Why was the Santa Fe Double Auction Tournament Held?

Why was the Santa Fe Double Auction Tournament Held?


In 1990, John Miller, Richard Palmer, and John Rust sponsored a double auction tournament at the Santa Fe Institute. $10,000 in prize money was offered to human contested who submitted computerized trading programs to play the role of buyers and sellers in the computerized Santa Fe double auction market. The purpose of the tournament was scientific: to study artificial versus human intelligence and the economics of price formation. Results of the double auction tournment are reported in the paper, Characterizing Effective Trading Strategies: Results from a Computerized Double Auction Tournament pubished in the Journal of Economic Dynamics and Control in 1993. While the double auction market has been studied theoretically by economists (see the book The Double Auction Market by Palmer, Rust and Miller (eds.), Westview Press, 1993, for more details), the standard way of modeling this in economics is as a continuous time game of incomplete information, and to date, nobody has been able to solve this game, or even compute rough approximations to its solution to determine how idealized rational traders should actually behave in this environment. Robert Wilson has made the most progress towards solving this problem in a paper, "Equilibria of Bid-Ask Markets" in Arrow and the Ascent of Economic Theory: Essays in Honor of Kenneth J. Arrow, G. Feiwel (ed.); Chapter 11, pp. 375-414. London and New York: Macmillan Press and New York University Press, 1987. In Wilson's characterization of the solution (for a special case where buyers and sellers each only have a single unit of the commodity to buy or sell), the double auction can be viewed as a type of "waiting game" where traders are reluctant to move too soon to place a bid or ask for fear of revealing their (private) valuations for the commodity, and thereby disadvantaging themselves relative to the other traders. However traders do not want to wait too long, since the double auction only lasts for a finite length of time, so delaying too long raises the risk that they might not trade at all. Wilson showed that there is an equilibrium to the double auction game (although in general there could be many different equilibria) that can recursively decomposed into a series of "subauctions". That is, after waiting a random length of time, the buyer with the highest valuation and the seller with the lowest cost call out bids and asks and trade with each other. This results in a double auction subgame with N-2 traders left, and after a further random delay the buyer with the next highest valuation trades with the seller with the next highest cost, and so on. To actually compute how much traders should bid, asks, or how long they should delay before placing them, whether to accept an existing outstanding bid or ask, etc. To determine these details, one would have to solve a very complicated system of nested partial differential equations. To our knowledge this has never been done, and so to date economists do not fully understand all the implications of these idealized game-theoretic models of the double auction market. further references Another approach to studing the double auction market is via laboratory experiments. A considerable amount has been learned about how actual humans actually play the double auction game, and what types of outcomes result from it.

In 1990, John Miller, Richard Palmer, and John Rust sponsored a double auction tournament at the Santa Fe Institute. $10,000 in prize money was offered to human contested who submitted computerized trading programs to play the role of buyers and sellers in the computerized Santa Fe double auction market. The purpose of the tournament was scientific: to study artificial versus human intelligence and the economics of price formation.

Results of the double auction tournment are reported in the paper, Characterizing Effective Trading Strategies: Results from a Computerized Double Auction Tournament pubished in the Journal of Economic Dynamics and Control in 1993.

While the double auction market has been studied theoretically by economists (see the book The Double Auction Market by Palmer, Rust and Miller (eds.), Westview Press, 1993, for more details), the standard way of modeling this in economics is as a continuous time game of incomplete information, and to date, nobody has been able to solve this game, or even compute rough approximations to its solution to determine how idealized rational traders should actually behave in this environment. Robert Wilson has made the most progress towards solving this problem in a paper, "Equilibria of Bid-Ask Markets" in Arrow and the Ascent of Economic Theory: Essays in Honor of Kenneth J. Arrow, G. Feiwel (ed.); Chapter 11, pp. 375-414. London and New York: Macmillan Press and New York University Press, 1987.

In Wilson's characterization of the solution (for a special case where buyers and sellers each only have a single unit of the commodity to buy or sell), the double auction can be viewed as a type of "waiting game" where traders are reluctant to move too soon to place a bid or ask for fear of revealing their (private) valuations for the commodity, and thereby disadvantaging themselves relative to the other traders. However traders do not want to wait too long, since the double auction only lasts for a finite length of time, so delaying too long raises the risk that they might not trade at all. Wilson showed that there is an equilibrium to the double auction game (although in general there could be many different equilibria) that can recursively decomposed into a series of "subauctions". That is, after waiting a random length of time, the buyer with the highest valuation and the seller with the lowest cost call out bids and asks and trade with each other. This results in a double auction subgame with N-2 traders left, and after a further random delay the buyer with the next highest valuation trades with the seller with the next highest cost, and so on. To actually compute how much traders should bid, asks, or how long they should delay before placing them, whether to accept an existing outstanding bid or ask, etc. To determine these details, one would have to solve a very complicated system of nested partial differential equations. To our knowledge this has never been done, and so to date economists do not fully understand all the implications of these idealized game-theoretic models of the double auction market. further references

Another approach to studing the double auction market is via laboratory experiments. A considerable amount has been learned about how actual humans actually play the double auction game, and what types of outcomes result from it.

Back to Main Page	Why a Sequel?