GAMES OF INCOMPLETE

INFORMATION…
-SIGNALING GAMES-

IN A GAME OF INCOMPLETE INFORMATION AT LEAST ONE

PLAYER LACKS INFORMATION ABOUT THE
OBJECTIVE/MOTIVES OF HER OPPONENT(S).
THAT IS, AT LEAST ONE PLAYER DOES NOT KNOW HER
OPPONENT(S) PAYOFFS …

A SIGNALING GAME IS AN INCOMPLETE INFORMATION GAME

WITH THE STRUCTURE GIVEN IN THE EXAMPLE BELOW

EXAMPLE:
PLAYER 1 MOVES FIRST, CHOOSING BETWEEN ACTION X AND
ACTION Y.
PLAYER 2 OBSERVES PLAYER 1’S ACTION AND THEN
CHOOSES BETWEEN LEFT AND RIGHT. THE GAME ENDS.

PAYOFFS: SUPPOSE PLAYER 1’S PAYOFFS CAN TAKE ONLY

TWO FORMS: AS GIVEN IN COLUMN “PLAYER 1(a)” (TYPE-a
PLAYER 1) OR IN COLUMN “PLAYER 1 (b)” (TYPE-b PLAYER 1).

PLAYER 2’S PAYOFFS MAY ALSO DEPEND ON THE TYPE OF

PLAYER 1.

OUTCOME PLAYER 1 (a) PLAYER 2 PLAYER 1 (b) PLAYER 2

(X, LEFT) 5 0 5 6
(X, RIGHT) 0 6 10 0
(Y,LEFT) 6 10 6 10
(Y, RIGHT) 0 0 10 0

(This is not the usual strategic (normal) form; it is just a presentation of the
payoffs)

1
INFORMATION STRUCTURE:
PLAYER 1 KNOWS HER PAYOFFS, THAT IS, HER TYPE (a OR b).
PLAYER 2 DOES NOT KNOW WHETHER HE IS PLAYING
AGAINST TYPE-a PLAYER 1 OR TYPE-b PLAYER 1.

THAT IS, PLAYER 2 HAS PRIOR BELIEFS ABOUT PLAYER 1’S TYPE:
Example: IT IS COMMON KNOWLEDGE THAT PLAYER 2
BELIEVES THAT PLAYER 1 IS OF TYPE-a WITH PROBABILITY 0.65, OF
TYPE-b WITH PROBABILITY 0.35.

ALTHOUGH THERE ARE TWO PLAYERS IN THIS GAME, WE

THINK OF THE GAME AS IF IT WERE PLAYED BY THREE
PLAYERS: Player 1-a, Player 1-b and Player 2.

The extensive form representation of this game is as follows:

Nature

{0.65} {0.35}
1a 1b

X Y Y X
Player 2

Player 2

Left Right
Left Right Left Right Left Right

(5,0) (0,6) (6,10) (0,0) (6,10) (10,0) (5,6) (10,0)

WE CAN ALSO REPRESENT THE SAME GAME AS FOLLOWS:

2
 (5,0) (6,10)
L {μ} 1a {β} L
X Y
(0,6) R R (0,0)
{0.65}

#2 Nature #2

{0.35}
(5,6) (6,10)
L X Y L
{1- μ} 1b {1-β}
R R
(10,0) (10,0)

REMARKS
1. Player 2 does not know whether he is playing against type-a
or type-b of Player 1. This fact is represented by the
information sets (broken lines connecting the two nodes
after X is played and the two nodes after Y is played).

2. Suppose Player 2 has observed that action X is chosen by

player 1. Player 2 will now choose between “Left” and
“Right”: We assume he has BELIEFS, represented by the
PROBABILITY {μ} which he assigns to the possibility that
he is facing type-a of Player 1.

3. When Player 2 finds herself at the information set following

action X (chosen by Player 1), we can compute Player 2’s
expected payoffs as a function of his beliefs:

If Player 2 plays “Left”, he gets:

3
 μ . (0) + (1 – μ) (6).
If Player 2 plays “Right”, he gets:
μ . (6) + (1 – μ) (0).

