University

of London EC2066 1

Game Theory
Essential reading
Subject guide chapter 4
N&S chapter 5

Introduction
• A game is an interaction between players (such as individuals or firms) in which players use
strategies.

• What is a strategy? A strategy is a battle plan that specifies the actions that a player will
make conditional on the information available at each move and for any possible contingency.

• For example, a firm may use a simple business strategy where it produces 100 units of output
regardless of what a rival does. In such a case, the strategy consists of a single action—
producing 100 units of output.

• However, a strategy can consist of a combination of actions, possibly contingent on what a

rival does. For example, a firm might decide to produce a small quantity as long as its rival
produced a small amount in the previous period, and a large quantity otherwise. Here, the
strategy consists of two actions—produce a small quantity or produce a large quantity.

• Players receive benefits from a game’s outcome, such as profits for firms or incomes or
utilities for individuals. This is known as payoffs1.

• Hence, a payoff function specifies each player’s payoff as a function of the strategies chosen
by all players.

Simultaneous-move or normal-form games

• Let us begin by examining static games, in which the players choose their actions
simultaneously, have complete information about the payoff function, and play the game once.

• Suppose there are two players, 1 and 2, and 1 can choose between two actions Up or
Down (so 1’s strategy set is {U, D}) and 2 can choose between two actions Left or
Right (so 2’s strategy set is {L, R})

• Then if 1 chooses Up and 2 chooses Left, we have the strategy profile {U, L}. There
are altogether 4 such strategy profiles: the one just mentioned {U, L}, plus{D, L},
{U, R} and {D, R}

• Once you understand what a strategy profile is, we can define the third element of a
game: a payoff function for each player. A payoff function for any player is defined
over the set of strategy profiles.For each strategy profile, each player gets a payoff.

• Let 𝑢! denote the payoff of player i.

1 We assume that rational players will try to obtain the highest payoffs they can.

University of London EC2066 2

• We write this in a convenient matrix form (known as the normal form) as follows.
The first number in each cell is the payoff of player 1 and the second number is the
payoff of player 2. Note that player 1 chooses rows and player 2 chooses
columns.

Dominant strategies

• Suppose a player has a strategy that gives a higher payoff compared to other
strategies irrespective of the strategy choices of others. Such a strategy is called a
dominant strategy. If a player has a dominant strategy, his problem is simple: he
should clearly play that strategy.

• Consider the following game. Each player has two strategies C (cooperate) and D
(defect, which means not to cooperate). There are 4 possible strategy profiles and
each profile generates a payoff for each player. The first number in each cell is the
payoff of player 1 and the second number is the payoff of player 2.

• Here each player has a dominant strategy, D. This is the well-known Prisoners'
Dilemma game. Rational players, playing in rational self-interest, get locked
into a dominant-strategy equilibrium that gives a lower payoff compared to the
situation where both players cooperate. However, cooperating cannot be part
of any equilibrium, since D is the dominant strategy.

• In the game above, the dominant strategy equilibrium is (D, D).

University of London EC2066 3

Dominated strategies and iterated elimination

• A dominated strategy for i is a strategy of i (say 𝑠! ) that yields a lower payoff

compared to another strategy (say 𝑠! ’) irrespective of what others are playing.
In other words, the payoff of i from playing 𝑠! is always (i.e. under all possible
choices of other players) lower than the payoff from playing 𝑠! ’

• Since 𝑠! is a dominated strategy, i would never play this strategy. Thus we can
eliminate dominated strategies.

• If in some game, all strategies except one for each player can be eliminated by
iteratively eliminating dominated strategies, the game is said to be dominance
solvable.

• Here is an example of a dominance solvable game. We find the equilibrium of

this game by iteratively eliminating dominated strategies.

We can eliminate dominated strategies iteratively as follows.

1. For player 1, Bottom is dominated by Top. Eliminate Bottom.

2. In the remaining game, for player 2, Right is dominated by Middle. Eliminate

Right.

