailers. However, in some case heterogeneity can be accommodated with a single contract.
The independence of a contract to some parameter is also advantageous if the supplier lacks
information regarding the parameter. For example, a supplier doesn’t need to know a retailer’s
demand distribution to coordinate the supply chain with a revenue-sharing contract, but would
need to know the retailer’s demand distribution with a quantity flexibility, sales rebate or
quantity-discount contract.

However the supplier may want to divide the retailers by offering a menu of contracts. Since
coordinating buy-back contracts are independent of the demand distribution, this separation
requires the supplier to offer non-coordinating buy-back contracts, i.e., supply chain efficiency
must be sacrificed to induce forecasting. Quantity-flexibility contracts do depend on the
demand distribution, so a menu can be constructed with two coordinating quantity-flexibility
contracts, i.e., supply chain efficiency need not be sacrificed. Surprisingly, unless forecasting is
very expensive, the supplier is still better off using the menu of buy-back contracts even though
this sacrifices some efficiency.

Coordinating the newsvendor with price dependent

demand
Model and analysis
The retailer now choose his price in addition to his order quantity. For initial insights, let’s
assume the retailer sets his price 𝑝 at the same time as his stocking decision 𝑞.

Π(𝑞, 𝑝) = (𝑝 − 𝑣 + 𝑔)𝑆(𝑞, 𝑝) − (𝑐 − 𝑣)𝑞 − 𝑔𝜇

Assume there exists a finite (but not necessarily unique) optimal quantity-price pair {𝑞°, 𝑝°}. Let
𝑝°(𝑞) be the supply chain optimal price for a given 𝑞.

10
Nathan Grognet LLSMS2035 2015 - 2016

A contract fails to coordinate if it’s unable to satisfy

𝜕Π(𝑞, 𝑝°(𝑞)) 𝜕𝑆(𝑞, 𝑝°(𝑞))

= 𝑆(𝑞, 𝑝°(𝑞)) + (𝑝°(𝑞) − 𝑣 + 𝑔) =0
𝜕𝑝 𝜕𝑝

At 𝑝°(𝑞) or if it’s able to do it but only with parameters that doesn’t coordinate the quantity
decision.

Quantity-flexibility contract
For price coordination the first order condition must hold (see page 8)

𝜕𝜋𝑟 (𝑞, 𝑝°(𝑞), 𝑤𝑞 , 𝛿) 𝜕𝑆(𝑞, 𝑝°(𝑞))

= 𝑆(𝑞, 𝑝°(𝑞)) + (𝑝°(𝑞) − 𝑣 + 𝑔𝑟 )
𝜕𝑝 𝜕𝑝
𝑞
𝜕𝐹(𝑦|𝑝°(𝑞))
⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡+(𝑤𝑞 + 𝑐𝑟 − 𝑣) ∫ 𝑑𝑦 = 0
(1−𝛿)𝑞 𝜕𝑝

The third term must be non-positive so coordination can only occur if 𝑔𝑠 = 0 and either
𝑤𝑞 = 𝑣 − 𝑐𝑟 (the supplier earns a negative profit) or 𝛿 = 0 (standard non-coordinating
wholesale price contract). The quantity-flexibility contract doesn’t coordinate with price-
dependent demand.

Sales-rebate contract
It doesn’t do fare better in this setting (see page 8)
𝑞
𝜕𝜋𝑟 (𝑞, 𝑝°(𝑞), 𝑤𝑠 , 𝑟, 𝑡) 𝜕𝑆(𝑞, 𝑝°(𝑞)) 𝜕𝐹(𝑦|𝑝°(𝑞))
= 𝑆(𝑞, 𝑝°(𝑞)) + (𝑝°(𝑞) − 𝑣 + 𝑔𝑟 ) −𝑟∫ 𝑑𝑦 = 0
𝜕𝑝 𝜕𝑝 𝑡 𝜕𝑝

Since the last term is negative when 𝑟 > 0 and 𝑡 < 𝑞, the retailer prices below the optimal price.
Coordination might be achieved if something is added to induce the retailer to a higher price.
A buy-back could provide that counterbalance because it reduces the cost of left over
inventory, so a retailer need not price as aggressively to generate sales.

Buy-back contract
Let’s consider a buy-back contract on its own (see page 6)

𝜕𝜋𝑟 (𝑞, 𝑝°(𝑞), 𝑤𝑏 , 𝑏) 𝜕𝑆(𝑞, 𝑝°(𝑞))

= 𝑆(𝑞, 𝑝°(𝑞)) + (𝑝°(𝑞) − 𝑣 + 𝑔𝑟 − 𝑏) =0
𝜕𝑝 𝜕𝑝

It holds only if 𝑏 = −𝑔𝑠 , it may violate the 𝑏 ≥ 0 constraint and the coordinating price isn’t
acceptable to the supplier, 𝑤𝑏 (−𝑔𝑠 ) = 𝑐𝑠 − 𝑔𝑠 . It doesn’t coordinate with price-dependent
demand, but a buy-back contract with 𝑏 > 0 may still perform better than a wholesale-price
contract.

It fails to coordinate in this setting because the parameters depend on the price, they are

𝑏 = (1 − 𝜆)(𝑝 − 𝑣 + 𝑔) − 𝑔𝑠
𝑤𝑏 = 𝜆𝑐𝑠 + (1 − 𝜆)(𝑝 + 𝑔 − 𝑐𝑟 ) − 𝑔𝑠

Those parameters are linear in 𝑝 for a given 𝜆. Hence, it coordinates with price-dependent
demand only if 𝑏 and 𝑤𝑏 are made contingent on the retail price chosen, or if they are chosen
after the retailers commits to a price (but before he chooses 𝑞).

11
Nathan Grognet LLSMS2035 2015 - 2016

1) Retailer chooses price 𝑝

2) Retailer chooses contract parameters 𝑏 and 𝑤𝑏
3) Retailer chooses order quantity 𝑞

This is the price-discount sharing contract also named “bill back”. Notice that the retailer gets
a lower wholesale price if he reduces his price (the supplier shares the cost of a price discount
with the retailer). For the retailer as well as the supplier, {𝑞°, 𝑝°} is optimal for 𝜆 ∈ [0,1].

Revenue-sharing contract
It coordinates if (see page 7)

𝜕𝜋𝑟 (𝑞, 𝑝°(𝑞), 𝑤𝑟 , 𝜙) 𝜕𝑆(𝑞, 𝑝°(𝑞))

= 𝑆(𝑞, 𝑝°(𝑞)) + (𝑝°(𝑞) − 𝑣 + 𝑔𝑟 /𝜙) =0
𝜕𝑝 𝜕𝑝

The first case to consider is when 𝑔𝑟 = 𝑔𝑠 = 0. Then we have 𝜕𝜋𝑟 (𝑞, 𝑝, 𝑤𝑟 , 𝜙)⁄𝜕𝑝 = 𝜕Π(𝑞, 𝑝)⁄𝜕𝑝
with any revenue-sharing contract. Thus, the retailer will choose 𝑝°(𝑞) no matter which
revenue-sharing contract is chosen. Therefore, we can coordinate to quantity’s decision with
precisely the same set of contracts used when the price is fixed.

When the price is fixed, the revenue-sharing and buy-back contracts are equivalent. Here, a
difference occurs because with a buy-back the retailer’s share of revenue (1 − 𝑏/𝑝) depends
on the price, whereas with revenue sharing it’s independent of the price. However, the price
contingent buy-back contract (or “price discount contract”) is again equivalent to revenue
sharing.

If 𝑔𝑟 = 𝑔𝑠 = 0, , the revenue sharing contract and the price contingent buy back contract yield
the same profit for any quantity and price

𝜋𝑟 (𝑞, 𝑝, 𝑤𝑟 , 𝜙) = 𝜋𝑟 (𝑞, 𝑝, 𝑏(𝑝), 𝑤𝑏 (𝑝)) = 𝜆Π(𝑞, 𝑝)

The second case is when 𝑔𝑟 > 0 or 𝑔𝑠 > 0, revenue sharing is less successful. Now, coordination
is only achieved if 𝜙 = 𝑔𝑟 /𝑔. There’s only one coordinating contract and one profit allocation.
This difficulty occurs again because of the dependence of the parameters to the retail price
𝜆𝑔 − 𝑔𝑟
𝜙=𝜆+
𝑝−𝑣
𝑤𝑟 = 𝜆(𝑐 − 𝑣) − 𝑐𝑟 + 𝜙𝑣

This dependence is only eliminated if 𝜙 = 𝜆 = 𝑔𝑟 /𝑔. Coordination for all profit allocations is
restored if we make the parameters contingent on the retail price (like with the buy-back
contract). In that case revenue sharing is again equivalent to the price-discount contract.

Quantity-discount contract
The retailer keeps all revenue, so the quantity-discount doesn’t distort the retailer’s pricing
decision. If 𝑔𝑠 = 0, then (see page 9)
𝜕𝜋𝑟 (𝑞, 𝑤𝑑 (𝑞), 𝑝) 𝜕𝑆(𝑞, 𝑝) 𝜕Π(𝑞, 𝑝)
= + (𝑝 − 𝑣 + 𝑔𝑟 )𝑆(𝑞, 𝑝) =
𝜕𝑝 𝜕𝑝 𝜕𝑝

So 𝑝°(𝑞) is optimal for the retailer, and 𝑞° is too for both of the firms. Coordination occurs
because the retailer’s pricing decision isn’t distorted, and the retailer’s quantity decision is

12
Nathan Grognet LLSMS2035 2015 - 2016

adjusted contingent that 𝑝° is chosen. If 𝑔𝑠 > 0, then the retailer’s pricing decision needs to be
distorted for coordination, which the quantity-discount doesn’t do.

