Professional Documents
Culture Documents
Anton J. Kleywegt
1
Introduction
Decision problems with a single decision maker are often
formulated as optimization problems
Let us review what is an Optimization Problem
Single decision maker
Set 𝑋𝑋 of feasible decisions
Objective function 𝑓𝑓: 𝑋𝑋 ↦ ℝ
For any 𝑥𝑥 ∈ 𝑋𝑋, 𝑓𝑓(𝑥𝑥) is the objective value that the decision maker
wants to maximize or minimize (let’s maximize here)
2
Introduction
Many decision making settings involve the following:
There are more than one decision maker
The objective function of some decision makers depend not
only on their own decision, but also on decisions of other
decision makers
If these conditions hold, then the decision making setting is
called a game (and not an optimization problem)
3
What is a Game?
Setup for many types of game:
There are 𝑛𝑛 decision makers, usually called players, indexed 𝑖𝑖 =
1, … , 𝑛𝑛
Each decision maker 𝑖𝑖 has feasible set 𝑋𝑋𝑖𝑖
Each decision maker 𝑖𝑖 has an objective function 𝑓𝑓𝑖𝑖 ∶ 𝑋𝑋1 × ⋯ ×
𝑋𝑋𝑛𝑛 ↦ ℝ
For any 𝑥𝑥1 ∈ 𝑋𝑋1 , … , 𝑥𝑥𝑛𝑛 ∈ 𝑋𝑋𝑛𝑛 , 𝑓𝑓𝑖𝑖 (𝑥𝑥1 , … , 𝑥𝑥𝑛𝑛 ) is the objective value that
decision maker 𝑖𝑖 wants to maximize
As the notation indicates, the dilemma of a decision maker is
that the objective value of the decision maker depends on
decisions of other decision makers
This dilemma makes it difficult for decision makers to make
good decisions, and for impartial analysts to predict outcomes
4
of games
What is a Game?
The study of games in the sense given above (multiple
decision makers) is called game theory
Purposes of game theory
Descriptive: To describe or explain observed phenomena
involving multiple decision makers
Predictive: To predict the outcome of decision making settings
involving multiple decision makers
Prescriptive: To inform decision makers how decisions should
be made in decision making settings involving multiple decision
makers
5
Types of Games
There are many types of games
The information structure of a game is an important
characteristic that distinguishes different types of games
Consider a game with 2 decision makers (often called players),
𝑖𝑖 = 1,2
If player 1 makes a decision without knowing the decision of
player 2, and player 2 makes a decision knowing the decision of
player 1, then the game is called a leader-follower game, or
Stackelberg game, or principal-agent game
Player 1 is called the leader or principal (usually imagined that
player 1 decides first), and player 2 is called the follower or
agent (player 2 decides second)
6
Types of Games
The information structure of a game (continued)
If all players make decisions knowing the decisions of all other
relevant players (so that players know all decisions made by
other players that affect their objective values), and all players
commit to their decisions simultaneously, then the game is
called a simultaneous play game
You may wonder how such a situation could come about – that
is a good question, that we will ask again later
7
Leader-Follower Games
Recall: A leader-follower game is a game in which the
leader makes a decision without knowing the decision of
the follower, or more precisely, without the follower
having committed to a decision, and the follower makes a
decision knowing the decision that the leader has
committed to make
There are 2 decision makers indexed 𝑖𝑖 = 1,2; suppose
that player 1 is the leader, and player 2 is the follower
Each decision maker 𝑖𝑖 has feasible set 𝑋𝑋𝑖𝑖
Each decision maker 𝑖𝑖 has objective function 𝑓𝑓𝑖𝑖 ∶ 𝑋𝑋1 ×
𝑋𝑋2 ↦ ℝ: For any 𝑥𝑥1 ∈ 𝑋𝑋1 , 𝑥𝑥2 ∈ 𝑋𝑋2 , 𝑓𝑓𝑖𝑖 (𝑥𝑥1 , 𝑥𝑥2 ) is the
objective value that decision maker 𝑖𝑖 wants to maximize
8
Leader-Follower Games
Since player 2 knows the decision 𝑥𝑥1 of player 1 at the
