cheap talk. when can cheap talk be believed? we have discussed costly signaling models like...

48
Cheap Talk

Upload: nickolas-woodcox

Post on 14-Dec-2015

217 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Cheap Talk

Page 2: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

When can cheap talk be believed?

• We have discussed costly signaling models like educational signaling.

• In these models, a signal of one’s type is credible if the cost of a signal differs between types and it doesn’t pay to send a false signal.

• But what can be learned if there is no cost to anyone from sending a signal.

• When will senders tell the truth and receivers believe what they are told?

Page 3: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Signaling intent

• Consider a simultaneous game in which one or more players are allowed to say how they are going to play.

• Will they tell the truth? • Will others pay attention to what they say?

Page 4: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Example:

• In Rock, Paper, Scissors, Bart gets to say what he is going to do on the next play, then gets to choose what to do.

• What would Bart do?• How would Lisa respond?

Page 5: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Babbling Equilibrium

• Message sender sends a completely uninformative message.

• Receiver ignores it. • In a pure conflict game, like RPS, this is the

only equilibrium. • If sender’s signal was at all informative, it

would be used to his disadvantage.

Page 6: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Common interest games

• In some games, the players have a common interest.

• If Player A gets a higher payoff when Player B knows how he will move than when Player B does not, it is in the interest of A to correctly inform B of what he will do and in the interest of B to believe A.

Page 7: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Dressing for the ball

Page 8: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

The story

• Players are the Countess and the Duchess• They are going to a formal ball.• Each has two favorite dresses, a red dress or a

blue dress. • Problem is they use the same designer. Their

red dresses are identical and so are their blues.• Both would be humiliated if they wore identical

dresses.

Page 9: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

A common interest game: Dressing for the Ball

Red Dress Blue Dress

Red Dress -10, -10 20, 20Blue Dress 20, 20 -10,-10

Duchess

Countess

Page 10: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Nash equilibrium

• There are two asymmetric equilibria in pure strategies.

• But if they play only once, how do they find it?• For single shot play, symmetric equilibrium

seems more likely.• Lets look for a symmetric Nash equilibrium in

mixed strategies.

Page 11: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

If the countess wears a red dress with probability ¾, the best response for the duchess is to wear a red dress with probability:

A) 1/4B) 3/4C) 1/2D) 0

Red Dress Blue DressRed Dress -10, -10 20, 20Blue Dress 20, 20 -10,-10

Page 12: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Symmetric equilibrium?

There is a symmetric Nash equilibrium in which duchess and countess each play the mixed strategy wear a red dress with probability p A) For any p less than 1/2B) For any p greater than 1/2C) Only if p=1/2D) Only if p=0E) Either if p=1 or p=0.

Page 13: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

What is the expected payoff to each player if each flips a fair coin to decide the color of her dress?

A) 15B) 5C) 12.5D) 10E) -5

Red Dress Blue DressRed Dress -10, -10 20, 20Blue Dress 20, 20 -10,-10

Page 14: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

How about messages?

• What do you think would happen if only the Countess can send a message?

• Countess decides what color to say she’s wearing and also what she does wear.

• Duchess hears Countess’s message and decides what to wear.

Page 15: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Possible Pure Strategies

• For Countess:Say Red, Wear RedSay Red, Wear BlueSay Blue, Wear RedSay Blue, Wear Blue

• For Duchess:Wear Blue if C says Red, Red if C says BlueWear Blue if C says Red, Blue if C says BlueWear Red if C says Red, Blue if C says BlueWear Red if C says Red, Red if C says Blue

Page 16: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Suggested exercise

• Draw an extensive form representation of this game

• Write out the strategic form and find subgame perfect Nash equilibria.

Page 17: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

A Nash equilibria

• Countess plays: Say “I’ll wear Red” and she wears Red

• Duchess plays: wear Blue if C says “I’ll wear Red”, and wear Red if C says “I’ll wear Blue”.

• Show that this is a N.E.• What other Nash equilibria can you find?

Page 18: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Another Nash equilibrium

• Countess says “I’ll wear red, then flips a coin to decide what to wear.

• Duchess pays no attention to what Countess says, flips a coin herself.

• This kind of equilibrium is known as a babbling equilibrium.