Discussion
Incentives to coordinate the retailer’s quantity decision may distort the retailer’s price decision.
This occurs with the buy-back, quantity-flexibility, and the sales-rebate contracts. Since the
quantity-discount leaves all revenue with the retailer, it doesn’t create such a distortion (when
𝑔𝑠 = 0). Revenue-sharing doesn’t distort the pricing decision when 𝑔𝑠 = 𝑔𝑟 = 0. In those
situations, the set of revenue-sharing contracts to coordinate the quantity decision with a fixed
price continue to coordinate the quantity decision with a variable price.

However, when there are goodwill costs, then the coordinating revenue-sharing parameters
generally depend on the retail price. This dependence is removed with only one revenue-
sharing contract, coordination is therefore achieved with a single profit allocation. Coordination
is restored with arbitrary profit allocation by making the parameters contingent on the retail
price chosen. This technique also applies to the buy-back contract: the price-discount sharing
contract coordinates the price setting newsvendor.

Coordinating the newsvendor with effort dependent

demand
The retailer can now take other actions to spur demand. No matter what level of effort the
retailer dedicates, the supplier prefers that he exerts even more effort. Problem is, those actions
benefit both firms but are costly to only one. Sharing the cost of effort is only solution, but
several conditions are required: the supplier must be able to observe the effort, the retailer’s
effort must be verifiable to the courts, and the activity must directly benefit the supplier.

Model and analysis

Let 𝑔(𝑒) be the retailer’s cost of exerting effort level 𝑒, and assume 𝑔𝑠 = 𝑔𝑟 = 0, 𝑣 = 0, and
𝑐𝑟 = 0. Suppose the retailer chooses his effort level at the same time as his order quantity, and
that the supplier cannot verify the retailer’s effort level.

Π(𝑞, 𝑒) = 𝑝𝑆(𝑞, 𝑒) − 𝑐𝑞 − 𝑔(𝑒)

𝜕Π(𝑞, 𝑒°(𝑞)) 𝜕𝑆(𝑞, 𝑒°(𝑞))

=𝑝 − 𝑔′ (𝑒°(𝑞)) = 0
𝜕𝑒 𝜕𝑒
Buy-back contract
𝜋𝑟 (𝑞, 𝑒, 𝑤𝑏 , 𝑏) = (𝑝 − 𝑏)𝑆(𝑞, 𝑒) − (𝑤𝑏 − 𝑏)𝑞 − 𝑔(𝑒)

For all 𝑏 > 0, we have 𝜕𝜋𝑟 (𝑞, 𝑒, 𝑤𝑏 , 𝑏)⁄𝜕𝑒 < 𝜕Π(𝑞, 𝑒)⁄𝜕𝑒. Thus 𝑒° cannot be the retailer’s optimal
effort when 𝑏 > 0, but this condition is required to coordinate the order quantity. The buy-
back contract doesn’t coordinate the supply chain.

Quantity-flexibility contract
𝑞
𝜋𝑟 (𝑞, 𝑒, 𝑤𝑞 , 𝛿) = 𝑝𝑆(𝑞, 𝑒) − 𝑤𝑞 (𝑞 − ∫ 𝐹(𝑦|𝑒)𝑑𝑦) − 𝑔(𝑒)
(1−𝛿)𝑞

13
Nathan Grognet LLSMS2035 2015 - 2016

For all 𝛿 > 0, we have 𝜕𝜋𝑟 (𝑞, 𝑒, 𝑤𝑞 , 𝛿)⁄𝜕𝑒 < 𝜕Π(𝑞, 𝑒)⁄𝜕𝑒. The retailer chooses a lower effort
than optimal. It doesn’t coordinate in this setting.

Revenue-sharing and sales-rebate contracts

For all 𝜙 < 1, we have 𝜕𝜋𝑟 (𝑞, 𝑒, 𝑤𝑟 , 𝜙)⁄𝜕𝑒 < 𝜕Π(𝑞, 𝑒)⁄𝜕𝑒 for the revenue-sharing contract. The
effort exerted by the retailer is lower than the optimum.

For all 𝑟 > 0 and 𝑞 > 𝑡, we have 𝜕𝜋𝑟 (𝑞, 𝑒, 𝑤𝑠 , 𝑟, 𝑡)⁄𝜕𝑒 > 𝜕Π(𝑞, 𝑒)⁄𝜕𝑒 for the sales-rebate
contract. The retailer exerts too much effort. This one can coordinate if it’s combined with a
buy-back contract to reduce the retailer’s incentive to exert effort. However, four parameters
makes for a complex contract.

Quantity-discount contract
Like with the price-dependent demand setting above, coordination can be achieved by letting
the retailer earn the entire reward for exerting effort.
𝑆(𝑞, 𝑒°) 𝑔(𝑒°)
𝑤𝑑 (𝑞) = (1 − 𝜆)𝑝 ( ) + 𝜆𝑐 + (1 − 𝜆)
𝑞 𝑞
𝜋𝑟 (𝑞, 𝑒) = 𝑝𝑆(𝑞, 𝑒) − (1 − 𝜆)𝑝𝑆(𝑞, 𝑒°) − 𝜆𝑐𝑞 − 𝑔(𝑒) + (1 − 𝜆)𝑔(𝑒°)

Given the optimal effort 𝑒°

𝜋𝑟 (𝑞, 𝑒°) = 𝜆𝑝𝑆(𝑞, 𝑒°) − 𝜆𝑐𝑞 − 𝜆𝑔(𝑒°) = 𝜆Π(𝑞, 𝑒°)

So 𝑞° is the optimal order quantity and any allocation of profit is feasible. This approach is so
powerful that it’s quite easy to design a quantity discount contract that coordinates the
newsvendor with demand dependent on price and effort. Because the retailer retains all
revenue, he optimizes price and effort.

Discussion
This setting is complex when the firms aren’t allowed to contract on the retailer’s effort level
directly, any contract that specifies effort level for the retailer is either unverifiable or
unenforceable. All contracts fail to coordinate because they distort the retailer’s marginal
incentive to exert effort, except for the quantity-discount contract because the retailer incurs
the entire cost and receives the entire benefit of effort.

The quality literature suggests that firms that cannot contract on effort directly can contract on
a proxy for effort (the frequency of internal or external failures), which is one solution around
potential observability or verifiability problems.

Coordination with multiple newsvendors

It is shown that retailers are biased towards ordering more inventory than optimal because of
a demand stealing effect: each retailer fails to account for the decrease in the other retailer’s
demands when the retailer increases his order quantity.

14
Nathan Grognet LLSMS2035 2015 - 2016

Competing newsvendors with a fixed retail price

Let’s assume 𝑐𝑟 = 𝑔𝑟 = 𝑔𝑠 = 𝑣 = 0, 𝑛 > 1 retailers, a total demand 𝐷, and a retail demand
divided between the 𝑛 firms proportionally to their stocking capacity (proportional allocation):
𝑞𝑖
𝐷𝑖 = ( ) 𝐷
𝑞

Demands at the retailers are perfectly correlated with the proportional allocation model. Either
every firm has excess demand (when 𝐷 > 𝑞) or every firm has excess inventory (when 𝐷 < 𝑞).

Given the proportional allocation rule, the integrated supply chain faces a single newsvendor
problem. We thus have: 𝐹(𝑞°) = (𝑝 − 𝑐)⁄𝑝.

Retailer 𝑖’s profit function with a buy back contract is

𝑞
𝑞𝑖
𝜋𝑖 (𝑞𝑖 , 𝑞−𝑖 ) = (𝑝 − 𝑤)𝑞𝑖 − (𝑝 − 𝑏) ( ) ∫ 𝐹(𝑥)𝑑𝑥
𝑞 0

Retailer 𝑖 has a unique optimal response to the other retailers’ strategies (their order quantities).
A set of order quantities, {𝑞1∗ , … , 𝑞𝑛∗ }, is a Nash equilibrium of the decentralized system if each
retailer’s order quantity is a best response. If there is a unique Nash equilibrium then that is
taken to be the predicted outcome of the decentralized game. In any case, we must satisfy
∗
𝜕𝜋𝑖 (𝑞𝑖 , 𝑞𝑗 ) 𝑝−𝑤 1 𝑞
= 𝑞∗ ( ) − 𝑞𝑖∗ 𝐹(𝑞 ∗ ) − 𝑞−𝑖
∗
( ∗ ∫ 𝐹(𝑥)𝑑𝑥 ) = 0
𝜕𝑞𝑖 𝑝−𝑏 𝑞 0

Substitute 𝑞−𝑖
∗
= 𝑞 ∗ − 𝑞𝑖∗ and solve for 𝑞𝑖∗ given a fixed 𝑞 ∗:
𝑝−𝑤 1 𝑞∗
( ) − ∗ ∫0 𝐹(𝑥)𝑑𝑥
𝑝−𝑏 𝑞
𝑞𝑖∗ = 𝑞 ∗
1 𝑞∗
𝐹(𝑞 ∗ ) − ∗ ∫0 𝐹(𝑥)𝑑𝑥
𝑞

The above gives each retailer’s equilibrium order conditional on 𝑞 ∗ being the equilibrium total
order quantity. Substitute into 𝑞 ∗ = 𝑛𝑞𝑖∗ and simplify:
∗
1 𝑛−1 1 𝑞 𝑝−𝑤
𝐹(𝑞 ∗ ) + ( ) ( ∗ ∫ 𝐹(𝑥)𝑑𝑥 ) =
𝑛 𝑛 𝑞 0 𝑝−𝑏

When 𝑏 < 𝑤 < 𝑝, there exists a unique 𝑞 ∗ that satisfies the above. In other words, there exists
a unique Nash equilibrium in which the total order quantity 𝑞 ∗ is given by the above equation
and each retailer’s order quantity is 𝑞𝑖∗ = 𝑞 ∗ /𝑛. Because the left side of the equation is
decreasing in 𝑛, 𝑞 ∗ is increasing in 𝑛: a single retailer that faces market demand 𝐷 purchases
less than multiple retailers facing the same demand. Competition makes the retailers order
more because of the demand stealing effect: each retailer ignores the fact that ordering more
means the other retailers’ demands stochastically decrease.

