• No results found

Additional Contribution Mechanism

In document Essays on Mechanism Design (Page 35-40)

II. A Prior Free Efficiency Comparison for the Public Good

2.7 Additional Contribution Mechanism

This section describes a mechanism, called the “additional contribution mecha-nism,” and discusses an equilibrium of this mechanism. Intuitively, this mechanism starts with fixed cost shares like the fixed contribution mechanism, but one agent can offer to pay a part of the fixed contribution of another agent (or agents) towards the production cost of the public good. Offering to pay part of another agent’s payment will potentially make the other agent willing to support the public good’s creation.

The additional contribution mechanism has the following structure. The mecha-nism designer sets initial shares s0 = (s01, ..., s0N) with PN

i=1s0i = c. Without loss of generality, assume s01 < v.

The mechanism has two stages:

1. Each agent i ≤ N learns her type. Agent 1 selects a vector b ∈ RN+ where for all i, 0 ≤ bi ≤ s0i − v, and b1 ≡ 0. This choice corresponds to how much agent 1 offers to pay on behalf of each other agent (hence the component corresponding to agent 1 herself is zero.)

2. Each agent i is given a final cost share ˆsi. For agent 1, that share is her initial share plus the sum of the components of b. For each agent 1 < i ≤ N , agent i’s new share is her old share minus bi, the corresponding component of b.

Formally, the final cost shares ˆsi are determined by the following formula:

ˆ

s1 = s01+

N

X

i=2

bi and for i > 1,

ˆ

si = s0i − bi

The agents play the fixed contribution mechanism with contribution shares ˆs.

It may seem counterintuitive that agent 1 would raise her own share by making an additional contribution to one or more other agents. However, if agent 1 believes that lowering another agent’s required contribution increases the likelihood the public good is created by enough to make up for the increased private cost if the public good is created, then raising her own required contribution can be incentive compatible. I show an example where this occurs in section 2.8. Here I show certain properties that will hold for any equilibrium of the mechanism that is in non-weakly dominated strategies.

For the following analysis, fix a type space T. A strategy for agent 1 is then a pair (βT, mT,1) where βT : T1 → ∆(RN) maps every type t1 ∈ T1 into a distribution over b vectors, and where mT,1 : T1× R → {“Yes”, “No”} maps every type t1 ∈ T1

and observed final share ˆs1 into an acceptance or rejection decision by agent 1. A strategy for an agent 1 < i ≤ N is mT,i: Ti× R → {“Yes”, “No”} which maps every type ti ∈ Ti and observed final share ˆsi into an acceptance or rejection decision by the agent. Technically, each agent’s strategy in the final stage could depend on the ˆs’s or

β’s, but because the fixed contribution mechanism is a dominant strategy mechanism, I can ignore the potential role of this information and focus on the dominant strategy equilibrium in the final stage.

Proposition II.17. There exists an equilibrium a of the additional contribution mechanism such that

ˆ

v(t) ≥ s0 ⇒ qM(aT(t)) = 1 (2.21) and

ˆ

v(t) ≥ ˆs ⇒ qM(aT(t)) = 1 (2.22) for all t ∈ T and all T ∈ Ω.

Proof. I construct the equilibrium as follows: for all i, mT,i is specified by every agent plays the equilibrium of the fixed contribution mechanism described in section 6 in the second stage. This is enough to prove equation (2.22). To complete the description of the equilibrium we have to specify agent 1’s choice βT of a b vector.

If ˆv1(t1) ≤ s01: In this case βT(t1) = ~0. Agent 1 has no incentive to deviate because agent 1 will always get a utility of zero in the second stage regardless of the b agent 1 chooses, as ˆs1 ≥ s01 ≥ ˆv1(t1). In the second stage, agent 1 will either be indifferent between “Yes” and “No” or will strictly prefer “No”.

If ˆv1(t1) > s01: In this case, βT,−1(t1) maximizes the following expression:

βT,−1(t1) ∈ argmax

0≤b−1≤~v−s0−1

ˆ

πt1s0−1− b−1≤ v−1(t−1) · vˆ1(t1) − (s01+

N

X

i=2

bi)

!

. (2.23)

Consider the function

F (b−1) = ˆπt1s0−1− b−1 ≤ v−1(t−1) · vˆ1(t1) − (s01+

N

X

i=2

bi)

!

(2.24)

which is the expression within the argmax of equation (2.23) and is the payoff agent 1 expects for each possible b−1. It is the product of F1(b−1) and F2(b−1), where

F1(b−1) = ˆπt1s0−1− b−1 ≤ v−1(t−1) = 1 − ˆπt1v−1(t−1) < s0−1− b−1

 (2.25)

which by inspection is everywhere non-negative, monotonically non-decreasing and right continuous in each dimension of b−1, and F2(b−1) is the continuous function,

F2(b−1) = ˆv1(t1) − (s01+

N

X

i=2

bi), (2.26)

equal to the difference between agent 1’s valuation and ˆs1 as a function of b−1, that is positive over some range.

I show F (b−1) has a maximum by contradiction. Assume F (b−1) has no maximum.

Then there exists a sequence of {bm−1} such that F (bm−1) > 0 is strictly increasing with m and there is no ˆb−1 such that F (ˆb−1) ≥ F (bm−1) for all m. The domain of possible b−1 is compact, so the sequence bm−1 has a convergent subsequence. Let the limit of this subsequence be b−1. By F1 monotonically non-decreasing and right-continuous in each dimension, F1(b−1) ≥ lim F1(bm−1). And by F2 continuous, F2(b−1) = lim F2(bm−1).

F (bm−1) > 0 implies F2(bm−1) > 0 for all m. Therefore, using that F1 is everywhere non-negative and F2 is positive for all bm−1,

F (b−1) = F1(b−1) · F2(b−1) ≥ lim F1(bm−1) · lim F2(bm−1) = lim F (bm−1) (2.27)

and by assumption F (bm−1) is strictly increasing in m, so for all m

F (b−1) ≥ F (bm−1). (2.28)

However, that contradicts the assumption that F (b−1) has no maximum. Therefore F (b−1) must have a maximum and so the argmax in equation (2.23) is well defined.

If there is more than one argmax, then an argmax is chosen arbitrarily, constrained to

s01+

N

X

i=2

βi < ˆv1(t1) (2.29)

Choosing a β(t1) that satisfies equation (2.29) ensures that the first part of the Proposition holds. There must be a b that is an argmax of equation (2.23) and satisfies equation (2.29) for the following reason: either t1’s expected utility at the argmax b is greater than zero, implying that agent 1 expects the public good will be produced with some probability and that equation (2.29) holds (that is, t1 pays less than ˆv1(t1)), or the expected utility at the argmax is 0, in which case b = ~0 is an argmax. In the latter case let β1(t1) = ~0. By assumption ˆv1(t1) > s01 so equation (2.29) is satisfied.

Then equation (2.21) follows from equation (2.29) and the fact that for all i > 1 ˆ

si ≤ s0i by construction, as well as the equilibrium actions in the second stage of the game.

I will refer to an equilibrium of the additional contribution mechanism that fits the description of Proposition II.17 as a “standard equilibrium” of the additional contri-bution mechanism.

In document Essays on Mechanism Design (Page 35-40)

Related documents