3. In the remaining game, for player 1, Top is dominated by Middle. Eliminate Top.

4. In the remaining game, for player 2, Middle is dominated by Left. Eliminate

Middle.

This gives us (Middle, Left) as the unique equilibrium.

University of London EC2066 4

Nash equilibrium

• In some games, players might not have dominant strategies; moreover none of
the strategies of any player might be dominated. The following game provides
an example.

• For games that do not have dominant or dominated strategies, the idea of
deriving an equilibrium using dominance arguments does not work.

• If we want to derive an equilibrium that does not rely on specific features such
as dominance, we need a concept of equilibrium that applies generally to all
games. As we show below, a Nash equilibrium (named after the
mathematician John Nash) is indeed such a solution concept.

• We start by analysing Nash equilibrium under pure strategies2.

• A strategy profile is a Nash equilibrium (in pure strategies) if it is a mutual

3
best response . (In other words, in a Nash equilibrium, no player can
have a strictly profitable deviation given what the others are doing).

• To find out the Nash equilibrium of the game above, we must look for the
mutual best responses. Let us check the best response of each player.

2 A pure strategy is a strategy that just chooses one action. In other words, a pure strategy
assigns a probability of 1 to a single action.
3 For a more formal definition of Nash Equilibrium, we can say that in a two-player game, a set of

strategies (a*, b*) is a Nash equilibrium if a* is player A ’s best response against b* and b* is player
B’s best response against a*.

University of London EC2066 5

Player 1’s best response is as follows:

Player 2’s strategy Player 1’s best response

A2 A1
B2 B1
C2 A1

Player 2’s best response:

Player 1’s strategy Player 2’s best response

A1 B2
B1 B2
C1 A2

Note from these that the only mutual best response is 𝐵! , 𝐵! . This is the only Nash
equilibrium in this game.

We can also find the Nash equilibrium by checking the boxes and see if any profitable
deviation exist. If a profitable deviation exists for either player, it will not be a Nash
equilibrium. If no profitable deviation exists for all players, then it will be a Nash
equilibrium.

For example, in the game above, suppose you want to check if (𝐴! , 𝐶! )is a Nash
equilibrium. You should check as follows.

If player 2 plays 𝐶! , player 1 cannot do any better by changing strategy from 𝐴! (4 is

better than 3 from 𝐵! or 3 from 𝐶! ). However, player 2 would not want to stay at
(𝐴! , 𝐶! )since player 2 can do better by switching to 𝐵! (3 from 𝐵! is better than 2
from 𝐶! ). We can therefore conclude that (𝐴! , 𝐶! )is not a Nash equilibrium.

You can similarly check that all boxes other than 𝐵! , 𝐵! have the property that some
player has an incentive to switch to another box.

However, if the players are playing 𝐵! , 𝐵! no player has an incentive to switch away.
Neither player can do better by switching given what the other player is playing.

Since player 2 is playing 𝐵! , player 1 gets 3 from 𝐵! which is better than 1 from 𝐴! or
2 from 𝐶! . So player 1 has no incentives to deviate.

Now, for player 2, since player 1 is playing 𝐵! , player 2 gets 1 from 𝐵! which is better
than 0 from 𝐴! or 0 from 𝐶! . So player 2 has no incentives to deviate.

Therefore 𝐵! , 𝐵! is a Nash equilibrium.

University of London EC2066 6

Nash equilibrium is not necessarily unique

Nash equilibrium is not necessarily unique. Consider the following game.

Note there are multiple pure strategy Nash equilibria. Both (A,A) and (B,B) are Nash
equilibria.

The Mixed Strategy Nash Equilibrium (MSNE)

Why Bother With MSNE?
Even before we begin discussing MSNE, or even, what a mixed strategy is, you
should be asking yourself why does game theory need another equilibrium concept,
when it already has pure strategy Nash equilibria? The straight forward answer is
some games might have no equilibrium in pure strategies. For example, consider the
following normal form game.

• Notice that this game has no pure strategy Nash equilibrium. However, we can
look for a mixed strategies4 Nash equilibrium, where players randomly select
from several possible actions.