1 𝑛−1 1 𝑞
𝑤
̂(𝑞) = 𝑝 (1 − ( ) 𝐹(𝑞) ⁡ − ( ) ( ∫ 𝐹(𝑥)𝑑𝑥 ))
𝑛 𝑛 𝑞 0

1 𝑞
∫ 𝐹(𝑥)𝑑𝑥 < 𝐹(𝑞)
𝑞 0

15
Nathan Grognet LLSMS2035 2015 - 2016

Due to the demand stealing effect the supplier can coordinate the supply chain and earn a
positive profit with just a wholesale price contract. If 𝑤
̂(𝑞°) is the coordinating wholesale price,
it can be shown that 𝑤 ̂(𝑞°) > 𝑐 when 𝑛 > 1. The supplier earns a positive profit with that
coordinating contract whereas in the single retailer model coordination was only achieved
when the supplier’s profit was zero (𝑤̂(𝑞°) = 𝑐). Nevertheless, this wholesale price contract isn’t
optimal for the supplier.

𝜋𝑠 (𝑞, 𝑤
̂(𝑞)) = 𝑞(𝑤
̂(𝑞°) − 𝑐)

𝜕𝜋𝑠 (𝑞°, 𝑤
̂(𝑞°)) 𝑞°𝑝𝑓(𝑞°)
=− <0
𝜕𝑞 𝑛

Rather than coordinating the supply chain with the wholesale price 𝑤 ̂(𝑞°), the supplier prefers
to charge a higher wholesale price and sell less than 𝑞° when 𝑛 > 1.

Now let 𝑤𝑏 (𝑏) be the wholesale price that coordinates the supply chain given the buy-back
rate. To coordinate the supply chain the supplier must use a wholesale price that is higher than
the coordinating wholesale price contract (since the buy-back provides an incentive to the
retailers to increase their order quantity). (see page 52 for demonstration)
𝑝−𝑏
𝜋𝑖 (𝑞𝑖∗ , 𝑞−𝑖
∗
)=( ) Π(𝑞°)
𝑝𝑛2
𝑝(𝑛 − 1) + 𝑏
𝜋𝑠 (𝑞°, 𝑤𝑏 (𝑏), 𝑏) = ( ) Π(𝑞°)
𝑝𝑛

The supplier can extract all supply chain profit with 𝑏 = 𝑝. The contract with 𝑏 = 0 provides a
lower bound for the supplier’s profit. The ratio of this lower bound to the supplier’s maximum
profit, Π(𝑞°), provides a measure of how much improvement is possible by using a coordinating
buy-back contract:
𝜋𝑠 (𝑞°, 𝑤𝑏 (0), 0) 𝑛 − 1
=
Π(𝑞°) 𝑛

As 𝑛 increases, the supplier’s potential gain decreases from using a coordinating buy back
contract rather than his optimal wholesale price contract.

Competing newsvendors with market clearing prices

Here, competition influences the retail price. The market price depends on the realization of
demand and the amount of inventory purchased. Suppose demand can be either low (𝑝𝑙 (𝑞) =
(1 − 𝑞)+ ) or high (𝑝ℎ (𝑞) = (1 − 𝑞/𝜃⁡)+ for 𝜃 > 1). There a continuum [0,1] of retailers that must
order inventory from a single supplier before the realization of the demand is observed, then
when it is, the market clearing price is determined. There’s no salvage value and production
cost is zero.

To set a benchmark, suppose a single monopolist controls the system and chooses how much
inventory he sells after demand is observed. He sells either 𝑞𝑙 = 1/2 or 𝑞ℎ = 𝜃/2 at 𝑝𝑙 = 𝑝ℎ =
1/2. It is shown (page 54) that ordering the greater amount, 𝜃/2, is optimal.
1 1 1 𝜃 1+𝜃
Π° = 0.5 ∗ ∗ + 0.5 ∗ ∗ =
2 2 2 2 8

16
Nathan Grognet LLSMS2035 2015 - 2016

Now we consider perfectly competitive retailers with just a wholesale price contract.
𝜃
𝜃≤3
2(1 + 𝜃)
𝜋𝑠 (𝑤 ∗ (𝜃)) =
𝜃
{ 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
8
It is shown (page 55) that, no matter the value of 𝜃, 𝜋𝑠 (𝑤 ∗ (𝜃)) < Π°, so the supplier doesn’t
capture the maximum possible profit with a wholesale price contract. The problem is that
competition leads the retailers to sell too much in the low demand state. The monopolist
doesn’t sell all of his inventory in the low demand state, but the perfectly competitive retailers
cannot be so restrained. The supplier must reduce the destructive competition that results from
having more inventory than the system needs. The supplier can implement resale price
maintenance so the retailers may not sell below a stipulated price 𝑝̅ . When 𝑝̅ is above the
market clearing price the retailers have unsold inventory, so demand so allocated among the
retailers.

It is shown (page 56) that with the (𝑝̅ , 𝑤

̅) resale price maintenance contract the retailers order
𝑞 = 𝜃/2, the optimal quantity is sold in either state, the retailers’ expected profit is zero, and
the supplier’s is Π°.

Another approach is offering a buy-back contract with 𝑏 = 𝑝̅ . The market price will then not
fall under 𝑏 because retailers can earn that much on each unit of inventory. It is shown (page
57) that the supplier maximizes the system’s profit with a buy-back that offers a full refund on
returns, that is 𝑏 = 𝑤.

Although they achieve the same objective, the supplier gets a higher wholesale price with the
buy-back contract: retailers don’t incur the cost of excess inventory in the low demand states
with a buy-back contract, but they do with resale price maintenance. A buy-back contract isn’t
the same as a revenue-sharing in this setting. It prevents the market clearing price from falling
below 𝑏 in the low demand state, but revenue-sharing doesn’t prevent destructive competition.

In the single newsvendor model the retailer controls the market price, whereas in this
competitive model the retailers do not. The wholesale price component of the contract always
reduces the retailer’s order quantity and the buy back component always increases the retailer’s
order quantity. Thus, depending on the relative strength of those two components, the buy-
back can either increase or decrease the retailers’ order quantities.

Discussion
For coordination, the supplier needs to reduce the retailers’ order quantities, which can be
done with just a wholesale price contract above marginal cost, but it’s not the supplier’s optimal
wholesale price contract. The supplier can do better with a buy-back contract, however the
incremental improvement over the simpler wholesale price contract decreases as retail
competition intensifies. With the demand stealing effect, the supplier needs to decrease the
retailers’ order quantities, while with a destructive competition effect, he needs to increase
them (this occurs when demand is uncertain and retail price is set to clear the market). When
demand is high the retailers earn a profit, but when demand is low deep discounting to clear

17
Nathan Grognet LLSMS2035 2015 - 2016

inventory leads to losses. The retailers anticipate this problem and respond by reducing their
inventory purchase. Both resale price maintenance and buy-back contracts prevent deep
discounting, and therefore alleviate the problem.

Setting Effect Supplier strategy

Fixed retail price Demand stealing effect Supplier reduces order quantities
Market clearing price Destructive competition effect Supplier increases order quantities

Coordinating the newsvendor with demand updating

The retailer might have a second opportunity to order inventory. The retailer would prefer to
delay all ordering to the second period while the supplier would prefer an earlier order
commitment.

Model and analysis

Let 𝜉 ≥ 0 be the realization of the demand signal, while 𝐺(∙) and 𝑔(∙) are its distribution and
density functions.

Let period 1 be the time before the demand signal and period 2 the time between the demand
signal and the start of the selling season. Let 𝑞𝑖 be the retailer’s total order as of period 𝑖. Early
production is cheaper, 𝑐1 < 𝑐2 . The supplier charges 𝑤𝑖 at period 𝑖. The supplier offers to buy
back all unsold units for 𝑏. The terms are set at the start of period 1 and stay unchanged. The
supplier operates under voluntary compliance, so he might deliver less or produce more in
period 1.

Period 2
The supply chain’s expected revenue minus the period 2 production cost is

Ω2 (𝑞2 |𝑞1 , 𝜉) = 𝑝𝑆(𝑞2 |𝜉) − 𝑐2 (𝑞2 − 𝑞1 ) = 𝑝𝑆(𝑞2 |𝜉) − 𝑐2 𝑞2 + 𝑐2 𝑞1

Let 𝑞2 (𝑞1 , 𝜉) be the supply chain’s optimal 𝑞2 given 𝑞1 and 𝜉. The retailer’s expected profit is

𝜋2 (𝑞2 |𝑞1 , 𝜉) = (𝑝 − 𝑏)𝑆(𝑞2 |𝜉) − 𝑤2 (𝑞2 − 𝑞1 ) + 𝑏𝑞2

⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡= (𝑝 − 𝑏)𝑆(𝑞2 |𝜉) − (𝑤2 − 𝑏)𝑞2 + 𝑤2 𝑞1

Where we assume to supplier delivers the retailer’s order in full. To coordinate we must choose

𝑝 − 𝑏 = 𝜆𝑝
𝑤2 − 𝑏 = 𝜆𝑐2
𝜋2 (𝑞2 |𝑞1 , 𝜉) = 𝜆(Ω2 (𝑞2 |𝑞1 , 𝜉) − 𝑐2 𝑞1 ) + 𝑤2 𝑞1

Hence, the contract coordinates the retailer’s period 2 decision.

Now consider whether the supplier indeed fills the retailer’s entire period 2 order.