time that player 2 has to make a decision, player 2 simply
solves the following optimization problem:
10
Solution Status of an Optimization Problem
An optimization problem can have multiple optimal
solutions
Example:
11
Solution Status of an Optimization Problem
An optimization problem can have no optimal solutions
Example 1:
12
Solution Status of an Optimization Problem
An optimization problem can have no optimal solutions
Example 2:
13
Leader-Follower Games
If the optimization problem of player 2 has a unique
optimal solution 𝑥𝑥2∗ (𝑥𝑥1 ), then we (and the leader player
1) can with some confidence predict how player 2 will
behave for each choice of decision 𝑥𝑥1 by player 1
If the optimization problem of player 2 has no optimal
solution or multiple optimal solutions, then we encounter
the problem that we cannot with confidence predict how
player 2 will behave for each choice of decision 𝑥𝑥1 by
player 1
14
Leader-Follower Games
Suppose that the optimization problem of player 2 has a
unique optimal solution 𝑥𝑥2∗ (𝑥𝑥1 ) for each choice of
decision 𝑥𝑥1 by player 1
In addition, suppose that player 1 is smart enough to
know the unique optimal solution 𝑥𝑥2∗ (𝑥𝑥1 ) of player 2 for
each choice of decision 𝑥𝑥1 by player 1
Then player 1 wants to solve the following optimization
problem:
15
Leader-Follower Games
Example:
A manufacturer produces widgets and sells them to a retailer
at a price of 𝑥𝑥1 each, chosen by the manufacturer
The retailer sells the widgets to consumers at a price of 𝑥𝑥2
each, chosen by the retailer
Consumer demand 𝑑𝑑(𝑥𝑥2 ) per unit time for widgets depend on
the retailer price 𝑥𝑥2 , as follows: 𝑑𝑑(𝑥𝑥2 ) = α − β𝑥𝑥2 , where α, β >
0, as long as 0 < 𝑥𝑥2 < α/β
The manufacturer has a cost of 𝑐𝑐1 per widget, and the retailer
has a cost of 𝑐𝑐2 per widget in addition to the price paid to the
manufacturer
What do you predict will happen in this simple market?
16
Leader-Follower Games
Example (continued):
First, note that the maximum price at which any widgets will be
sold to consumers is α/β
Thus, if the total supply chain cost 𝑐𝑐1 + 𝑐𝑐2 > α/β, then the
manufacturer and the retailer cannot both be profitable, and
thus there will be no widget supply chain
For the rest of the example, suppose that 𝑐𝑐1 + 𝑐𝑐2 ≤ α/β, so
that there can be a profitable supply chain (it is no guarantee
that there will be a profitable supply chain)
As is typical, suppose that the manufacturer chooses a price 𝑥𝑥1
and reveals it to the retailer before the retailer decides how
many widgets to buy (it would seem unreasonable to expect
the retailer to decide how much to buy before the retailer
knows what price the retailer will have to pay, but many health
17
care providers expect patients to do exactly that)
Leader-Follower Games
Example (continued):
Thus, the manufacturer is the leader, and the retailer is the
follower
As we did in our earlier analysis, we first look at the follower’s
optimization problem for any given value of the leader’s
decision 𝑥𝑥1
Assume that the retailer wants to maximize the retailer’s profit
per unit time
Thus the retailer’s optimization problem is
18
Leader-Follower Games
Example (continued):
19
Leader-Follower Games
Example (continued):
What do we need for 𝑥𝑥2∗ to be feasible?
Recall that it is required that 0 < 𝑥𝑥2 < α/β
Thus it is required that 𝑐𝑐2 + 𝑥𝑥1 < α/β
Thus, for any manufacturer price 𝑥𝑥1 < α/β − 𝑐𝑐2 , we have that
the retailer’s best response decision is
20
Leader-Follower Games
Example (continued):
Assume that the manufacturer wants to maximize the
manufacturer’s profit per unit time
Thus the manufacturer’s optimization problem is
21
Leader-Follower Games
Example (continued):
22
Leader-Follower Games
Example (continued):
What do we need for 𝑥𝑥1∗ to be feasible?
