Professional Documents
Culture Documents
Bayesian implementation
Leandro Gorno
December 2, 2014
1 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
2 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Definition
The (type-contingent) strategy profile s ∗ ∈ S Θ is a Bayesian
Nash equilibrium (henceforth BNE) of the mechanism
Γ = (S1 , ..., SI , g ) if
3 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
4 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Bayesian implementation
7 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Two definitions
Definition
A social choice function is implementable in BNE if there exists a
mechanism Γ = (S1 , ..., SI , g ) with a BNE s ∗ ∈ S Θ such that
f (θ) = g (s ∗ (θ)) for all θ ∈ Θ.
Definition
A social choice function is Bayesian incentive compatible if
truthtelling is a BNE of the associated direct revelation
mechanism.
8 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Proof.
Sufficiency of Bayesian incentive compatibility is obvious.
Suppose f is implementable in BNE using Γ = (S1 , ..., SI , g ).
Then, almost as in the dominant strategies case, we have
9 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
10 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
11 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Quasilinear environments
Recall that:
ui (x, θi ) = vi (k, θi ) + mi + ti .
12 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Independence assumption
Notation: Y
φ−i (θ−i ) = φj (θj ).
j6=i
13 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Mechanism specification
Transfers:
Z X
ti (θ) = vj (k ∗ (θi , θ̃−i ), θ̃j )φ−i (θ̃−i )d θ̃−i + hi (θ−i ) (2)
j6=i
14 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Implementability result
Theorem
Consider the quasilinear environment and assume that types are
independent. Then, every social choice function satisfying
Equations 1 and 2 is Bayesian incentive-compatible.
15 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Implementability result
Proof.
( I )
X
∗ ∗
E {vi (k (θ), θi ) + ti (θ)|θi } = E vj (k (θ), θj ) + hi (θ−i )
j=1
( I )
X
≥E vj (k ∗ (θi0 , θ−i ), θj ) + hi (θ−i )
j=1
16 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Choose X
1
hi (θ−i ) = − ξj (θj ),
I −1 j6=i
where
Z X
ξi (θi ) = vj (k ∗ (θi , θ̃−i ), θ̃j )φ−i (θ̃−i )d θ̃−i .
j6=i
17 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Note that
PI 1
PI P
i=1 hi (θ−i ) = − I −1 j6=i ξj (θj )
1
Pi=1
I
= − I −1 i=1 (I − 1)ξi (θi )
PI
= − i=1 ξi (θi ).
Hence,
I
X I
X I
X
ti (θ) = ξi (θi ) + hi (θ−i ) = 0
i=1 i=1 i=1
19 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Linear utility
20 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Other mechanisms?
ui (x, θi ) = θi vi (k) + mi + ti
21 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Definitions
and
Ui (θi ) = θi v i (θi ) + t i (θi ).
22 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Full characterization
Theorem
The social choice function f (·) = (k(·), t1 (·), ..., tI (·)) is Bayesian
incentive compatible if and only if, for every i = 1, ..., I ,
v i (·) is non-decreasing.
Rθ
Ui (θi ) = Ui (θi ) + θ i v i (s)ds for all θi ∈ Θi .
i
Proof.
For necessity, pick θi0 > θi . Incentive compatibility implies
Linear utility
Proof.
Combining the two inequalities:
Ui (θi0 ) − Ui (θi )
v i (θi ) ≥ ≥ v i (θi0 ).
θi0 − θi
24 / 25
Bayesian Nash equilibrium Bayesian implementation Expected externality mechanism Linear utility
Linear utility
Proof.
For sufficiency, suppose i is of type θi and consider her
incentives to claim a different type θi0 . Without loss, assume
θi > θi0 . Then,
Z θi Z θi
Ui (θi )−Ui (θi0 ) = v i (s)ds ≥ v i (θi0 )ds = (θi −θi0 )v i (θi0 )
θi0 θi0
Z θi Z θi
Ui (θi )−Ui (θi0 ) = v i (s)ds ≤ v i (θi )ds = (θi −θi0 )v i (θi )
θi0 θi0
25 / 25