• Let 𝑥 be the total inventory in the supply chain at the start of period 2 (𝑥 ≥ 𝑞1 )
• The supplier’s inventory at the start of period 2 is 𝑥 − 𝑞1
• Let 𝑦 be the inventory at the retailer after the supplier’s delivery in period 2
• The supplier completely fills the retailer’s order when 𝑦 = 𝑞2

18
Nathan Grognet LLSMS2035 2015 - 2016

• The supplier clearly delivers the retailer’s full order when 𝑥 ≥ 𝑞2 because there is no
reason to partially fill the retailer’s order and have left over inventory
• If 𝑞2 > 𝑥, the supplier must produce additional units to deliver the retailer’s complete
order
• Let Π2 (𝑦|𝑥, 𝑞1 , 𝜉) be the supplier’s profit, where 𝑥 ≤ 𝑦 ≤ 𝑞2

Π2 (𝑦|𝑥, 𝑞1 , 𝑞2 , 𝜉) = 𝑏𝑆(𝑦|𝜉) − 𝑏𝑦 + 𝑤2 (𝑦 − 𝑞1 ) − (𝑦 − 𝑥)𝑐2

⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡= (1 − 𝜆)(Ω2 (𝑦|𝑞1 , 𝜉) − 𝑐2 𝑞1 ) + 𝑐2 𝑥 − 𝑤2 𝑞1

In period 2, the retailer orders the supply chain optimal quantity and the supplier fills the order
entirely, even with voluntary compliance and no matter how much inventory the supplier
carries between periods (as long as the retailer orders rationally).

Period 1
Assuming {𝑤2 , 𝑏} is chosen, the retailer’s expected profit is (recall that 𝑏 = 𝑤2 − 𝜆𝑐2 )

𝜋1 (𝑞1 ) = −(𝑤1 − 𝑏)𝑞1 + 𝜆𝐸[Ω2 (𝑞2 (𝑞1 , 𝜉)|𝑞1 , 𝜉)]

⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡= −(𝑤1 − 𝑤2 + 𝜆𝑐2 )𝑞1 + 𝜆𝐸[Ω2 (𝑞2 (𝑞1 , 𝜉)|𝑞1 , 𝜉)]

The supply chain’s expected profit is

Ω1 (𝑞1 ) = −𝑐1 𝑞1 + 𝐸[Ω2 (𝑞2 (𝑞1 , 𝜉)|𝑞1 , 𝜉)]

We must then choose 𝑤1 so that 𝑤1 − 𝑤2 + 𝜆𝑐2 = 𝜆𝑐1 , because then 𝜋1 (𝑞1 ) = 𝜆Ω1 (𝑞1 ) and
therefore the retailer order the supply chain’s optimal quantity 𝑞1° , and any portion of the supply
chain can be allocated to the retailer.

With centralized operations it doesn’t matter whether inventory is left at the supplier in period
1 because the supply chain moves all inventory to the retailer in period 2. With decentralized
control supply chain coordination is only achieved if the supplier doesn’t hold inventory
between periods. It is also possible that the supplier might attempt to use cheaper period 1
production to profit from a possible period 2 order.

With a coordinating {𝑤1 , 𝑤2 , 𝑏} contract the supplier produces just enough inventory to cover
the retailer’s period 1 order. Overall, those contracts coordinate the supply chain and arbitrarily
allocate profits. Interestingly, with a coordinating contract the supplier’s margin in period 2 is
actually lower than in period 1: 𝑤2 − 𝑐2 = 𝑤1 − (𝜆𝑐1 + (1 − 𝜆)𝑐2 ) < 𝑤1 − 𝑐1

Coordination with internal markets

The transfer payment rates are now contingent on the realization of some random event that
occurs after the contract is signed.

Model and analysis

Suppose there is one supplier, one production manager, and two retailers. The following
occurs:

1) The production manager chooses a production input level 𝑒, which yields an output of
𝑄 = 𝑌𝑒 finished units, where 𝑌 ∈ [0,1] is a random variable
2) The production manager incurs cost 𝑐(𝑒)

19
Nathan Grognet LLSMS2035 2015 - 2016

3) Retailer 𝑖 observes 𝛼𝑖 , the realization of the random variable 𝐴𝑖 > 0

4) Each retailer submits an order to the supplier
5) The supplier allocates 𝑞𝑖 units to retailer 𝑖, being sure that 𝑞𝑖 doesn’t exceed retailer 𝑖’s
order and 𝑞1 + 𝑞2 ≤ 𝑄
−1⁄𝜂
6) Finally, retailer 𝑖 earns revenue 𝑞𝑖 𝑝𝑖 (𝑞𝑖 ), where 𝑝𝑖 (𝑞𝑖 ) = 𝛼𝑖 𝑞𝑖 and 𝜂 is demand
elasticity

Let 𝜃 be the realization of 𝑌, 𝐴 = {𝐴1 , 𝐴2 } and 𝛼 = {𝛼1 , 𝛼2 }. Let 𝛾 be the fraction of 𝑄 allocated
to retailer one. Let 𝜋(𝛾, 𝛼, 𝑄) be total retailer revenue if retailer one is allocated 𝛾𝑄 units and
retailer two (1 − 𝛾)𝑄:

𝜋𝑟 (𝛾, 𝛼, 𝑄) = 𝑝1 (𝑞1 )𝑞1 + 𝑝2 (𝑞2 )𝑞2

(𝜂−1)⁄𝜂 (𝜂−1)⁄𝜂
⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡= 𝛼1 𝑞1 + 𝛼2 𝑞2
(𝜂−1)⁄𝜂
⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡= (𝛼1 𝛾 + 𝛼2 (1 − 𝛾)(𝜂−1)⁄𝜂 )𝑄 (𝜂−1)⁄𝜂

The optimal allocation of production to the two retailers depends on the demand realizations
but not on the production output. Let 𝛾°(𝛼) be the optimal share to allocate to retailer one:
𝜂
𝛼1
𝛾°(𝛼) = 𝜂 𝜂
𝛼1 + 𝛼2
𝜂 𝜂 1⁄𝜂
𝜋𝑟 (𝛼, 𝑄) = 𝜋𝑟 (𝛾°(𝛼), 𝛼, 𝑄) = (𝛼1 + 𝛼2 ) 𝑄(𝜂−1)⁄𝜂

Total expected supply chain profit is thus

Π(𝑒, 𝐴, 𝑌) = 𝐸[𝜋(𝐴, 𝑌𝑒)] − 𝑐(𝑒)

There is a unique optimal production effort level 𝑒° that coordinate.

Now consider decentralized operations. To achieve channel coordination it must be that the
retailers purchase all of the supplier’s output and that output must be allocated to the retailers
properly. A fixed per unit wholesale price cannot achieve those tasks: for any fixed wholesale
price there is some realization of 𝑌 such that the retailers don’t purchase all of the supplier’s
output. In addition, with a fixed wholesale price it’s possible the retailers desire more than the
supplier’s output, in which case the supplier must implement some allocation rule by making
the transfer price contingent on the realization of 𝐴 and 𝑄.
(𝜂−1)⁄𝜂
𝜋𝑖 (𝑞𝑖 , 𝑤) = 𝛼𝑖 𝑞𝑖 − 𝑤𝑞𝑖

𝜕𝜋𝑖 (𝑞𝑖∗ , 𝑤) 𝜂−1

=( ) 𝛼𝑖 (𝑞𝑖∗ )−1⁄𝜂 − 𝑤 = 0
𝜕𝑞𝑖 𝜂

It follows that 𝑞𝑖∗ = 𝛾°(𝛼)𝑄 when 𝑤 = 𝑤(𝛼, 𝑄)

𝜂−1 𝜂 𝜂 1⁄𝜂 𝜕𝜋(𝛼, 𝑄)
𝑤(𝛼, 𝑄) = ( ) (𝛼1 + 𝛼2 ) 𝑄 −1⁄𝜂 =
𝜂 𝜕𝑄

Hence, when the supplier charges 𝑤(𝛼, 𝑄) the retailers order exactly 𝑄 units in total and the
allocation of inventory between them maximizes supply chain revenue. Note that 𝑤(𝛼, 𝑄) is
the marginal value of additional production.

20
Nathan Grognet LLSMS2035 2015 - 2016

The unique market clearing price is 𝑤(𝛼, 𝑄), and so the market optimizes the supply chain’s
profit without the supplier observing 𝐴. It remains to determine the supplier’s compensation
scheme for his production manager. The production effort 𝑒 is assumed unobservable and thus
non-contractible by the supplier. But the supplier observes 𝑌𝑒. Suppose the supplier pays the
manager
𝑛−1 𝜂 𝜂 1⁄𝜂
( 𝑛 ) (𝑒°)−1⁄𝜂 𝐸 [(𝐴1 + 𝐴2 ) 𝑌 (𝜂−1)⁄𝜂 ] 𝐸[𝑄𝑤(𝐴, 𝑄)|𝑒°]
=
𝐸[𝑌] 𝐸[𝑄|𝑒°]

He will pay the production manager the expected shadow price of capacity and sell capacity
to the retailers for the realized shadow price of capacity. The production manager’s optimal
effort is 𝑒°.

The supplier earns zero profit in expectation from the internal market: 𝐸[𝑄𝑤(𝐴, 𝑄)|𝑒°] is the
expected revenue from the retailers, and it’s also the expected payout to the production
manager (see above). To earn a positive profit the supplier can charge the production manager
and/or the retailers fixed fees. In fact, the supplier breaks even or loses money with the internal
market approach to supply chain coordination. Hence, this is a viable strategy for the supplier
only if it’s coupled with fixed fees.

Discussion
One agent produces a resource (production manager) that has uncertain value to the other
part of the supply chain (retailers). Coordination therefore requires the proper incentives to
produce and consume the resource. Production can be coordinated with a single price
(expected value of output) but its consumption requires a state dependent price, which can be
provided via a market mechanism. Interestingly, the expected revenue from selling the resource
via the market may be less than the expected cost to purchase the resource. Hence, there must
be a market maker (supplier) that stands between producers and consumers of the resource
and he must be ready to lose money in this market. So the market maker is willing to participate
only if there exists some other instrument (fixed fees) to extract rents from the participants.
The market maker uses the market to align incentives, but doesn’t directly profit from the
market.

Asymmetric information
Here, coordination requires that the supplier takes the correct action and that an accurate
demand forecast is shared. This model highlights the issue of contract compliance: with forced
compliance coordination and accurate forecast sharing are possible, however with voluntary
compliance information sharing is possible but only if optimal supply chain performance is
sacrificed.