Recall that it is required that 𝑐𝑐2 + 𝑥𝑥1 < α/β
23
Leader-Follower Games
Example (continued):
24
Supply Chain Structure
Questions:
How does the length of a supply chain affect the supply chain’s
total profit?
How does the length of a supply chain affect the supply chain’s
consumer surplus?
What are the effects of vertical integration versus outsourcing
from the supply chain’s point of view?
25
Supply Chain Structure
26
Supply Chain Structure
27
Supply Chain Structure
28
Best Response Pricing of Sellers in a
Supply Chain
29
Best Response Pricing of Sellers in a
Supply Chain
30
Best Response Pricing of Sellers in a
Supply Chain
31
Best Response Pricing of Sellers in a
Supply Chain
32
Best Response Pricing of Sellers in a
Supply Chain
33
Best Response Pricing of Sellers in a
Supply Chain
34
Best Response Pricing of Sellers in a
Supply Chain
35
Best Response Pricing of Sellers in a
Supply Chain
36
Best Response Pricing of Sellers in a
Supply Chain
37
Best Response Pricing of Sellers in a
Supply Chain
38
Double Marginalization in a Supply Chain
39
Double Marginalization in a Supply Chain
40
Optimal Supply Chain
41
Optimal Supply Chain
42
Double Marginalization in a Supply Chain
43
Vertical Integration and Outsourcing in
a Supply Chain
Lessons learned:
If a supply chain becomes longer, for example through
outsourcing, then the end consumer price increases, the
equilibrium quantity decreases, the total supply chain profit
decreases, and the consumer surplus decreases
Unless outsourcing greatly reduces supply chain cost, outsourcing
is bad for sellers and bad for consumers
If a supply chain becomes shorter, for example through
vertical integration, then the end consumer price decreases,
the equilibrium quantity increases, the total supply chain
profit increases, and the consumer surplus increases
Unless vertical integration greatly increases supply chain cost,
vertical integration is good for sellers and good for consumers
44
Solutions of Games
Often the solution of a game is not desirable, and a
solution that is better for everyone is not a solution of
the game
Thus, Adam Smith’s invisible hand makes mistakes
Once we acknowledge the fact that different decision
makers make decisions in pursuit of their own objectives
that turn out not to be the best for the entire supply
chain, a natural question is whether one can design the
decision making environment or the game, through
contracts or laws, in such a way that when people make
decisions in pursuit of their own objectives, the outcome
will be good overall, for example, such that the desirable
outcome will be a predictable solution of the game
46
Supply Chain Coordination
Supply chain coordination is the design of the rules of the
supply chain, usually through contracts, in such a way that
when members of the supply chain make decisions in
pursuit of their own objectives, the outcome will be good
for the supply chain
If the design leads to an outcome that is optimal for the
supply chain, then we say that the supply chain is
coordinated, or that the design coordinates the supply
chain
47
Supply Chain Coordination
Question:
Consider a supply chain with 2 members, say a wholesaler and
a retailer
Can a fixed wholesale price coordinate the supply chain?
That is, does there exist a contract, specifying the wholesale
price, that is acceptable to both the wholesaler and the retailer,
such that the outcome under the contract will be optimal for
the supply chain?
48
Supply Chain Coordination
49
Supply Chain Coordination
50
Supply Chain Coordination
51
Supply Chain Coordination
52
Supply Chain Coordination
53
Supply Chain Coordination
54
Supply Chain Coordination
55
Supply Chain Coordination
56
Supply Chain Coordination
57
Supply Chain Coordination
Next we show how to design a 2-parameter contract called a
buy-back contract such that the supply chain will be
coordinated, and such that the members of the supply chain will
be better off with such a contract than without a contract
The purpose of this lesson is not to make a case for buy-back
contracts in particular, the purpose is to demonstrate the
concepts and calculations associated with the design of
contracts to coordinate the supply chain, and we use buy-back
contracts to facilitate the demonstration – there are many
other contract types that will also coordinate the supply chain
in such a way that the members of the supply chain will be
better off with such a contract than without a contract
58
Supply Chain Coordination
What is a buy back contract?
