A peer-to-peer agent auction

In this work we examine a peer-to-peer agent continuous double auction. We compare agents trading using peer-to-peer communications with agents using the same trading strategy in an auction that makes use of a centralized auctioneer to disseminate information. We present simulation data for these two auctions running with 2,500 to 160,000 agents. We find that the peer-to-peer auction is able to display price convergence behavior similar to that of the centralized auction. Further, the data shows that the peer-to-peer system has a constant cost in the number of message rounds needed to find the market equilibrium price as the number of traders is increased, in contrast to the linear cost incurred by the central auctioneer. Considering the above message costs, the peer-to-peer system outperformed the simple central auction by at least 100 times in our simulations. We further calculate that for a distributed hierarchical set of auctioneers, for which the message rounds cost of finding equilibrium are reduced to logarithmic in the number of traders, the peer-to-peer system will still produce better performance for systems with more than 5,000 traders.