Human traders don't have a chance against bots which have been competing among themselves for a decade, resulting in a super-bot whose adaptive aggression annihilates all animates. A decade has passed since IBM's seminal demonstration that software bots could beat humans at securities and commodities trading. Since then the bots have been competing among themselves until now, on the 10-year anniversary of bot's supremacy, an architecture called AA (adaptive-aggressive) has emerged as king-of-the-hill.
The International Joint Conference on Artificial Intelligence (IJCAI), held last month in Barcelona, marked a decade since its 2001 meeting where software robots (bots) proved their supremacy over humans at stock-market trading. Technically called a continuous double auction--since both buyers and sellers set prices on their offers--CDA today dominates the trading activity on nearly all commodity and stock exchanges. Most institutional investors use bots that are variations on the well-known CDA strategies which were outlined a decade ago in an IBM paper at IJCAI 2001 entitled: Agent-Human Interactions in the Continuous Double Auction.
In that paper, software bots were shown to consistently outperform human traders on real-time CDA markets, in particular by using two trading-agent strategies called ZIP (zero-intelligence plus) and GD (after the economists Steven Gjerstad and John Dickhaut). In subsequent work, the same research group at Thomas J. Watson Research Center reported an improved strategy they called GDX, which outperformed both ZIP and GD when bots were pitted against other bots (as is the case today on most modern stock and commodities exchanges).
The performance of IBM's GDX continues to be used as a figure-of-merit for commercial bidding strategies, but according to a new paper presented this year at IJCAI 2011, a newer strategy called Adaptive Aggressive (AA) outperforms ZIP, GD and GDX in bot-versus-bot trading. The new paper entitled: Human-Agent Auction Interactions: Adaptive-Aggressive Agents Dominate was presented last month at the conference by professor Dave Cliff and doctoral candidate Marco De Luca, both of the University of Bristol (England).
The AA algorithm differs for the other algorithms in that it aggressively trades off profit for the ability to make a transaction, it then adapts its aggressiveness after each completed buy or sell. By mimicking the method used in IBM's original paper to judge the bots, Cliff and De Luca claim to show that AA outperforms all the other strategies in both bot-to-bot and human-versus-bot trading.
Further Reading