The supplier financially compensates the retailer for every unit of the
SKU that remains unsold at the end of the selling season
It does not necessarily mean that the retailer returns the unsold goods
to the possession of the supplier, often the retailer salvages the unsold
goods himself, so that the “buy back” just means that the supplier gives
some financial compensation to the retailer for every unit unsold at
the end of the selling season, that is, the supplier supplements the
retailer’s salvage value, and nothing more
59
Supply Chain Coordination
Question:
Consider a supply chain with 2 members, say a wholesaler and a
retailer
Can a buy back contract coordinate the supply chain?
That is, does there exist a 2-parameter contract, specifying the
wholesale price and the buy back price, that is acceptable to both the
wholesaler and the retailer, such that the outcome under the contract
will be optimal for the supply chain?
60
Supply Chain Coordination
61
Supply Chain Coordination
62
Supply Chain Coordination
63
Supply Chain Coordination
64
Supply Chain Coordination
65
Supply Chain Coordination
66
Supply Chain Coordination
67
Supply Chain Coordination
68
Supply Chain Coordination
69
Supply Chain Coordination
70
Supply Chain Coordination
71
Supply Chain Coordination
72
Supply Chain Coordination
73
Supply Chain Coordination
74
Supply Chain Coordination
75
Supply Chain Coordination
76
Supply Chain Coordination
77
Supply Chain Coordination
78
Supply Chain Coordination
79
Supply Chain Coordination
80
Supply Chain Coordination
81
Supply Chain Coordination
82
Supply Chain Coordination
83
Supply Chain Coordination
84
Supply Chain Coordination
85
Supply Chain Coordination
86
Supply Chain Coordination
87
Supply Chain Coordination
88
Supply Chain Coordination
89
Supply Chain Coordination
90
Supply Chain Coordination
91
Supply Chain Coordination
92
Supply Chain Coordination
93
Supply Chain Coordination
A 2-parameter contract such as a buy-back contract can coordinate a
supply chain consisting of a wholesaler and a retailer who faces a
newsvendor problem in each sales season
We determined how to design a buy-back contract such that
the expected total supply chain profit under such a contract is equal to the
maximum expected total supply chain profit, and
both the retailer and the wholesaler are better off under the contract than
without a contract
There is a range of such contracts that we can choose from, but some
are better for the retailer and others are better for the wholesaler
Results from cooperative game theory tell us how to choose from
the range of such contracts
These ideas are very general, but the calculations depend on the
particular situation at hand
94
Simultaneous Play Games
If all players make decisions knowing the decisions of all
other relevant players (so that players know all decisions
made by other players that affect their objective values),
and all players commit to their decisions simultaneously,
then the game is called a simultaneous play game
It is a good question how such a situation could come about
95
Simultaneous Play Games
More precise statement of a simultaneous play game:
96
Simultaneous Play Games
Statement of a simultaneous play game (continued):
97
Simultaneous Play Games
Solution concept of a simultaneous play game:
98
Simultaneous Play Games
Solution concept of a simultaneous play game
(continued):
Equilibrium of a game: Given the decisions of all the players, no
player has an incentive to deviate from the player’s decision if
the other players will remain committed to their decisions
That is, given the decisions of all the players, each player’s
decision optimizes that player’s objective function
That does not imply that an equilibrium is a desirable outcome
for the players (more about that later)
Concept of equilibrium used in a particular economic context
by Cournot in 1838 (more about that later)
Concept of equilibrium studied by Von Neumann and
Morgenstern in 1944, and generalized by Nash in 1951
99 (sometimes called Nash equilibrium)
Simultaneous Play Games
Solution concept of a simultaneous play game
(continued):
100
Simultaneous Play Games
Questions:
Does a game always have an equilibrium?
If no, what if a game has no equilibrium?
Does a game always have a unique equilibrium?
If no, what if a game has multiple equilibria?
Is an equilibrium always a desirable outcome?
Does an equilibrium of a game describe the outcome of a
game between human decision makers, that is, what is the
descriptive and predictive accuracy of the equilibrium concept?
How can game theory and equilibria be used in practice?
