on the equivalence of bayesian and dominant strategy implementation

54
On the Equivalence of Bayesian and Dominant Strategy Implementation Alex Gershkov, Jacob K. Goeree, Alexey Kushnir, Benny Moldovanu, Xianwen Shi NES 20th Anniversary Conference, December 2012

Upload: new-economic-school

Post on 14-Apr-2017

447 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: On the Equivalence of Bayesian and Dominant Strategy Implementation

On the Equivalence of Bayesian andDominant Strategy Implementation

Alex Gershkov, Jacob K. Goeree, Alexey Kushnir,Benny Moldovanu, Xianwen Shi

NES 20th Anniversary Conference, December 2012

Page 2: On the Equivalence of Bayesian and Dominant Strategy Implementation

This paper subsumes

Gershkov, Moldovanu, and Shi “Bayesian and Dominant StrategyImplementation Revisited” (2011)

Goeree and Kushnir “On the Equivalence of Bayesian and Domi-nant Strategy Implementation in a General Class of Social ChoiceProblems” (2011)

...forthcoming in Econometrica

Page 3: On the Equivalence of Bayesian and Dominant Strategy Implementation

This paper subsumes

Gershkov, Moldovanu, and Shi “Bayesian and Dominant StrategyImplementation Revisited” (2011)

Goeree and Kushnir “On the Equivalence of Bayesian and Domi-nant Strategy Implementation in a General Class of Social ChoiceProblems” (2011)

...forthcoming in Econometrica

Page 4: On the Equivalence of Bayesian and Dominant Strategy Implementation

Motivation

Bayesian Implementation

VS

Dominant Strategy Implementation

Page 5: On the Equivalence of Bayesian and Dominant Strategy Implementation

Motivation

Wilson (1987)’s critique of Bayesian approach: ”... it (game theory) is

deficient to the extent it assumes other features to be common knowledge,such as one agent’s probability assessment about another’s preferences orinformation”.

Page 6: On the Equivalence of Bayesian and Dominant Strategy Implementation

Main result

For any Bayesian IC and interim IR mechanismthere exists a dominant strategy IC and ex post IRmechanism that delivers

a) the same interim expected utilities to agents,

b) the same ex ante expected social surplus.

Environment: social choice, linear utilities, and independent,one-dimensional, private values

Page 7: On the Equivalence of Bayesian and Dominant Strategy Implementation

Related literature

I Mookherjee and Reichelstein (1992)

I single crossing utility functionsI any monotone BIC social choice rule is DIC implementable

I Vincent and Manelli (2010)I 1-unit auctions, IPV, symmetric and asymmetricI for any BIC mechanism there is DIC mechanism that yields

I the same interim expected allocation probabilitiesI the same interim expected utilities of agents

I Goeree and Kushnir (2012)I a geometric approach to mechanism designI characterize the set of interim expected agent values implementable with

an IC and IR mechanisms

Page 8: On the Equivalence of Bayesian and Dominant Strategy Implementation

Related literature

I Mookherjee and Reichelstein (1992)

I single crossing utility functionsI any monotone BIC social choice rule is DIC implementable

I Vincent and Manelli (2010)I 1-unit auctions, IPV, symmetric and asymmetricI for any BIC mechanism there is DIC mechanism that yields

I the same interim expected allocation probabilitiesI the same interim expected utilities of agents

I Goeree and Kushnir (2012)I a geometric approach to mechanism designI characterize the set of interim expected agent values implementable with

an IC and IR mechanisms

Page 9: On the Equivalence of Bayesian and Dominant Strategy Implementation

Related literature

I Mookherjee and Reichelstein (1992)

I single crossing utility functionsI any monotone BIC social choice rule is DIC implementable

I Vincent and Manelli (2010)I 1-unit auctions, IPV, symmetric and asymmetricI for any BIC mechanism there is DIC mechanism that yields

I the same interim expected allocation probabilitiesI the same interim expected utilities of agents

I Goeree and Kushnir (2012)I a geometric approach to mechanism designI characterize the set of interim expected agent values implementable with

an IC and IR mechanisms

Page 10: On the Equivalence of Bayesian and Dominant Strategy Implementation

Model

Page 11: On the Equivalence of Bayesian and Dominant Strategy Implementation

Model

I Set of agents I = {1, ..., I}

I Set of social alternatives K = {1, ..., K}

I uki (xi , ti ) = aki xi + ti

I aki ≥ 0I xi distributed according to λi with support Xi = [x

¯i , xi ]

