Unformatted text preview: s show clear support for the bounded-rational compatible 15 properties of ascending-price and other iterative auctions, rst identi ed by Parkes et al. 17 as important for on-line auctions. It is interesting that the revenue-equivalence theorem 12 fails when agents have hard valuation problems and limited or costly computation. The ascending-price auction dominates the sealed-bid auction, in terms of allocative e ciency and revenue, when there are more than a few agents in the market e.g. N 5. This holds even when agents are uninformed about 12 the values of other agents in the ascending-price auction but informed in the sealed-bid auction. Iterative auctions do more than provide agents with useful information about the values of other agents, they can also reduce an agent's uncertainty about the outcome, i.e. about the nal prices and allocation. Posted-price auctions are often necessary with many agents e.g. N 50 because they eliminate a bidder's uncertainty about the outcome of the auction. An agent knows that if it accepts the price of an item it will de nitely receive the item, and at that price. This simpli es an agent's metadeliberation problem. However, with small numbers of agents posted-price auctions are typically less e cient than auctions with dynamic pricing unless the seller is very well informed about agents' values. Agents with information about the likely outcome of an auction can deliberate more e ciently about the values of di erent outcomes, for example quickly eliminating from their possibility set all bundles of items that are very expensive. Iterative auctions provide agents with this information dynamically during the auction as bids are received. The right" agents can deliberate for the right" amount of time, shifting deliberation away from agents with low values and towards agents with high values, and towards values for bundles that t into good global solutions. Another approach to improve deliberation e ciency could provide agents with historical info...
View Full Document
This document was uploaded on 03/19/2014 for the course COMP CS286r at Harvard.
- Fall '13