Page 19: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

An “eccentric” Nash equilibrium

• Duchess says “I’ll wear red”, then wears blue.• Countess plays “Wear color that Duchess

claims she will wear.”• This is an equilibrium. Duchess always “lies”Countess believes that duchess will “lie” and acts accordingly. • What does it mean when Duchess says “Red”?

Page 20: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Simultaneous messages

• Why should one of them get to move first?

• Suppose that the duchess and the countess each get to send one message to the other.

• Neither knows what the other’s message says when she sends hers.

Page 21: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Single messages sent simultaneously

• A symmetric Nash equilibrium:• Each flips a coin and tells the other “I will

wear red” or “I will wear blue” with probability ½. If they each said a different color, they wear what they said they would. If they said the same color, they each toss a coin to decide what to wear.

• Check that this is a Nash equilibrium

Page 22: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

If they each use the single message strategy discussed in previous slide, what is the probability

that they wear different colors to the ball?

A) ½B) 1C) ¼D) ¾E) 2/3

Page 23: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

A second message?

• Suppose that if they say same color on first message, they get a chance to send a second message in an attempt to coordinate.

• What would a symmetric equilibrium look like?

• What would be the chances of wearing different dresses?

Page 24: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Conflicting InterestsDressing for the Ball

Red Dress Blue Dress

Red Dress 10, -10 0, 10Blue Dress 0, 10 10,-10

Duchess

Social Climber

What are the equilibria if there is no pre-ball communication?

Page 25: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

One player sends signal

Suppose Duchess sends a message to the social climber saying what she will wear.

Can the duchess gain by lying? What will the social climber make of what she says?

Is any informative message an equilibrium?

What about babbling?

Page 26: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Partially Conflicting InterestsRed preferred

Red Dress Blue Dress

Red Dress -10, -10 20, 0Blue Dress 0, 20 -10,-10

Duchess

Countess

What is the mixed strategy equilibrium if there is no pre-ball communication?

Page 27: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Finding symmetric mixed equilibrium

• Payoff to countess if duchess wears red with probability p – Wearing red: -10p+20(1-p)=20-30p– Wearing blue 0p-10(1-p)= 10p-10

• Countess will mix if 20-30p= 10p-10, so p=3/4.• By symmetry, each will mix if the other wears red with

probability ¾.• In this equilibrium, each gets a payoff of 10p-10= -2.5

Page 28: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Simultaneous message case

• Suppose each sends a message, “Red” or “Blue”.– If messages are different, each wears what she said– If messages are the same, each wears red with

probability ¾.

Page 29: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Strategic form with one round of talk

Say Red Say Blue

Say Red -2.5, -2.5 20, 0

Say Blue 0, 20 -2.5,-2.5

• Expected payoffs to countess if duchess says “red” with probability p,• Say “red” -2.5p+20(1-p)=20-22.5p• Say “blue” 0p-2.5(1-p)= 2.5p-2.5

• These are equal when 20-22.5p= 2.5p-2.5 or 25p=22.5, which implies p=9/10.

Page 30: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Alice and Bob without talk

Bob

Go to A Go to B

Go to A

Alice Alice

Go to B Go to A Go to B

23 0

011

32

Page 31: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Alice and Bob

Go to Movie A Go to Movie B

Go to Movie A 3,2 1,1

Go to Movie B 0,0 2,3

Alice

Bob

• Mixed strategy equilibrium: Alice goes to A with probability p such that 2p= p+3(1-p), so p=3/4.• Similar reasoning finds Bob goes to B with

probability 3/4

Page 32: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Nash equilibrium

• Mixed strategy equilibrium. Bob goes to B with p=3/4, Alice goes to A with probability 3/4.

• Probabilities: Meet at A 3/16: Meet at B 3/16

• Probability they find each other is only 3/8.• Expected payoff to each is (3/16)3+(3/16)2+(9/16)1+(1/16)0=3/2

Page 33: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Talking it over

Suppose Bob gets to say where he is going and Alice doesn’t get to say anything.

What do you think would be an equilibrium?

Page 34: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Two-way conversation, single message

• Each gets to send the other a single message, suggesting which movie to go to, then decide where to go.

• Suggested equiibrium: If both say same movie, they both go there. If they name different movies, they play original mixed strategy game.