The capacity procurement game

There is a manufacturer M and a single supplier S. Let 𝐷𝜃 be demand, where 𝜃 ∈ {ℎ, 𝑙}. While
with full information both firms observe 𝜃, only M observes 𝜃 with asymmetric information.

21
Nathan Grognet LLSMS2035 2015 - 2016

Stage one
M gives S a demand forecast and offers S a contract which includes an initial order 𝑞𝑖 . Assuming
S accept the contract, he then constructs 𝑘 units of capacity at a cost 𝑐𝑘 > 0 per unit.

Stage two
M observes 𝐷𝜃 and places her final order 𝑞𝑓 with S where the contract specifies the set of
feasible final orders. Then S produces min{𝐷𝜃 , 𝑘} units at a cost of 𝑐𝑝 > 0 per unit and delivers
those units to M. Finally, M pays S based on the agreed contract and M earns 𝑟 > 𝑐𝑝 + 𝑐𝑘 per
unit of demand satisfied.

Full information
Assume both firms observe 𝜃. After observing demand, min{𝑘, 𝐷𝜃 } units will be produced. We
must now see how much capacity to build. Let 𝑆𝜃 (𝑥) be expected sales with 𝑥 units of capacity:
𝑥
𝑆𝜃 (𝑥) = 𝑥 − 𝐸[(𝑥 − 𝐷𝜃 )+ ] = 𝑥 − ∫ 𝐹𝜃 (𝑥)𝑑𝑥
0

The expected profit of the supply chain with 𝑘 units of capacity is

Ω𝜃 (𝑘) = (𝑟 − 𝑐𝑝 )𝑆𝜃 (𝑘) − 𝑐𝑘 𝑘

Ω′𝜃 (𝑘) = (𝑟 − 𝑐𝑝 )𝑆𝜃′ (𝑘) − 𝑐𝑘 = 0

𝑐𝑘
𝑆𝜃′ (𝑘𝜃° ) = 𝐹̅𝜃 (𝑘𝜃° ) =
𝑟 − 𝑐𝑝

The optimal capacity 𝑘𝜃° satisfies the newsvendor critical ratio, so Ω°𝜃 = Ω𝜃 (𝑘𝜃° ). Supply chain
coordination is achieved if the supplier builds 𝑘𝜃° units of capacity and defers all production
until after receiving the manufacturer final order.

Consider an option contract: M purchases 𝑞𝑖 options for 𝑤𝑜 per option at stage 1 and then pays
𝑤𝑒 to exercise each option at stage 2.

𝑇(𝑤𝑜 , 𝑤𝑒 , 𝑞𝑖 ) = 𝑤𝑜 𝑞𝑖 + 𝑤𝑒 𝑆𝜃 (𝑞𝑖 )

It could be seen as a buy-back contract: M pays 𝑤 = 𝑤𝑜 + 𝑤𝑒 at stage 1 for an order quantity

of 𝑞𝑖 and S pays 𝑏 = 𝑤𝑒 per unit in stage 2 that M “returns”. The manufacturer’s expected profit
is

Π𝜃 (𝑞𝑖 ) = (𝑟 − 𝑤𝑒 )𝑆𝜃 (𝑞𝑖 ) − 𝑤𝑜 𝑞𝑖

If contract parameters are chosen so that (𝑟 − 𝑤𝑒 ) = 𝜆(𝑟 − 𝑐𝑝 ) and 𝑤𝑜 = 𝜆𝑐𝑘

Π𝜃 (𝑞𝑖 ) = 𝜆(𝑟 − 𝑐𝑝 )𝑆𝜃 (𝑞𝑖 ) − 𝜆𝑐𝑘 𝑞𝑖 = 𝜆Ω𝜃 (𝑞𝑖 )

Hence, 𝑞𝑖 = 𝑘𝜃° is the manufacturer’s optimal order, the supply chain is coordinated and the
profit can be arbitrarily allocated between firms.

Now suppose the manufacturer is unable to verify the supplier actually builds 𝑘 = 𝑞𝑖 .

22
Nathan Grognet LLSMS2035 2015 - 2016

We assume 𝑘 < 𝑞𝑖 and the demand expectation of the supplier is 𝜏

𝜋(𝑘, 𝑞𝑖 , 𝜏) = (𝑤𝑒 − 𝑐𝑝 )𝑆𝜏 (𝑘) + 𝑤𝑜 𝑞𝑖 − 𝑐𝑘 𝑘

⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡= (1 − 𝜆)(𝑟 − 𝑐𝑝 )𝑆𝜏 (𝑘) − 𝑐𝑘 (𝑘 − 𝜆𝑞𝑖 )

𝜕𝜋(𝑘𝜃° , 𝑘𝜃° , 𝜃)
<0
𝜕𝑘
Thus 𝑘𝜃° doesn’t maximize the supplier’s profit if 𝑞𝑖 = 𝑘𝜃° . The problem is that the above cost
function assumes the manufacturer pays 𝑤𝑜 per option no matter what capacity is constructed.
Therefore, only 𝑤𝑒 impacts the supplier’s decision on the margin, he sets his capacity as if he
is offered just a wholesale price contract.

Forced compliance is when the supplier must choose 𝑘 = 𝑞𝑖 and voluntary compliance is when
he can choose 𝑘 < 𝑞𝑖 even though the manufacturer pays 𝑤𝑜 𝑞𝑖 for 𝑞𝑖 options. Even if a penalty
is implemented with voluntary compliance, the supplier is likely to build less capacity. Suppose
that the supplier must refund the manufacturer (min{𝑞𝑖 , 𝐷𝜃 } − 𝑘)+ 𝑤𝑜 if 𝐷𝜃 > 𝑘, even then the
supplier has a 1 − 𝐹(𝑘) chance of pocketing the fee for (𝑞𝑖 − 𝑘) options without having built
the capacity to cover those options: if 𝐷𝜃 < 𝑘, then the manufacturer wouldn’t know the
supplier was unable to cover all options.

To summarize, with forced compliance the manufacturer can use a number of contracts to
coordinate the supply chain and divide its profit. However, coordination with those contracts
isn’t assured with anything less than forced compliance. With voluntary compliance, the
manufacturer’s initial order has no impact on the supplier’s capacity decision, in order to
influence that decision the manufacturer is relegated to a contract based on his final order 𝑞𝑓 .
With a wholesale price contract, the supplier’s profit is

𝜋𝜃 (𝑘) = (𝑤 − 𝑐𝑝 )𝑆𝜃 (𝑘) − 𝑐𝑘 𝑘

There exists a unique wholesale price for any 𝑘 such that 𝑘 is optimal for the supplier. Let 𝑤𝜃 (𝑘)
be that wholesale price:
𝑐𝑘
𝑤𝜃 (𝑘) = + 𝑐𝑝
𝐹̅𝜃 (𝑘)

The manufacturer’s profit function is now

Π𝜃 (𝑘) = (𝑟 − 𝑤𝜃 (𝑘))𝑆𝜃 (𝑘)

The manufacturer can choose a desired capacity by offering the wholesale price contract 𝑤𝜃 (𝑘).
There is a unique capacity 𝑘𝜃∗ and a unique wholesale price 𝑤𝜃∗ = 𝑤(𝑘𝜃∗ ) that induces that
capacity. The supply chain isn’t coordinated because 𝑘𝜃∗ < 𝑘𝜃° .

Forecast sharing
Suppose now that the supplier doesn’t observe 𝜃. A high demand forecast from the
manufacturer is suspect since a low demand manufacturer could offer a high demand forecast
to get more capacity. The manufacturer will share his demand forecast by offering the right
contract. In this model, forecast sharing takes place via a signaling equilibrium.

23
Nathan Grognet LLSMS2035 2015 - 2016

There are two types of signaling equilibria:

• Separating equilibrium: either the contract is a high or low demand type, but no
contract is offered by both types
• Pooling equilibrium: either the contract is offered by a low type or it is offered by both
types

Forecasting only occurs with a separating equilibrium: a high type M offers the best contract
among those designated for high types and a low type M offers the best contract among those
designated for low types. The aim here is that the low type must not prefer to mimic a high
type by choosing a high type contract.

Forced compliance
Let 𝜋̂ be the supplier’s minimum acceptable profit. The supplier must be diligent about biased
forecast because a manufacturer would prefer a larger share of sub-optimal profit over a
smaller share of optimal profit.

Suppose the low type manufacturer offers an options contract with

𝜋̂
𝜆𝑙 = 1 −
Ω°𝑙

And the high type offers min{𝜆ℎ , 𝜆̂ℎ }, where

̂
𝜋 Ω°𝑙 −𝜋
̂
𝜆ℎ = 1 − and 𝜆̂ℎ =
Ω°ℎ Ω𝑙 (𝑘ℎ° )

Since min{𝜆ℎ , 𝜆̂ℎ } > 𝜆𝑙 , the high type isn’t interested in offering the low type’s contract. The
low type has no interest in offering the high type’s contract because the low type is
indifferent between earning her low type profit Ω°𝑙 − 𝜋̂ and 𝜆̂ℎ percent of the high type profit
Ω𝑙 (𝑘ℎ° ). As long as the high type doesn’t capture more than 𝜆̂ℎ percent of the supply chain’s
profit, the low type has no interest in pretending to be a high type. Thus, the above contracts
are a separating equilibrium.

With low demand the supplier earns 𝜋̂. With high demand he earns the same amount if 𝜆ℎ <
𝜆̂ℎ , otherwise he earns more. The manufacturer would prefer to take more from the supplier
when 𝜆ℎ > 𝜆̂ℎ , but then the supplier cannot trust the manufacturer’s forecast (because a low
type would gladly accept some sub-optimal supply chain performance in exchange for a
large fraction of the profit). Even though the manufacturer cannot drive the supplier’s profit
down to 𝜋̂, the supply chain is coordinated in all situations.

Voluntary compliance
Here we have a wholesale price contract. If 𝑤ℎ∗ > 𝑤𝑙∗ then it’s possible the low type will not
mimic the high type. But if that’s not the case, the high type needs to supplement with some
additional transfer payment.

