3.

Perfect-information Games

3.1 Trees, frames and games

Often interactions are not simultaneous but sequential. For example, in the game of
Chess the two players, White and Black, take turns moving pieces on the board, having
full knowledge of the opponent’s (and their own) past moves. Games with sequential
interaction are called dynamic games or games in extensive form. This chapter is devoted to
the subclass of dynamic games characterized by perfect information, namely the property
that, whenever it is her turn to move, a player knows all the preceding moves.

Perfect-information games are represented by means of rooted directed trees.

Definition 3.1.1 A rooted directed tree consists of a set of nodes and directed edges
joining them.
• The root of the tree has no directed edges leading to it (has indegree 0), while
every other node has exactly one directed edge leading to it (has indegree 1).
• There is a unique path (that is, a unique sequence of directed edges) leading
from the root to any other node. A node that has no directed edges out of it (has
outdegree 0) is called a terminal node, while every other node is called a decision
node.
• We shall denote the set of nodes by X, the set of decision nodes by D and the set
of terminal nodes by Z. Thus X = D ∪ Z.
76 Chapter 3. Perfect-information Games
Definition 3.1.2 A finite extensive form (or frame) with perfect information consists of
the following items.
• A finite rooted directed tree.
• A set of players I = {1, . . . , n} and a function that assigns one player to every
decision node.
• A set of actions A and a function that assigns one action to every directed edge,
satisfying the restriction that no two edges out of the same node are assigned the
same action.
• A set of outcomes O and a function that assigns an outcome to every terminal
node.
 Example 3.1 Amy (Player 1) and Beth (Player 2) have decided to dissolve a business
partnership whose assets have been valued at $100,000. The charter of the partnership
prescribes that the senior partner, Amy, make an offer concerning the division of the assets
to the junior partner, Beth. The junior partner can Accept, in which case the proposed
division is implemented, or Reject, in which case the case goes to litigation.
- Litigating involves a cost of $20,000 in legal fees for each partner and the typical verdict
assigns 60% of the assets to the senior partner and the remaining 40% to the junior
partner.
- Suppose, for simplicity, that there is no uncertainty about the verdict (how to model
uncertainty will be discussed in a later chapter). Suppose also that there are only two
possible offers that Amy can make: a 50-50 split or a 70-30 split.
This situation can be represented as a finite extensive form with perfect information as
shown in Figure 3.1. Each outcome is represented as two sums of money: the top one is
what Player 1 gets and the bottom one what Player 2 gets. 

1
Offer Offer
50-50 70-30

2 2

Accept Reject Accept Reject

o o o o
1 2 3 4
$50,000 $40,000 $70,000 $40,000
$50,000 $20,000 $30,000 $20,000

Figure 3.1: A perfect-information extensive form representing the situation described in

Example 3.1
3.1 Trees, frames and games 77

What should we expect the players to do in the game of Figure 3.1? Consider the
following reasoning, which is called backward induction reasoning, because it starts from
the end of the game and proceeds backwards towards the root:
- If Player 2 (the junior partner) is offered a 50-50 split then, if she accepts, she will
get $50,000, while, if she rejects, she will get $20,000 (the court-assigned 40%
minus legal fees in the amount of $20,000); thus, if rational, she will accept.
- Similarly, if Player 2 is offered a 70-30 split then, if she accepts, she will get $30,000,
while, if she rejects, she will get $20,000 (the court-assigned 40% minus legal fees
in the amount of $20,000); thus, if rational, she will accept.
- Anticipating all of this, Player 1 realizes that, if she offers a 50-50 split then she
will end up with $50,000, while if she offers a 70-30 split then she will end up with
$70,000; thus, if Player 1 is rational and believes that Player 2 is rational, she will
offer a 70-30 split and Player 2, being rational, will accept.

The above reasoning suffers from the same flaw as the reasoning described in Chapter
2: it is not a valid argument because it is based on an implicit assumption about how Player
2 ranks the outcomes, which may or may not be correct. For example, Player 2 may feel
that she worked as hard as her senior partner and the only fair division is a 50-50 split;
indeed she may feel so strongly about this that – if offered an unfair 70-30 split – she
would be willing to sacrifice $10,000 in order to “teach a lesson to Player 1”; in other
words, she ranks outcome o4 above outcome o3 .

Using the terminology introduced in Chapter 2, we say that the situation represented in
Figure 3.1 is not a game but a game-frame. In order to convert that frame into a game we
need to add a ranking of the outcomes for each player.

Definition 3.1.3 A finite extensive game with perfect information is a finite extensive
form with perfect information together with a ranking %i of the set of outcomes O, for
every player i ∈ I.
78 Chapter 3. Perfect-information Games

As usual, it is convenient to represent the ranking of Player i by means of an ordinal

utility function Ui : O → R. For example, take the extensive form of Figure 3.1 and assume
that Player 1 is selfish and greedy, that is, her ranking is:

best o3
o1 (or, in the alternative notation, o3 1 o1 1 o2 ∼1 o4 ).
worst o2 , o4

while Player 2 is concerned with fairness and her ranking is:

best o1
o2 , o4 (or, in the alternative notation, o1 2 o2 ∼2 o4 2 o3 )
worst o3

Then we can represent the players’ preferences using the following utility functions:

outcome → o1 o2 o3 o4
utility function ↓

U1 (Player 1) 2 1 3 1
U2 (Player 2) 3 2 1 2

and replace each outcome in Figure 3.1 with a pair of utilities or payoffs, as shown in
Figure 3.2, thereby obtaining one of the many possible games based on the frame of Figure
3.1.

Now that we have a game (rather than just a game-frame), we can apply the backward-
induction reasoning and conclude that Player 1 will offer a 50-50 split, anticipating that
Player 2 would reject the offer of a 70-30 split, and Player 2 will accept Player 1’s 50-50
offer. The choices selected by the backward-induction reasoning have been highlighted in
Figure 3.2 by doubling the corresponding edges.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 3.6.1 at the end of this chapter.
3.1 Trees, frames and games 79

1
Offer Offer
50-50 70-30

2 2

Accept Reject Accept Reject

2 1 3 1
3 2 1 2

Figure 3.2: A perfect-information game based on the frame of Figure 3.1

 80 Chapter 3. Perfect-information Games

3.2 Backward induction

The backward-induction reasoning mentioned above can be formalized as an algorithm for
solving any finite perfect-information game. We say that a node is marked if a utility vector
is associated with it. Initially all and only the terminal nodes are marked; the following
procedure provides a way of marking all the nodes.

Definition 3.2.1 The backward-induction algorithm is the following procedure for

solving a finite perfect-information game:
1. Select a decision node x whose immediate successors are all marked. Let i be
the player who moves at x. Select a choice that leads to an immediate successor
of x with the highest payoff (or utility) for Player i (highest among the utilities
associated with the immediate successors of x). Mark x with the payoff vector
associated with the node that follows the selected choice.
2. Repeat the above step until all the nodes have been marked.

Note that, since the game is finite, the above procedure is well defined. In the initial
steps one starts at those decision nodes that are followed only by terminal nodes, call
them penultimate nodes. After all the penultimate nodes have been marked, there will
be unmarked nodes whose immediate successors are all marked and thus the step can be
repeated. Note also that, in general, at a decision node there may be several choices that
maximize the payoff of the player who moves at that node. If that is the case, then the
procedure requires that one such choice be selected. This arbitrary selection may lead to
the existence of several backward-induction solutions.

1
a b
2 2
x y
c d e f

3
Player 1's payoff 2 0 3
g h
Player 2's payoff 1 0 1
Player 3's payoff 0 2 0

1 0
2 0
1 1

Figure 3.3: A perfect-information game with multiple backward-induction solutions.