101
Simultaneous Play Games and Equilibria
Example: Bertrand Duopoly
Two sellers sell competing products
Each seller chooses the price of its product
Each seller knows how demand depends on the prices of both
sellers’ products
Each seller wants to maximize its own profit
103
Simultaneous Play Games and Equilibria
Example: Bertrand Duopoly (continued)
104
Simultaneous Play Games and Equilibria
Example: Bertrand Duopoly (continued)
105
Simultaneous Play Games and Equilibria
Example: Merger of Bertrand Duopolists (Horizontal
Integration)
Two sellers merge to form one company
Such a merger of sellers of competing products is called
horizontal integration, to distinguish it from vertical integration
involving a merger of a supplier and its customer
The merged company chooses the prices of the products of
both original companies to maximize its profit
106
Simultaneous Play Games and Equilibria
Example: Merger of Bertrand Duopolists (continued)
107
Simultaneous Play Games and Equilibria
Example: Merger of Bertrand Duopolists (continued)
108
Simultaneous Play Games and Equilibria
Example: Merger of Bertrand Duopolists (continued)
109
Simultaneous Play Games and Equilibria
Example: Merger of Bertrand Duopolists (continued)
110
Simultaneous Play Games and Equilibria
Example: Cournot Duopoly
Two sellers produce products that consumers regard as
practically identical
Each seller chooses how much to produce
The market clearing price depends on the total amount
produced (and offered in the market)
Each seller knows how much the other seller produces, and
how price depends on the total amount produced
Each seller wants to maximize its own profit
112
Simultaneous Play Games and Equilibria
Example: Cournot Duopoly (continued)
113
Simultaneous Play Games and Equilibria
Example: Merger of Cournot Duopolists (Horizontal
Integration)
Two sellers merge to form one company
The merged company chooses the total quantity of the
product to produce to maximize its profit
114
Simultaneous Play Games and Equilibria
Example: Merger of Cournot Duopolists (continued)
115
Simultaneous Play Games and Equilibria
Example: Merger of Cournot Duopolists (continued)
116
Simultaneous Play Games and Equilibria
Example: Merger of Cournot Duopolists (continued)
117
Simultaneous Play Games and Equilibria
Example: Traffic equilibrium
118
Simultaneous Play Games and Equilibria
Example: Traffic equilibrium (continued)
119
Simultaneous Play Games and Equilibria
Example: Traffic equilibrium (continued)
120
Simultaneous Play Games and Equilibria
Example: Traffic equilibrium (continued)
121
Simultaneous Play Games and Equilibria
Example: Traffic equilibrium (continued)
122
Simultaneous Play Games and Equilibria
Example: Traffic equilibrium (continued)
123
Simultaneous Play Games and Equilibria
Example: Traffic equilibrium (continued)
124
Simultaneous Play Games and Equilibria
Example: Traffic equilibrium (continued)
125
Simultaneous Play Games and Equilibria
Example: Traffic equilibrium (continued)
126
Simultaneous Play Games and Equilibria
The prisoner’s dilemma game
Two complicit conspirators are apprehended
The prosecutor needs one of the two prisoners to testify
against the other one to get a strong conviction
The prosecutor makes an offer to each of the two prisoners:
If the prisoner testifies against the other prisoner, and the other one
does not testify, then the testifying prisoner will go free, while the
other one gets a stiff sentence
If both testify, then both get moderate sentences
If neither testifies, both will be kept in custody while the court case
drags on, but eventually both will be set free
127
Simultaneous Play Games and Equilibria
The prisoner’s dilemma game (continued)
Alternatives: T = testify, N = not testify
128
Simultaneous Play Games
Questions:
Does a game always have an equilibrium?
If no, what if a game has no equilibrium?
Does a game always have a unique equilibrium?
If no, what if a game has multiple equilibria?
Is an equilibrium always a desirable outcome?
Does an equilibrium of a game describe the outcome of a
game between human decision makers, that is, what is the
descriptive and predictive accuracy of the equilibrium concept?
How can game theory and equilibria be used in practice?
What have we learned from these examples?
