Chapter 1

Introduction

From ancient times, human history has been dotted with numerous instances of conflict and
cooperation. Why do humans conflict or cooperate? With the invention of science1 , the focus
of modern science shifts from observing natural and social phenomena to constructing theories
to explain the observed phenomena. Game theory is one such theory, rooted in mathematics,
to explain conflict and cooperation. According to Morgenstern (1968), the modern origins of
game theory dates back to Leibnitz who, in 1710, emphasized the need of modeling conflict
situations. In 1712, Waldegrave formulated an initial version of the minimax strategy. In 1881,
Edgeworth showed the congruence between game theory and economics. One of the early and an
important result is the minimax theorem by von Neumann (1928). The publication of “Theory
of Games and Economic Behavior” by von Neumann and Morgenstern in 19442 provided a
logical foundation to game theory and its application in economic and social sciences. Nash
(1950) introduces the concept of Nash equilibrium that forms the cornerstone of applications of
game theory in multiple disciplines from economics to biology.
The primitives of game theory include players (or agents) who are rational, self-interested,
and have free will. Each player is endowed with an action set. There are rules (or protocols)
The Primitives
that enable interaction among the players. Rules include the protocols of communication among
the players. The other primitive is information about the environment, in which a game is
embedded. These primitives of game theory capture the strategic situations of conflict or
cooperation. Given the rules and the environment, joint action of players lead to an outcome,
and each player has a preference over outcomes. Each player is strategic—chooses the action
that leads to her preferred outcome, knowing that other players are doing likewise. Each player
has partial control over the outcomes. In game theory, we study solution concepts—given the
primitives, which outcomes are reasonable? Traditionally, game theory is divided into two
strands—noncooperative games and cooperative games 3 . In noncooperative games, modeling
unit is a single player—each player chooses noncooperatively. In noncooperative games, solution
concepts are called equilibrium outcomes. Nash equilibrium and its refinement are examples of
1
Wootton (2016) is an excellent introduction to the philosophy of science.
2
In 1994, to mark 50 years of publication of this book, Nobel Prize Committee decided to award the Nobel
Prize for Economic Sciences to game theory. John Nash, John Harsanyi, and Reinhard Selten shared the Nobel
Prize in 1994. We discuss the contributions of these laureates in the subsequent chapters.
3
This nomenclature is misleading, as cooperation is possible in noncooperative games like in repeated games
(discussed in Chapter 8) . Cooperative games are also called coalitional games.

3
4 CHAPTER 1. INTRODUCTION

solution concepts for noncooperative games. Communication among the players is modeled in
noncooperative games. In cooperative games, modeling unit is a coalition of players—what
a coalition of players can achieve? In a cooperative game, communication among the players
is exogenous to the game. The examples of solution concepts for cooperative games are the
Shapley value—axiomatic fair division of the payo↵ to a coalition; and, the core—how to divide
the payo↵ to a coalition so that it remains stable?
Noncooperative games are represented in normal-form (or strategic-form) form and extensive-
form, discussed in Chapters 2 and 4, respectively. The most common representation of coop-
erative games is characteristic function form, discussed in Chapter 6. Richer representations
of both nonoperative and cooperative games are discussed in Chapter 8. Let us consider an
example of a noncooperative game.

Example 1.1. There are two players—1 and 2. Player 1 and 2 are endowed with action sets
{T, B} and {L, R}, respectively. Each player is self-interested and rational, and knows that
the other player is also rational and self-interested. The combinations of their actions lead to
outcomes, as shown in the matrix below (these games are called bi-matrix games). In each
box of the matrix, the first number is the payo↵ to player 1 and second number is the payo↵
to player 2. The payo↵ matrix is known to both players (a game of complete information).
Each player has a preference over the outcome, however each player has partial control over
the outcomes. The rules mandate that each player should choose an action from the action set
simultaneously, and there is no communication between the players. In this game setting, what
are reasonable outcomes?