Therefore Player 2 will play

“left” if μ < 1/2
“left” with any probability if μ = 1/2
“right” if μ > 1/2 .

(This means: If player 2 believes it is more likely that Player 1

is of type-a when he observes X is played, then Player 2 should
play “right”; otherwise he should play “left”).

4. Similarly, suppose Player 1 plays Y. Let Player 2’s belief

that Player 1 is of type-a be represented by the probability
{β}. We can compute Player 2’s expected payoffs as a
function of his beliefs:

If Player 2 plays “Left”, he gets:

β . (10) + (1 – β) (10) = 10.
If Player 2 plays “Right”, he gets:
β . (0) + (1 – β) (0) = 0.

Note that Player 2’s optimal action when he observes that

player 1 has played Y does not depend on his beliefs: No
matter player 1’s type (a or b) Player 2 will get 10 if he plays
“Left”, 0 if he plays “right”. Therefore Player 2 plays “left” if
he observes that player 1 played “Y”.

5. BUT... Where do Player 2’s beliefs come from? What

criterion shall we use in saying, for example “Player 2
believes that he faces type-a of player 1 with probability
0.75, when X is played”? The answer is that we try to infer
Player 2’s beliefs from the strategies of the two types of
player 1.

4
Consider for example the following strategies for the two types
of Player 1:
Type-a: “Play X with probability 0.75, play Y with probability
0.25”
Type-b: “Play X with probability 1, play Y with probability 0.”
The question for player 2 is:

What is the probability that Player 1 is of type-a, given that X

has been played?

μ = Prob(type-a | X) =

Probability that #1 is type-a AND plays X

Probability that X is played (by both types)

(0.65).(0.75) 0.4875
= = = 0.582
(0.65).(0.75) + (0.35)(1) 0.8375

The rule we used above to compute prob(type-a | X) is called

Bayes’ Rule. WHEN PLAYER 2’S BELIEFS ARE OBTAINED
THROUGH BAYES’ RULE (USING PLAYER 1’S
STRATEGY AND PRIOR BELIEFS) WE SAY THAT
PLAYER 2’S BELIEFS ARE CONSISTENT WITH
STRATEGIES.

In the calculation above,

(Probability that #1 is of type-a AND plays X) =
(Probability that #1 is of type-a).(Probability that type-a
plays X) = (0.65).(0.75) = 0.4875.

Probability that X is played (by both types) =

(Probability that #1 is of type-a AND plays X) +
(Probability that #1 is of type-b AND plays X) =

5
(0.65)(0.75) + (0.35)(1) = 0.8375

Under these strategies of the two types of Player 1, the rational

Player 2 should believe that he faces type-a with probability
0.582 when he observes X is played.

Note that we have Prob(type-b | X) = 1 - Prob(type-a | X) .

(Given that X is observed by #2, the probability that #2 faces

type-b is equal to 1 - Prob(type-a | X) ).

Under the strategies for the two types of player 1, what should
Player 2 believe when he observes that Player 1 has played Y?
Using Bayes’ rule once again, we get:

β = Prob(type-a | Y) =

Probability that #1 is type-a AND plays Y

Probability that Y is played (by both types)

(0.65).(0.25) 0.4875
= = = 1
(0.65).(0.25) + (0.35)(0) 0.4875

Thus, if “Y” is observed, under the strategies of the two types,

Player 2 must infer that he definitely is playing against type-a.

6
THE GAME PRESENTED ABOVE BELONGS TO A CLASS
OF GAMES CALLED SIGNALING GAMES.

A Signaling game
(i) is played by two players;
(ii) one player knows his payoffs (is the informed player),
the other does not know the informed player’s payoffs
(is the uninformed player); both players know the
uninformed player’s payoffs. This is represented by a
move by “Nature” at the beginning of the game,
choosing the informed player’s payoffs (type) which is
observed only by the informed player.
(iii) The uninformed player has “prior beliefs” about the
informed player’s type (payoffs), given by the
probability with which Nature determines the types of
the informed player.
(iv) The informed player moves first, and chooses an
action.
(v) The uninformed player observes this action (tries to
infer the informed player’s type) and acts next, by
choosing an action himself.
(vi) The game ends.