Some games have a unique equilibrium
An equilibrium is not always a desirable outcome
129
Dominant and Dominated Decisions
Decision 𝑥𝑥𝑖𝑖′ strictly dominates decision 𝑥𝑥𝑖𝑖 for player 𝑖𝑖 if
decision 𝑥𝑥𝑖𝑖′ has strictly better objective value for player 𝑖𝑖
than decision 𝑥𝑥𝑖𝑖 regardless of the other players’ decisions:
𝑓𝑓𝑖𝑖 𝑥𝑥1 , … , 𝑥𝑥𝑖𝑖−1 , 𝑥𝑥𝑖𝑖′ , 𝑥𝑥𝑖𝑖+1 , … 𝑥𝑥𝑛𝑛 > 𝑓𝑓𝑖𝑖 (𝑥𝑥1 , … , 𝑥𝑥𝑖𝑖−1 , 𝑥𝑥𝑖𝑖 , 𝑥𝑥𝑖𝑖+1 , … 𝑥𝑥𝑛𝑛 )
for all 𝑥𝑥1 ∈ 𝑋𝑋1 , … , 𝑥𝑥𝑖𝑖−1 ∈ 𝑋𝑋𝑖𝑖−1 , 𝑥𝑥𝑖𝑖+1 ∈ 𝑋𝑋𝑖𝑖+1 , … , 𝑥𝑥𝑛𝑛 ∈ 𝑋𝑋𝑛𝑛
In shorthand: 𝑓𝑓𝑖𝑖 𝑥𝑥𝑖𝑖′ , 𝑥𝑥−𝑖𝑖 > 𝑢𝑢𝑖𝑖 (𝑥𝑥𝑖𝑖 , 𝑥𝑥−𝑖𝑖 ) for all 𝑥𝑥−𝑖𝑖 ∈ 𝑋𝑋−𝑖𝑖
′
𝑥𝑥𝑖𝑖 is said to be strictly dominated by 𝑥𝑥𝑖𝑖 in 𝑋𝑋
Strictly dominant decision:
Decision 𝑥𝑥𝑖𝑖′ of player 𝑖𝑖 is called a strictly dominant decision
if decision 𝑥𝑥𝑖𝑖′ strictly dominates all other decisions 𝑥𝑥𝑖𝑖 of player
𝑖𝑖: 𝑓𝑓𝑖𝑖 𝑥𝑥𝑖𝑖′ , 𝑥𝑥−𝑖𝑖 > 𝑓𝑓𝑖𝑖 𝑥𝑥𝑖𝑖 , 𝑥𝑥−𝑖𝑖 for all (𝑥𝑥𝑖𝑖 , 𝑥𝑥−𝑖𝑖 ) ∈ 𝑋𝑋 with 𝑥𝑥𝑖𝑖 ≠ 𝑥𝑥𝑖𝑖′
130
Equilibrium in Strictly Dominant Decisions
Decision profile 𝑥𝑥 = (𝑥𝑥1 , … , 𝑥𝑥𝑛𝑛 ) of a game is called an
equilibrium in strictly dominant decisions if each
player’s decision 𝑥𝑥𝑖𝑖 is a strictly dominant decision for
player 𝑖𝑖.
A rational player does not choose a strictly dominated decision
A rational player chooses a strictly dominant decision, even
without requiring to know what other players will do
So, if such an outcome of strictly dominant decisions exists (it
rarely does), then it is a very strong prediction of what the
players will do
Prisoners’ Dilemma has an equilibrium in strictly dominant
decisions
If an ESDD exists, can there be more than one ESDDs?
131
Dominant and Dominated Decisions
Player 2
Player 1 Left Center Right
Up 16, 2 1, 5 7, 3
Middle 4, 1 3, 4 6, 3
Down 3, 9 -1, 8 8, -1
Does Player 1 have a strictly dominant decision? Player 2?
No for both
Does Player 1 have a strictly dominated strategy?
No.
Does Player 2 have a strictly dominated strategy?
Yes, Right is strictly dominated by Center, so eliminate Right!