• Suppose both players simply flip the penny and play whatever side turns up.
The result of this strategy is a random choice of Heads with probability 1/2
and Tails with probability ½ (given that the penny is a fair penny).

4 A mixed strategy is a strategy in which the player chooses among possible actions according to

the probabilities the player assigns.

University of London EC2066 7

This set of strategies, with both playing Heads or Tails with equal chance, is in
fact the mixed-strategy Nash equilibrium of the game. How do we know that?
To verify this, we just need to show that both players’ strategies are best
responses to each other (so that no profitable deviation exist for both players).

• Given that both players play head or tails with equal chance of ½, therefore, all four
outcomes corresponding to the four boxes are equally likely to occur, each occurring
with probability ¼.

• A’s expected payoff is:

¼ x (1) + ¼ x (-1) + ¼ x (-1) + ¼ x (1) = 0

• Similarly, B’s expected payoff is:

¼ x (1) + ¼ x (-1) + ¼ x (-1) + ¼ x (1) = 0

• Now, let us check if any player can deviate to a strategy that produces a strictly
higher payoff than 0.

• Given that A plays Heads with probability ½ and Tails with probability ½, can B do
better by choosing a different strategy?

• No. B will always get an expected payoff of 0. For example, if A plays Heads with
probability ½ and Tails with probability ½ and B plays Heads as a pure strategy5, B’s
expected payoff will be ½ x (-1) + ½ (1) = 0

• How about for A? Given that B plays Heads with probability ½ and Tails with
probability ½, can A do better by choosing a different strategy?

• No. A will always get an expected payoff of 0. For example, if B plays Heads with
probability ½ and Tails with probability ½ and A plays Tails as a pure strategy6, A’s
expected payoff will be ½ x (-1) + ½ (1) = 0

• Since neither player has a strictly profitable deviation, we have a mixed strategy Nash
equilibrium with both players playing Heads and Tails with equal probabilities.

5 In other words, B plays Heads with probability 1 and Tails with probability 0.
6 In other words, A plays Tails with probability 1 and Heads with probability 0.

University of London EC2066 8

Let us now see how we can compute the mixed strategies in the next example
below.

• Notice that this game has no pure strategy Nash equilibrium. However, as you
will see below, the game does have a Nash equilibrium in mixed strategies.

• Suppose we have an equilibrium in which both players play strictly mixed

strategies: player 1 plays 𝐴! with probability p and 𝐵! with probability (1 − p)
where 0 < p < 1, and player 2 plays 𝐴! with probability q and 𝐵! with
probability (1 − q) where 0 < q < 1.

In this case, whenever player 1 plays 𝐴! , she gets an expected payoff of:

𝜋! 𝐴! = 3𝑞 + 2 1 − 𝑞

Whenever player 1 plays 𝐵! , she gets an expected payoff of:

𝜋! 𝐵! = 2𝑞 + 3 1 − 𝑞

• Suppose 𝜋! (𝐴! ) > 𝜋! (𝐵! ). Then clearly player 1 should simply play 𝐴! , rather
than any strictly mixed strategy,

• In other words, player 1's best response in this case would be to choose p = 1,
rather than a strictly mixed strategy p < 1. Thus, we do not have a mixed
strategy Nash equilibrium.

• Similarly, if 𝜋! (𝐴! ) < 𝜋! (𝐵! ), player 1’s best response would be to choose
p = 0 (i.e. just play 𝐵! ), and again we cannot have a mixed strategy Nash
equilibrium.

• So if player 1 is going to play a mixed strategy in equilibrium it must be that

she is indifferent between the two strategies.

University of London EC2066 9

• In other words, in equilibrium q must be such that 𝜋! (𝐴! ) = 𝜋! (𝐵! ), i.e. we

have:

3𝑞 + 2 1 − 𝑞 = 2𝑞 + 3 1 − 𝑞

which implies 𝑞 = 1 2

• But if player 2 is going to choose q = 1/2 it must be that he is indifferent

between 𝐴! and 𝐵! (otherwise player 2 would not want to mix, but would
play a pure strategy).