TYPES OF EQUILIBRIA IN SIGNALING GAMES:

1. POOLING EQUILIBRIA
2. SEPARATING EQUILIBRIA

7
 1. IN A POOLING EQUILIBRIUM,

BOTH TYPES OF THE INFORMED PLAYER CHOOSE

THE SAME ACTION... Therefore the uninformed player
learns nothing new about the type of the informed player.
(In our example, type-a and type-b of player 1 both choose
the same action (X or Y) with probability one.)

2. IN A SEPARATING EQUILIBRIUM,

EACH TYPE OF THE INFORMED PLAYER CHOOSES A

DIFFERENT ACTION... Therefore the uninformed player
learns the type of the informed player, after observing the
action chosen by Player 1.
(In our example, the two types of player 1 choose different
actions; for instance, type-a plays X and type-b plays Y with
probability one.)

EQUILIBRIUM CONCEPT FOR SIGNALING GAMES

A SEQUENTIAL EQUILIBRIUM (or, Perfect Bayesian

Equilibrium) consists of:
 a strategy for each type of player 1 (informed player)
 a strategy for player 2 (uninformed player)
 Player 2’s beliefs about player 1’s type when he moves
SUCH THAT:

* Player 2’s strategy is optimal GIVEN his beliefs

* Player 2’s beliefs are CONSISTENT with strategies of player
1’s types,
* The strategy of each type of Player 1 is optimal given Player 2’s
strategy.

8
LET US SEARCH FOR A SEQUENTIAL
EQUILIBRIUM IN OUR GAME:
(5,0) (6,10)
L {μ} 1a {β} L
X Y
(0,6) R R (0,0)
{0.65}

#2 Nature #2

{0.35}
(5,6) (6,10)
L X Y L
{1- μ} 1b {1-β}
R R
(10,0) (10,0)

Recall, we found the optimal strategy of Player 2 as a function

of his beliefs:
When he observes “Y” he plays “Left” no matter his beliefs;
When he observes “X”, he plays “Left” only if μ < 1/2.

Can there be a Pooling equilibrium where both types of player

1 play “X”?
If both types of player 1 play “X”, player 2’s CONSISTENT
belief about player 1’s type are computed using Bayes’ rule:

μ = (0.65)(1) / [(0.65)(1) + (0.35)(1)] = 0.65.

Then, Player 2’s optimal response is to play “Right” because μ

> 0.5.
Type-a of Player 1 gets 0, type-b gets 10.

9
 To have a sequential Pooling equilibrium, neither type-a nor
type-b of Player 1 should not deviate… Where? There is only
one alternative: to “Y”.
Obviously, type-b will not deviate because her payoff is the
largest payoff she can get in this game.
But look at type-a: If she deviates to “Y”, Player 2 will play
Left, and so type-a will get 6. This is larger than “0”, so type-a
will deviate.
We conclude that we don’t have a pooling equilibrium in
which both types of player 1 play “X”.

Can there be a Pooling equilibrium where both types of player 1

play “Y”?

If both types of player 1 play “Y,” Player 2 gets 10, type-a gets 6
and type-b gets 6.

Obviously type-a will not deviate to “X” because she cannot hope
to get a payoff larger than 6 (the maximum she can get is 5).

Consider type-b of player 1: She would prefer deviating to “X” if

she knows that player 2 will play “right” when he observes “X”
(so that she gets 10). BUT what will player 2 do if he observes
“X”?

We have shown that his optimal action depends on his beliefs

(choose left only if μ < 0.5).

Can we know what his beliefs are when X is chosen?