I ti agent’s transfer

I x ∈ X = Πi∈IXi , vectors are bold-faced

Page 12: On the Equivalence of Bayesian and Dominant Strategy Implementation

Model

I Set of agents I = {1, ..., I}

I Set of social alternatives K = {1, ..., K}

I uki (xi , ti ) = aki xi + ti

I aki ≥ 0I xi distributed according to λi with support Xi = [x

¯i , xi ]

I ti agent’s transfer

I x ∈ X = Πi∈IXi , vectors are bold-faced

Page 13: On the Equivalence of Bayesian and Dominant Strategy Implementation

Model

I Set of agents I = {1, ..., I}

I Set of social alternatives K = {1, ..., K}

I uki (xi , ti ) = aki xi + ti

I aki ≥ 0I xi distributed according to λi with support Xi = [x

¯i , xi ]

I ti agent’s transfer

I x ∈ X = Πi∈IXi , vectors are bold-faced

Page 14: On the Equivalence of Bayesian and Dominant Strategy Implementation

Model

I Set of agents I = {1, ..., I}

I Set of social alternatives K = {1, ..., K}

I uki (xi , ti ) = aki xi + ti

I aki ≥ 0I xi distributed according to λi with support Xi = [x

¯i , xi ]

I ti agent’s transfer

I x ∈ X = Πi∈IXi , vectors are bold-faced

Page 15: On the Equivalence of Bayesian and Dominant Strategy Implementation

Model

I Direct mechanism (q, t):I qk : X → [0, 1], k ∈ K, ∑k∈K qk (x) = 1I ti : X → R, i ∈ I

I Qki (x

′i ) = Ex−i (q

k(x ′i , x−i )) interim probability alternative k is chosen

I Ti (x ′i ) = Ex−i (ti (x′i , x−i )) the expected transfer to agent i

I Agent i ’s interim expected utility from truthful reporting:

ui (xi ) = Ex−i (∑k∈K aki qk(x)xi + ti (x))

= ∑k∈K aki Qki (xi )xi + Ti (xi )

Page 16: On the Equivalence of Bayesian and Dominant Strategy Implementation

Model

I Direct mechanism (q, t):I qk : X → [0, 1], k ∈ K, ∑k∈K qk (x) = 1I ti : X → R, i ∈ I

I Qki (x

′i ) = Ex−i (q

k(x ′i , x−i )) interim probability alternative k is chosen

I Ti (x ′i ) = Ex−i (ti (x′i , x−i )) the expected transfer to agent i

I Agent i ’s interim expected utility from truthful reporting:

ui (xi ) = Ex−i (∑k∈K aki qk(x)xi + ti (x))

= ∑k∈K aki Qki (xi )xi + Ti (xi )

Page 17: On the Equivalence of Bayesian and Dominant Strategy Implementation

Model

I Direct mechanism (q, t):I qk : X → [0, 1], k ∈ K, ∑k∈K qk (x) = 1I ti : X → R, i ∈ I

I Qki (x

′i ) = Ex−i (q

k(x ′i , x−i )) interim probability alternative k is chosen

I Ti (x ′i ) = Ex−i (ti (x′i , x−i )) the expected transfer to agent i

I Agent i ’s interim expected utility from truthful reporting:

ui (xi ) = Ex−i (∑k∈K aki qk(x)xi + ti (x))

= ∑k∈K aki Qki (xi )xi + Ti (xi )

Page 18: On the Equivalence of Bayesian and Dominant Strategy Implementation

Model

I Direct mechanism (q, t):I qk : X → [0, 1], k ∈ K, ∑k∈K qk (x) = 1I ti : X → R, i ∈ I

I Qki (x

′i ) = Ex−i (q

k(x ′i , x−i )) interim probability alternative k is chosen

I Ti (x ′i ) = Ex−i (ti (x′i , x−i )) the expected transfer to agent i

I Agent i ’s interim expected utility from truthful reporting:

ui (xi ) = Ex−i (∑k∈K aki qk(x)xi + ti (x))

= ∑k∈K aki Qki (xi )xi + Ti (xi )

Page 19: On the Equivalence of Bayesian and Dominant Strategy Implementation

Model

I To simplify notation:

vi (x) = ∑k∈K aki qk(x)

Vi (xi ) = ∑k∈K aki Qki (xi )

and v = (v1, ..., vI ), V = (V1, ..., VI ).