Player 2
L R
T 10, 10 0, 11
Player 1
B 11, 0 3, 3

Player 1’s preference for outcomes are (11, 0) �1 (10, 10) �1 (3, 3) �1 (0, 11). Similarly, player
2’s preferences are (0, 11) �2 (10, 10) �2 (3, 3) �2 (11, 0). You can observe that the preferences
are conflicting. The best outcome for player 1 is the worst outcome for player 2.
If player 1 tries to achieve outcome (11, 0) by playing B, player 2 responds by playing R. If
player 1 tries to achieve her second best outcome (10, 10), player 2 again responds by playing
R. Conjecturing player 2’s responses, player 1 prefers to choose B. Similarly, conjecturing
player 1’s responses, player 2 prefers to choose R. Hence, the reasonable outcome is (3, 3). This
outcome is self enforcing, and is not the best outcome for any player. This is also the Nash
equilibrium4 of this game. If the players can coordinate their choices, they can achieve (10, 10),
which is a preferred outcome over (3, 3) for both players. ◁

1.1 Preferences and Utility

In the preceding discussion, we mention that “each player is self-interested and rational”. What
is self-interest? Who is rational? In this section, we define these notions. Preferences over
4
Desiderata of Nash equilibrium is discussed in Chapter 2.
1.2. REPRESENTATIONS OF GAMES 5

the set of outcomes are fundamental in defining these notions. Self-interest means that these
preferences are derived from the player’s attitude towards the outcomes. This does not preclude
a possibility when a player prefers an outcome where other players are benefiting.
In game theory, players often decide under uncertainty. As in Example 1.1, player 1 makes
her choice when she has uncertainty about the choice of player 2, and vice versa. Furthermore,
Rationality
in some cases, a player is uncertain about the environment (games of incomplete information);
or, may be imperfectly aware of what had transpired earlier in the game (games of imperfect
information). von Neumann and Morgenstern (1944) propose an axiomatic theory of decision
making under uncertainty. They represent the uncertainty in the form of lottery—list of out-
comes and their associated probability. Each agent has a preference over the set of lotteries.
von Neumann and Morgenstern (1944) assume that the preferences satisfy a set of axioms to
represent preferences in the form of a utility function; and the agents behave as if they are
maximizing the expected value of the utility function. The “expected utility theory” is the
workhorse of decision making under uncertainty, and the axioms describe rational behavior.
We are not listing the axioms (we refer Kreps (1988) for an excellent introduction to this topic);
however, the essence of the axioms is that a rational agent is intelligent enough to rank all the
outcomes, forms expectations when deciding under uncertainty, and is consistent. In this book,
we assume that the players are rational, as per the expected utility theory.
The axioms of von Neumann-Morgenstern expected utility theory have been tested for de-
scribing the observed behavior of the decision makers under uncertainty. Since Allais (1953),
many papers have reported violations of the axioms. Various behavioral theories that are better
equipped to describe the observed behavior like the “prospect theory” (Kahneman and Tver-
sky, 1979) and the “rank-dependent utility theory” (Quiggin, 1993) have been proposed in the
literature. Gilboa (2009) provides a good introduction to the theories of decision making under
uncertainty, and Camerer (2003) integrates behavioral theories and game theory.

1.2 Representations of Games

In this section, we study di↵erent representations of noncooperative and cooperative games.

1.2.1 Normal-Form (or Strategic-Form)

Normal-form (or strategic-form) is a representation of noncooperative games. In this repre-

sentation, each strategic player chooses her plan of action independently, and without any
communication with other players. The outcome is uncertain as it depends on the joint actions
of all the players.