Bayes’ rule does not help us because according to the strategies of

the two types of player 1, action “X” is played with probability
zero! In fact, applying Bayes’ rule yields

10
μ = (0.65)(0) / [(0.65)(0) + (0.35)(0)] = 0/0 = ? Indeterminate…

When Bayes’ rule is of no help to us in computing beliefs, we have

to postulate them. We then ask:

“Can we find beliefs for player 2, such that NEITHER type-a

NOR type-b of Player 1 will not deviate to X?”

The answer is “yes”… If Player 2 believes μ < 0.5, Player 2 will

choose “left” when he observes X. Then, if type-b deviates she
will get 5, instead of 6.

We found a pooling sequential equilibrium:

Type-a of player 1: Play Y with probability one;

Type-b of player 1: Play Y with probability one;
Player 2: Play “left” if Y is observed, play “left” if X is observed;
Beliefs: If X is observed, any μ < 0.5; If Y is observed, β = 0.65.

HOW DO WE VERIFY EXISTENCE OF POOLING

EQUILIBRIA?
For each strategy of the informed player (Player 1), we
check whether we can construct an equilibrium in which
-Player 2 responds optimally to Player 1’s strategy, given
that her beliefs do not change (because all types of the informed player
play the same strategy; this common strategy yields no new information)
-None of the types of the informed player deviates, given
player 2’s optimal response.

DID WE CHECK ALL POSSIBLE POOLING EQUILIBRIA IN

THIS GAME?

11
 YES…
Because there are two strategies Player 1 can play, X and Y, and
we checked whether a Pooling equilibrium exists for each of these
strategies.

Can there be a Separating equilibrium (where each type

of player 1 choose a distinct strategy?

(5,0) (6,10)
L {μ} 1a {β} L
X Y
(0,6) R R (0,0)
{0.65}

#2 Nature #2

{0.35}
(5,6) (6,10)
L X Y L
{1- μ} 1b {1-β}
R R
(10,0) (10,0)

Remark: In a separating equilibrium, the uninformed player

learns the type of player 1 by observing Player 1’s action. This
means that Player 2’s updated beliefs μ and β are either equal to 1
or 0.

How do we proceed? By trial and error. SO:

POSSIBILITY 1: Type-a plays X while type-b plays Y in a
sequential equilibrium (?)

12
 If type-a plays X and type-b plays Y, and if this is an
equilibrium, Player 2 must believe μ = 1 when she observes X and
β = 0 when she observes Y. Therefore (according to her best
response given these beliefs),
Player 2 should play “Right” if she observes X, (which leads
to the payoffs (0) for type-a and (6) to Player 2), “Left” if she
observes Y (which leads to the payoffs (6) for type-b and (10) for
Player 2).

We now turn to the two types of Player 1 and check whether

they would deviate from their strategies given this expected
response of Player 2.
By deviating to “Y”, type-a can get 6 (instead of 0, which he
gets by playing “X”): Therefore type-a will deviate. We cannot
have such a separating equilibrium… No need to check anything
else.

POSSIBILITY 2: Type-a plays Y while type-b plays X in a

sequential equilibrium (?)

If this is what Player 1’s two types play, then μ = 0 when Player 2
observes X and β = 1 when Player 2 observes Y. Therefore
(according to her best response given these beliefs),
Player 2 should play “Left” if she observes X, (which leads
to the payoffs (5) for type-b and (6) to Player 2), “Left” if she
observes Y (which leads to the payoffs (6) for type-a and (10) for
Player 2).
But observe that now type-b will deviate, from “X” to “Y”
and get (6) instead of (5). (see the lower part of the game below)

(5,6) (6,10)
L X Y L
{1- μ} 1b {1-β}
R R
(10,0) (10,0)

13
Conclusion: This game has no separating equilibria.
In other words, there is no equilibrium in which Player 2 can
learn the type of Player 1 by observing Player 1’s action (X or Y).
We say that “there is no signaling of information.”

The game has one pooling equilibrium in which both types of

Player 1 play “Y”, to which Player 2 responds with “Left”.