I Agent i ’s interim expected utility from truthful reporting:

ui (xi ) = Vi (xi )xi + Ti (xi )

Page 20: On the Equivalence of Bayesian and Dominant Strategy Implementation

Model

Definition (Equivalent mechanisms)

Two mechanisms (q, t) and (q, t) are equivalent if they deliver

I the same interim expected utilities to all agents,

I the same ex ante expected social surplus.

Page 21: On the Equivalence of Bayesian and Dominant Strategy Implementation

Model

Definition (Equivalent mechanisms)

Two mechanisms (q, t) and (q, t) are equivalent if they deliver

I the same interim expected utilities to all agents,

I the same ex ante expected social surplus.

Page 22: On the Equivalence of Bayesian and Dominant Strategy Implementation

Model

Definition (Equivalent mechanisms)

Two mechanisms (q, t) and (q, t) are equivalent if they deliver

I the same interim expected utilities to all agents,

I the same ex ante expected social surplus.

Page 23: On the Equivalence of Bayesian and Dominant Strategy Implementation

Main theorem

Page 24: On the Equivalence of Bayesian and Dominant Strategy Implementation

Preliminary steps

Fact 1. (e.g. Myerson, 1981) A mechanism is BIC iff (i) for all i ∈ I andxi ∈ Xi , Vi (xi ) is non-decreasing in xi and (ii) agents interimexpected utilities satisfy

ui (xi ) = ui (x i ) +∫ xi

x i

Vi (t)dt

Fact 2. (e.g. Laffont and Maskin, 1980) A mechanism is DIC iff (i) for alli ∈ I and xi ∈ Xi , vi (xi , x−i ) is non-decreasing in xi and (ii)agents utilities satisfy

ui (xi , x−i ) = ui (x i , x−i ) +∫ xi

x i

vi (t, x−i )dt

Page 25: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence

Theorem

Let (q, t) be a BIC and interim IR mechanism. An equivalent DIC and expost IR mechanism is given by (q, t), where q solves

min{qk}k∈K

Ex||v(x)||2

s.t.

qk(x) ≥ 0 ∀k, x

∑k qk(x) = 1 ∀x

Vi (xi ) = Vi (xi ) ∀i , xi

(Program)

and transfers t follow from

ti (xi , x−i ) = ti (x i , x−i ) + vi (x¯ i , x−i )x i − vi (xi , x−i )xi +

∫ xi

x i

vi (t, x−i )dt.

where ti (x i , x−i ) = (vi (x i , x−i )/Vi (x i ))Ti (x i ).

Page 26: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence

Theorem

Let (q, t) be a BIC and interim IR mechanism. An equivalent DIC and expost IR mechanism is given by (q, t), where q solves

min{qk}k∈K

Ex||v(x)||2

s.t.

qk(x) ≥ 0 ∀k, x

∑k qk(x) = 1 ∀x

Vi (xi ) = Vi (xi ) ∀i , xi

(Program)

and transfers t follow from

ti (xi , x−i ) = ti (x i , x−i ) + vi (x¯ i , x−i )x i − vi (xi , x−i )xi +

∫ xi

x i

vi (t, x−i )dt.

where ti (x i , x−i ) = (vi (x i , x−i )/Vi (x i ))Ti (x i ).

Page 27: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof.

I Prove vi (xi , x−i ) = ∑k∈K aki qk(xi , x−i ) is non-decreasing in xifor q – solution to (Program),

I Steps of the proof:

1. Discrete and uniformly distributed types (Lemma 1)I an extension of the theorem due to Gutmann et al. (1991)

2. Continuous and uniformly distributed types (Lemma 2)

3. Arbitrary type distributions (Lemma 3)

Page 28: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof.

I Prove vi (xi , x−i ) = ∑k∈K aki qk(xi , x−i ) is non-decreasing in xifor q – solution to (Program),

I Steps of the proof:

1. Discrete and uniformly distributed types (Lemma 1)I an extension of the theorem due to Gutmann et al. (1991)

2. Continuous and uniformly distributed types (Lemma 2)

3. Arbitrary type distributions (Lemma 3)

Page 29: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof.