Definition 1.1 (Normal-form game). A normal-form game is a tuple �N, (Ai ), (ui )�. N is
a finite set of n players, indexed by i. Ai is an action set of player i. The combination of
actions chosen by each player gives the set of action profiles A = A1 × � × An . Each action
profile gives a unique outcome. Each player has a utility function ui ∶ A → R, in the spirit of
von Neumann-Morgenstern theory, which captures the preferences of player i over the set of
outcomes. △
6 CHAPTER 1. INTRODUCTION

In this book, we use this representation to model simultaneous-move game. Example 1.1 is a
normal-form game. Chapter 2 discusses normal-form games, with applications of normal-form
games in operations management are discussed in Chapter 3. This representation is generic
enough to model sequential-move games using “induced normal-form”; however, extensive-form
games are more efficient way to model games with sequential moves.
This representation can also model strategic situations when the players have incomplete in-
formation about the environment—“games of incomplete information” that are usually modeled
as Bayesian games, discussed in Chapter 8. Simultaneous auction, like first-price sealed-bid auc-
tion, is an example of a game of incomplete information where each player know her valuation
for the auctioned object, but is not aware of the valuations of the other players.

1.2.2 Extensive-Form
Extensive-form game is another representation of noncooperative games that can model sequen-
tial moves of players in a game, as a game tree.

Example 1.2. Assume in Example 1.1, if the rules are amended such that player 1 chooses from
the action set {T, B}, then player 2 makes the choice from the action set {L, R}, and the choice
of player 1 is known to player 2. This strategic situation can be shown as an extensive-form
game, in the form of a game tree (Figure 1.1).

1
T B
2 2

L R L R

(10, 10) (0, 11) (11, 0) (3, 3)

Figure 1.1: The Game Tree

◁

Definition 1.2 (Extensive-form game). An extensive-form game has N —a finite set of n play-
ers, indexed by i, and a game tree with the following structure:

• It has a set of nodes x ∈ X and a binary relation � called precedence. x � x′ means x

precedes x′ in a game tree. Each node in X has only one predecessor. � is transitive
(x � x′ , x′ � x′′ ⇒ x � x′′ ), asymmetric (x � x′ ⇒ x′ � x), and incomplete (not every pair
of nodes x, y can be ordered).

• The set of nodes with no predecessors are called initial nodes. The set of nodes that do
not precede other nodes are called terminal nodes, denoted as Z ⊂ X.

• Every node x ∈ X � Z is assigned to a player i ∈ N , or to nature5 . Ai (x) is the action set

5
In game theory, nature is a non-strategic player that captures exogenous uncertainty—uncertainty that does
not depend on the choices of strategic players.
1.2. REPRESENTATIONS OF GAMES 7

of player i, if node x is assigned to player i. A probability distribution is known for the

moves of nature.

• Each player has a utility function ui ∶ Z → R. ui (z) is the payo↵ to player i when the
game terminates at z ∈ Z. △

In extensive-form game with perfect information, a player knows her position perfectly in
the game tree. In extensive-form game with imperfect information, a player is imperfectly aware
of her position in the game tree. In Example 1.2, if the choice of player 1 is not revealed to
player 2; then player 2, when making his choice, is not sure whether he is on left node or right
mode in the game tree in Figure 1.1. Information sets capture knowledge of a player about her
position in the game tree.

Definition 1.3 (Information Sets). The decision nodes in X are partitioned into information
sets, such that the following conditions are satisfied for any x and x′ in the same information
set:

• x � x and x′ � x.

• The same player i is assigned to x and x′ .

• Ai (x) = Ai (x′ ). △

It means that the player cannot distinguish among the nodes in her information set. For a
game of perfect information, each information set is a singleton. In a game tree, the nodes in
the same information set are joined by a dashed line. The modified game tree of Example 1.2
when player 2 cannot distinguish between nodes x and x′ in the same information set, is shown
in Figure 1.2.

1
T B
x 2 x′

L R L R

(10, 10) (0, 11) (11, 0) (3, 3)

Figure 1.2: Game of Imperfect Information

In extensive-form games, the players are intelligent enough to make assessment of what
had transpired earlier in the game tree; and, using the assessment and conjecturing what will
transpire at the subsequent decision nodes, choose the best action at each information set.
We use extensive-form representation to model strategic situations with sequential-moves in
Chapter 4. The examples of such strategic situations include bargaining, sequential auctions,
and signaling games. This representation can be enriched to model repeated games—when
players play a game repeatedly—discussed in Chapter 8.