• How can such indifference come about? Well, player 1 must choose p in such
a way so as to make player 2 indifferent between 𝐴! and 𝐵! . In other words,
player 1’s choice of p is such that 𝜋! (𝐴! ) = 𝜋! (𝐵! ) where

𝜋! 𝐴! = 1 𝑝 + 1 1 − 𝑝 = 1
and

𝜋! 𝐵! = 3 𝑝 + 0 1 − 𝑝 = 3𝑝

So for player 2 to be indifferent between 𝐴! and 𝐵! , we need

𝜋! (𝐴! ) = 𝜋! (𝐵! )

1 = 3𝑝

which implies 𝑝 = 1/3.

• Therefore, the mixed strategy Nash equilibrium is as follows. Player 1 plays

𝐴! with probability 1/3 and 𝐵! with probability 2/3, while player 2 plays 𝐴!
with probability 1/2 and 𝐵! with probability 1/2.

University of London EC2066 10

We can also show this in a diagram.

University of London EC2066 11

Activity 4.1 Find the mixed strategy Nash equilibrium in the following
game. Also show all Nash equilibria of the game in a diagram by drawing
the best response functions.

Player 2
A B
A 2, 1 0, 0
Player 1 B 0, 0 1, 2

University of London EC2066 12

Sequential-move or extensive-form games

Let us now consider sequential move games. In this case, we need to draw a game
tree to depict the sequence of actions. Games depicted in such a way are also known
as extensive-form games7.

To start with, we assume that each player can observe the moves of players who act
before them.

Consider the following extensive-form game. Each player has two actions: player
1’s actions are a1 and a2 and player 2’s actions are b1 and b2.

Player 1 moves before player 2. Player 2 can observe player 1’s action and,
therefore, can vary his action depending on the action of player 1.

For each profile of actions by player 1 and player 2 there are payoffs at the
end. As usual, the first number is the payoff of the first mover (in this case
player 1) and the second number is the payoff of the second mover (here
player 2).

7 Extensive-form game is also known as sequential-move game

University of London EC2066 13

Actions and strategies

The notion of a strategy is fairly straightforward in a normal form game.

However, for an extensive-form game, it is a little bit more

complicated. A strategy in an extensive form game is a complete plan of
actions. In other words, a strategy for player i must specify an action at
every node at which i can possibly move.

Consider the game above. As already noted, each player has 2 actions.
For player 1, the set of actions and the set of strategies is the same.
Player 1 can simply decide between a1 and a2. Therefore, the strategy
set of player 1 is simply {a1, a2}.

Player 2, on the other hand, must plan for two different contingencies.
He must decide what to do if player 1 plays a1, and what to do if player
1 plays a2. Note that such decisions must be made before the game is
actually played. Essentially, game theory tries to capture the process of
decision-making of individuals. Faced with a game such as the one
above, player 2 must consider both contingencies. This is what we
capture by the notion of a strategy. It tells us what player 2 would do in
each of the two possible cases.

Now player 2 can choose 2 possible actions at the left node (after player
1 plays a1) and 2 possible actions at the right node (after a2). So there
are 2 x 2 = 4 possible strategies for player 2. These are:

1. If player 1 plays a1, play b1 and if player 1 plays a2, play b1.
2. If player 1 plays a1, play b1 and if player 1 plays a2, play b2.
3. If player 1 plays a1, play b2 and if player 1 plays a2, play b1.
4. If player 1 plays a1, play b2 and if player 1 plays a2, play b2.

For the sake of brevity of notation, we write these as follows. Just as

we read words from left to right, we read strategies from left to right.
So we write the strategy ‘if player 1 plays a1, play b2 and if player 1
plays a2, play b1’ as b2b1.Reading from left to right, this implies that
the plan is to play b2 at the left node and play b1 at the right node.
This is precisely what the longer specification says.