24
Nathan Grognet LLSMS2035 2015 - 2016

There are three suggestions:

• The high type offers a fixed fee 𝐴, then we must have

Π𝑙 (𝑤𝑙∗ , 𝑙) ≥ Π𝑙 (𝑤ℎ∗ , ℎ) − 𝐴

Πℎ (𝑤ℎ∗ , ℎ) − 𝐴 > Πℎ (𝑤𝑙∗ , 𝑙)

Πℎ (𝑤ℎ∗ , ℎ) − Πℎ (𝑤𝑙∗ , 𝑙) > Π𝑙 (𝑤ℎ∗ , ℎ) − Π𝑙 (𝑤𝑙∗ , 𝑙)

• The high type can offer a higher wholesale price

𝜕Πℎ (𝑤ℎ∗ ,ℎ) 𝜕Π𝑙 (𝑤ℎ∗ ,ℎ)
𝜕𝑤
=0 and 𝜕𝑤
<0

• The high can also offer a firm commitment. In stage 1, he commits to purchase at least
𝑚 units in stage 2, 𝑞𝑓 ≥ 𝑚. It’s a “capacity reservation contract” and it’s not desirable
with full information because it may lead to 𝑚 − 𝐷𝜃 units wasted.

Discussion
Coordination is achieved only if the demand forecast is shared accurately. An options contract
can be used with forced compliance to coordinate the supplier’s action and share information.
Sharing information is more costly with voluntary compliance. The high demand manufacturer
may wish to consider options other than signaling:

• He can pay the supplier for units and take delivery at stage 1. This option completely
disregard the benefit of deferring production.
• He can choose a contract associated with a pooling equilibrium. The supplier evaluates
the contract as if he is dealing with a representative manufacturer. The terms aren’t as
good but this option is cheaper.

Conclusion
Several key conclusions:

• Coordination failure is common. Poor supply chain performance may be due to

conflicting incentives that can be managed.
• There are multiple kids of contracts that achieve coordination and arbitrarily divide
profit.
• The consequence of coordination failure is context specific.
• Managing incentive conflicts can lead to Pareto improvements or “win-win” situations
instead of “zero sum game”.

25
Nathan Grognet LLSMS2035 2015 - 2016

Game Theory in Supply Chain Analysis

Non-cooperative static games
The players choose strategies simultaneously and are thereafter committed to their chosen
strategies.

Game setup
Each strategy is defined on a set 𝑋𝑖 , 𝑥𝑖 ∈ 𝑋𝑖 , so we call the Cartesian product 𝑋1 × 𝑋2 × … × 𝑋𝑛
the “strategy space”. Each player’s set of feasible strategies must be independent of the
strategies chosen by the other players. A game can also be designed in the extensive form
(tree) where actions are chosen only as needed, so sequential choices are possible. The
distinction is that in the normal form (matrix) a player is able is commit to all future decisions.
We only consider pure strategies.

In a non-cooperative game the players are unable to make binding commitments before
choosing their strategies. In a cooperative game players can make side-payments and form
coalitions.

Retailer 𝑖 must maximizes his profit.

+
𝜋𝑖 (𝑄𝑖 , 𝑄𝑗 ) = 𝐸𝐷 [𝑟𝑖 min (𝐷𝑖 + (𝐷𝑗 − 𝑄𝑗 ) , 𝑄𝑖 ) − 𝑐𝑖 𝑄𝑖 ]

Best response functions and the equilibrium of the game

Definition 1:
Given a n-player game, player 𝑖’s best response (function) to the strategies 𝑥−𝑖 of the other players
is the strategy 𝑥𝑖∗ that maximizes player 𝑖’s payoff 𝜋𝑖 (𝑥𝑖 , 𝑥−𝑖 ):

𝑥𝑖∗ (𝑥−𝑖 ) = arg max 𝜋𝑖 (𝑥𝑖 , 𝑥−𝑖 )

𝑥𝑖

The best response functions are

𝑟𝑖 − 𝑐𝑖
𝑄𝑖∗ (𝑄𝑗 ) = 𝐹 −1 + ( )
𝐷𝑖 +(𝐷𝑗 −𝑄𝑗 ) 𝑟𝑖

Taken together, the two best response functions form a best response mapping 𝑅 2 → 𝑅 2 .

𝜕 2 𝜋𝑖
𝜕𝑄𝑖∗ (𝑄𝑗 ) 𝜕𝑄𝑖 𝜕𝑄𝑗 𝑟𝑖 𝑓𝐷 +(𝐷 −𝑄 )+|𝐷 >𝑄 (𝑄𝑖 ) Pr(𝐷𝑗 > 𝑄𝑗 )
𝑖 𝑗 𝑗 𝑗 𝑗
=− 2 =− <0
𝜕𝑄𝑗 𝜕 𝜋𝑖 𝑟𝑖 𝑓𝐷 +(𝐷 −𝑄 )+ (𝑄𝑖 )
𝑖 𝑗 𝑗
𝜕𝑄𝑖2

The slopes of the best response functions are negative, which implies an intuitive result that
each player’s best response is monotonically decreasing in the other player’s strategy. The
equilibrium is located on the intersection of the best responses.

Definition 2:
An outcome (𝑥1∗ , 𝑥2∗ , … , 𝑥𝑛∗ ) is a Nash equilibrium of the game if 𝑥𝑖∗ is a best response to 𝑥−𝑖
∗
for
all 𝑖 = 1,2, … , 𝑛.

26
Nathan Grognet LLSMS2035 2015 - 2016

A Nash equilibrium must satisfy the system of equations 𝜕𝜋𝑖 ⁄𝜕𝑥𝑖 = 0. Hence, 𝑥 ∗ solves the
first-order conditions if and only if it is a fixed point of mapping 𝑓(𝑥) defined above. No player
wants to unilaterally deviate from it since such a behavior would lead to lower payoffs. Two
particularly vexing problems are the non-existence of equilibrium and the multiplicity of
equilibria. The players may choose strategies from different equilibria. An interesting feature
of the NE concept is that the system optimal solution needn’t be a NE. Hence, decentralized
decision making generally introduces inefficiency in the supply chain. In fact, a NE may not
even be on the Pareto frontier, it can be Pareto inferior.

Existence of equilibrium
A NE is a solution to a system of 𝑛 first-order conditions, so an equilibrium may not exist. The
simplest way to demonstrate the existence of NE is through verifying concavity of the players’
payoffs.

Theorem 1:
Suppose that for each player the strategy space is compact (closed and bounded) and convex and
the payoff function is continuous and quasi-concave with respect to each player’s own strategy.
Then there exists at least one pure strategy NE in the game.

Theorem 2:
Suppose that a game is symmetric and for each player the strategy space is compact and convex
and the payoff function is continuous and quasi-concave with respect to each player’s own
strategy. Then there exists at least on symmetric pure strategy NE in the game.

Non-quasi-concave payoffs may lead to non-existence of NE (best response functions don’t

intersect anywhere). It’s nevertheless easy to verify that the newsvendor’s objective function is
concave (hence quasi-concave) and that a NE exists.

If the quasi-concavity cannot be verified, there exists another way to prove the existence of NE.

Definition 3:
A twice continuously differentiable payoff function 𝜋𝑖 (𝑥1 , … , 𝑥𝑛 ) is supermodular (submodular) if
𝜕 2 𝜋𝑖 ⁄𝜕𝑥𝑖 𝜕𝑥𝑗 ≥ 0 (≤ 0) for all 𝑥 and all 𝑗 ≠ 𝑖. The game is called supermodular if the players’
payoffs are supermodular.

Supermodularity essentially means complementarity between any two strategies and isn’t
linked directly to either convexity, concavity or even continuity. While in most situations the
positive sign of the second derivative can be used to verify supermodularity, sometimes it’s
necessary to utilize supermodularity-preserving transformations to show that payoffs are
supermodular.

Theorem 3:
In a supermodular game there exists at least one NE.

The newsvendor game is submodular (see second-order cross-partial derivative at page 25)
thus existence of NE cannot be assured. However we can redefine the ordering of the players’
strategies.

27
Nathan Grognet LLSMS2035 2015 - 2016

Let 𝑦 = −𝑄𝑗 so that

𝜕 2 𝜋𝑖
= 𝑟𝑖 𝑓𝐷 +(𝐷 +𝑦)+|𝐷 >𝑄 (𝑄𝑖 ) Pr(𝐷𝑗 > −𝑦) > 0
𝜕𝑄𝑖 𝜕𝑄𝑗 𝑖 𝑗 𝑗 𝑗

And the game becomes supermodular in (𝑥𝑖 , 𝑦) so existence of NE is assured. We only altered
the ordering of one player’s strategy space. We can state that in general NE exists in games
with decreasing best responses (submodular games) with two players. One way is to consider
an 𝑛-player game where best responses are functions of aggregate actions of all other players.
If best responses in such a game are decreasing, then NE exists.

Roughly speaking, Tarski’s fixed point theorem (above) only requires best response mappings
to be non-decreasing for the existence of equilibrium and doesn’t require quasi-concavity of
the players’ payoffs and allows for jumps in best responses.

If 𝑓(𝑥) is non-decreasing but possibly with jumps up then it’s not possible to derive a situation
without an equilibrium. However, when 𝑓(𝑥) jumps down, non-existence is possible. Hence,
increasing best response functions is the only major requirement for an equilibrium to exist.
Players’ objectives don’t have to be quasi-concave or even continuous. Jumps in best responses
happen when the objective function is bi-modal (or more generally multi-modal).

Uniqueness of equilibrium
Method 1 – Algebraic argument
Sometimes one can ascertain that the solution is unique by simply looking at the optimality
conditions (one player may have a unique closed-form solution that doesn’t depend on the
other player’s strategy). One can also assure uniqueness by analyzing geometrical properties
of the best response functions and arguing that they intersect only once. Finally, it may be
possible to use a contradiction argument: assume that there is more than one equilibrium and
prove that it leads to a contradiction.