A SPECIAL SIGNALING GAME: (CHEAP TALK)

The general idea behind the examples considered below is the

following:
THERE IS ABSOLUTELY NO DIFFERENCE BETWEEN
PLAYING X AND Y FOR EACH TYPE OF PLAYER 1:
PAYOFFS OF THE INFORMED PLAYER DEPEND ONLY ON
THE UNINFORMED PLAYER’S ACTION. IN THIS CASE, WE
SAY THAT THE ACTIONS OF THE INFORMED PLAYER
ARE “CHEAP TALK.”

THE QUESTION IS: CAN THE INFORMED PLAYER’S

ACTION REVEAL ANY INFORMATION TO THE
UNINFORMED PLAYER (about the informed player’s type)?

In the game above, this means that type-a’s and type-b’s payoffs
depend only on the action of Player 2 (“left” or “right,”): Type-a
gets the same payoff from X and Y, as long as Player 2 chooses
“left,” and he gets the same payoff from X and Y as long as
Player 2 chooses “right”. The same is true for type-b.

(Remark: There are other definitions of “cheap talk:” Some game

theorist call “cheap talk” any communication before the game is
played. Whatever the definition, one common element stands out:

14
Cheap talk itself is a costless action, and a costless action cannot
have a direct impact on your payoff.)

A CHEAP-TALK RESULT:
Suppose that Player 1’s strategies are “cheap talk.” If type-a and
type-b both prefer Player 2 to play the same strategy, then player
1’s strategy is irrelevant for the equilibrium outcome. It has no
information content: Cheap talk CANNOT signal anything in this
case.

(5,0) (5,10)
L {μ} 1a {β} L
X Y
(0,6) R R (0,0)
{0.65}

#2 Nature #2

{0.35}
(6,6) (6,10)
L X Y L
{1- μ} 1b {1-β}
R R
(0,0) (0,0)

In this game, Player 1’s strategy is “cheap talk.” For example,

given that Player 2 plays “left,” type-a of Player 1 gets the same
payoff (5) from playing X and Y. The same is true for type-b (she
always gets (6) if player 2 plays “left”).
In the sequential equilibrium of this game, the action of
Player 1 cannot reveal any information about the type of Player
1. Therefore, any sequential equilibrium of this game must be of
the “POOLING” type.

15
Real-life example: Player 1 is an accused suspect (informed about
whether he is guilty or not) and Player 2 is the judge. Player 1 has
two possible types: Type-a is guilty, type-b is innocent. Each type
of Player 1 has two actions, “X = confess” and “Y = deny.” The
judge, Player 2, has two actions, “Convict” and “Acquit.” Both
types of player 1 only care about whether they are acquitted of
convicted, and moreover they both prefer to be acquitted. This
game does not have any separating equilibria. See the
representation below:

(5,0) (5,-2)
Acquit {μ} Guilty {β} Acquit
Confess Deny
(0,6) Convict Convict (0,6)
{0.65}

Judge Nature Judge

{0.35}
(6,6) (6,6)
Acquit Confess Deny Acquit
{1- μ} Innocent {1-β}
Convict Convict
(0,0) (0,0)

As an exercise, (i) show that this game has no separating

equilibria and (ii) find its pooling equilibrium (or equilibria).

CHEAP-TALK: A SIGNALING GAME THAT REVEALS US

THE IMPORTANCE OF INTERPRETATION OF LANGUAGE:
Suppose that Player 1’s strategies are “cheap talk.” Suppose,
further, that type-a prefers that Player 2 plays “left” while type-b
prefers that Player 2 plays “right.” Finally, Player 2 obtains a
higher payoff by playing left against type-a, right against type-b.

16
 Under these preferences (payoffs), IF player 2 KNEW the
type of Player 1, the games that Player 2 plays against each type
of Player 1 would have different SPE outcomes:

Type-a Type-b

X Y X Y
#2 #2 #2 #2
left right left right
left right left right

In the game at the left, if Player 2 knows that she plays against
type-a, then she will play left. Since type-a’s action is cheap talk,
he only cares about the response of player 2; given that player 2
plays left, he is indifferent between X and Y. So, (X,left) and
(Y,left) are both SPE outcomes. For the game at the right,
(X,right) and (Y,right) are both SPE outcomes.