So the strategy set of player 2 is {(b1 b1 ), (b1 b2 ), (b2 b1 ), (b2 b2 )}.

University of London EC2066 14

To find Nash equilibria in an extensive-form game, the most convenient method is to

transform it into its normal form. This is shown below. Note that player 1 has two
strategies and player 2 has four strategies. Therefore, we have a 2-by-4matrix of
payoffs as follows

Note that if we pair, say, a1 with b2b1, only the first component of player
2’s strategy is relevant for the payoff. In other words, since player 1 plays
a1, the payoff is generated by player 2’s response to a1, which in this
case is b2. Similarly, if we pair a2 with b2b1, the second component of
player 2’s strategy is relevant for the payoff. Since player 1 plays a2, we
need player 2’s response to a2, which in this case is b1, to determine the
payoff.

Once we write down the normal form, it is easy to find Nash equilibria.
Here let us only consider pure strategy Nash equilibria. There are three
pure strategy Nash equilibria in this game. Finding these is left as an
exercise (see Example 4.1 below).

Example 4.1Find the pure strategy Nash equilibria of the extensive-

form game above.
It is easiest to check each box. If we start at the top left-hand box, player
1 would switch to a2. So this is not a Nash equilibrium. From (a2, b1b1)
no player can do better by deviating. Therefore, (a2, b1b1) is a Nash
equilibrium.

Next try (a1, b1b2). This is indeed a Nash equilibrium. Note that you
must write down the full strategy of player 2. It is not enough to write
(a1, b1). Unless we know what player 2 would have played in the node
that was not reached (in this case the node after a2 was not reached), we
cannot determine whether a strategy is part of a Nash equilibrium. So
while (a1, b1b2) is indeed a Nash equilibrium, (a1, b1b1) is not.

Finally, (a2, b2b1) is also a Nash equilibrium. These are the three pure
strategy Nash equilibria of the game.

University of London EC2066 15

Imperfect information: information sets

So far we have assumed perfect information: each player is perfectly informed of all
previous moves. Let us now see how to represent imperfect information: players
may not be perfectly informed about some of the (or all of the) previous moves.

Consider the extensive-form game introduced at the start of this section. Suppose at
the time of making a decision, player 2 does not know what strategy player 1 has
chosen.Since player 2 does not know what player 1 has chosen, at the time of taking
an action player 2 does not know whether he is at the left node or at the right node.
To capture this situation of imperfect information in our game-tree, we say that the
two decision nodes of player 2 are in an information set. We represent this
information set as in Figure 4.3: by connecting the two nodes by a dotted line
(another standard way to do this is to draw an elliptical shape around the two
nodes – you will see this in N&S).

University of London EC2066 16

Note that player 2 knows he is at the information set (after player 1 moves), but does
not know where he is in the information set (i.e. he does not know what player 1 has
chosen). Since player 2 cannot distinguish between the two nodes inside his information
set, he cannot take different actions at the two nodes. Therefore, the strategy set of 2 is
simply {b1, b2}. In other words, now, for both players the strategy set coincides with
the action set. This is not surprising: since player 2 takes an action without knowing
what player 1 has done, the game is the same as a simultaneous-move game. Indeed,
the normal form of the game above coincides with a game in which the two players
choose strategies simultaneously. This is shown below. Now the Nash equilibrium is
simply (a2, b1).

Player 2
b1 b2
a1 3, 1 1, 0
Player 1 a2 4, 1 0, 1

University of London EC2066 17

Incredible threats in Nash equilibria and subgame

perfection

• Consider the game shown in Figure 4.4. Firm E, where E stands for entrant, is
deciding whether to enter a market. The market has an incumbent firm (Firm
I). If the entrant enters, the incumbent firm must decide whether to fight (start
a price war, say) or accommodate the entrant. The sequence of actions and
payoffs is as follows.

• What will be the Nash equilibrium of this game? To answer this, we can write
down the normal form of the game.

The normal form is as follows.