3.2 Backward induction 81

For example, in the game of 3.3 starting at node x of Player 2 we select choice c (since
it gives Player 2 a higher payoff than d). Then we move on to Player 3’s node and we find
that bothTHEORY
GAME choices –there are Bonanno
Giacomo payoff maximizing for Player 3; thus there are two ways to
proceed, as shown in the next two figures.
In Figure 3.4 we show the steps of the backward-induction algorithm with the selection of
choice g, while Figure 3.5 shows the steps of the algorithm with the selection of choice
h. As before, the selected choices are shown by double edges. In Figures 3.4 and 3.5 the
marking of nodes is shown explicitly, but later on we will represent the backward-induction
solution more succinctly by merely highlighting the selected choices.

1 1
a b a b
2
1
2 2 2
1
2 2
0 x y 0 x y
c d e f c d e f
1
3 3 2
1
2 0 3 2 0 3
g h g h
1 0 1 1 0 1
0 2 0 0 2 0

1 0 1 0
2 0 2 0
1 1 1 1
STEP 1 STEP 2
2 2
1 1 1 1
0 0
a b a b
2 2 2 1 2
1
2 2 1
1 2 2
0 x y 1 0 x y 1

c d e f c d e f
1 1
3 2
1
3 2
1
2 0 3 2 0 3
g h g h
1 0 1 1 0 1
0 2 0 0 2 0

1 0 1 0
2 0 2 0
1 1 1 1
STEP 3 LAST STEP

Figure 3.4: One possible output of theFigure

backward-induction
2.4 algorithm applied to the game
of Figure 3.3 One possible output of the backward-induction
algorithm to the game of Figure 2.3.

6
 GAME THEORY – Giacomo Bonanno

82 Chapter 3. Perfect-information Games

1 1
a b a b
2
1
2 2 2
1
2 2
0 x y 0 x y
c d e f c d e f
0
3 3 0
1
2 0 3 2 0 3
g h g h
1 0 1 1 0 1
0 2 0 0 2 0

1 0 1 0
2 0 2 0
1 1 1 1
STEP 1 STEP 2

1 1 3
1
0
a b a b
2
1
2 2 3 2
1
2 2 3
1 1
0 x y 0 0 x y 0

c d e f c d e f
0 0
3 0
1
3 0
1
2 0 3 2 0 3
g h g h
1 0 1 1 0 1
0 2 0 0 2 0

1 0 1 0
2 0 2 0
1 1 1 1
STEP 3 LAST STEP

Figure 2.5
Figure 3.5: AnotherAnother
possible outputoutput
possible of theofbackward-induction algorithm applied to the
the backward-induction
game of Figure 3.3 algorithm to the game of Figure 2.3

HowHowshould
shouldone
onedefine
define the output
output ofofthe
thebackward-induction
backward-inductionalgorithm
algorithmand
and
thethe
notion
notion of backward-induction solution? What kind of objects are they? Before
of backward-induction solution? What kind of objects are they? Before we answer this we
answer we
question thisneed
question we need
to introduce theto introduce
notion the notion
of strategy of strategy in a perfect-
in a perfect-information game.
information game.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 3.6.2 at the end of this chapter.

7
 3.3 Strategies in perfect-information games 83

3.3 Strategies in perfect-information games

A strategy for a player in a perfect-information game is a complete, contingent plan on how
to play the game. Consider, for example, the game shown in Figure 3.6 (which reproduces
3.3) and let us focus on Player 2.

1
a b
2 2
x y
c d e f

3
Player 1's payoff 2 0 3
g h
Player 2's payoff 1 0 1
Player 3's payoff 0 2 0

1 0
2 0
1 1

Figure 3.6: Copy of the game of Figure 3.3

Before the game is played, Player 2 does not know what Player 1 will do and thus a
complete plan needs to specify what she will do if Player 1 decides to play a and what she
will do if Player 1 decides to play b. A possible plan, or strategy, is “if Player 1 chooses
a then I will choose c and if Player 1 chooses b then I will choose e”, which we can
denote more succinctly as (c, e). The other possible plans, or strategies, for Player 2 are
(c, f ), (d, e) and (d, f ). The formal definition of strategy is as follows.
Definition 3.3.1 A strategy for a player in a perfect-information game is a list of
choices, one for each decision node of that player.

For example, suppose that Player 1 has three decision nodes in a given game: at one node
she has three possible choices, a1 , a2 and a3 , at another node she has two possible choices,
b1 and b2 , and at the third node she has four possible choices, c1 , c2 , c3 and c4 . Then a
strategy for Player 1 in that game can be thought of as a way of filling in three blanks:
 
__________
| {z } , __________
| {z } , __________
| {z } .
one of a1 ,a2 ,a3 one of b1 ,b2 one of c1 ,c2 ,c3 ,c4

Since there are 3 choices for the first blank, 2 for the second and 4 for the third, the total
number of possible strategies for Player 1 in this case would be 3 × 2 × 4 = 24. One
strategy is (a2 , b1 , c1 ), another strategy is (a1 , b2 , c4 ), etc.
84 Chapter 3. Perfect-information Games

It should be noted that the notion of strategy involves redundancies. To see this,
consider the game of Figure 3.7. In this game a possible strategy for Player 1 is (a, g),
which means that Player 1 is planning to choose a at the root of the tree and would choose
g at her other node. But if Player 1 indeed chooses a, then her other node will not be
reached and thus why should Player 1 make a plan on what to do there? One could justify
this redundancy in the notion of strategy in a number of ways:
1. Player 1 is so cautious that she wants her plan to cover also the possibility that she
might make mistakes in the implementation of parts of her plan (in this case, she
allows for the possibility that – despite her intention to play a – she might end up
playing b), or
2. we can think of a strategy as a set of instructions given to a third party on how to
play the game on Player 1’s behalf, in which case Player 1 might indeed worry
about the possibility of mistakes in the implementation and thus want to cover all
contingencies.
An alternative justification relies on a different interpretation of the notion of strategy: not
as a plan of Player 1 but as a belief in the mind of Player 2 concerning what Player 1 would
do. For the moment we will set this issue aside and simply use the notion of strategy as
given in Definition 4.2.1.

1
a b
2 2
x y
c d e f

1
2 0 3
g h
1 0 1

1 1
2 0

Figure 3.7: A perfect-information game

Using Definition 4.2.1, one can associate with every perfect-information game a
strategic-form (or normal-form) game: a strategy profile determines a unique terminal node
that is reached if the players act according to that strategy profile and thus a unique vector
of payoffs. Figure 3.8 shows the strategic-form associated with the perfect-information
game of Figure 3.7, with the Nash equilibria highlighted.
3.3 Strategies in perfect-information games 85

Play er 2
ce cf de df
ag 2 1 2 1 0 0 0 0
ah 2 1 2 1 0 0 0 0
Play er 1
bg 3 1 1 2 3 1 1 2
bh 3 1 1 0 3 1 1 0

Figure 3.8: The strategic form of the perfect-information game of Figure 3.7 with the Nash
equilibria highlighted

Because of the redundancy discussed above, the strategic form also displays redundan-
cies: in thisTHEORY
GAME case the– top twoBonanno
Giacomo rows are identical.
Armed with the notion of strategy, we can now revisit the notion of backward-induction
solution. Figure 3.9 shows the two backward-induction solutions of the game of Figure
3.7.

1 1
a b a b
2 2 2 2
x y x y

c e f c d e f
d
1 1
2 0 3 2 0 3
g h g h
0 1 1 0 1
1

1 1
1 1
2 0
2 0
(a) (b)

Figure 2.8 now

Figure 3.9: The backward-induction Fig 9 of the game of Figure 3.7
solutions
The backward-induction solutions of the game of Figure 2.6.