Given this, let us now consider the “cheap talk

signaling game” under incomplete information:
Type-a: prefers to be employed in the finance department
Type-b: prefers to be employed in the accounting department
X = To say “Employ me in the finance department”
Y = To say “Employ me in the Accounting department”

Player 2 (the boss) prefers to employ player 1 in the department

in which he is more effective.
Left = employ player 1 in finance department
Right = employ player 1 in accounting department

Suppose that the wages in both departments are the same.

17
This game can be represented as follows (exact values of payoffs
is not important, so long as the order of the players’ preferences
are respected)

(5,5) (5,5)
L {μ} 1a {β} L
X Y
(0,2) R R (0,2)
{0.65}

#2 Nature #2

{0.35}
(0,1) (0,1)
L X Y L
{1- μ} 1b {1-β}
R R
(6,6) (6,6)

This game has MANY equilibria:

Pooling Equilibrium 1: Both types of player 1 play X, so that (by

Bayes’ rule) μ = 0.65. Then, player 2’s optimal response is to play
“L” because
“L” yields the expected payoff (5)(0,65) + (1)(0.35) = 3.60,
“R” yields the expected payoff (2)(0.65) + (6)(0.35) = 3.40.
As a result, type-a obtains (5) and type-b obtains (0).

Player 2 must also have a plan of action when he observes “Y.” In

a sequential equilibrium, this plan of action must be optimal
GIVEN her beliefs {β} about player 1’s type when she observes
“Y”. Let us find that optimal response (plan): By playing “L”
player 2 expects the payoff (5)(β) + (1)(1-β), whereas “R” yields
(2)(β) + (6)(1-β). Therefore Player 2 plays “L” if β > 5/8.

18
Since neither type of player 1 plays “Y,” we cannot compute β by
using Bayes’ rule; we are free to assume values for β. So we can
ask, can we find values for β such that neither type-a
nor type-b will deviate from “X” to “Y”? If yes, the
strategies above will form a pooling equilibrium.
Type-a of player 1 has no gain from deviating to “Y” no
matter what player 2 does when she observes “Y.” But type-b
would prefer to deviate to “Y” IF player 2 plays “R” when she
observes “Y.” Player 2 will not play “R” for beliefs β > 5/8.

The pooling equilibrium is:

Type-a: play “X”
Type-b: play “X”
Player 2: If player 1 plays “X”, play “L.” If player 1 plays “Y,”
play “L.”
Beliefs: μ = 0.65, and β > 5/8.

This is an equilibrium because no player deviates given others’

strategies and beliefs, AND the beliefs of player 2 are consistent
with the strategies (they are computed using Bayes’ rule
whenever possible)

Interpretation: What sense can we make of this pooling

equilibrium? When both types of player 1 say ““Employ me in
the Accounting department,” the boss (player 2) believes that
player 1 is more effective in finance department with probability
at least 5/8, therefore, employs player 1 in the finance
department. (this may sound rather odd to you, but what we
understand from language is a cultural matter.) Given this belief
and behavior of the boss, both types of player 1 play X: they
declare “Employ me in the Finance department,” and the boss
employs player 1 in the finance department (no matter his type).

19
Pooling Equilibrium 2: Both types of player 1 play Y, so that (by
Bayes’ rule) β = 0.65. Then, player 2’s optimal response is to play
“L” because
“L” yields the expected payoff (5)(0,65) + (1)(0.35) = 3.60,
“R” yields the expected payoff (2)(0.65) + (6)(0.35) = 3.40.
As a result, type-a obtains (5) and type-b obtains (0).

It is an exercise left to you to show that the following strategies

and beliefs constitute a pooling equilibrium:

Type-a: play “Y”

Type-b: play “Y”
Player 2: If player 1 plays “X”, play “L.” If player 1 plays “Y,”
play “L.”
Beliefs: β = 0.65, and μ > 5/8.

