Behavior of Trading Automata in a Computerized Double Auction Market

This paper reports the results of a series of tournaments held at the Santa Fe Institute beginning in March, 1990 in which computer programs played the roles of buyers and sellers in a simplified synchronized double auction market. We show that despite the decentralized nature of the trading process and traders' incomplete information about supply and demand, transaction price trajectories for a heterogeneous collection of computer programs typically converged to the competitive equilibrium, resulting in allocations that were nearly 100% efficient. We also show that a very simple trading strategy is a higWy effective and robust performer in these markets. A simple rule-of-thumb was able to outperform more complex algoritllms that used statistically-based predictions of future transaction prices, explicit optimizing principles, and sophisticated "learning algorithms". Acknowledgements: This paper is forthcoming in Tile Double Auction Ma.rket: Institutions, Tlleories and Evidence D. Friedman, J. Geanakoplos, D. Lane and J. Rust (eds.) Addison Wesley, 1992. We are grateful for the generosity of the Santa Fe Institute and its Economics program for providing the facilities, salary, administrative, and computational support that made this research possible. John Rust is grateful to the Alfred Sloan Foundation for providing tournament prize money, and National Science Foundation grants SES-8721 199 for computer hardware, and SES-901oo46 for funding of matching human experiments at the Experimental Science Laboratory at the University of Arizona (Vernon Smith, Co-PI). International Business Machines provided funding for tournament organizational expenses. John Miller would like to acknowledge an equipment grant from Sun Microsystems. The general idea of the computerized double auction toumament emerged from discussions in a March 1988 meeting at the Sante Fe Institute, including Phil Anderson, Ken Arrow, Brian Arthur, John Holland, Tim Kehoe, Richard Palmer, John Rust, Tom Sargent, and Eugenia Singer. We also acknowledge helpful discussions with Robert Axelrod, Charles Plott, Vernon Smith, and Shyam Sunder, and extend our tllanks to particular individuals whose expertise made this tournament possible, including Michael Angerman, Marcella Austin, Ronda Butler-Villa, Steven Pope, Ginger Richardson, Dan Schneidewend, Andi Sutherland, George Tsibouris, and Della Ulibarri. We are grateful for helpful comments from Dan Friedman and seminar participants at the Universities of Oslo and Stockholm, and the London School of Economics. Finally we would like to thank all participants of tlle DA tournament for tlleir willingness to commit time and effort in developing an ingenious collection of trading programs.