University of London EC2066 18

• Note that there are two pure strategy Nash equilibria: (In, A) and (Out, F).

• However, the latter equilibrium involves an incredible threat. Clearly, if the

entrant does decide to enter the market, the incumbent has no incentive to
choose F. Hence the threat of F is incredible.

• Yet, Nash equilibrium cannot prevent this possibility. Out is the best response
to F and once Firm E decides to stay out, anything (and in particular F) is a
best response for Firm I.

• Let us now describe a solution concept that imposes extra conditions (i.e.
further to the requirement that strategies be mutual best responses) for
equilibrium and leads to a refinement of the set of Nash equilibria, namely
subgame perfection.

• To understand the refinement, you first need to understand the idea of a

subgame.

• A subgame is a part of a game that starts from a node which is a singleton (i.e.
a subgame does not start at an information set), and includes all successors of
that node.If one node in an information set belongs to a subgame, so do all
other nodes in that information set. In other words, you cannot cut an
information set so that only part of it belongs to a subgame. That would
clearly alter the information structure of the game, which is not allowed.

• We can now define a subgame perfect equilibrium.

Subgame perfect Nash equilibrium

A strategy combination is a subgame perfect Nash equilibrium (SPNE) if:

it is a Nash equilibrium of the whole game

it induces a Nash equilibrium in every subgame.

Perfect information: backward induction

• In perfect information games (recall that, in a game of perfect information,

each player knows all past moves of other players), this is easy. Subgame
perfect Nash equilibria can be derived simply by solving backwards, i.e. by
using backward induction.

• Solving backwards in the entry game, we see that Firm I would choose A if
Firm E chose In. Knowing this, Firm E would compare 0 (from Out) with 2
(from In and A), and choose In. Therefore, the SPNE is (In, A).

University of London EC2066 19

• Let us see that this equilibrium derived using backward induction fits with the
definition of SPNE given above.

• The game has a subgame starting at the node after Firm E plays In (also, the
whole game is always trivially a subgame).

• In the subgame after E plays In, there is only one player (Firm I), and the Nash
equilibrium in this subgame is simply the optimal action of Firm I, which in
this case is to choose A.

• Therefore the Nash equilibrium in the subgame is A. It follows that any Nash
equilibrium of the whole game that involves playing A in the subgame is a
subgame perfect Nash equilibrium.

• Here, the only Nash equilibrium of the whole game that satisfies this property
is (In, A). Therefore, this is the only SPNE.

• Next, consider the game in which player 1 chooses between L, R and player 2
moves second and chooses between l, r. In this game there are two strict
subgames, one starting after each action of player 1. In the left subgame,
player 2’s optimal choice is l, and in the right subgame player 2’s optimal
choice is r. Given this, player 1 would compare 3 from R and 0 from L, and
choose R. The choices obtained by backward induction are shown in Figure
4.6.

• It follows that the SPNE is (R, lr).

University of London EC2066 20

Activity 4.3 Derive the pure strategy subgame perfect Nash equilibria of the game
in Figure 4.2.

University of London EC2066 21

Subgame perfection under imperfect information

• Backward induction need not work under imperfect information: you cannot
fold backwards when you come up against an information set.

• Consider the imperfect information game below. Note that this game does not
have any strict subgames (recall that you cannot start a subgame from an
information set or cut an information set), so the only subgame is the whole
game. Therefore, any Nash equilibrium of the whole game is trivially subgame
perfect.

• The pure strategy Nash equilibrium in this game is (a2, b1). This is also the
subgame perfect Nash equilibrium.

University of London EC2066 22

Next, consider the game in Figure 4.7.

• Initially player 1 decides whether to come in (and play some game with player
2) or stay out (in which case player 2 plays no role).

• If the choice is to come in, player 1 decides between A and B, and player 2
decides between C and D.

• When player 2 makes his decision, he knows that player 1 has decided to
come in (if not then player 2 would not have been asked to play), but without
knowing what player 1 has chosen between A and B.