It is clear from the definition of backward-induction algorithm (Definition 3.2.1) that

the procedure selects a choice at every decision node and thus yields a strategy profile for
the entire game: the backward-induction solution shown in Panel (a) of Figure 3.9 is the
strategy profile ((a, g), (c, f )), while the backward-induction solution shown in Panel (b)
is the strategy profile ((b, h), (c, e)). Both of them are Nash equilibria of the corresponding
strategic-form game, but not all the Nash equilibria correspond to backward-induction
solutions. The relationship between the two concepts is explained in the next section.
 86 Chapter 3. Perfect-information Games

R A backward-induction solution is a strategy profile. Since strategies contain a

description of what a player actually does and also of what the player would do in
circumstances that do not arise, one often draws a distinction between the backward-
induction solution and the backward-induction outcome which is defined as the
sequence of actual moves. For example, the backward-induction outcome associated
with the solution ((a, g), (c, f )) is the play ac with corresponding payoff (2, 1), while
the backward-induction outcome associated with the solution ((b, h), (c, e)) is the
play be with corresponding payoff (3, 1).

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 3.6.3 at the end of this chapter.

3.4 Relationship between backward induction and other solutions

If you have gone through the exercises for the previous three sections, you will have seen
that in all those games the backward-induction solutions are also Nash equilibria. This is
always true, as stated in the following theorem.

Theorem 3.4.1 Every backward-induction solution of a perfect-information game is a

Nash equilibrium of the associated strategic form.

In some games the set of backward-induction solutions coincides with the set of Nash
equilibria (see, for example, Exercise 3.9), but typically the set of Nash equilibria is larger
than (is a proper superset of) the set of backward-induction solutions (for example the
game of Figure 3.7 has two backward-induction solutions – shown in Figure 3.9 – but five
Nash equilibria, shown in Figure 3.8).
Nash equilibria that are not backward-induction solutions often involve incredible
threats. To see this, consider the following game.
An industry is currently a monopoly and the incumbent monopolist is making a profit of
$5 million. A potential entrant is considering whether or not to enter this industry.
- If she does not enter, she will make $1 million in an alternative investment.
- If she does enter, then the incumbent can either fight entry with a price war whose outcome
is that both firms make zero profits, or it can accommodate entry, by sharing the market
with the entrant, in which case both firms make a profit of $2 million.
This situation is illustrated in Figure 3.10 with the associated strategic form. Note that we
are assuming that each player is selfish and greedy, that is, cares only about its own profit
and prefers more money to less.
The backward-induction solution is (in, accommodate) and it is also a Nash equilibrium.
However, there is another Nash equilibrium, namely (out, fight). The latter should be
discarded as a “rational solution” because it involves an incredible threat on the part of the
incumbent, namely that it will fight entry if the potential entrant enters.
- It is true that, if the potential entrant believes the incumbent’s threat, then she is better off
staying out; however, she should ignore the incumbent’s threat because she should realize
that – when faced with the fait accompli of entry – the incumbent would not want to carry
out the threat.
3.4 Relationship between backward induction and other solutions 87

Potential Incumbent
entrant in
2
accommodate
2
out fight

1 0
5 0

Chain store
fight accommodate

Potential In 0,0 2,2

entrant
Out 1,5 1,5

Figure 3.10: The entry game

Reinhard Selten (who shared the 1994 Nobel Memorial prize in economics with two
other game theorists, John Harsanyi and John Nash) discussed a repeated version of the
above entry game, which has become known as Selten’s Chain Store Game. The story is
as follows:

- A chain store is a monopolist in an industry. It owns stores in m different towns

(m ≥ 2).

- In each town the chain store makes $5 million if left to enjoy its privileged position
undisturbed.

- In each town there is a businesswoman who could enter the industry in that town, but
earns $1 million if she chooses not to enter; if she decides to enter, then the monopolist
can either fight the entrant, leading to zero profits for both the chain store and the entrant
in that town, or it can accommodate entry and share the market with the entrant, in
which case both players make $2 million in that town.

Thus, in each town the interaction between the incumbent monopolist and the potential
entrant is as illustrated in Figure 3.10.
88 Chapter 3. Perfect-information Games

However, decisions are made sequentially, as follows:

At date t(t = 1, . . . , m) the businesswoman in town t decides whether or not to enter and if
she enters then the chain store decides whether or not to fight in that town.
What happens in town t at date t becomes known to everybody. Thus, for example, the
businesswoman in town 2 at date 2 knows what happened in town 1 at date 1 (either that
there was no entry or that entry was met with a fight or that entry was accommodated).
Intuition suggests that in this game the threat by the incumbent to fight early entrants might
be credible, for the following reason. The incumbent could tell Businesswoman 1 the
following:

“It is true that, if you enter and I fight, I will make zero profits, while by
accommodating your entry I would make $2 million and thus it would seem
that it cannot be in my interest to fight you. However, somebody else is
watching us, namely Businesswoman 2. If she sees that I have fought your
entry then she might fear that I would do the same with her and decide to
stay out, in which case in town 2, I would make $5 million, so that my total
profits in towns 1 and 2 would be (0 + 5) = $5 million. On the other hand, if I
accommodate your entry, then she will be encouraged to entry herself and I
will make $2 million in each town, for a total profit of $4 million. Hence, as
you can see, it is indeed in my interest to fight you and thus you should stay
out.”

Does the notion of backward induction capture this intuition? To check this, let us consider
the case where m = 2, so that the extensive game is not too large to draw. It is shown in
Figure 3.11, where at each terminal node the top number is the profit of the incumbent
monopolist (it is the sum of the profits in the two towns), the middle number is the profit
of Businesswoman 1 and the bottom number is the profit of Businesswoman 2. All profits
are expressed in millions of dollars. We assume that all the players are selfish and greedy,
so that we can take the profit of each player to be that player’s payoff. The backward-
induction solution is unique and is shown by the thick directed edges in Figure 3.11.
The corresponding outcome is that both businesswomen will enter and the incumbent
monopolist accommodates entry in both towns.
3.4 Relationship between backward induction and other solutions 89

$0 $2 $5
$0 $0 $0
$0 $2 $1

acquiesce
fight
$10m chain store
$1m buswmn 1
chain store
$1m buswmn2
in out

businesswoman 2
out
fight businesswoman 1
chain store businesswoman 2
in out
in
acquiesce

businesswoman 2 chain store

fight acquiesce
in out
chain store
$5 $7
fight acquiesce $1 $1
$0 $2

$2 $4 $7
$2 $2 $2
$0 $2 $1

Figure 3.11: Selten’s Chain-Store game

 90 Chapter 3. Perfect-information Games

Thus the backward-induction solution does not capture the “reputation” argument outlined
above. However, the backward-induction solution does seem to capture the notion of
rational behavior in this game. Indeed, Businesswoman 1 should reply to the incumbent
with the following counter-argument:

“Your reasoning is not valid. Whatever happens in town 1, it will be common

knowledge between you and Businesswoman 2 that your interaction in town
2 will be the last; in particular, nobody else will be watching and thus there
won’t be an issue of establishing a reputation in the eyes of another player.
Hence in town 2 it will be in your interest to accommodate entry, since in
essence you will be playing the one-shot entry game of Figure 3.10. Hence a
rational Businesswoman 2 will decide to enter in town 2 whatever happened in
town 1: what you do against me will have no influence on her decision. Thus
your “reputation” argument does not apply and it will in fact be in your interest
not to fight my entry: your choice will be between a profit of $(0 + 2) = $2
million, if you fight me, and a profit of $(2 + 2) = $4 million, if you don’t
fight me. Hence I will enter and you will not fight me.”