Method 2 – Contraction mapping argument

We want to show that the best response mapping is a contraction, which then implies the
mapping has a unique fixed point. Suppose we want to find a solution to the following fixed
point equation:

𝑥 = 𝑓(𝑥), 𝑥 ∈ 𝑅1

To do so an algorithm generates a sequence of values

{𝑥 (1) , 𝑥 (2) , 𝑥 (3) , … } where 𝑥 (1) is picked arbitrarily and
𝑥 (𝑡) = 𝑓(𝑥 (𝑡−1) ). We hope that this sequence will
converge to a unique fixed point. For that, we need
|𝑓 ′ (𝑥)| < 1 in some vicinity of 𝑥 ∗ = 𝑓(𝑥 ∗ ).

The iterative scheme in the figure is a contraction

mapping.

Definition 4:
Mapping 𝑓(𝑥), 𝑅 𝑛 → 𝑅 𝑛 is a contradiction if ||𝑓(𝑥1 ) − 𝑓(𝑥2 )|| ≤ 𝛼||𝑥1 − 𝑥2 ||, ∀𝑥1 , 𝑥2 , 𝛼 < 1.

28
Nathan Grognet LLSMS2035 2015 - 2016

Theorem 4:
If the best response mapping is a contraction on the entire strategy space, there is a unique NE
in the game.

If the best response mapping is a contraction, the NE obtained as a result of such iterative play
is stable but the opposite isn’t necessarily true. A major restriction is that the contraction
mapping condition must be satisfied everywhere. The best response mapping may be a
contraction locally but not outside of a specific 𝜀-neighborhood. The entire strategy space must
be considered.

Theorem 4 doesn’t actually explain how to validate that a best reply mapping is a contraction.
We must define the matrix of derivatives of the best response functions:

The spectral radius 𝜌(𝐴) = {max|𝜆| : 𝐴𝑥 = 𝜆𝑥, 𝑥 ≠ 0} is equal to the largest absolute eigenvalue.

Theorem 5:
The mapping 𝑓(𝑥), 𝑅 𝑛 → 𝑅 𝑛 is a contraction if and only if 𝜌(𝐴) < 1 everywhere.

The spectral radius rule is an extension of the requirement |𝑓′(𝑥)| < 1 above. We use the fact
that the largest eigenvalue is bounded above by any of the matrix norms. It is sufficient to
verify that (for every 𝑘) no column sum or no row sum of matrix 𝐴 exceeds one:
𝜕𝑓𝑘 𝜕𝑓𝑖
∑𝑛𝑖=1 |
𝜕𝑥𝑖
| <1 or ∑𝑛𝑖=1 |
𝜕𝑥𝑘
| <1

Thanks to the “Implicit Function Theorem”, theorem 5 can be restated as

𝑛
𝜕 2 𝜋𝑘 𝜕 2 𝜋𝑘
∑ | | < | 2 | , ∀𝑘
𝜕𝑥𝑘 𝜕𝑥𝑖 𝜕𝑥𝑘
𝑖=1,𝑖≠𝑘

This condition is also known as “diagonal dominance” because the diagonal of the Hessian
dominates the off-diagonal entries:

In games with only two players, the condition simplifies to

𝜕𝑓 𝜕𝑓
|𝜕𝑥1 | < 1 and |𝜕𝑥2 | < 1
2 1

29
Nathan Grognet LLSMS2035 2015 - 2016

The slopes of the best response functions are less than one. Thus, if they cross at one point
then they cannot cross at an additional point. In our newsvendor model, we have found that
the slopes of the best response function are less than one.

𝜕𝑄𝑖∗ (𝑄𝑗 ) 𝑓𝐷 +(𝐷 −𝑄 )+|𝐷 >𝑄 (𝑄𝑖 ) Pr(𝐷𝑗 > 𝑄𝑗 )

𝑖 𝑗 𝑗 𝑗 𝑗
| |=| |<1
𝜕𝑄𝑗 𝑓𝐷 +(𝐷 −𝑄 )+ (𝑄𝑖 )
𝑖 𝑗 𝑗

Hence, the best response mapping in the newsvendor game is a contraction and the game
has a unique and stable NE.

Method 3 – Univalent mapping argument

We can also verify if the best response mapping is one-to-one: that is, if 𝑓(𝑥) is a 𝑅 𝑛 → 𝑅 𝑛
mapping, then 𝑦 = 𝑓(𝑥) implies that for all 𝑥 ′ ≠ 𝑥, 𝑦 ≠ 𝑓(𝑥 ′ ). Clearly, if the best response
mapping is one-to-one then there can be at most one fixed point of such mapping. This
property requires quasi-concavity of the mapping which translates into quasi-definiteness of
the Hessian.

Theorem 6:
Suppose the strategy space of the game is convex and all equilibria are inferior. Then if the
determinant |𝐻| is negative quasi-definite on the players’ strategy set, there is a unique NE.

Notice that the univalent mapping argument is somewhat weaker than the contraction
mapping argument. The newsvendor game satisfies the univalence theorem. In the case of just
two players the univalence theorem can be written as

𝜕 2 𝜋2 𝜕 2 𝜋1 𝜕 2 𝜋1 𝜕 2 𝜋2
| + | ≤ 2√| 2 ∙ | , ∀𝑥1 , 𝑥2
𝜕𝑥2 𝜕𝑥1 𝜕𝑥1 𝜕𝑥2 𝜕𝑥1 𝜕𝑥22

Method 4 – Index theory approach

We now require a certain sign from the Hessian, but this requirement need hold only at the
equilibrium point.

Theorem 7:
Suppose the strategy space of the game is convex and all payoff functions are quasi-concave.
Then if (−1)𝑛 |𝐻| is positive whenever 𝜕𝜋𝑖 ⁄𝜕𝑥𝑖 = 0, all 𝑖, there is a unique NE.

This method is also weaker (the condition is satisfied if |𝐻| is negative definite which is implied
above). Plus, this only needs to hold at the equilibrium. For a two-player game we have

𝜕 2 𝜋1 𝜕 2 𝜋1
2 𝜕𝜋1 𝜕𝜋2
| 𝜕 𝑥1 𝜕𝑥1 𝜕𝑥2 |
> 0, ∀𝑥1 , 𝑥2 : = 0, =0
| 𝜕2𝜋 2
𝜕 𝜋2 | 𝜕𝑥1 𝜕𝑥2
2
𝜕𝑥2 𝜕𝑥1 𝜕 2 𝑥2

Which can be interpreted as meaning the multiplication of the slopes of best response
functions shouldn’t exceed one at the equilibrium:
𝜕𝑓1 𝜕𝑓2
< 1 at 𝑥1∗ , 𝑥2∗
𝜕𝑥2 𝜕𝑥1

30
Nathan Grognet LLSMS2035 2015 - 2016

Let 𝑔(𝑥1 ) = 𝑓1−1 (𝑥1 ) − 𝑓2 (𝑥1 ) be an auxiliary function that measures the distance between two
best responses. It can be shown that this function crosses zero only from below if the slope at
the crossing point is positive:

𝜕𝑔(𝑥1 ) 𝜕𝑓1−1 (𝑥1 ) 𝜕𝑓2 (𝑥1 ) 1 𝜕𝑓2 (𝑥1 )

= − = − >0
𝜕𝑥1 𝜕𝑥1 𝜕𝑥1 𝜕𝑓2 (𝑥2 ) 𝜕𝑥1
𝜕𝑥2

Which holds if the previous equation holds. It holds in the newsvendor game since each slope
is less than one and hence the multiplication of slopes is less than one as well everywhere.

Multiple equilibria
It’s entirely possible that a non-equilibrium outcome results because one player plays one
equilibrium strategy while a second player chooses a strategy associated with another
equilibrium. However, if a game is repeated, then it’s possible that the players eventually find
themselves in one particular equilibrium. Furthermore, that equilibrium may not be the most
desirable one.

If the players have unidimensional strategies, then the system of 𝑛 first-order conditions
reduces to a single equation and one need only show that there is a unique solution to that
equation to prove the symmetric equilibrium is unique.

An alternative method to rule out some equilibria is to focus only on the Pareto optimal
equilibrium, of which there may be only one.

Dynamic games
Sequential moves: Stackelberg equilibrium concept
First we find the solution for the second player as a response to any decision made by the first:
𝜕𝜋2 (𝑥2 , 𝑥1 )
𝑥2∗ (𝑥1 ): =0
𝜕𝑥2

Next, find the solution for the first player anticipating the response by the second player:

𝑑𝜋1 (𝑥1 , 𝑥2∗ (𝑥1 )) 𝜕𝜋1 (𝑥1 , 𝑥2∗ ) 𝜕𝜋1 (𝑥1 , 𝑥2 ) 𝜕𝑥2∗
= + =0
𝑑𝑥1 𝜕𝑥1 𝜕𝑥2 𝜕𝑥1

The first player chooses the best possible point on the second player’s best response function,
so he’s at least as well off as he would be in NE. Consider now the newsvendor game:
𝑟2 − 𝑐2
𝑄2∗ (𝑄1 ) = 𝐹𝐷−1 +(𝐷 −𝑄 ) +( )
2 1 1 𝑟2

𝑑𝜋1 (𝑄1 , 𝑄2∗ (𝑄1 ))

= 𝑟1 Pr(𝐷1 + (𝐷2 − 𝑄2 )+ > 𝑄1 ) − 𝑐1
𝑑𝑄1
𝜕𝑄2∗
⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡−⁡𝑟1 Pr(𝐷1 + (𝐷2 − 𝑄2 )+ < 𝑄1 , 𝐷2 > 𝑄2 ) =0
𝜕𝑄1

31
Nathan Grognet LLSMS2035 2015 - 2016

Existence of a Stackelberg equilibrium is easy to demonstrate. However, uniqueness may be

considerably harder to demonstrate. A sufficient condition is quasi-concavity of the leader’s
profit function, 𝜋1 (𝑥1 , 𝑥2∗ (𝑥1 )).