University of London EC2066 23

1. Pure strategy Nash equilibria.

• Let us first identify the pure strategy Nash equilibria.

Note that player 1 has 4 strategies: Out A, Out B, In A and In B, while player
2 has 2 strategies: C and D. You might think strategies like Out A do not make
sense, but in game theory we try to model the thought process of players, and
even if player 1 stays out, she would do so only after thinking about what she
would have done had she entered.
Strategies reflect such thinking (it is as if player 1 is saying ‘I have decided to
finally stay out, but had I come in I would have chosen A’).

Let us now write down the normal form of the game.

You should be able to see from this that the pure strategy Nash equilibria are
(Out A, C), (Out B, C) and (In A, D)

2. Pure strategy subgame perfect Nash equilibria.

• We now proceed to find the subgame perfect Nash equilibria.

• Note that backward induction does not work here: we cannot fold back given
the information set of player 2. However, subgame perfection still works.

• Note that apart from the whole game, there is just one strict subgame, which
starts at the node after player 1 chooses In. Below we write down the normal
form of the strict subgame.

• The subgame has two pure strategy Nash equilibria: (B, C) and (A, D)

University of London EC2066 24

• Now (B, C) and (A, D) are the Nash equilibria in the subgame.
Therefore, it must be that any Nash equilibria of the whole game that
involves playing either (B, C) and (A, D) in the subgame are subgame
perfect.

• Thus, the pure strategy SPNE of the whole game are (Out B, C) and
(In A, D).

• For example, suppose (B, C) is played in the subgame, player 1

compares 1 (from Out) with 0 (from In followed by (B, C)) and decides
to stay out. Therefore, a SPNE of the whole game is (Out B, C).

• If, on the other hand, (A, D) is played in the subgame, player 1

compares 1 (from out) with 2 (from In followed by (A, D)) and decides
to come in. Therefore, another SPNE of the whole game is (In A,D).

University of London EC2066 25

Repeated Prisoners’ Dilemma

Consider the following Prisoners’ Dilemma game:

• In a one-shot game, rational players simply play their dominant strategies. So

(D, D) is the only possible equilibrium8. Suppose the game is repeated. Can
we say something about the behaviour of players in such ‘supergames’ that
differs from the behaviour in the one-shot game?

• First, consider the case of a finite number of repetitions. Say the game
is played twice. Would anything change? The answer is no. In the
second round, players simply face a one-shot game and they would
definitely play their dominant strategies. Given that (D, D) will be
played in the next period, playing anything other than D today makes
no sense. Therefore, in each period players would play (D, D). But this
logic extends to any finite number of repetitions. If the game is played a
100 times, in the last period (D, D) will be played. This implies that
(D, D) will be played in the 99th period, and so on.

• This implies that cooperation cannot be sustained if the number of repetition is

finite. However, if the games are infinitely repeated, as we will see below,
cooperation can be sustained.

• Assume now that when the game is repeated many times, players play them as
if the games are infinitely repeated.

• Let δ denote the common discount factor across players, where 0 < δ < 1. If
today’s date is 0, and a player receives x in period t, the present value of that
payoff is 𝛿 ! x. The discount factor can reflect players’ time preference. This
can also arise from a simple rate of interest calculation, in which case δ can
be interpreted as 1/(1 + r), where r is the rate of interest. Note that higher
values of δ indicate that players are more patient (i.e. value future payoffs
more).

8 Recall from above that this is a prisoners' dilemma game - rational players playing
in rational self-interest, get locked into a dominant-strategy equilibrium that gives a
lower payoff compared to the situation where both players cooperate.

University of London EC2066 26

Cooperation through trigger strategies

Start by playing C (that is, cooperate at the very first period (period 0), when
there is no history yet).

In period t ≥1:
• if (C, C) was played last period, play C
• if anything else was played last period, play D.

Suppose each player follows this strategy. Note that anything other
than (C, C) implies playing (D,D) next period, once a switch to (D, D)
has been made, there is no way back: the players must play (D, D)
forever afterwards. This is why this is a trigger strategy.