In order to capture the reputation argument described above we need to allow for some
uncertainty in the mind of some of the players, as we will show in a later chapter. In a
perfect-information game uncertainty is ruled out by definition.
By Theorem 3.4.1 the notion of backward induction can be seen as a refinement of
the notion of Nash equilibrium. Another solution concept that is related to backward
induction is the iterated elimination of weakly dominated strategies. Indeed the backward-
induction algorithm could be viewed as a step-wise procedure that eliminates dominated
choices at decision nodes, and thus strategies that contain those choices. What is the
relationship between the two notions? In general this is all that can be said: applying
the iterated deletion of weakly dominated strategies to the strategic form associated with
a perfect-information game leads to a set of strategy profiles that contains at least one
backward-induction solution; however,
(1) it may also contain strategy profiles that are not backward-induction solutions, and
(2) it may fail to contain all the backward-induction solutions, as shown in Exercise 3.8.

3.5 Perfect-information games with two players

We conclude this chapter with a discussion of finite two-player extensive games with
perfect information.
We will start with games that have only two outcomes, namely “Player 1 wins” (denoted
by W1 ) and “Player 2 wins” (denoted by W2 ). We assume that Player 1 strictly prefers
W1 to W2 and Player 2 strictly prefers W2 to W1 . Thus we can use utility functions with
values 0 and 1 and associate with each terminal node either the payoff vector (1, 0) (if the
outcome is W1 ) or the payoff vector (0, 1) (if the outcome is W2 ). We call these games
win-lose games. An example of such a game is the following.
3.5 Perfect-information games with two players 91

 Example 3.2 Two players take turns choosing a number from the set {1, 2, . . . , 10}, with
Player 1 moving first. The first player who brings the sum of all the chosen numbers to
100 or more wins. 

The following is one possible play of the game (the bold-face numbers are the ones
chosen by Player 1 and the underlined numbers the ones chosen by Player 2):

10, 9, 9, 10, 8, 7, 4, 10, 1, 8, 3, 3, 8, 10.

In this play Player 2 wins: before her last move the sum is 90 and with her final choice of
10 she brings the total to 100. However, in this game Player 1 has a winning strategy, that
is, a strategy that guarantees that he will win, no matter what numbers Player 2 chooses.
To see this, we can use backward-induction reasoning. Drawing the tree is not a practical
option, since the number of nodes is very large: one needs 10,000 nodes just to represent
the first 4 moves! But we can imagine drawing the tree, placing ourselves towards the
end of the tree and ask what partial sum represents a “losing position”, in the sense that
whoever is choosing in that position cannot win, while the other player can then win
with his subsequent choice. With some thought one can see that 89 is the largest losing
position: whoever moves there can take the sum to any number in the set {90, 91, . . . , 99},
thus coming short of 100, while the other player can then take the sum to 100 with an
appropriate choice. What is the largest losing position that precedes 89? The answer is 78:
whoever moves at 78 must take the sum to a number in the set {79, 80, . . . , 88} and then
from there the other player can make sure to take the sum to 89 and then we know what
happens from there! Repeating this reasoning we see that the losing positions are: 89, 78,
67, 56, 45, 34, 23, 12, 1. Since Player 1 moves first he can choose 1 and put Player 2 in
the first losing position; then, whatever Player 2 chooses, Player 1 can put her in the next
losing position, namely 12, etc. Recall that a strategy for Player 1 must specify what to do
in every possible situation in which he might find himself. In his game Player 1’s winning
strategy is as follows:

Start with the number 1. Then, at every turn, choose the number
(11 − n), where n is the number that was chosen by Player 2 in the
immediately preceding turn.

Here is an example of a possible play of the game where Player 1 employs the winning
strategy and does in fact win:

1, 9, 2, 6, 5, 7, 4, 10, 1, 8, 3, 3, 8, 9, 2, 5, 6, 1, 10

We can now state a general result about this class of games.

Theorem 3.5.1 In every finite two-player, win-lose game with perfect information one
of the two players has a winning strategy.
92 Chapter 3. Perfect-information Games

Although we will not give a detailed proof, the argument of the proof is rather simple.
By applying the backward-induction algorithm we assign to every decision node either the
payoff vector (1, 0) or the payoff vector (0, 1). Imagine applying the algorithm up to the
point where the immediate successors of the root have been assigned a payoff vector. Two
cases are possible.

Case 1: at least one of the immediate successors of the root has been assigned the payoff
vector (1, 0). In this case Player 1 is the one who has a winning strategy and his initial
choice should be such that a node with payoff vector (1, 0) is reached and then his future
choices should also be such that only nodes with payoff vector (1, 0) are reached.

Case 2: all the immediate successors of the root have been assigned the payoff vector
(0, 1). In this case it is Player 2 who has a winning strategy. An example of a game where
it is Player 2 who has a winning strategy is given in Exercise 3.11.

We now turn to finite two-player games where there are three possible outcomes:
“Player 1 wins” (W1 ), “Player 2 wins” (W2 ) and “Draw” (D). We assume that the rankings
of the outcomes are as follows: W1 1 D 1 W2 and W2 2 D 2 W1 .
Examples of such games are Tic-Tac-Toe (http://en.wikipedia.org/wiki/Tic-tac-toe),
Draughts or Checkers (http://en.wikipedia.org/wiki/Draughts) and Chess (al-
though there does not seem to be agreement as to whether the rules of Chess guarantee
that every possible play of the game is finite). What can we say about such games? The
answer is provided by the following theorem.

Theorem 3.5.2 Every finite two-player, perfect-information game with three outcomes:
Player 1 wins (W1 ), Player 2 wins (W2 ) and Draw (D), and preferences W1 1 D 1 W2
and W2 2 D 2 W1 , falls within one of the following three categories:
1. Player 1 has a strategy that guarantees outcome W1 .
2. Player 2 has a strategy that guarantees outcome W2 .
3. Player 1 has a strategy that guarantees that the outcome will be W1 or D and Player
2 has a strategy that guarantees that the outcome will be W2 or D, so that, if both
players employ these strategies, the outcome will be D.
3.5 Perfect-information games with two players 93

The logic of the proof is as follows. By applying the backward-induction algorithm we

assign to every decision node either the payoff vector (2, 0) (corresponding to outcome
W1 ) or the payoff vector (0, 2) (corresponding to outcome W2 ) or the payoff vector (1, 1)
(corresponding to outcome D). Imagine applying the algorithm up to the point where the
immediate successors of the root have been assigned a payoff vector. Three cases are
possible.

Case 1: at least one of the immediate successors of the root has been assigned the payoff
vector (2, 0); in this case Player 1 is the one who has a winning strategy.

Case 2: all the immediate successors of the root have been assigned the payoff vector
(0, 2); in this case it is Player 2 who has a winning strategy.

Case 3: there is at least one immediate successor of the root to which the payoff vector
(1, 1) has been assigned and all the other immediate successors of the root have been
assigned either (1, 1) or (0, 2). In this case we fall within the third category of Theorem
3.5.2.

Both Tic-Tac-Toe and Checkers fall within the third category (http://en.wikipedia.
org/wiki/Solved_game#Solved_games). As of the time of writing this book, it is not
known to which category the game of Chess belongs.

Test your understanding of the concepts introduced in this section, by

going through the exercises in Section 3.6.4 at the end of this chapter.
 94 Chapter 3. Perfect-information Games

3.6 Exercises
3.6.1 Exercises for Section 3.1: Trees, frames and games
The answers to the following exercises are in Section 4.7 at the end of this chapter.

