Distributed Algorithmic Mechanism Design (DAMD) combines theoretical computer science’s traditional focus on computational tractability with its more recent interest in incentive compatibility and distributed computing. The Internet’s decentralized nature, in which distributed computation and autonomous agents prevail, makes DAMD a very natural approach for many Internet problems. This paper first outlines the basics of DAMD and then reviews previous DAMD results on multicast cost sharing and interdomain routing. The remainder of the paper describes several promising research directions and poses some specific open problems.

The mechanism design literature assumes too much common knowledge of the environment among the players and planner. We relax this assumption by studying implementation on richer type spaces, with more higher order uncertainty. We study the "ex post equivalence" question: when is interim implementation on all possible type spaces equivalent to requiring ex post implementation on the space of payoff types? We show that ex post equivalence holds when the social choice correspondence is a function and in simple quasi-linear environments. When ex post equivalence holds, we identify how large the type space must be to obtain the equivalence. We also show that ex post equivalence fails in general, including in quasi-linear environments with budget balance. For quasi-linear environments, we provide an exact characterization of when interim implementation is possible in rich type spaces. In this environment, the planner can fully extract players’ belief types, so the incentive constraints reduce to conditions distinguishing types with the same beliefs about others’ types but different payoff types.

The implementation problem is the problem of designing a mechanism (game form) such that the equilibrium outcomes satisfy some criterion of social optimality. The early literature assumed that each agent would simply report his

Abstract not found

In this paper, we introduce the notion of transferable control, defined as a situation where one party (the principal, say) can transfer control to another party (the agent) but cannot commit herself to do so. One theoretical foundation for this notion of transferable control builds on the distinction between formal and real authority introduced by Aghion and Tirole (1997), in which the actual exercise of authority may require noncontractible information, absent which formal control rights are vacuous. We then use this notion to study the extent to which control transfers may allow an agent to reveal information regarding his ability or willingness to cooperate with the principal in the future. We show that the distinction between contractible and transferable control can drastically influence how learning takes place: with contractible control, information about the agent can often be acquired through revelation mechanisms that involve communication and message-contingent control allocations; in contrast, when control is transferable but not contractible, it can be optimal to transfer control unconditionally

We analyze situations in which a group of agents (and possibly a designer) have to reach a decision that will affect all the agents. Examples of such scenarios include the selection of a candidate, the location of a nuclear reactor or the siting of a major sports event. To address the problem of reaching a decision, we propose a one-stage multi-bidding mechanism in which agents submit bids to determine the winning project. All Nash equilibria of this mechanism are efficient and immune to any coalitional deviations. Moreover, the payoffs attained in equilibrium by the agents satisfy intuitively appealing lower bounds. (JEL D78, D62) 1 Reaching decisions about the location of noxious facilities, such as dump-sites, environmentally hazardous plants, nuclear power generators and the like, is a highly contentious issue. For instance, in February 2000, the US Senate decided that nationwide nuclear waste would be shipped to the Yucca mountain site in Nevada (conditional to it being approved as a high-level nuclear waste repository). Despite the attractive compensation package, the State of Nevada voiced vehement opposition. President

was written while the first author was on sabbatical at the London School of We view a contract as a list of outcomes. Ex ante, the parties commit not to consider outcomes not on the list, i.e., these are “ruled out”. Ex post, they freely bargain over outcomes on the list, i.e., the contract specifies no mechanism to structure their choice; in this sense outcomes on the list are not “ruled in”. A “loose” contract (long list) maximizes flexibility but may interfere with ex ante investment incentives. When these incentives are important enough, the parties may write a “tight ” contract (short list), even though this leads to ex post inefficiency. 1

A social choice function is robustly implemented if every equilibrium on every type space achieves outcomes consistent with a social choice function. We identify a robust monotonicity condition that is necessary and (with mild extra assumptions) sufficient for robust implementation. Robust monotonicity is strictly stronger than both Maskin monotonicity (necessary and almost sufficient for complete information implementation) and ex post monotonicity (necessary and almost sufficient for ex post implementation). It is equivalent to Bayesian monotonicity on all type spaces. It requires that there not be too much interdependence of types. We characterize robust monotonicity for some interesting economic environments. We identify conditions where, if robust implementation is possible, it is possible in a direct mechanism. We identify conditions where, if robust implementation is not possible, virtual robust implementation is not possible either.

A social choice function is robustly implementable if there is a mechanism under which the process of iteratively eliminating strictly dominated messages leads to outcomes that agree with the social choice function for all beliefs at every type pro…le. In an interdependent value environment with single crossing preferences, we identify a contraction property on the preferences which together with strict ex post incentive compatibility is su ¢ cient to guarantee robust implementation in the direct mechanism. Strict ex post incentive compatibility and the contraction property are also necessary for robust implementation in any mechanism, including indirect ones. The contraction property requires that the interdependence is not too large. In a linear signal model, the contraction property is equivalent to an interdependence matrix having all eigenvalues smaller than one.

Allocation and exchange of many discrete resources – such as kidneys or school seats – is conducted via direct mechanisms without monetary transfers. A primary concern in designing such mechanisms is the coordinated strategic behavior of market participants and its impact on resulting allocations. To assess the impact of this implementation constraint, we construct the full class of group dominant-strategy incentive compatible and Pareto efficient mechanisms. We call these mechanisms “Trading Cycles. ” This class contains new mechanisms as well as such previously studied mechanisms as top trading cycles, serial dictatorships, and hierarchical exchange. In some problems, the new trading-cycles mechanisms perform better than all previously known mechanisms. Just as importantly, knowing that all group incentive-compatible and efficient mechanisms can be implemented as trading cycles allows us to determine easily which efficient