I Prove vi (xi , x−i ) = ∑k∈K aki qk(xi , x−i ) is non-decreasing in xifor q – solution to (Program),

I Steps of the proof:

1. Discrete and uniformly distributed types (Lemma 1)I an extension of the theorem due to Gutmann et al. (1991)

2. Continuous and uniformly distributed types (Lemma 2)

3. Arbitrary type distributions (Lemma 3)

Page 30: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof.

I Prove vi (xi , x−i ) = ∑k∈K aki qk(xi , x−i ) is non-decreasing in xifor q – solution to (Program),

I Steps of the proof:

1. Discrete and uniformly distributed types (Lemma 1)I an extension of the theorem due to Gutmann et al. (1991)

2. Continuous and uniformly distributed types (Lemma 2)

3. Arbitrary type distributions (Lemma 3)

Page 31: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof of Lemma 1.

Lemma 1. Xi is discrete, λi is uniform. Let q be a solution of (Program)then vi (xi , x−i ) is non-decreasing in xi for all i ∈ I , x ∈ X .

Proof outline.

1. The set of solutions to (Program) is not empty.

2. Assume vi (xi , x−i ) is not non-decreasing.

3. Construct feasible q′ that delivers lower value to Ex||v(x)||2.

4. Contradiction.

Page 32: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof of Lemma 1.

Lemma 1. Xi is discrete, λi is uniform. Let q be a solution of (Program)then vi (xi , x−i ) is non-decreasing in xi for all i ∈ I , x ∈ X .

Proof outline.

1. The set of solutions to (Program) is not empty.

2. Assume vi (xi , x−i ) is not non-decreasing.

3. Construct feasible q′ that delivers lower value to Ex||v(x)||2.

4. Contradiction.

Page 33: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof of Lemma 1.

Lemma 1. Xi is discrete, λi is uniform. Let q be a solution of (Program)then vi (xi , x−i ) is non-decreasing in xi for all i ∈ I , x ∈ X .

Proof outline.

1. The set of solutions to (Program) is not empty.

2. Assume vi (xi , x−i ) is not non-decreasing.

3. Construct feasible q′ that delivers lower value to Ex||v(x)||2.

4. Contradiction.

Page 34: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof of Lemma 1.

Lemma 1. Xi is discrete, λi is uniform. Let q be a solution of (Program)then vi (xi , x−i ) is non-decreasing in xi for all i ∈ I , x ∈ X .

Proof outline.

1. The set of solutions to (Program) is not empty.

2. Assume vi (xi , x−i ) is not non-decreasing.

3. Construct feasible q′ that delivers lower value to Ex||v(x)||2.

4. Contradiction.

Page 35: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof of Lemma 1.

Lemma 1. Xi is discrete, λi is uniform. Let q be a solution of (Program)then vi (xi , x−i ) is non-decreasing in xi for all i ∈ I , x ∈ X .

Proof outline.

1. The set of solutions to (Program) is not empty.

2. Assume vi (xi , x−i ) is not non-decreasing.

3. Construct feasible q′ that delivers lower value to Ex||v(x)||2.

4. Contradiction.

Page 36: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof of Lemma 1.

Proof.

1. The set of solutions of (Program) is not empty.

min{qk}k∈K

Ex||v(x)||2 (Program)

s.t.

qk(x) ≥ 0 ∀k, x

∑k qk(x) = 1 ∀xVi (xi ) = Vi (xi ) ∀i , xi

2. Suppose vj (xj , x−j ) > vj (x ′j , x−j ) for some j , x ′j > xj , and some x−j .

3. {qk}k∈K being BIC means Vj (xj ) = Vj (xj ) = Ex−j v(xj , x−j ) isnon-decreasing.

4. ∃ x′−j such that vj (xj , x′−j ) < vj (x ′j , x′−j ). Consider allocations at 4

type-points.

Page 37: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof of Lemma 1.

Proof.

1. The set of solutions of (Program) is not empty.

min{qk}k∈K

Ex||v(x)||2 (Program)

s.t.

qk(x) ≥ 0 ∀k, x

∑k qk(x) = 1 ∀xVi (xi ) = Vi (xi ) ∀i , xi

2. Suppose vj (xj , x−j ) > vj (x ′j , x−j ) for some j , x ′j > xj , and some x−j .

3. {qk}k∈K being BIC means Vj (xj ) = Vj (xj ) = Ex−j v(xj , x−j ) isnon-decreasing.

