Efficient mechanisms with small subsidies

We address the problem of designing efficient mechanisms that never yield revenue, instead requiring small subsidies. Such mechanisms will be pertinent for settings in which taxing agents is undesirable or impractical—imagine, e.g., a government or private philanthropist that seeks to make a minimal monetary contribution that will allow a group of individuals to reach an efficient decision without being stripped of any of the surplus. Our approach is a close analog of [1], where structure in agent valuations is used to arrive at agent-independent revenue lower bounds for the VCG mechanism that form the basis for redistributing VCG revenue back to the agents without distorting incentives; in the current paper we use valuation structure to obtain revenue upper bounds that form the basis for a second stage of redistribution, returning the revenue of the redistribution mechanism of [1] such that revenue is non-positive but still close to zero. The mechanism we propose is applicable to arbitrary decision problems, always achieving dominant strategy efficiency, ex post individual rationality, and no-revenue. In single-item allocation settings it is asymptotically strongly budget-balanced as the population size grows; we show empirically that for standard distributions over valuations it requires subsidies that are less than 5% of social value, in expectation, for groups of more than 5 agents.