Unformatted text preview: s between the bounds; 2 the true value is uniformly distributed between the bounds over all possible stochastic sequences of deliberations.4 E ciency is computed as the true value for the good of the agent that wins the auction, as a fraction of the maximum value over all agents. Revenue is
The agents need to know how many agents remain active to implement the Nash equilibrium. We could ask the agents to pay a small participation fee" in each round of the auction to remain active, so that the auctioneer can report this information in each round. 4 Actually, this requires that the new expected value for the good in simulation is not uniformly distributed with respect to the current bounds, but favors more central values.
3 7 computed as the price paid for the good, as a fraction of the maximum value over all agents. Finally, the average utility to an agent for participation in the auction is computed as the surplus vi , p to agent i that wins the auction for price p, minus the total cost of deliberation for all agents, and divided by the number of agents in the auction. Agents can lose utility from participation because of assumptions made in deriving metadeliberation strategies, for example about the bids of other agents and an agent's own deliberation procedure.5 We check that the utility for participation is positive to validate agent strategies. We compare the performance of each auction as we vary the number of agents, N , and the computational e ectiveness 1 , and cost C of agents' deliberation procedures. All results are averaged over at least 1000 trials. E ciency is often the primary performance measure of mechanism design, but if the auctioneer is also the seller then revenue can be important. We write M1 M2 if mechanism M1 dominates M2 in terms of e ciency and revenue, or M1 M2 if the mechanisms have con icting ordering for revenue and e ciency, or very similar performance; A denotes the ascending-price auction, S the sealed-bid auction, and P the posted-price auction. 4.1 Ad...
View Full Document
This document was uploaded on 03/19/2014 for the course COMP CS286r at Harvard.
- Fall '13