Interpretation: Similar to the previous pooling equilibrium. Now,

when both types of player 1 say ““Employ me in the finance
department,” the boss believes that player 1 is more effective in
finance department with probability at least 5/8; given this belief,
it becomes optimal for the boss to employ player 1 in the finance
department when he is told X = “employ me in the finance
department”.
In this equilibrium, both types of player 1 declare Y =
“Employ me in the Accounting department,” and the boss
employs player 1 in the finance department (no matter his type).
Since both types of player 1 declare this same wish, the boss
learns nothing new about the type of player 1. Her initial beliefs
are unchanged (continues to believe that player 1 is more
effective in finance department with probability 0.65). Given this
belief, it is optimal for the boss to employ player 1 in the finance
department.

I urge you to look at the payoffs in the signaling game and

think more about plausibility of this pooling equilibrium.

20
Separating Equilibrium: Type-a of player 1 plays X, and type-b
plays “Y.” By Bayes’ rule, β = 0 and μ = 1. Then, player 2’s
optimal response is to play “L” if she observes “X” and “R” if she
observes “Y.” The equilibrium path is given below:

(5,5) (5,5)
L {μ=1} 1a {β=0} L
X Y
(0,2) R R (0,2)
{0.65}

#2 Nature #2

{0.35}
(0,1) (0,1)
L X Y L
{1- μ} 1b {1-β}
R R
(6,6) (6,6)

If 1a deviates to “Y”, he gets 0 instead of 5, and if 1b deviates to

“X” he gets 1 instead of 6. The equilibrium is:

Type-a: play “X”

Type-b: play “Y”
Player 2: If player 1 plays “X”, play “L.” If player 1 plays “Y,”
play “R.”
Beliefs: β = 0, and μ = 1.

Interpretation: In this equilibrium, type-a, who is more effective

in the finance department declares X = “Employ me in the
Finance department,” while type-b, who is more effective in the
accounting department declares Y = “Employ me in the
accounting department.”

21
 The boss employs type-a in the finance department, being
sure that only type-a can ask to be employed in the finance
department; she employs type-b in the accounting department.
IN THIS EQUILIBRIUM, THE TYPES OF PLAYER 1 TELL
THE TRUTH ABOUT THEIR PREFERENCES.

CONCLUSION FOR THE ENTIRE ANALYSIS OF THIS

GAME:
HOW PLAYERS INTERPRET LANGUAGE IS
COMMON KNOWLEDGE IN A SPECIFIC CULTURE. THIS
EXAMPLE HAS SHOWN THAT MANY DIFFERENT TYPES
OF EQUILIBRIA MAY EMERGE, DEPENDING ON THE
COMMON INTERPRETATION OF LANGUAGE.
NEVERTHELESS, THE SEPARATING EQUILIBRIUM IS
EFFICIENT, WHILE THE POOLING EQUILIBRIA ARE
NOT… CULTURES SHOULD ADOPT THE MOST
EFFECTIVE INTERPRETATION OF LANGUAGE, THE ONE
WE HAVE IN THE SEPARATING EQUILIBRIUM.

OTHER EXAMPLES OF SIGNALING GAMES

International Relations example:

Player 1 is Saddam Hussein (before the Iraq invasion) and
Player 2 is George Bush.
Saddam has two possible types: Type-a (SaddamWep) has
weapons of mass destruction while type-b (SaddamNoWep) has
none.
Bush does not know Saddam’s type; he will choose between
attacking and not attacking after observing Saddam’s action and
updating his beliefs about Saddam’s type. Saddam has to possible
actions: “X = Admit inspectors of weapons,” and “Y = Claiming
no possession of weapons and rejecting inspectors.”

22
 Both types of Saddam prefer Bush not to attack Iraq. But
Bush obtains different payoffs from attack and not attack,
depending also on whether Saddam has or not mass destruction
weapons. First number is Saddam’s payoff, second is Bush’s
payoff.