Exercise 3.1 How could they do that! They abducted Speedy, your favorite tortoise!
They asked for $1,000 in unmarked bills and threatened to kill Speedy if you don’t pay.
Call the tortoise-napper Mr. T. Let the possible outcomes be as follows:

o1 : you don’t pay and speedy is released

o2 : you pay $ 1,000 and speedy is released
o3 : you don’t pay and speedy is killed
o4 : you pay $ 1,000 and speedy is killed

You are attached to Speedy and would be willing to pay $1,000 to get it back. However,
you also like your money and you prefer not to pay, conditional on each of the two
separate events “Speedy is released” and “Speedy is killed”. Thus your ranking of the
outcomes is o1 you o2 you o3 you o4 . On the other hand, you are not quite sure of
what Mr. T’s ranking is.
(a) Suppose first that Mr T has communicated that he wants you to go to Central Park
tomorrow at 10:00 a.m. and leave the money in a garbage can; he also said that,
two miles to the East and at the exact same time, he will decide whether or not to
free Speedy in front of the police station and then go and collect his money in
Central Park. What should you do?
(b) Suppose that Mr T is not as dumb as in part (a) and instead gives you the following
instructions: first you leave the money in a garbage can in Central Park and then
he will go there to collect the money. He also told you that if you left the money
there then he will free Speedy, otherwise he will kill it. Draw an extensive form
or frame to represent this situation.
(c) Now we want to construct a game based on the extensive form of part (b). For
this we need Mr T’s preferences. There are two types of criminals in Mr T’s line
of work: the professionals and the one-timers. Professionals are in the business
for the long term and thus, besides being greedy, worry about reputation; they
want it to be known that (1) every time they were paid they honored their promise
to free the hostage and (2) their threats are to be taken seriously: every time they
were not paid, the hostage was killed. The one-timers hit once and then they
disappear; they don’t try to establish a reputation and the only thing they worry
about, besides money, is not to be caught: whether or not they get paid, they
prefer to kill the hostage in order to eliminate any kind of evidence (DNA traces,
fingerprints, etc.). Construct two games based on the extensive form of part (b)
representing the two possible types of Mr T.

 3.6 Exercises 95

Exercise 3.2 A three-man board, composed of A, B, and C, has held hearings on a

personnel case involving an officer of the company. This officer was scheduled for
promotion but, prior to final action on his promotion, he made a decision that cost the
company a good deal of money. The question is whether he should be (1) promoted
anyway, (2) denied the promotion, or (3) fired. The board has discussed the matter at
length and is unable to reach unanimous agreement. In the course of the discussion it
has become clear to all three of them that their separate opinions are as follows:
• A considers the officer to have been a victim of bad luck, not bad judgment, and
wants to go ahead and promote him but, failing that, would keep him rather than
fire him.
• B considers the mistake serious enough to bar promotion altogether; he’d prefer
to keep the officer, denying promotion, but would rather fire than promote him.
• C thinks the man ought to be fired but, in terms of personnel policy and morale,
believes the man ought not to be kept unless he is promoted, i.e., that keeping an
officer who has been declared unfit for promotion is even worse than promoting
him.
PROMOTE KEEP FIRE

A: best middle worst

B: worst best middle

C: middle worst best

Assume that everyone’s preferences among the three outcomes are fully evident as a
result of the discussion. The three must proceed to a vote.
Consider the following voting procedure. First A proposes an action (either promote
or keep or fire). Then it is B’s turn. If B accepts A’s proposal, then this becomes the
final decision. If B disagrees with A’a proposal, then C makes the final decision (which
may be any of the three: promote, keep or fire). Represent this situation as an extensive
game with perfect information. (Use utility numbers from the set {1, 2, 3}.) 

3.6.2 Exercises for Section 3.2: Backward induction

The answers to the following exercises are in Section 4.7 at the end of this chapter.

Exercise 3.3 Apply the backward-induction algorithm to the two games of Exercise
3.1 Part (c). 

Exercise 3.4 Apply the backward-induction algorithm to the game of Exercise 3.2. 
 96 Chapter 3. Perfect-information Games

3.6.3 Exercises for Section 3.3: Strategies in perfect-information games

The answers to the following exercises are in Section 4.7 at the end of this chapter.

Exercise 3.5 Write the strategic form of the game of Figure 3.2, find all the Nash
equilibria and verify that the backward-induction solution is a Nash equilibrium. 

Exercise 3.6 Write the strategic form of the game of Figure 3.3, find all the Nash
equilibria and verify that the backward-induction solutions are Nash equilibria. 

Exercise 3.7 Consider the game of Exercise 3.2.

(a) Write down all the strategies of Player B.
(b) How many strategies does Player C have?


Exercise 3.8 Consider the perfect-information game shown in Figure 3.12.

(a) Find the backward-induction solutions.
(b) Write down all the strategies of Player 1.
(c) Write down all the strategies of Player 2.
(d) Write the strategic form associated with this game.
(e) Does Player 1 have a dominant strategy?
(f) Does Player 2 have a dominant strategy?
(g) Is there a dominant-strategy equilibrium?
(h) Does Player 1 have any dominated strategies?
(i) Does Player 2 have any dominated strategies?
(j) What do you get when you apply the iterative elimination of weakly dominated
strategies?
(k) What are the Nash equilibria?

3.6 Exercises 97

1
L R
2 2
a b c d e
1
2 4 2 3 W E
1 0 0 2

1 0
2 3

Figure 3.12: The perfect-information game for Exercise 3.8

Exercise 3.9 Consider an industry where there are two firms, a large firm, Firm 1, and
a small firm, Firm 2. The two firms produce identical products.
- Let x be the output of Firm 1 and y the output of Firm 2. Industry output is Q = x + y.
- The price P at which each unit of output can be sold is determined by the inverse
demand function P = 130 − 10Q. For example, if Firm 1 produces 4 units and Firm 2
produces 2 units, then industry output is 6 and each unit is sold for P = 130 − 60 = $70.
- For each firm the cost of producing q units of output is C(q) = 10q + 62.5.
- Each firm is only interested in its own profits.
- The profit of Firm 1 depends on both x and y and is given by

Π1 (x, y) = x [130 − 10(x + y)] − (10x + 62.5)

| {z } | {z }
revenue cost
and similarly the profit function of Firm 2 is given by

Π2 (x, y) = y [130 − 10(x + y)] − (10y + 62.5).

| {z } | {z }
revenue cost
- The two firms play the following sequential game. First Firm 1 chooses its own output
x and commits to it; then Firm 2, after having observed Firm 1’s output, chooses its own
output y; then the price is determined according to the demand function and the two
firms collect their own profits. In what follows assume, for simplicity, that x can only
be 6 or 6.5 units and y can only be 2.5 or 3 units.
(a) Represent this situation as an extensive game with perfect information.
(b) Solve the game using backward induction.
(c) Write the strategic form associated with the perfect-information game.
(d) Find the Nash equilibria of this game and verify that the backward-induction
solutions are Nash equilibria.

 98 Chapter 3. Perfect-information Games

Exercise 3.10 Consider the perfect-information game shown in Figure 3.13 where x is
an integer.
(a) For every value of x find the backward induction solution(s).
(b) Write the corresponding strategic-form and find all the Nash equilibria.


A B

2 2

C D E F

layer 1's payoff 2 1 1 x

layer 2's payoff 5 0 0 2

Figure 3.13: A perfect-information game

3.6.4 Exercises for Section 3.5: Two-player games

The answers to the following exercises are in Section 4.7 at the end of this chapter.

