MS&E 246: Lecture 16 Signaling games - Stanford...
Transcript of MS&E 246: Lecture 16 Signaling games - Stanford...
![Page 1: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/1.jpg)
MS&E 246: Lecture 16Signaling games
Ramesh Johari
![Page 2: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/2.jpg)
Signaling games
Signaling games are two-stage games where:
• Player 1 (with private information)moves first.His move is observed by Player 2.
• Player 2 (with no knowledge of Player 1’s private information) moves second.
• Then payoffs are realized.
![Page 3: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/3.jpg)
Dynamic games
Signaling games are a key example of dynamic games ofincomplete information.
(i.e., a dynamic game where the entire structure is not common knowledge)
![Page 4: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/4.jpg)
Signaling games
The formal description:Stage 0:
Nature chooses a random variable t1,observable only to Player 1,from a distribution P(t1).
![Page 5: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/5.jpg)
Signaling games
The formal description:Stage 1:
Player 1 chooses an actiona1 from the set A1.
Player 2 observes this choice of action.(The action of Player 1 is also called a “message.”)
![Page 6: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/6.jpg)
Signaling games
The formal description:Stage 2:
Player 2 chooses an actiona2 from the set A2.
Following Stage 2, payoffs are realized:Π1(a1, a2 ; t1) ; Π2(a1, a2 ; t1).
![Page 7: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/7.jpg)
Signaling games
Observations:• The modeling approach follows Harsanyi’s
method for static Bayesian games.• Note that Player 2’s payoff depends
on the type of player 1!• When Player 2 moves first,
and Player 1 moves second,it is called a screening game.
![Page 8: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/8.jpg)
Application 1: Labor markets
A key application due to Spence (1973):Player 1: worker
t1 : intrinsic abilitya1 : education decision
Player 2: firm(s)a2 : wage offered
Payoffs: Π1 = net benefitΠ2 = productivity
![Page 9: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/9.jpg)
Application 2: Online auctions
Player 1: sellert1 : true quality of the gooda1 : advertised quality
Player 2: buyer(s)a2 : bid offered
Payoffs: Π1 = profitΠ2 = net benefit
![Page 10: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/10.jpg)
Application 3: Contracting
A model of Cachon and Lariviere (2001):Player 1: manufacturer
t1 : demand forecasta1 : declared demand forecast,
contract offerPlayer 2: supplier
a2 : capacity builtPayoffs: Π1 = profit of manufacturer
Π2 = profit of supplier
![Page 11: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/11.jpg)
A simple signaling game
Suppose there are two types for Player 1,and two actions for each player:
• t1 = H or t1 = LLet p = P(t1 = H)
• A1 = { a, b }• A2 = { A, B }
![Page 12: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/12.jpg)
A simple signaling game
Nature moves first:
Nature
![Page 13: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/13.jpg)
A simple signaling game
Nature moves first:
Nature
H
L
![Page 14: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/14.jpg)
A simple signaling game
Player 1 moves second:
Nature
1.1
1.2
H
L
![Page 15: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/15.jpg)
A simple signaling game
Player 1 moves second:
Nature
1.1
1.2
H
L
a
a b
b
![Page 16: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/16.jpg)
A simple signaling game
Player 2 observes Player 1’s action:
Nature
1.1
1.2
H
L
a
a b
b
2.22.1
![Page 17: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/17.jpg)
A simple signaling game
Player 2 moves:
Nature
1.1
1.2
H
L
a
a b
b
2.22.1
A
B
A
B
A
B
A
B
![Page 18: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/18.jpg)
A simple signaling game
Payoffs are realized: Πi(a1, a2; t1)
Nature
1.1
1.2
H
L
a
a b
b
2.22.1
A
B
A
B
A
B
A
B
![Page 19: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/19.jpg)
Perfect Bayesian equilibrium
Each player has 2 information sets,and 2 actions in each, so 4 strategies.
A PBE is a pair of strategies and beliefssuch that:-each players’ beliefs are derived from strategies using Bayes’ rule (if possible)-each players’ strategies maximize expected payoff given beliefs
![Page 20: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/20.jpg)
Pooling vs. separating equilibria
When player 1 plays the same action,regardless of his type,it is called a pooling strategy.
When player 1 plays different actions,depending on his type,it is called a separating strategy.
![Page 21: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/21.jpg)
Pooling vs. separating equilibria
In a pooling equilibrium,Player 2 gains no information about t1from Player 1’s message⇒ P2(t1 = H | a1) = P(t1 = H) = p
In a separating equilibrium,Player 2 knows Player 1’s type exactlyfrom Player 1’s message⇒ P2(t1 = H | a1) = 0 or 1
![Page 22: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/22.jpg)
An eBay‐like model
Suppose seller has an item with quality either H (prob. p) or L (prob. 1 - p).
Seller can advertise either H or L.Assume there are two bidders.Suppose that bidders always bid truthfully,
given their beliefs.(This would be the case if the seller used a second price auction.)
![Page 23: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/23.jpg)
An eBay‐like model
Suppose seller always advertises high.Then: buyers will never “trust” the seller,
and always bid expected valuation.This is the pooling equilibrium:s1(H) = s1(L) = H.sB(H) = sB(L) = p H + (1 - p) L.
![Page 24: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/24.jpg)
An eBay‐like model
Is there any equilibrium where s1(t1) = t1? (In this case the seller is truthful.)In this case the buyers bid:sB(H) = H, sB(L) = L.
But if the buyers use this strategy,the seller prefers to always advertise H!
![Page 25: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/25.jpg)
An eBay‐like model
Now suppose that if the seller lies when the true value is L,there is a cost c (in the form of lower reputation in future transactions).
If H - c < L,then the seller prefers to tell the truth⇒ separating equilibrium.
![Page 26: MS&E 246: Lecture 16 Signaling games - Stanford Universityweb.stanford.edu/~Rjohari/Teaching/Notes/246_Lecture16_2007.pdfSignaling games Observations: • The modeling approach follows](https://reader035.fdocuments.us/reader035/viewer/2022070708/5eb75c6460d4d620a536b6c4/html5/thumbnails/26.jpg)
An eBay‐like model
This example highlights the importance ofsignaling costs:To achieve a separating equilibrium, there must be a difference in the costsof different messages.
(When there is no cost, the resulting message is called “cheap talk.”)