For you all. Interesting because it would appear to be a real live marketplace, with real live money, tho no crypto per se in there. --Ed ------- Forwarded Message From: rkeca04@uts.manchester-computing-centre.ac.uk (Thomas Krichel) Subject: Arizona token exchange To: cti-econ@mailbase.ac.uk Date: Thu, 10 Dec 92 11:55:03 GMT Greetings, this is an item from corryfee. Apologies to those who have already seen it. I thought it was sufficiently interesting to pass it on. BTW, the some of the results of the Santa Fe experiment, to which the text refers, will be published in a forthcomming book by John Rust and Dan Friedman "Double Auction Markets: Theory, Institutions, Evidence". Publisher is Addison-Wesley (Santa Fe Institute series On Studies in the Science of Complexity). Cheers, ______ / / Department of Economics / /_ ________ __. _ University of Keele / / /_(_) / / <_(_/|_/_) ST5 5BG _ , _ Great Britain ' ) / / // Tel.: 44 - 782 - 714847 /-< __ o _. /_ _ // Fax: 44 - 782 - 717577 ./ )/ (_<_(__/ /_</_</_ E-mail: Krichel@uts.mcc.ac.uk

ANNOUNCING THE OPENING OF THE

ARIZONA TOKEN EXCHANGE (AZTE)

January 1, 1993

Earn cash profits by competing against computerized program traders

THE CHALLENGE

Is "artificial intelligence" superior to human intelligence? In some domains such as chess, computer programs now outperform all but the very best human players. However in other domains such as speech, handwriting, and other kinds of pattern recognition, computers lag far behind human beings. On Wall Street computer "program traders" are becoming increasingly common, yet there is substantial controversy over their performance -- they have even been blamed as a factor in the October 1987 stock market crash. The purpose of this study, co-sponsored by the University of Arizona's Economic Science Laboratory and the Santa Fe Institute, is to compare the performance of human and program traders to see whether humans can learn to exploit the limitations and idiosyncracies of computers in repeated interactions.

THE ARIZONA TOKEN EXCHANGE

To compare the performance of human and program traders we have created a computerized market, the Arizona Token Exchange (AZTE), in which a fictional commodity, "tokens", are traded. The market is a simplified version of commodity exchanges such as the Chicago Board of Trade where buyers and sellers are able to call out bids and asks to buy or sell units of the commodity. In each trading session on AZTE traders are assigned the role of buyer or seller and are given an allocation of tokens. A seller's objective is to sell their tokens for as much as possible above the token cost and a buyer's objective is to buy tokens as cheaply as possible below their redemption value.

By ranking the token costs and redemption values, well-defined supply and demand curves can be constructed. The intersection of these curves defines the so-called competitive equilibrium (CE) price and quantity, at which neoclassical economic theory predicts all trading will occur. The complication is that in the AZTE, each trader's token costs and redemption values are private information and differ from trader to trader. Thus traders in the AZTE face a complex sequential decision problem: how much should they bid or ask for their own tokens, how soon should they place a bid or ask, and under what circumstances should they accept an outstanding bid or ask of some other trader? An additional complication is that each trading session runs for a fixed amount of time. This creates a difficult trade-off, for if traders spend too much time looking for a good deal, they may find themselves locked out of the market without trading anything.

HOW IS TRADER PERFORMANCE EVALUATED?

In the AZTE there is a well-defined performance measure: trading efficiency, EFF. This is the ratio of profits a trader actually earns divided by the profits it would have made if all trades took place at the competitive equilibrium level. Thus, if a trader's EFF is greater than 100% they are earning more than their "fair" share of the profits. The use of EFF is more desirable performance measure than simply using trading profits, since profits depend on the token allocations which are allocated at random from a known distribution.

After each trading session, participants will earn cash profits equal to the following linear function of their efficiency:

$ payments = a + b(EFF-100)

The term a represents a fixed fee paid for participating in the trading session and the term b(EFF-100) represents a bonus (penalty) for trading above (or below) 100% efficiency. Thus, it is possible to lose money in any particular trading session. Dollar earnings are cumulated over successive trading sessions and subjects are eligible to "cash out" at any time after participating in a minimum number of trading sessions (provided cumulative earnings are positive).

THE OPPONENTS: COMPUTER PROGRAM TRADERS

Unlike real commodities markets where most traders are humans, in the AZTE all of your opponents will be computer programs. The opponent programs will be selected from a field of over 30 different trading strategies including winners of the Santa Fe Institute's Double Auction Tournament held in March, 1990. The program traders range in sophistication from simple rules of thumb (such as Gode-Sunder "Zero-Intelligence" strategy) to sophisticated optimizing/learning algorithms (such as neural nets and genetic algorithms) developed from the recent literature on artificial intelligence. The identities of your opponents will (usually) be revealed to you at the start of each trading session. You will also be informed about other market characteristics such as the number of buyers and sellers, the number of tokens, and the joint distribution from which token values are drawn.

SETTING UP AN ACCOUNT

To trade on the AZTE you will need a Unix or PC-compatible computer linked to the Internet computer network. We provide the trading interface software that allows you to log on and trade at any time you like and for as long as you like (subject to general restrictions). To qualify for an AZTE trading account you need to file an application providing information on your address, phone, and email address, and a release form stating whether or not you want to remain anonymous in published analyses of the outcome of this experiment. Upon receipt of the application we will set up a trading account and access password. Your dollar earnings will cumulate in your account until you decide to cash out, at which time we will close your account and mail you a check for the total amount of your earnings.

The software and ASCII traders' manual (including the application form) is available via anonymous ftp on "fido.econ.arizona.edu", in the azte sub-directory. The manual (azte.man) explains how the software works and what is required to use it. We suggest you ftp this first and read through it, then get the appropriate trading interface for your system. The DOS interface requires VGA graphics resolution and the use of Clarkson packet drivers for the network interface card. The Clarkson drivers are also available via ftp on "fido.econ.arizona.edu".

If you don't have access to anonymous ftp, we will mail you a diskette containing the software and trader's manual. To cover the costs of a diskette and surface mail, send $5.00 to:

Shawn LaMaster Manager, Economic Science Systems Development Economic Science Laboratory McClelland Hall, Room #116 University of Arizona Tucson, Arizona 85719 (602) 621-6218

Internet: lamaster@ziggy.econ.arizona.edu

We will assist in ftp and setting up the Clarkson packet drivers, just give us a call.

The AZTE software was co-developed by:

Sean Coates Economic Science Laboratory, University of Arizona Shawn LaMaster Economic Science Laboratory, University of Arizona Richard G. Palmer Duke University John Rust University of Wisconsin Vernon L. Smith Economic Science Laboratory, University of Arizona

------- End of Forwarded Message