Exercise 3.11 Consider the following perfect-information game. Player 1 starts by

choosing a number from the set {1, 2, 3, 4, 5, 6, 7}, then Player 2 chooses a number from
this set, then Player 1 again, followed by Player 2, etc. The first player who brings the
cumulative sum of all the numbers chosen (up to and including the last one) to 48 or
more wins. By Theorem 3.5.1 one of the two players has a winning strategy. Find out
who that player is and fully describe the winning strategy. 
3.6 Exercises 99

Exercise 3.12 Consider Figure 3.14 and the following two-player, perfect-information
game. A coin is placed in the cell marked ‘START’ (cell A1). Player 1 moves first
and can move the coin one cell up (to A2) or one cell to the left (to B1) or one cell
diagonally in the left-up direction (to B2). Then Player 2 moves, according to the same
rules (e.g. if the coin is in cell B2 then the admissible moves are shown by the directed
edges). The players alternate moving the coin. Black cells are not accessible (so that,
for example, from A3 the coin can only be moved to A4 or B3 and from F3 it can only
be moved to G4, as shown by the directed edge). The player who manages to place the
coin in the cell marked ‘END’ wins.
(a) Represent this game by means of an extensive form with perfect information by
drawing the initial part of the tree that covers the first two moves (the first move
of Player 1 and the first move of Player 2).
(b) Suppose that the coin is currently in cell G4 and it is Player 1’s turn to move.
Show that Player 1 has a strategy that allows her to win the game starting from
cell G4. Describe the strategy in detail.
(c) Describe a play of the game (from cell A1) where Player 1 wins (describe it by
means of the sequence of cells visited by the coin).
(d) Describe a play of the game (from cell A1) where Player 2 wins (describe it by
means of the sequence of cells visited by the coin).
(e) Now go back to the beginning of the game. The coin is in cell A1 and player
1 has the first move. By Theorem 3.5.1 one of the two players has a winning
strategy. Find out who that player is and fully describe the winning strategy.


Figure 3.14: The coin game

100 Chapter 3. Perfect-information Games

Exercise 3.13 — ? ? ? Challenging Question ? ? ?. .

Two women, Anna and Bess, claim to be the legal owners of a diamond ring that -
each claims - has great sentimental value. Neither of them can produce evidence of
ownership and nobody else is staking a claim on the ring. Judge Sabio wants the ring to
go to the legal owner, but he does not know which of the two women is in fact the legal
owner. He decides to proceed as follows. First he announces a fine of $F > 0 and then
asks Anna and Bess to play the following game.
Move 1: Anna moves first. Either she gives up her claim to the ring (in which case
Bess gets the ring, the game ends and nobody pays the fine) or she asserts her claim, in
which case the game proceeds to Move 2.
Move 2: Bess either accepts Anna’s claim (in which case Anna gets the ring, the game
ends and nobody pays the fine) or challenges her claim. In the latter case, Bess must
put in a bid, call it B, and Anna must pay the fine of $F to Sabio. The game goes on to
Move 3.
Move 3: Anna now either matches Bess’s bid (in which case Anna gets the ring, Anna
pays $B to Sabio in addition to the fine that she already paid and Bess pays the fine of
$F to Sabio) or chooses not to match (in which case Bess gets the ring and pays her bid
of $B to Sabio and, furthermore, Sabio keeps the fine that Anna already paid).
Denote by CA the monetary equivalent of getting the ring for Anna (that is, getting the
ring is as good, in Anna’s mind, as getting $CA ) and CB the monetary equivalent of
getting the ring for Bess. Not getting the ring is considered by both to be as good as
getting zero dollars.
(a) Draw an extensive game with perfect information to represent the above situation,
assuming that there are only two possible bids: B1 and B2 . Write the payoffs to
Anna and Bess next to each terminal node.
(b) Find the backward-induction solution of the game you drew in part (a) for the
case where B1 > CA > CB > B2 > F > 0.
Now consider the general case where the bid B can be any non-negative number and
assume that both Anna and Bess are very wealthy. Assume also that CA ,CB and F are
positive numbers and that CA and CB are common knowledge between Anna and Bess .
We want to show that, at the backward-induction solution of the game, the ring always
goes to the legal owner. Since we (like Sabio) don’t know who the legal owner is, we
must consider two cases.
Case 1: the legal owner is Anna. Let us assume that this implies that CA > CB .
Case 2: the legal owner is Bess. Let us assume that this implies that CB > CA .
(c) Find the backward-induction solution for Case 1 and show that it implies that the
ring goes to Anna.
(d) Find the backward-induction solution for Case 2 and show that it implies that the
ring goes to Bess.
(e) How much money does Sabio make at the backward-induction solution? How
much money do Ann and Bess end up paying at the backward-induction solution?

 3.7 Solutions to exercises 101

3.7 Solutions to exercises

Solution to Exercise 3.1.
(a) For you it is a strictly dominant strategy to not pay and thus you should not pay.
(b) The extensive form is shown in Figure 3.15.
(c) For the professional, concern with reputation implies that o2 MrT o4 and o3 MrT o1 .
If we add the reasonable assumption that after all money is what they are after, then
we can take the full ranking to be o2 MrT o4 MrT o3 MrT o1 .
Representing preferences with ordinal utility functions with values in the set {1, 2, 3, 4},
we have
outcome → o1 o2 o3 o4
utility function ↓
Uyou 4 3 2 1
UMrT 1 4 2 3

The corresponding game is obtained by replacing in Figure 3.15 o1 with the payoff
vector (4,1), o3 with the payoff vector (2,2), etc.

For the one-timer, the ranking can be taken to be (although this is not the only
possibility) o4 MrT o2 MrT o3 MrT o1 , with corresponding utility representation:

outcome → o1 o2 o3 o4
utility function ↓
Uyou 4 3 2 1
UMrT 1 3 2 4
The corresponding extensive-form game is shown in Figure 3.16. 

You You
not pay not pay
pay pay

Mr T Mr T Mr T Mr T
release kill release kill release kill release kill

o1 o3 o2 o4 4 2 3 1
1 2 3 4
Figure 3.15: The game-
frame for Part (b) of Exer- Figure 3.16: The game for Part ((c) of
cise 3.1 Exercise 3.1 when Mr T is a one-timer
102 Chapter 3. Perfect-information Games

Solution to Exercise 3.2. The game is shown in Figure 3.17 (‘P’ stands for promote, ‘K’
for keep (without promoting), ‘F’ for fire). 

A
P F
B K B
accept
P accept
3 F 1
1 t B
c ep 2
2 reject ac reject reject 3
C K C C

F 2
P P F P F
K 3 K K
1

P K F P K F P K F
3 2 1 3 2 1 3 2 1
1 3 2 1 3 2 1 3 2
2 1 3 2 1 3 2 1 3

Figure 3.17: The game for Exercise 3.2

Solution to Exercise 3.3. The application of the backward-induction algorithm is shown

by double edges in Figure 3.18 for the case of a professional Mr. T and in Figure 3.19
for the case of a one-timer Mr. T. Thus, against a professional you will pay and against a
one-timer you would not pay. With the professional you would get Speedy back, with the
one-timer you will hold a memorial service for Speedy. 

You You
not pay not
pay pay
pay

Mr T Mr T Mr T Mr T
release kill release kill release kill release kill

4 2 3 1 4 2 3 1

1 2 4 3 1 2 3 4

Figure 3.18: The game for Part (b) Figure 3.19: The game for Part ((c)
of Exercise 3.3 of Exercise 3.3

Solution to Exercise 3.4. The backward-induction algorithm yields two solutions, shown
in Figures 3.20 and 3.21. The difference between the two solutions lies in what Player B
would do if Player A proposed F. In both solutions the officer is kept without promotion.
3.7 Solutions to exercises 103

A
P F
B B
accept K
accept
3 1
B
1 e pt 2
2 reject a cc
reject reject 3
C C C
2
P F 3 F F
K P P
1 K K

3 2 1 3 2 1 3 2 1
1 3 2 1 3 2 1 3 2
2 1 3 2 1 3 2 1 3

Figure 3.20: The first game for Exercise 3.4

A
P F
B B
accept K accept
3 1
B
1 e pt 2
2 reject a cc
reject reject 3
C C C
2
P F 3 F F
K P P
1 K K

3 2 1 3 2 1 3 2 1
1 3 2 1 3 2 1 3 2
2 1 3 2 1 3 2 1 3

Figure 3.21: The second game for Exercise 3.4

104 Chapter 3. Perfect-information Games

Solution to Exercise 3.5. The game of Figure 3.2 is reproduced in Figure 3.22, with the
unique backward-induction solution marked by double edges. The corresponding strategic
form is shown In Figure 3.23 (for each of Player 2’s strategies, the first element in the
pair is what Player 2 would do at her left node and the second element what she would do
at her right node). The Nash equilibria are highlighted. One Nash equilibrium, namely
(Offer 50-50,(Accept,Reject)), corresponds to the backward induction solution, while the
other Nash equilibrium, namely (Offer 70-30,( Reject,Reject)) does not correspond to a
backward-induction solution. 

