THE UNIVERSITY OF SYDNEY

SCHOOL OF ECONOMICS
Abhijit Sengupta

1.

ECOS3012: Strategic Behaviour

Handout 1

Semester 2, 2015

Describing Games: Extensive and Strategic Forms

As an informal description, a group of agents may be considered players in a game if

the outcome from an action by an agent depends on what other agents might do, and
if all agents are sophisticated enough to recognize that interdependence.
In such a setting, an agents best course of action typically depends on what actions the other agents take.
This is of course a characteristic of games like chess or tennis. An observation with far-reaching implication
is that this is also a characteristic of most economic and business interactionsindeed, of most social and
many biological interactions.
To see some immediate examples, we need only note that how successful a particular choice of action
would be for a big bank (e.g., increasing mortgage interest rate), or a car manufacturer (e.g., introducing a
sportier model), or a political party (e.g., adopting a particular policy position), or an animal seeking food
(e.g., aggressive posturing) almost always depends (in part) on what its competitors might do; moreover, the
relevant players all recognize that fact. Thus, competition among banks, car manufacturers, political parties,
and animals can be viewed as examples of games. This is why game theory has emerged as a fundamental way
of thinking about social and biological interactionsin economics and business, political science, psychology,
sociology, philosophy, and evolutionary biology. Although in this course we will be mostly concerned with
the application of game-theoretic ideas to economic interactions, we will look at a few examples from other
fields.
Our first task in studying game theory is to find a systematic way of representing these informal descriptions. The goal, then, is to build a theory that will allow us to predict what would happen when any
particular game is played.
There are three main ways of describing a game that have been found useful: (a) the extensive form, (b)
the strategic form (also called the normal form), and (c) the coalitional form.

1.1.

The Extensive Form

Definition 1. [This is, in fact, not a precise definition, but an informal description that will be adequate
for our purposes. A precise definition can be found on page 227 of MWG; however, you are not responsible
for that definition in this course.]
A game in extensive form consists of the following specifications:
(a) the set of players, N = {1, . . . , n},
(b) the order in which the players take actions (equivalently, move),
(c) the set of actions (also called choices) that are available to each player whenever it is the players turn
to move,
(d) the information available to a player whenever it is the players turn to move,
(e) for each player i in N , the players payoff, ui , as a function of the profile of actions taken by all of the
players; the objective of each player is to maximize her payoff,
(f) actions, if any, by chancerepresenting possible events outside the control of any of the playersand
their probabilities.
A game in extensive form is represented as a game tree. Lets look at some simple examples.
Example 1. Suppose that the market for a certain specialized instrument has only one manufacturer and
one corporate buyer. The manufacturer first chooses whether to produce one high quality unit (call it action
h) or one low quality unit (action l): the respective costs of production are ch and cl . The buyer then must

ECOS3012: Strategic Behaviour

Handout 1

Semester 2, 2015

decide whether to buy the unit (call it action b) at a previously fixed price p or not (action n) without
observing the quality of the product. The product would have a value of vh or vl dollars to the buyer
depending on whether it is of high or low quality. The payoff of the manufacturer is simply its profits; the
payoff of the buyer is the difference between her valuation and the price she has to pay if she purchases the
product, and zero, if she does not purchase the product. Assume that
vh > vl > p > ch > cl > 0.
The above description can be represented as a game in extensive form as follows:
1
h

2
b

2
n

b
b

p ch , vh p

ch , 0

p cl , vl p

cl , 0

A game tree is shown as a collection of ordered nodes. The unique node of a game tree which has no
predecessor is called the initial node; every other node has a unique immediate predecessor. The nodes with
no successors are called the terminal nodes; payoffs are distributed at the terminal nodes. We write the
payoffs as a vector: the i-th component is the payoff of player i. All nonterminal nodes are called decision
nodes.
Each decision node is associated with a unique player (shown circled), the player who takes an action
at that node. The set of actions available to a player at a decision node of the player are represented by
branches connecting the node to an immediate successor (h and l for player 1).
The set of all decision nodes associated with each player is partitioned into information sets of the player.
An information set of a player represents the collection of decision nodes that the player cannot distinguish
between. An information set of a player containing multiple decision nodes is indicated by a dashed line
connecting the nodes (or as an ellipse containing the nodes).
In Example 1, each player has one information set; player 1s information set is a singleton set (that is,
it has only one decision node), 2s information set has two decision nodes. It indicates that player 2, when
choosing between the actions b and n, does not know whether player 1 chose the action h or the action l.
Observation 1. At different decision nodes within a given information set of a player, the set of actions
available to the player and the players choice of action must be precisely the same (since the player cannot
tell the nodes apart).
The structure of the extensive form will always be assumed to be common knowledge: that is, every
player knows it, everyone knows that everyone knows it, everyone knows that everyone knows that everyone
knows it, . . .
As we shall see later in the course, situations involving incomplete informationi.e., situations in which
a player may not know some relevant attribute of another playersurprisingly, can still be handled within
the common knowledge framework.
Further, in this course, we will also assume that each player has perfect recall of past actions that the
player herself has taken.