132
Elimination of strictly dominated decisions
Rational players do not choose decisions that are strictly
dominated
So we can eliminate any strictly dominated decision and
consider only the reduced game
This process can be repeated until no strictly dominated
decision in reduced games can be found
This process of repeated elimination of strictly dominated
decisions is called iterated elimination of strictly
dominated decisions
133
Iterated Elimination of Strictly
Dominated Decisions
Player 2 Player 2
Left Center Right Left Center
Player 1
Player 1
Up 16, 2 1, 5 7, 3 Up 16, 2 1, 5
Middle 4, 1 3, 4 6, 3 Middle 4, 1 3, 4
Down 3, 9 -1, 8 8, -1 Down 3, 9 -1, 8
Player 2 Player 2
Center Left Center
Player 1
Player 1
Up 1, 5 Up 16, 2 1, 5
Middle 3, 4 Middle 4, 1 3, 4
Player 2
Player 1
Center
Middle 3, 4
134
Iterated Elimination of Strictly
Dominated Decisions
If iterated elimination of strictly dominated decisions
leads to a single decision profile (may not always!), it may
be a reliable prediction of the game outcome
Let’s formalize the process:
For 𝑡𝑡 = 0, let 𝑋𝑋𝑖𝑖0 = 𝑋𝑋𝑖𝑖 for each player 𝑖𝑖
𝑡𝑡
For 𝑡𝑡 ≥ 1, let 𝑋𝑋𝑖𝑖 denote those decisions of player 𝑖𝑖 surviving
after the 𝑡𝑡-th round of elimination
i.e. 𝑥𝑥𝑖𝑖 ∈ 𝑋𝑋𝑖𝑖𝑡𝑡 if 𝑥𝑥𝑖𝑖 ∈ 𝑋𝑋𝑖𝑖𝑡𝑡−1 and 𝑥𝑥𝑖𝑖 is not strictly dominated in 𝑋𝑋 𝑡𝑡
A decision 𝑥𝑥𝑖𝑖 for player 𝑖𝑖 is called iteratively strictly
undominated in 𝑋𝑋 if 𝑥𝑥𝑖𝑖 ∈ 𝑋𝑋𝑖𝑖𝑡𝑡 for all 𝑡𝑡 ≥ 1, i.e. 𝑥𝑥𝑖𝑖
survives all rounds of iterated elimination of strictly
dominated decisions
135
Weakly Dominated Decisions
Player 2
Player 1 L R
U 1, 1 0, 0
D 0, 0 0, 0
137
Simultaneous Play Games
Question: Does a game always have an equilibrium?
Each game considered so far has an equilibrium
Consider the game of matching pennies:
2 Players
Each player has 2 alternatives (heads or tails)
Player 1 wants to choose an alternative different from that of player 2,
and player 2 wants to choose the same alternative as player 1
Applications:
Penalty shootout in soccer: Striker wants to kick where goalkeeper
will not cover, and goalkeeper wants to cover where striker kicks
Operation of weigh stations on highways
Screening of imported containers for dangerous contents
138
Simultaneous Play Games
Question: Does a game always have an equilibrium?
The game of matching pennies
139
Simultaneous Play Games
Question: Does a game always have an equilibrium?
The game of rock-paper-scissors
140
Simultaneous Play Games
The games of matching pennies, rock-paper-scissors, and
others like it, have no equilibrium because the feasible
sets of the players are not convex
One proposed remedy: convexify the feasible sets of the
players
The way this is usually described is that randomized
decisions are allowed
In this context, randomization is a description of
convexification, and has nothing to do with confusing or
misleading an opponent
Having acknowledged that, feel free to think of
randomized decisions if you prefer to do so
141
Simultaneous Play Games
Simultaneous play game with randomized decisions
(mixed strategies)
142
Simultaneous Play Games
Simultaneous play game with randomized decisions
(mixed strategies)
143
Simultaneous Play Games
Solution concept of a simultaneous play game with
randomized decisions (mixed strategies)
144
Simultaneous Play Games
Properties of mixed strategy equilibria
Every (deterministic/pure strategy) equilibrium is a special case
of a mixed strategy equilibrium
If each player has a finite number of alternatives, then the game
always has at least one mixed strategy equilibrium
If a game has no (deterministic/pure strategy) equilibria, then it
has a unique mixed strategy equilibrium
If in a mixed strategy equilibrium at least one player strictly
randomizes, that is, for some player 𝑖𝑖, for some feasible
decision 𝑥𝑥𝑖𝑖 ∈ 𝑋𝑋𝑖𝑖 , the mixed strategy equilibrium 𝑝𝑝∗ has 0 <
𝑝𝑝𝑖𝑖∗ (𝑥𝑥𝑖𝑖 ) < 1, then the mixed strategy equilibrium can be called
a strict mixed strategy equilibrium
145
Simultaneous Play Games
Properties of mixed strategy equilibria (continued)
146
Simultaneous Play Games
Mixed strategy equilibrium
The game of matching pennies
147
Simultaneous Play Games
Properties of mixed strategy equilibria (continued)
A player is indifferent between all decisions on which the
player puts positive probabilities – all those decisions have
exactly the same objective value for the player
Conceptual shortcoming of mixed strategy equilibria: If the
player is indifferent between those decisions, why would the
player choose the particular probabilities prescribed by the
mixed strategy equilibrium, why not choose one of the
decisions deterministically?