(-3,2) (-2,-1)
Attack {μ} SaddamWep {β} Attack
Admit Deny
(1,1) Don’t Don’t (8,-4)
{0.75}

Bush Nature Bush

{0.25}
(-1,0) (-1,1)
Attack Admit Deny Attack
{1- μ} SaddamNoWep {1-β}
Don’t Don’t
(3,2) (5,-2)

SaddamWep: prefers to deny inspectors whether or not Bush

attacks.
SaddamNoWep prefers to deny inspectors if Bush attacks for
sure, prefers to admit inspectors if Bush does not attack for sure.

Bush: prefers to attack if Saddam denies. If Saddam admits,

Bush prefers to attack only if Saddam has weapons of mass
destruction.

A Pooling equilibrium:

SaddamWep: deny inspectors

SaddamNoWep: deny inspectors
Bush: Attack if Saddam denies or admits.

23
Beliefs of Bush: β = 0.75, μ > 2/3.

Exercise: Show that this game has no pooling equilibria in which

both types of Saddam admit inspectors.

Exercise: Show that this game has no separating equilibria.

Labor Market Signaling example

Type-a of player 1: A high-ability person (potential worker)

Type-b of player 1: A low-ability person (potential worker)
Player 2: A firm

X = Get university education

Y = Don’t get university education

The firm does not know the ability of the worker, observes the
education choice of the worker and offers a competitive wage.

(Under a competitive wage offer the firm makes zero profits from
hiring the worker… This is approximately the case if there are
many firms competing for the worker).

Payoffs:
Worker: Wage – cost of education (if any) – Z
Where Z is the disutility of working.

Firm: 20 – wage (if the worker has high ability)

10 - wage (if the worker has low ability)

24
Thus, if the firm believes that the worker has high ability with
probability p, its expected payoff from hiring the worker is
Firm’s expected profit: 20.p + 10.(1-p) – wage.

The competitive wage is the wage which makes this expected

profit equal to zero:

Wage = 20.p + 10.(1-p).

Suppose Z = 5, and
the cost of university education is:
6 if the worker has high ability,
11 if the worker has low ability.
We assume for simplicity that education does not increase the
ability of the worker; ability is innate.

Can there be a Separating Equilibrium in which “High

Ability” chooses education and “Low Ability” chooses no
education?

L {μ=1} High ability {β=0}

wage Education No wage
(wage-11, (wage-5,
20-wage) {0.65} 20-wage)

Firm Nature Firm

{0.35}
wage Education No wage
(wage-16, {1- μ} Low ability {1-β} (wage-5,
10-wage) 10-wage)

25
High ability: get university education
Low ability: DON’T get university education
Firm: Offer the competitive wage W = 20 if you observe that the
worker has university education, Offer the competitive wage W =
10 if you observe that the worker does not have university
education.
Beliefs: β = 0, μ = 1.

Let us verify that these strategies and belief system constitute a

separating equilibrium:
(i) Given the firm’s wages and beliefs, the high-ability
worker is getting the payoff 9 = 20 –11. If he deviates to
no education, he gets 10 –5 = 5. Thus, high-ability
worker will not deviate. The low-ability type is getting
5= 10 – 5, if he deviates to getting university education
his payoff becomes 4 = 20 –16. Therefore the worker
will not deviate.
(ii) The firm is offering a competitive wage 20 when μ = 1,
the wage 10 when β = 0.

THUS, CHOICE OF AN EDUCATION LEVEL CAN SERVE TO

SIGNAL INNATE ABILITIES…

Question:
We assumed in the analysis that the firm makes a
competitive wage offer (and justified this by saying that there
could be many other firms competing for the same worker).
What would happen to the separating equilibrium above if,
instead, we assume that the firm makes a “take-it-or-leave-it”
(monopsony) offer?
A “take-it-or-leave-it” wage offer is a wage that leaves the worker
indifferent between working in the firm and not working (assume
that by not working, the worker can guarantee the payoff “0”). In
this example, the “take-it-or-leave-it” wage offer is wage = 5.

26