Page 2 of 7

ECOS3012: Strategic Behaviour

Handout 1

Semester 2, 2015

Example 2. Let us now vary Example 1 to suppose that the buyer, before making her decision, can observe
the quality of the product.
1
h

2.1
b

2.2
n

b
b

p ch , vh p

ch , 0

p cl , vl p

cl , 0

Although player 1 still has one information set, player 2 now has two information sets, shown as 2.1 and 2.2.
Example 3. Each of two firms must plan whether to introduce (action I for player 1, action i for 2) a new
computer game next year or not (action N for 1, n for 2). It turns out that the products are very similar, and
each firm will have to make the decision without knowing the decision of the competitor. Also, at this time,
neither firm knows the size of demand that the product will have: depending on factors out of their control,
the demand might be large (L) or small (S), each with probability 1/2. However, firm 2 has commissioned
a market survey and will know the size of demand before putting the product on market.
Suppose that a firms profit is zero if it decides not to introduce the product. The profit from being the
only firm with the product is (in millions of dollars) 8 if demand is large, and 2 if it is small. The profit to
each if both introduce the product is 4 if demand is large and -1 if it is small.
We represent the exogenous uncertainty as a move of chance (shown as player 0).
0
1
2

1
2

2.1
i

2.1
n

2.2
n

2.2
n

b
b

b
b

b
b

4, 4

8, 0

0, 8

0, 0

1, 1

2, 0

0, 2

0, 0

Here, player 1 has one information set while 2 has two information sets, 2.1 and 2.2.
The same description can also be equivalently represented by different game trees: further discussion in
class.
Example 4. Now lets vary the story of Example 3. Suppose neither firm will find out the size of demand
for their products before they launch them; but firm 1 moves first, and firm 2 knows 1s choice before it
makes its own decision. Everything else is as before. The corresponding extensive form then is:

Page 3 of 7

ECOS3012: Strategic Behaviour

Handout 1

Semester 2, 2015

0
1
2

1
2

2.1

2.2

2.1

2.2

b
b

b
b

b
b

4, 4

8, 0

0, 8

0, 0

1, 1

2, 0

0, 2

0, 0

In Example 4, again, 1 has one information set while 2 has two information sets.
Example 5. Of course, a game may have more than two players or a given player may take action more
than once in the game, as in the next example.
1.1
a

2.1

2.2

3.1
h

1.2
i

1.3

k l

1.3

n o

3.2

n o

b
b

b
b

b
b

b
b

5
1
1

2
3
2

6
2
0

0
0
4

1
4
3

4
5
5

3
1
7

1
4
0

0
2
0

2
2
4

5
3
1

3
7
2

1
0
0

Definition 2. A game in extensive form is said to be a game of perfect information if every information set
of each player consists of a single decision node; otherwise, it is a game of imperfect information. It is said
to be a simultaneous-move game if all players move only once and choose their actions without observing
the choices of actions by the other players.
Example 2 is a game of perfect information; the other examples so far are games of imperfect information;
moreover, Example 1 is a simultaneous-move game.

Page 4 of 7

ECOS3012: Strategic Behaviour

1.2.

Handout 1

Semester 2, 2015

The Notion of Strategies

A central concept in game theory is the notion of a strategy.

Definition 3. A (pure) strategy of a player specifies a choice of an action by the player at each information
set of the player. Thus, a strategy of a player is a complete plan, formulated at the beginning of the game,
about how the player would play the game in every possible contingency.
In Example 1, player 1 has 2 strategies: h and l. Player 2 has 2 strategies: b and n.
However, in Example 2, although player 1 still has 2 strategies, h and l, player 2 has 4 strategies:
choose b if 1 chose h and choose b if 1 chose l :

in notation, bb

choose b if 1 chose h and choose n if 1 chose l :

choose n if 1 chose h and choose b if 1 chose l :

in notation, bn
in notation, nb

choose n if 1 chose h and choose n if 1 chose l :

in notation, nn.