1
Offer Offer
50-50 70-30

2 2

Accept Reject Accept Reject

2 1 3 1
3 2 1 2

Figure 3.22: The extensive-form game for Exercise 3.5

Player 2
(Accept,Accept) (Accept,Reject) (Reject,Accept) (Reject,Reject)
offer 50-50 2 3 2 3 1 2 1 2
Player 1
offer 70-30 3 1 1 2 3 1 1 2

Figure 3.23: The strategic-form game for Exercise 3.5

Solution to Exercise 3.6. The game of Figure 3.3 is reproduced in Figure 3.24 with the
two backward-induction solutions marked by double edges. The corresponding strategic
form is shown in Figure 3.25. The Nash equilibria are highlighted. The backward-
induction solutions are (a, (c, f ), g) and (b, (c, e), h) and both of them are Nash equilibria.
There are three more Nash equilibria which are not backward-induction solutions, namely
(b, (d, f ), g), (a, (c, f ), h) and (b, (d, e), h). 
3.7 Solutions to exercises 105
GAME THEORY – Giacomo Bonanno

1 1
a b a b
2 2 2 2
x y x y
c d e f c e f
d
3 3
2 0 3
g h 2 0 3
1 0 1 g h
1 0 1
0 2 0
0 2 0

1 0
1 0
2 0
2 0
1 1
1 1

THIS IS Fig 24 which is Fig3again

Figure 3.24: The extensive-form game for Exercise 3.6

The corresponding strategic form is as follows. The Nash equilibria are highlighted.
The backward-induction solutions are (a , ( c , f ) , g ) and (b , ( c , e ) , h ) and both of them
are Nash equilibria. There are three more PlayerNash 2 equilibria which are not backward-
induction solutions,
ce namely (b , ( d , fcf
) , g ), (b , ( c , e ) , h ) de
and (b , ( d , e ) , h ). df
Play er a 2 1 0 2 1 Player
0 20 0 2 0 0 2
1 b 3 1 ce 0 1 2 cf 1 3 1 de 0 1 2 df 1
Play er a 2 1 0 2 1 0 0 0 2 0 0 2
1 b 3 1 0 1 Player 2 3:1 g 3 1 0 1 2 1

Player 3: g
Player 2
ce cf de df
Player 2
Play er a 2 1 0 2 1 0 0 0 2 0 0 2
ce cf de df
1 b 3 1 0 0 0 1 3 1 0 0 0 1
Play er a 2 1 0 2 1 0 0 0 2 0 0 2
1 b 3 1 0 0 Player
0 3:1 h 3 1 0 0 0 1

Player 3: h
Figure 3.25: The strategic-form game for Exercise 3.6
Exercise 2.7. The game of Exercise 2.2 is reproduced below:
106 Chapter 3. Perfect-information Games

Solution to Exercise 3.7. The game of Exercise 3.2 is reproduced in Figure 3.26.
(a) All the possible strategies of Player B are shown in Figure 3.27.
(b) Player C has three decision nodes and three choices at each of her nodes. Thus she
has 3 × 3 × 3 = 27 strategies. 

A
P F
B K B
accept
accept
3 P F 1
1 B
pt 2
ce
2 reject ac reject reject 3
C K C C

F 2
P P F P F
K 3 K K
1

P K F P K F P K F
3 2 1 3 2 1 3 2 1
1 3 2 1 3 2 1 3 2
2 1 3 2 1 3 2 1 3

Figure 3.26: The extensive-form game for Exercise 3.7

If A chooses P If A chooses K If A chooses F

1 accept accept accept
2 accept accept reject
3 accept reject accept
4 accept reject reject
5 reject accept accept
6 reject accept reject
7 reject reject accept
8 reject reject reject

Figure 3.27: The eight strategies of Player B

3.7 Solutions to exercises 107

Solution to Exercise 3.8.

(a) One backward-induction solution is the strategy profile ((L,W ), (a, e)) shown by
double edges in Figure 3.28. The corresponding backward-induction outcome is the
play La with associated payoff vector (2, 1). The other backward-induction solution
is the strategy profile ((R,W ), (a, d)) shown in Figure 3.29. The corresponding
backward-induction outcome is the play Rd with associated payoff vector (3, 2).

1
L R
2 2
a b c d e
1
2 4 2 3 W E
1 0 0 2

1 0
2 3

Figure 3.28: One backward-induction solution of the game of Part (a) of Exercise 3.8

1
L R
2 2
a b c d e
1
2 4 2 3 W E
1 0 0 2

1 0
2 3

Figure 3.29: A second backward-induction solution of the game of Part (a) of Exercise 3.8

(b) Player 1 has four strategies: LW, LE, RW and RE.

(c) Player 2 has six strategies: ac, ad, ae, bc, bd and be.
108 Chapter 3. Perfect-information Games

(d) The strategic form is shown in Figure 3.30.

2
ac ad ae bc bd be
LW 2 , 1 2 , 1 2 , 1 4 , 0 4 , 0 4 , 0
1 LE 2 , 1 2 , 1 2 , 1 4 , 0 4 , 0 4 , 0
RW 2 , 0 3 , 2 1 , 2 2 , 0 3 , 2 1 , 2
RE 2 , 0 3 , 2 0 , 3 2 , 0 3 , 2 0 , 3

Figure 3.30: The strategic-form game for Part (d) of Exercise 3.8

(e) Player 1 does not have a dominant strategy.

(f) For Player 2 ae is a weakly dominant strategy.
(g) There is no dominant strategy equilibrium.
(h) For Player 1 RE is weakly dominated by RW (and LW and LE are equivalent).
(i) For Player 2 ac is weakly dominated by ad (and ae), ad is weakly dominated by ae,
bc is (strictly or weakly) dominated by every other strategy, bd is weakly dominated
by be (and by ae and ad), be is weakly dominated by ae.
Thus the dominated strategies are: ac, ad, bc, bd and be.
(j) The iterative elimination of weakly dominated strategies yields the following reduced
game (in Step 1 eliminate RE for Player 1 and ac, ad, bc, bd and be for Player 2; in
Step 2 eliminate RW for Player 1):

Player 2
ae
Player LW 2 , 1
1 LE 2 , 1

Thus we are left with one of the two backward-induction solutions, namely
((L,W ), (a, e)) but also with ((L, E), (a, e)) which is not a backward-induction solu-
tion.

(k) The Nash equilibria are highlighted in Figure 3.31.

There are five Nash equilibria: (LW, ac), (LE, ac), (RW, ad), (LW, ae) and (LE, ae).
namaste 
3.7 Solutions to exercises 109

2
ac ad ae bc bd be
LW 2 , 1 2 , 1 2 , 1 4 , 0 4 , 0 4 , 0
1 LE 2 , 1 2 , 1 2 , 1 4 , 0 4 , 0 4 , 0
RW 2 , 0 3 , 2 1 , 2 2 , 0 3 , 2 1 , 2
RE 2 , 0 3 , 2 0 , 3 2 , 0 3 , 2 0 , 3

Figure 3.31: The highlighted cells are the Nash equilibria (for Part (k) of Exercise 3.8)

Solution to Exercise 3.9.