Empirical shortcoming of mixed strategy equilibria: Mixed
strategy equilibria sometimes describe how people make
decisions for repeated games, but they hardly ever describe
how people make decisions for single stage games
148
Simultaneous Play Games
Questions:
Does a game always have an equilibrium? No
If no, what if a game has no equilibrium? Look for mixed
strategy equilibria
Does a game always have a unique equilibrium?
If no, what if a game has multiple equilibria?
Is an equilibrium always a desirable outcome?
Does an equilibrium of a game describe the outcome of a
game between human decision makers, that is, what is the
descriptive and predictive accuracy of the equilibrium concept?
How can game theory and equilibria be used in practice?
149
Simultaneous Play Games
Questions:
Does a game always have an equilibrium?
If no, what if a game has no equilibrium?
Does a game always have a unique equilibrium?
If no, what if a game has multiple equilibria?
Is an equilibrium always a desirable outcome?
Does an equilibrium of a game describe the outcome of a
game between human decision makers, that is, what is the
descriptive and predictive accuracy of the equilibrium concept?
How can game theory and equilibria be used in practice?
150
Simultaneous Play Games
Question: Does a game always have a unique equilibrium?
The previously considered games always have unique equilibria
Consider the game of battle of the sexes (coordination game):
2 Players, man and woman
Each player has 2 alternatives for an evening of entertainment
(football or opera)
Man prefers football, woman prefers opera, and both prefer to spend
the evening together rather than separate
151
Simultaneous Play Games
The game of battle of the sexes (coordination game)
152
Simultaneous Play Games
The game of battle of the sexes (coordination game)
The game of battle of the sexes has two (deterministic/pure
strategy) equilibria: (𝑂𝑂, 𝑂𝑂) and (𝐹𝐹, 𝐹𝐹)
The game of battle of the sexes also has a strict mixed strategy
equilibrium
Recall the property that if in equilibrium a player puts positive
probability on 2 alternatives, then both alternatives must have
exactly the same objective value for the player (given the
mixed strategy of the other player)
153
Simultaneous Play Games
The game of battle of the sexes (coordination game)
The strict mixed strategy equilibrium of the game of battle of
the sexes:
154
Simultaneous Play Games
A general 2 × 2 simultaneous play game:
155
Simultaneous Play Games
A general 2 × 2 simultaneous play game (continued):
156
Simultaneous Play Games
A general 2 × 2 simultaneous play game (continued):
157
Simultaneous Play Games
A general 2 × 2 simultaneous play game (continued):
158
Simultaneous Play Games
A general 2 × 2 simultaneous play game (continued):
159
Simultaneous Play Games
The game of battle of the sexes (coordination game)
The game of battle of the sexes has two pure strategy
equilibria as well as a strict mixed strategy equilibrium
The two pure strategy equilibria (𝑂𝑂, 𝑂𝑂) and (𝐹𝐹, 𝐹𝐹) are efficient
– they produce Pareto efficient outcomes in which you cannot
increase one player’s benefit without reducing another player’s
benefit
The two pure strategy equilibria (𝑂𝑂, 𝑂𝑂) and (𝐹𝐹, 𝐹𝐹) are in some
sense unfair – in each such pure strategy equilibrium, one
player gets much more benefit than the other player
The strict mixed strategy equilibrium is inefficient – it puts
positive probability on action pairs that are not Pareto efficient
160
Simultaneous Play Games
The game of battle of the sexes (coordination game)
161
Simultaneous Play Games
A correlated equilibrium for the game of battle of the
sexes:
162