Notation. We first number every information set of each player: this can be done in arbitrary order. Once
the information sets are numbered, we follow the convention that the k-th component of a players strategy
represents the action chosen at the k-th numbered information set of the player. Thus, the strategy bn of
player 2 means 2 chooses b at information set 2.1 and n at information set 2.2.
Observation 2. Note the very important distinction between actions and strategies. Suppose player i in
an extensive form game has l information sets: i.1, . . . , i.l and that she has a choice of mk actions at her
k-th information set. Then, player i has m1 m2 ml strategies.
If an agent has only one information set, strategies and actions are synonymous for that player. In
particular, for a simultaneous-move game, the notions of strategies and actions coincide for each player.

1.3.

Games in Strategic Form

A second way to represent a situation as a game, more concise than the extensive form, is to write down
the strategies of the players and the corresponding payoffs.
Definition 4. A game in strategic form (also called the normal form) consists of the following specifications:
(a) the set of players, N = {1, . . . , n},
(b) for each player i in N , the set of strategies, Si , available to the player,
(c) for each strategy profile, s = (s1 , . . . , sn ), where si Si , the payoff of player i, ui (s1 , . . . , sn ).
Example 1 in strategic form is given by:
N = {1, 2},
u1 (h, b) = p ch ,
u1 (l, b) = p cl ,

S1 = {h, l},

u2 (h, b) = vh p,
u2 (l, b) = vl p,

S2 = {b, n},
u1 (h, n) = ch ,
u1 (l, n) = cl ,

u2 (h, n) = 0,
u2 (l, n) = 0.

The payoffs in the above game can be more compactly represented by the following bi-matrix form:

Player 1 h
l

Player 2
b
n
p ch , vh p
ch , 0
p cl , vl p
cl , 0

Page 5 of 7

ECOS3012: Strategic Behaviour

Handout 1

Semester 2, 2015

The strategic form of Example 2 is given by:

N = {1, 2},

Player 1 h
l

S1 = {h, l},

S2 = {bb, bn, nb, nn},

Player 2
bn
nb
p ch , vh p
ch , 0
cl , 0
p cl , vl p

bb
p ch , vh p
p cl , vl p

nn
ch , 0
cl , 0

The set of players and the set of strategies for Example 3 are:
N = {1, 2},

S1 = {I, N },

S2 = {ii, in, ni, nn}.

[Aside: If you find it intuitively puzzling that player 2 has 4 strategies, recall that strategies are complete
contingent plans drawn up at the beginning of the game.]
However, because of the chance move, here the relevant payoffs are expected payoffs. Thus,
u1 (I, in) =

1
1
(4) + (2) = 3,
2
2

u2 (I, in) =

1
1
(4) + (0) = 2.
2
2

You should verify that the strategic form for Example 3 is given by
Player 2
Player 1 I
N

ii
3/2, 3/2
0, 5

in
3, 2
0, 4

ni
7/2, 1/2
0, 1

nn
5, 0
0, 0

Observation 3. To every game in extensive form, there corresponds a unique game in strategic form.
However, different extensive forms may generate the same strategic form. For instance, all of the following
three extensive games [we will later study this example and explain the symbols E, N,. . . ]
1

1
N

E
2

2
b

0, 4
A

b
b

b
b

b
b

b
b

2, 2

1, 1

2
2

1
1

0
4

0
4

2
2

0
4

1
1

0
4

lead to the same strategic form:

N = {1, 2},

S1 = {E, N },

Player 1 E
N

S2 = {A, F },

Player 2
A
F
2, 2
1, 1
0, 4
0, 4

Page 6 of 7

ECOS3012: Strategic Behaviour

Handout 1

Semester 2, 2015

Example 6. Games with three or more players can also be represented compactly in strategic form. The
following is a representation of a game with
N = {1, 2, 3},

Player 1 T
B

S1 = {T, B},

S2 = {L, R},

Player 2
L
R
7, 8, 2 5, 9, 1
8, 5, 1 2, 4, 0

Player 1 T
B

S3 = {l, r}.
Player 2
L
R
5, 5, 3 3, 9, 2
4, 8, 4 2, 2, 1
r

Now that we have seen how to formally represent various scenarios as games, we are ready to pursue the
main goal of game theory: to formulate principles for predicting the outcome of games played by intelligent
and experienced players. That is, we seek a theory that will allow us to predict, for any given game, what
strategy each player will choose.

Page 7 of 7