(a) The extensive game is shown in Figure 3.32.

x=6 x=6.5

2 2

y=2.5 y=3 y=2.5 y=3

147.5 117.5 132.5 100

25 27.5 12.5 12.5

Figure 3.32: The extensive-form game for Exercise 3.9

110 Chapter 3. Perfect-information Games

(b) There are two backward-induction solutions.

The first is the strategy profile shown in Figure 3.33. The corresponding backward-
induction outcome is given by Firm 1 producing 6 units and Firm 2 producing 3
units with profits 117.5 for Firm 1 and 27.5 for Firm 2.
The other backward-induction solution is the strategy profile shown in Figure 3.34.
The corresponding backward-induction outcome is given by Firm 1 producing 6.5
units and Firm 2 producing 2.5 units with profits 132.5 for Firm 1 and 12.5 for Firm
2.

x=6 x=6.5

2 2

y=2.5 y=3 y=2.5 y=3

147.5 117.5 132.5 100

25 27.5 12.5 12.5

Figure 3.33: One backward-induction solution of the game of Figure 3.32

x=6 x=6.5

2 2

y=2.5 y=3 y=2.5 y=3

147.5 117.5 132.5 100

25 27.5 12.5 12.5

Figure 3.34: A second backward-induction solution of the game of Figure 3.32

(c) The strategic form is shown in Figure 3.35.

3.7 Solutions to exercises 111

Firm 2
(2.5,2.5) (2.5,3) (3,2.5) (3,3)

6 147.5 25 147.5 25 117.5 27.5 117.5 27.5

Firm 1
6.5 132.5 1 2.5 100 1 2.5 132.5 1 2.5 100 1 2.5

Figure 3.35: The strategic-form game for Part (d) of Exercise 3.9

(d) The Nash equilibria are highlighted in Figure 3.35. In this game the set of Nash
equilibria coincides with the set of backward-induction solutions. 

Solution to Exercise 3.10. The game under consideration is shown in Figure 3.36, where
x is an integer.

A B

2 2

C D E F

2 1 1 x
5 0 0 2

Figure 3.36: The extensive-form game for Part (a) of Exercise 3.10

(a) The backward-induction strategy of Player 2 is the same, no matter what x is, namely
(C, F). Thus the backward induction solutions are as follows.
• If x < 2, there is only one: (A, (C, F)).
• If x = 2 there are two: (A, (C, F)) and (B, (C, F)).
• I f x > 2, there is only one: (B, (C, F)).
112 Chapter 3. Perfect-information Games

(b) The strategic form is shown in Figure 3.37. First note that (A, (C, E)) is a Nash
equilibrium for every value of x. Now, depending on the value of x the other Nash
equilibria are as follows:
• If x < 1, (A, (C, F)).
• If 1 ≤ x < 2, (A, (C, F)) and (B, (D, F)).
• If x = 2, (A, (C, F)), (B, (C, F)) and (B, (D, F)).
• If x > 2, (B, (C, F)) and (B, (D, F)). 

Player 2
CE CF DE DF

A 2 5 2 5 1 0 1 0
Player 1
B 1 0 x 2 1 0 x 2

Figure 3.37: The strategic-form game for Part (b) of Exercise 3.10

Solution to Exercise 3.11. Let us find the losing positions. If Player i, with his choice,
can bring the sum to 40 then he can win (the other player with her next choice will take
the sum to a number between 41 and 47 and then Player i can win with his next choice).
Working backwards, the previous losing position is 32 (from here the player who has to
move will take the sum to a number between 33 and 39 and after this the opponent can
take it to 40). Reasoning backwards, the earlier losing positions are 24, 16, 8 and 0. Thus
Player 1 starts from a losing position and therefore it is Player 2 who has a winning strategy.
The winning strategy is: at every turn, if Player 1’s last choice was n then Player 2 should
choose (8 − n). 

Solution to Exercise 3.12.

(a) The initial part of the game is shown in Figure 3.38.

1
A2 B2
B1
2 2 2
C3
A3 B3 B2 C2 B3
B2 C1 C2

Figure 3.38: The initial part of the game of Part (a) of Exercise 3.12
3.7 Solutions to exercises 113

(b) From G4 Player 1 should move the coin to H5. From there Player 2 has to move it to
H6 and Player 1 to H7 and Player 2 to H8 and from there Player 1 wins by moving
it to H9.
1 2 1 2 1 2 1 2 1
(c) A1 → B2 → C3 → D4 → E5 → F5 → G6 → H7 → H8 → H9.
1 2 1 2 1 2 1 2 1 2
(d) A1 → B2 → C3 → D4 → E5 → F5 → G6 → G7 → H7 → H8 → H9.

(e) Using backward induction we can label each cell with a W (meaning that the player
who has to move when the coin is there has a winning continuation strategy) or with
an L (meaning that the player who has to move when the coin is there can be made
to lose).
If all the cells that are accessible from a given cell are marked with a W then that
cell must be marked with an L.
If from a cell there is an accessible cell marked with an L then that cell should be
marked with a W. See Figure 3.39.
From the picture it is clear that it is Player 1 who has a winning strategy. The
winning strategy of Player 1 is: move the coin to cell B1 and from then on, after
every move of Player 2, move the coin to a cell marked L. 

Figure 3.39: Solution for the coin game

114 Chapter 3. Perfect-information Games

Solution to Exercise 3.13.

(a) The game is shown in Figure 3.40.

give up
Anna claim
0
CB
assert
claim

accept
Bess CA
0
challenge and
challenge and
choose B 1
choose B 2

Anna Anna
don't don't
match match match
match

F C A B 1 F F C A B 2 F
C B B 1 F C B B 2 F

Figure 3.40: The extensive-form game for Part (a) of Exercise 3.13

(b) The backward-induction solution is marked by thick arrows in Figure 3.40.

(c) The sequence of moves is shown in Figure 3.41.
Suppose that Anna is the legal owner and values the ring more than Bess does:
CA > CB . At the last node Anna will choose “match” if CA > B and “don’t match” if
B > CA . In the first case Bess’s payoff will be −F, while in the second case it will
be CB − B, which is negative since B > CA and CA > CB . Thus in either case Bess’s
payoff would be negative. Hence at her decision node Bess will choose “accept”
(Bess can get the ring at this stage only if she bids more than the ring is worth to
her). Anticipating this, Anna will assert her claim at the first decision node. Thus at
the backward-induction solution the ring goes to Anna, the legal owner. The payoffs
are CA for Anna and 0 for Bess. Note that no money changes hands.
3.7 Solutions to exercises 115

(d) Suppose that Bess is the legal owner and values the ring more than Anna does:
CB > CA . At the last node Anna will choose "match" if CA > B and "don’t match"
if B > CA . In the first case Bess’s payoff will be −F, while in the second case it
will be CB − B, which will be positive as long as CB > B. Hence at her decision
node Bess will choose to challenge and bid any amount B such that CB > B > CA .
Anticipating this, at her first decision node Anna will give up (and get a payoff of 0),
because if she asserted her claim then her final payoff would be negative. Thus at
the backward-induction solution the ring goes to Bess, the legal owner. The payoffs
are 0 for Anna and CB for Bess. Note that no money changes hands.
(e) As pointed out above, in both cases no money changes hands at the backward-
induction solution. Thus Judge Sabio collects no money at all and both Ann and
Bess pay nothing. 

give up
Anna claim
0
CB
assert
claim

accept
Bess CA
0

challenge and
choose B

Anna
don't
match
match

F C A B  F

C B B F

Figure 3.41: The extensive-form game for Part ((c) of Exercise 3.13