You are on page 1of 12

...

PBE Refinements
Swapnendu Banerjee
Jadavpur University
Kolkata, INDIA
2X2X2 Signalling Model
✘ Some PBE that satisfies Req. 1 to 4 rests on unreasonable beliefs.

✘ Further refinements can be done .

✘ Rule out those PBE that rests on unreasonable beliefs.

✘ To do that we need some additional Requirements.

✘ Look at the pooling on L equilibrium of the following game.

2
Example
(3, 2) (1,0)
u u
L 𝜃1 R
p q
(2,0) d 0.5 d
(0,1)
Receiver Nature Receiver

(1,0) (2,1)
u 0.5 u
1-p 1-q
L 𝜃2 R
d d
(1,1) (0,0)

The following strategies and beliefs {(L,L) ; (u, d) ; p = 0.5; q > 1/2}
constitute a set of pooling PBE if we go by requirements 1 to 4.
3
But this equilibria suffers from a problem.

The off-the-equilibrium belief q > ½ is not reasonable for the receiver.

Note that the 𝜃1 -type will never play R since R is strictly dominated for that
type. So any belief of the receiver that puts q > 0 is unreasonable.

But requirements 1 to 4 is not sufficient to rule these equilibria out.

So we need additional requirements.

4
Definition: In a signalling game, the message 𝑚𝑗 is dominated for type 𝜃𝑖 if
there exists another message 𝑚𝑗′ such that lowest possible payoff from 𝑚𝑗′ is
greater than 𝜃𝑖 ’ s highest possible payoff from 𝑚𝑗 .

Requirement 5 (Dominance based criteria):


If the information set following 𝑚𝑗 is off-the-equilibrium-path and
𝑚𝑗 is dominated for type 𝜃𝑖 then (if possible) the Receiver’s belief 𝜇 𝜃𝑖 𝑚𝑗
should place zero probability on type 𝜃𝑖 , i.e. 𝜇 𝜃𝑖 𝑚𝑗 = 0. (This is possible
provided 𝑚𝑗 is not dominated for all types)

Given this q=0 and therefore {(L,L) ; (u, d) ; p = 0.5; q > 1/2} are struck down.
5
Change the previous game slightly:

(3, 2) (1,0)
u u
L 𝜃1 R
p q
(2,0) d 0.5 d
(0,1)
Receiver Nature Receiver

(1,0) (2,0)
u 0.5 u
1-p 1-q
L 𝜃2 R
d d
(1,1) (0,1)

Now look {(L,L) ; (u, d) ; p = 0.5; q } constitute a set of pooling PBE


if we go by requirements 1 to 4. (for any q within 0 to 1)
6
But only {(L,L) ; (u, d) ; p = 0.5; q=0 } satisfies requirement 1 to 5.

From infinite number of equilibria we get only one that satisfies requirement
5 => Huge refinement.

But there are games where even requirement 5 fails to remove all PBE with
unreasonable beliefs.

To illustrate we look at the following game.

7
Look at this new game:

(1, 1) (0,1)
u u
L 𝜃1 R
p q
(3,0) d 0.1 d
(2,0)
Receiver Nature Receiver

(0,-1) (1,-1)
u 0.9 u
1-p 1-q
L 𝜃2 R
d d
(2,0) (3,0)

Now look, {(L,L) ; (d, u) ; p = 0.1; 𝒒 ≥ 𝟏/𝟐} constitute a set of


pooling PBE if we go by requirements 1 to 5.
8
But these equilibria also suffer from a problem.

The off-the-equilibrium belief q > ½ is not reasonable either.

Note that the 𝜃1 -type on-the-equilibrium path gets the equilibrium payoff of
3 whereas if she plays R at most she can get 2.

So R is “equilibrium Dominated” although NOT ‘fully dominated’.

Thus 𝜃1 -type should not play R.

9
So any belief of the receiver that puts q > 0 is unreasonable although R is
not ‘fully dominated’.

Thus requirement 5 will not help since it talks about ‘dominance’ not
‘equilibrium dominance’.

So we need additional requirement(s).

This takes us to Requirement 6.

10
Definition: Given a PBE in a signalling game, the message 𝑚𝑗 is ‘equilibrium
dominated’ for type 𝜽𝒊 if 𝜃𝑖 ’s equilibrium payoff is greater than 𝜃𝑖 ’s highest
possible payoff from 𝑚𝑗 .

Requirement 6 (Equilibrium Dominance based criteria, Intuitive Criterion):


If the information set following 𝑚𝑗 is off-the-equilibrium-path and
𝑚𝑗 is ‘equilibrium dominated’ for type 𝜽𝒊 then (if possible) the Receiver’s
belief 𝜇 𝜃𝑖 𝑚𝑗 should place zero probability on type 𝜃𝑖 , i.e. 𝜇 𝜃𝑖 𝑚𝑗 =0. (This is
possible provided 𝑚𝑗 is not ‘equilibrium dominated’ for all types)

Given this q=0 and therefore {(L,L) ; (d, u) ; p = 0.5; 𝒒 ≥ 𝟏/𝟐} are struck down.

11
This Equilibrium Dominance based criteria is also known as the Intuitive
Criterion. (Credit to Cho and Kreps (1987)).

In 2X2 Games Intuitive Criterion removes all pooling PBE that rests on
unreasonable off-the-equilibrium path beliefs.

The problem arises from the freedom that is given for off-the-equilibrium
path beliefs.

What about refinement of Separating equilibria? Any Idea?

Reference: Gibbons, Chapter 4, pages 235-238.


12

You might also like