4. ∃ x′−j such that vj (xj , x′−j ) < vj (x ′j , x′−j ). Consider allocations at 4

type-points.

Page 38: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof of Lemma 1.

Proof.

1. The set of solutions of (Program) is not empty.

min{qk}k∈K

Ex||v(x)||2 (Program)

s.t.

qk(x) ≥ 0 ∀k, x

∑k qk(x) = 1 ∀xVi (xi ) = Vi (xi ) ∀i , xi

2. Suppose vj (xj , x−j ) > vj (x ′j , x−j ) for some j , x ′j > xj , and some x−j .

3. {qk}k∈K being BIC means Vj (xj ) = Vj (xj ) = Ex−j v(xj , x−j ) isnon-decreasing.

4. ∃ x′−j such that vj (xj , x′−j ) < vj (x ′j , x′−j ). Consider allocations at 4

type-points.

Page 39: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof of Lemma 1.

Proof.

1. The set of solutions of (Program) is not empty.

min{qk}k∈K

Ex||v(x)||2 (Program)

s.t.

qk(x) ≥ 0 ∀k, x

∑k qk(x) = 1 ∀xVi (xi ) = Vi (xi ) ∀i , xi

2. Suppose vj (xj , x−j ) > vj (x ′j , x−j ) for some j , x ′j > xj , and some x−j .

3. {qk}k∈K being BIC means Vj (xj ) = Vj (xj ) = Ex−j v(xj , x−j ) isnon-decreasing.

4. ∃ x′−j such that vj (xj , x′−j ) < vj (x ′j , x′−j ). Consider allocations at 4

type-points.

Page 40: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof of Lemma 1.

Page 41: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof of Lemma 1.

Does q′ satisfy:qk(x) ≥ 0 ∀k, x

∑k qk(x) = 1 ∀xVi (xi ) = Vi (xi ) ∀i , xi

?

Page 42: On the Equivalence of Bayesian and Dominant Strategy Implementation

BIC-DIC equivalence. Proof of Lemma 1.

I The above argument shows q′ is feasible.

I Direct calculations also show

Ex (||v′(x)||2 − ||v(x)||2) < 0

I This contradicts to {qk}k∈K being a solution to (Program). QED

Page 43: On the Equivalence of Bayesian and Dominant Strategy Implementation

The limits of BIC-DIC equivalence

Page 44: On the Equivalence of Bayesian and Dominant Strategy Implementation

The limits of BIC-DIC equivalence

I Stronger notion of the equivalenceI counterexample to equivalence based on Qk

i (xi ) = Qki (xi )

I Correlated values (see Cremer and McLean, 1988)

I Multidimensional types

I Non-linear utilities

I Interdependent values

Page 45: On the Equivalence of Bayesian and Dominant Strategy Implementation

Interdependent Valuesbased on ”A Geometric Approach to Mechanism Design” (2012) Goeree and Kushnir

I A discrete version of Maskin (1992)’s example

I Single-unit auction, 2 bidders, 2 types {x , x}

I K = 3 states: assign to bidder 1, bidder 2, or not at all

I Bidder i ’s value vi (xi , xj ) = xi + αxj

I Compare BIC with EPIC (ex post incentive compatibility)

v

vα = 0

EPIC

BIC

v

vα = 1

2

EPIC

BIC

v

vα = 2

EPIC

BIC

Feasible (yellow shaded), BIC (light gray), and EPIC (dark gray)

Page 46: On the Equivalence of Bayesian and Dominant Strategy Implementation

Interdependent Valuesbased on ”A Geometric Approach to Mechanism Design” (2012) Goeree and Kushnir

I A discrete version of Maskin (1992)’s example

I Single-unit auction, 2 bidders, 2 types {x , x}

I K = 3 states: assign to bidder 1, bidder 2, or not at all

I Bidder i ’s value vi (xi , xj ) = xi + αxj

I Compare BIC with EPIC (ex post incentive compatibility)

v

vα = 0

EPIC

BIC

v

vα = 1

2

EPIC

BIC

v

vα = 2

EPIC

BIC

Feasible (yellow shaded), BIC (light gray), and EPIC (dark gray)