Simultaneous moves: repeated and stochastic games

Repeated games haven’t found many applications in the SCM, because it’s awkward to assume
that nothing links successive games (there is some transfer of inventory and/or backorders
between periods). It’s important to notice that the set of equilibria in repeated games is much
larger than the set of equilibria in a static game and may include equilibria that aren’t possible
in the static game.

Players can use trigger strategies: the player will choose one strategy until the opponent
changes his play at which point the first player will change the strategy. This threat of reverting
to a different strategy may even induce players to achieve the best possible outcome, the
centralized solution, which is called an implicit collusion. To separate credible threats from non-
credible, Selten introduced the notion of subgame-perfect equilibrium.

In time-dependent multi-period games players’ payoffs in each period depend on the actions
in the previous as well as current periods. Typically, the payoff structure doesn’t change from
period to period, they are stationary. We will only discuss one type of time-dependent multi-
period games, stochastic games or Markov games. The setup of the stochastic game is
essentially a combination of a static game and a Markov Decisions Process: we have a set of
states and a transition mechanism 𝑝(𝑠 ′ |𝑠, 𝑥), probability that we transition from state 𝑠 to state
𝑠′ given action 𝑥. In game theory setting, non-stationary equilibria (trigger strategies for
example) are possible.

With the assumption that the policy is stationary the stochastic game reduces to an equivalent
static game and equilibrium is found as a sequence of NE in an appropriately modified single-
period game. To illustrate, consider an infinite-horizon variant of the newsvendor game with
lost sales in each period and inventory carry-over to the subsequent period. In addition to the
unit-revenue 𝑟 and the unit-cost 𝑐, we introduce the inventory holding cost ℎ and a discount
factor 𝛽. We also denote 𝑥𝑖𝑡 the inventory position at the beginning of the period and 𝑦𝑖𝑡 the
order-up-to quantity.
∞
+ + +
𝜋𝑖 (𝑥 1 ) = 𝐸 ∑ 𝛽𝑡𝑡−1 ⁡[𝑟𝑖 min (𝑦𝑖𝑡 , 𝐷𝑖𝑡 + (𝐷𝑗𝑡 − 𝑦𝑗𝑡 ) ) − ℎ𝑖 (𝑦𝑖𝑡 − 𝐷𝑖𝑡 − (𝐷𝑗𝑡 − 𝑦𝑗𝑡 ) ) − 𝑐𝑖 𝑄𝑖𝑡 ]
𝑡=1

+ +
𝑥𝑖𝑡+1 = (𝑦𝑖𝑡 − 𝐷𝑖𝑡 − (𝐷𝑗𝑡 − 𝑦𝑗𝑡 ) )

Using the standard manipulations from Heymann and Sobel, we have

∞

𝜋𝑖 (𝑥 1 ) = 𝑐𝑖 𝑥𝑖1 + ∑ 𝛽𝑖𝑡−1 𝐺𝑖𝑡 (𝑦𝑖𝑡 )

𝑡=1

32
Nathan Grognet LLSMS2035 2015 - 2016

Where 𝐺𝑖𝑡 (𝑦𝑖𝑡 ) is a single-period objective function

+ + +
𝐺𝑖𝑡 (𝑦𝑖𝑡 ) = 𝐸 [(𝑟𝑖 − 𝑐𝑖 ) (𝐷𝑖𝑡 + (𝐷𝑗𝑡 − 𝑦𝑗𝑡 ) ) − (𝑟𝑖 − 𝑐𝑖 ) (𝐷𝑖𝑡 + (𝐷𝑗𝑡 − 𝑦𝑗𝑡 ) − 𝑦𝑖𝑡 )
+ +
− (ℎ𝑖 + 𝑐𝑖 (1 − 𝛽𝑖 )) (𝑦𝑖𝑡 − 𝐷𝑖𝑡 − (𝐷𝑗𝑡 − 𝑦𝑗𝑡 ) ) ]

Since the single-period game is the same in each period and demand is stationary, by
restricting consideration to the stationary inventory policy 𝑦𝑖 = 𝑦𝑖𝑡 , we can find the solution
𝑟𝑖 − 𝑐𝑖
𝑦𝑖∗ = 𝐹 −1 + ( )
𝐷𝑖 +(𝐷𝑗 −𝑦𝑗∗ ) 𝑟𝑖 + ℎ𝑖 − 𝑐𝑖 𝛽𝑖

Except for the additional inventory carry-over and holding costs, the solution is essentially the
same as in the static game (page 25).

Cooperative games
Games in characteristic form and the core of the game
The cooperative game is also called the characteristic form. It consists of the set of players 𝑁
with subsets or coalitions 𝑆 ⊆ 𝑁 and a characteristic function 𝑣(𝑆) that specifies a (maximum)
value created by any subset of players in 𝑁 (the total pie that members of a coalition can create
and divide). The characteristic function only defines the total value that can be created by
utilizing all players’ resources.

Definition 5:
The utility vector 𝜋1 , … , 𝜋𝑁 is in the core of the cooperative game if ∀𝑆 ⊂ 𝑁, ∑𝑖∈𝑆 𝜋𝑖 ≥ 𝑣(𝑆) and
∑𝑖∈𝑁 𝜋𝑖 ≥ 𝑣(𝑁).

A utility vector is in the core if the total utility of every possible coalition is at least as large as
the coalition’s value, there doesn’t exist a coalition of players that could make all of its members
at least as well off and one member strictly better off.

As it’s true for NE, the core may not exist, it may be empty, and it’s often not unique. With an
empty core it’s difficult to predict what coalitions would form and what value each player would
receive.

Shapley value
This solution concept is based on three axioms:

1) The value of a player shouldn’t change due to permutations of players

2) If a player’s added value to the coalition is zero then this player shouldn’t get any profit
from the coalition
3) Additivity of payoffs is required: if 𝑣1 and 𝑣2 are characteristic functions in any two
games, and if 𝑞1 and 𝑞2 are a player’s Shapley value in these two games, then the
player’s Shapley value in the composite game, 𝑣1 + 𝑣2 , must be 𝑞1 + 𝑞2 .

There’s a unique Shapley value for each player that satisfies those three axioms.

33
Nathan Grognet LLSMS2035 2015 - 2016

Theorem 11:
The Shapley value, 𝜋𝑖 , for player 𝑖 in an 𝑁-person non-cooperative game with transferable utility
is:
|𝑆|! (|𝑁| − |𝑆| − 1)!
𝜋𝑖 = ∑ (𝑣(𝑆 ∪ {𝑖}) − 𝑣(𝑆))
|𝑁|!
𝑆⊆𝑁\𝑖

If the orderings are equally likely, there is a probability of |𝑆|! (|𝑁| − |𝑆| − 1)!⁄|𝑁|! that when
player 𝑖 is picked he will find 𝑆 players in the coalition already. The marginal contribution of
adding player 𝑖 to coalition 𝑆 is 𝑣(𝑆 ∪ {𝑖}) − 𝑣(𝑆). Hence, the Shapley value is nothing more
than a marginal expected contribution of adding player 𝑖 to the coalition. The Shapley value
needn’t be in the core, hence, although the Shapley value is appealing from the perspective of
fairness, it may not be a reasonable prediction of the outcome of a game.

Biform games
A biform game can be thought of as a non-cooperative game with cooperative games as
outcomes and those cooperative games lead to specific payoffs. The core of each possible
cooperative game is non-empty, but it’s also unlikely to be unique. The proposed solution is
that each player is assigned a confidence index, 𝛼𝑖 ∈ [0,1], each player then expects to earn in
each possible cooperative game a weighted average of the minimum and maximum values in
the core with 𝛼𝑖 being the weight. Once a specific value is assigned to each player for each
cooperative sub-game, the first stage non-cooperative game can be analyzed just like any
other non-cooperative game.

Signaling and screening

Signaling game
A player has more information than the other and it’s the player with more information that
makes the first move. At the start of the game, the supplier and the manufacturer know the
probability distribution over the set of types. We refer to this probability distribution as the
supplier’s belief regarding the types of manufacturers. If the supplier’s belief regarding the
manufacturer’s type is resolved to a single type after observing the manufacturer’s action, then
the manufacturer has signaled a type to the supplier. If the manufacturer’s action doesn’t
provide the supplier with additional information, we have a pooling equilibrium. It is not
desirable because the supplier won’t choose the correct capacity (the supplier might have more
capacity than needed, or he could underinvest in capacity).

In contrast to the pooling equilibrium, there is the separating equilibrium (also called signaling
equilibrium). The key condition is that only one manufacturer type is willing to choose the
action designated for that type. An ideal action for a high demand manufacturer is one that
costlessly signals his high demand forecast. If a costless signal doesn’t exist, then the goal is to
seek the lowest cost signal. With forced compliance, there are many costless signals for the
manufacturer. One solution is to give a sufficiently large lump sum payment to the supplier (it
is called “burning money”). A better signal is a contract offer that is costless to a high demand
manufacturer but expensive to a low demand manufacturer. A good example of that is a
minimum commitment that is costly only if realized demand is lower than the commitment.

34
Nathan Grognet LLSMS2035 2015 - 2016

Screening
Here the player that lacks information is the first to move. The supplier offers a menu of
contracts with the intention of getting the manufacturer to reveal his type via the contract
selected, it’s referred as “mechanism design”. The “revelation principle” begins with the
presumption that a set of optimal mechanisms exists. Associated with each of these
mechanisms is a NE that specifies which contract each manufacturer type chooses and the
supplier’s action given the chosen contract. Nevertheless, it states that an optimal mechanism
that involves deception can be replaced by a mechanism that doesn’t involve deception, there
exists an equivalent mechanism that is truth-telling.

The gap between what a manufacturer earns with the menu of contracts and what he would
earn if the supplier knew his type is called an “information rent”. A manufacturer’s private
information allows him to keep some rent that he wouldn’t be able to keep if the supplier knew
his type. Hence, even though there may be no cost to information revelation with a signaling
game, the same isn’t true with a screening game.

35