Resilient Provision of a Public Good

Silvio Micali,Jing Chen

2008-01-01

Abstract:We present two resilient mechanisms for the provision of a public good. Both mechanisms adopt a knowledge-based benchmark. Introduction Our first mechanism is appropriate to the case when the players —i.e., the potential beneficiaries of the good— are few in number and/or know each each other quite well. (In this case, for concreteness, we envision the provisioning to occur in a laboratory, and refer to the good as —a new piece of— equipment, to the players as members of the lab, and to the potential provisioner as the lab director.) Our second mechanism is more appropriate when the players are quite numerous and/or may have only local knowledge, that is, each player only knows a few of the other players. (In this case, for concreteness, we envision the provisioning to occur in a city, and refer to the good as a public park, to the players as citizens, and to the potential provisioner as the mayor.) Notation. Let N = {1, . . . , n} be a set of players, and γ ∈ R+ the cost (to the ”potential provider”) of provisioning the good. A player i’s valuation of the good is a non-negative real. The profile of all possible valuations of the players is denoted by V. The profile of the players’ true valuations is denoted by TV . An outcome is a pair (x, P ), where x is a bit indicating whether the good will be provided (x = 1) or not (x = 0), and P is a profile of prices (real numbers). A payer’s utility is TVi · x − Pi. A player i’s general external knowledge, denoted by GKi, is i’s information about TV−i. A player i’s relevant external knowledge, denoted by RKi, is a subprofile in V−i such that, for each j 6= i, RKi j is the maximum value with GKi and less than TVj . All knowledge of a player is private to him. In the two mechanisms below, ”numbered steps are performed by players, and bullet ones by the mechanism.” 1 Our First Mechanism Mechanism M1 • Set x = 0 and Pi = 0 for each player i. 1. Each player i simultaneously and publicly announces a valuation subprofile V i for players in −i. • Set: γi = ∑ j∈−i V i j for each player i, and ? = arg maxi γi. (We shall refer to player ? as the “star player”.) • If γ? 0 publicly and simultaneously announce YES or NO. • If some player announces NO, reset P? = P? + V ? i for each player i who announces NO, and HALT. • (If all players announce YES) Reset: (1) x = 1; (2) P? = γ − γ?; and (3) Pi = V ? i for each player i 6= ?. Variant. In the last mechanism step replace instruction 2 with the following instruction (2′) P? = α·(γ−γ?), where the coefficient α is a constant between 0 and 1 (so as to generate a “surplus” for the lab). 2 Our Second Mechanism Mechanism M2 • Set x = 0 and Pi = 0 for each player i. 1. Each player i simultaneously and publicly announces (A) a subset of players Si ⊂ −i and (B) a valuation subprofile V i for the players in Si. • ∀j: If j 6∈ Si for all i 6= j, then set EVj = 0; else, nj = arg maxi 6=j V i j , and set EVj = V ni j . Set K = ∑ j EVj . • If K 0 publicly and simultaneously announces YES or NO. • If some player announces NO: ∀k such that player k announces NO, reset Pnk = EVk. HALT. • (All players announce YES) (1) reset x = 1 and Pj = EVj for all j; (2) ∀j, reset Pnj = Pnj−(K−γ) EVj K . Variant. In the last mechanism step replace instruction 2 with the following instruction (2′) Pnj = Pnj − α · (K − γ) EVj K where the coefficient α is a constant between 0 and 1 (so as to generate a “surplus” for the city).