Page 47: On the Equivalence of Bayesian and Dominant Strategy Implementation

Interdependent Valuesbased on ”A Geometric Approach to Mechanism Design” (2012) Goeree and Kushnir

I A discrete version of Maskin (1992)’s example

I Single-unit auction, 2 bidders, 2 types {x , x}

I K = 3 states: assign to bidder 1, bidder 2, or not at all

I Bidder i ’s value vi (xi , xj ) = xi + αxj

I Compare BIC with EPIC (ex post incentive compatibility)

v

vα = 0

EPIC

BIC

v

vα = 1

2

EPIC

BIC

v

vα = 2

EPIC

BIC

Feasible (yellow shaded), BIC (light gray), and EPIC (dark gray)

Page 48: On the Equivalence of Bayesian and Dominant Strategy Implementation

Interdependent Valuesbased on ”A Geometric Approach to Mechanism Design” (2012) Goeree and Kushnir

I A discrete version of Maskin (1992)’s example

I Single-unit auction, 2 bidders, 2 types {x , x}

I K = 3 states: assign to bidder 1, bidder 2, or not at all

I Bidder i ’s value vi (xi , xj ) = xi + αxj

I Compare BIC with EPIC (ex post incentive compatibility)

v

vα = 0

EPIC

BIC

v

vα = 1

2

EPIC

BIC

v

vα = 2

EPIC

BIC

Feasible (yellow shaded), BIC (light gray), and EPIC (dark gray)

Page 49: On the Equivalence of Bayesian and Dominant Strategy Implementation

Interdependent Valuesbased on ”A Geometric Approach to Mechanism Design” (2012) Goeree and Kushnir

I A discrete version of Maskin (1992)’s example

I Single-unit auction, 2 bidders, 2 types {x , x}

I K = 3 states: assign to bidder 1, bidder 2, or not at all

I Bidder i ’s value vi (xi , xj ) = xi + αxj

I Compare BIC with EPIC (ex post incentive compatibility)

v

vα = 0

EPIC

BIC

v

vα = 1

2

EPIC

BIC

v

vα = 2

EPIC

BIC

Feasible (yellow shaded), BIC (light gray), and EPIC (dark gray)

Page 50: On the Equivalence of Bayesian and Dominant Strategy Implementation

Interdependent Valuesbased on ”A Geometric Approach to Mechanism Design” (2012) Goeree and Kushnir

I A discrete version of Maskin (1992)’s example

I Single-unit auction, 2 bidders, 2 types {x , x}

I K = 3 states: assign to bidder 1, bidder 2, or not at all

I Bidder i ’s value vi (xi , xj ) = xi + αxj

I Compare BIC with EPIC (ex post incentive compatibility)

v

vα = 0

EPIC

BIC

v

vα = 1

2

EPIC

BIC

v

vα = 2

EPIC

BIC

Feasible (yellow shaded), BIC (light gray), and EPIC (dark gray)

Page 51: On the Equivalence of Bayesian and Dominant Strategy Implementation

Conclusion

Page 52: On the Equivalence of Bayesian and Dominant Strategy Implementation

Conclusion

BIC-DIC equivalence for social choice problems insettings with independent private values, linear util-ities, one-dimensional types, and for general typedistributions (including discrete type-space).

The proof is short and constructive.

Identify limits of BIC-DIC equivalence: stronger no-tion of equivalence, interdependent values, correlatedvalues, multi-dimensional types, non-linear utilities.

Page 53: On the Equivalence of Bayesian and Dominant Strategy Implementation

Conclusion

BIC-DIC equivalence for social choice problems insettings with independent private values, linear util-ities, one-dimensional types, and for general typedistributions (including discrete type-space).

The proof is short and constructive.

Identify limits of BIC-DIC equivalence: stronger no-tion of equivalence, interdependent values, correlatedvalues, multi-dimensional types, non-linear utilities.

Page 54: On the Equivalence of Bayesian and Dominant Strategy Implementation

Conclusion

BIC-DIC equivalence for social choice problems insettings with independent private values, linear util-ities, one-dimensional types, and for general typedistributions (including discrete type-space).

The proof is short and constructive.

Identify limits of BIC-DIC equivalence: stronger no-tion of equivalence, interdependent values, correlatedvalues, multi-dimensional types, non-linear utilities.