• Draw extensive form tree.

Page 35: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Game of simultaneous messages

• Pure strategies, at first decision node– Say I am going to A– Say I am going to B

• After hearing other person’s message (and one’s own) go to one movie or the other.

• Sample strategy for BobSay A, If Alice says A, go to A. If Alice says B, go to to B with probability ¾.

Page 36: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

A Nash equilibriumin mixed strategies

• With probability p, say “I am going to A” and with probability (1-p) say “I am going to B.”

• If both say they are going to same place, they both go there. If they say different things, they ignore the conversation and play mixed strategy for which movie to attend.

Page 37: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Talking game: Abbreviated payoff matrix

Say Movie A Say Movie B

Say Movie A 3, 2 3/2,3/2Say Movie B 3/2,3/2 2,3Alice

Bob

If both say same movie, they both go there. If they say different movies,They play original mixed strategy game.

Page 38: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Mixed strategy equilibrium for this game

• If Alice says Movie A with probability p,Then Bob’s payoff from saying “movie A” is 2p+(3/2)(1-p) and his payoff from saying “Movie B” is 3(1-p)+(3/2)p.These are equal if 3/2+1/2p=3-(3/2)p, which implies p=3/4.

Alice

Bob

Page 39: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Payoffs

• With probability 3/16, they both say A and go to A

• with probability 3/16, they both say B and go to B.

• With probability 10/16, they say different things from each other and play original mixed strategy equilibrium.

• Expected payoffs :3(3/16)+2(3/16)+1.5(10/16)=15/8>3/2.

Page 40: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Mixed strategy equilibrium for Talking Game

• If Bob says “movie A” with probability q, when will Alice be willing to use a mixed strategy?

• Her expected payoff from saying Movie A is 3q+3/2(1-q) and her expected payoff from saying B is 3/2q+2(1-q).

• These are equalized when q=1/4. In a mixed strategy equilibrium, Bob says A with probability ¼ and B with probability ¾.

• Symmetric argument shows that Alice says A with probability ¾ and B with probability ¼.

• Probability they both say the same thing is therefore 3/16+3/16=3/8.

Page 41: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

What is probability they get together?

• With probability 3/8, they agree on where to go. If they don’t agree, then they play the no communications mixed strategy equilibrium and meet with probability 3/8.

• So probability they meet is 3/8+5/8(3/8)=39/64Simple talk helped, but didn’t completely solve the problem.Would more talk help?

Page 42: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Adding further rounds of discussion

• Suppose that if first set of messages do not say same place, they try again.

• Then if second set do not coincide they try yet again, and so on.

• Is this a reasonable model of an argument?

Page 43: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Things to think about

• Why bother to talk? Only pays if others listen.• Why listen if all you hear is nonsense or lies.• Why do politicians lie?• Do some voters pay attention to what they

say?• How did language evolve. Prevalence of

common interest games?• Why don’t more animals have more language?

Page 44: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Aesop’s Reason for Truth-telling

• There was once a young Shepherd Boy who tended his sheep at the foot of a mountain near a dark forest. It was rather lonely for him all day, so he thought upon a plan by which he could get a little company and some excitement. He rushed down towards the village calling out “Wolf, Wolf,” and the villagers came out to meet him, and some of them stopped with him for a considerable time.

Page 45: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

• This pleased the boy so much that a few days afterwards he tried the same trick, and again the villagers came to his help.

• But shortly after this a Wolf actually did come out from the forest, and began to worry the sheep, and the boy of course cried out “Wolf, Wolf,” still louder than before. But this time the villagers, who had been fooled twice before, thought the boy was again deceiving them, and nobody stirred to come to his help.

Page 46: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

The moral of the story

• . So the Wolf made a good meal off the boy’s flock, and when the boy complained, the wise man of the village said:

“A liar will not be believed, even when he speaks the truth.”

Page 47: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

Lesson for Game theory

A truth-telling equilibrium is more difficult to find in games that are played only once.

In the wolf story, the reason for being truthful when it is not too costly is that you are more likely to be believed when it is very important to be believed.

Page 48: Cheap Talk. When can cheap talk be believed? We have discussed costly signaling models like educational signaling. In these models, a signal of one’s

That’s all for today…