Let us see if this will sustain cooperation. Suppose a player deviates in period t.
We only need to consider what happens from t onwards. The payoff starting at
period t is given by:
𝛿
3 + 𝛿 + 𝛿! + ⋯ = 3 +
1−𝛿

If the player did not deviate in period t, the payoff from t onwards would be:

2
2 + 2𝛿 + 2𝛿 ! + ⋯ =
1−𝛿

For deviation to be suboptimal, we need:

2 𝛿
≥3+
1−𝛿 1−𝛿

which implies:
1
𝛿≥
2

University of London EC2066 27

Thus if the players are patient enough, cooperation can be sustained in

equilibrium. In other words, playing (C, C) always can be the outcome of an
equilibrium if the discount factor δ is at least 1/2.

University of London EC2066 28

Folk theorem
We showed above that the cooperative payoff (2, 2) can be sustained in equilibrium
for high enough values of δ. However, this is not the only possible equilibrium
outcome. Indeed, many different payoffs can be sustained in equilibrium.

For example, note that always playing (D, D) is an equilibrium no matter what the
value of δ is. Each player simply adopts the strategy ‘play D initially, and at any
period 𝑡 > 1, play D irrespective of history’.

Note that both players adopting this strategy is a mutual best response. Therefore, we
can sustain (1, 1) in equilibrium.

In fact, by using suitable strategies, we can sustain many more – in fact infinitely
more – payoffs as equilibrium outcomes. This is the idea behind the folk theorem.

The folk theorem implies that anything from no cooperation to full cooperation can be
sustained through the use of trigger strategies for high enough values of δ.

In other words, the folk thereom says that not only is full cooperation possible, but
partial cooperation is also possible (so long as δ is high enough). By partial
cooperation, we mean strategies that result in payoffs for the two players that give
more than the non-cooperative payoff of 1.

This is shown by the shaded part in figure 4.8.

It should not be too difficult to see how any of these payoffs in the shaded part can be
sustained.

Pick any payoff combination in the shaded area. By definition, these payoffs are
greater than what a player could get under non-cooperation.

Define a trigger strategy that says ‘Play this sequence as long as the other player plays
his part, otherwise, switch forever to D’. You should be able to see that for high
enough values of δ, all the payoffs in the shaded part can be sustained in equilibrium.

University of London EC2066 29

Let us now state the folk theorem formally.

To state this, we will need to compare payoffs in the repeated game to payoffs in the
one-shot game that is being repeated. The easiest way to do that is to normalise the
repeated game payoff by multiplying by (1 − δ). Then if a player gets 2 every period,
the repeated game payoff is (1 − δ) × 2/(1 − δ) = 2.

As you can see, this normalization implies that the set of normalised repeated game
payoffs now coincide with the set of payoffs in the underlying one-shot game. So now
we can just look at the set of payoffs of the one-shot game and ask which of these are
sustainable as the normalised payoff in some equilibrium of the repeated game.

Note that in this game, a player can always get at least 1 by simply playing D. It
follows that a player must get at least 1 as a (normalised) repeated game payoff.

To see the whole set of payoffs that can be sustained, let us plot the payoffs from the
four different pure strategy profiles. These are shown in Figure 4.8 above. Now join
the payoffs and form a convex set as shown in the figure. We now have a set of
payoffs that can arise from pure or mixed strategies.

The Folk theorem is the following claim. Consider any pair of payoffs
(𝜋1, 𝜋 2) such that 𝜋 i > 1 for i = 1, 2. Any such payoff can be supported as
an equilibrium payoff for high enough δ.
As noted above, for the Prisoners’ Dilemma game we can also sustain
the payoff (1, 1) as an equilibrium outcome irrespective of the value of
δ.
The set of payoffs that can be supported as equilibrium payoffs in our
example is shown as the shaded part in Figure 4.8.

University of London EC2066 30

Tutorial

Sample Examination Questions (from subject guide)

University of London EC2066 31

University of London EC2066 32

University of London EC2066 33