Web game reveals market sense

By Kimberly Patch, Technology Research News

The exact workings of the financial markets are a mystery. It is clear that the collective decisions of many traders affect financial markets, but it is less clear how traders make decisions, and how these decisions affect each other.

Researchers from the University of Fribourg in Switzerland have tapped the Internet to investigate speculative trading behavior and found that people tend to employ one of two distinct strategies depending on the complexity of a financial market. The results also show that humans are good at filtering information.

In addition to ferreting out information about markets and human behavior, the method could eventually be used to train financial traders, said Joseph Wakeling, a researcher at the University of Fribourg.

The researchers used a Web-based financial game to gain results from several hundred people playing several tens of thousands of game turns against computer-controlled agents.

Playing the game is very simple, said Wakeling. It provides a market price history and asks players to predict if the next price movement will be a rise or fall.

The underlying mechanism that determines what happens is less simple, Wakeling said. For each person there are 94 computer-controlled players. Each player independently chooses to be a member of one of two groups -- those predicting a rise, or those predicting a fall. Whoever is in the smaller of the groups -- the minority group -- wins that round and gains points. Those in the majority group lose points.

The price movement in the game is the difference in size between the two groups, said Wakeling. "If one group -- we can call them buyers -- is bigger, then the price rises by the size difference. If the other [group] -- sellers -- is bigger, the market falls by the difference," he said.

The only information the human and computer-controlled agents have about the market is the correct choices from the past few rounds. Predicting the market actually means predicting which of the two groups will be larger, said Wakeling. "We then assume that [players] will want to join the other group, which they think will be smaller, and so by doing this they affect the actual outcome of the market," he said.

The computer-controlled agents act as controls and make decisions using simple, well-defined strategies. The approach allows researchers to investigate the behavior of a single human in an environment that involves collective actions.

The results showed that human players are "quite good at spotting and exploiting market inefficiencies; they're also good at spotting what information is superfluous and not using more than is necessary," said Wakeling.

When the market complexity is below a certain level most players are able to use a logical, deductive approach to get the better of the market, said Wakeling. As market complexity increases, however, there is an observable limit to humans' ability to cope logically, he said. Beyond this threshold, people have to find other methods of decision-making.

That players' logical capacity should break down like this is not surprising, said Wakeling. What happens next is, however. "People are quite literally repeating the same prediction many times in succession," he said.

More surprising, the strategy performs better than random decision-making, said Wakeling. The open questions are what triggers the behavior change and why the repetitive strategy works.

The researchers have two ideas that may explain the behavior. It may be that as market complexity increases, the number of patterns the player must bear in mind to make a logical decision simply becomes too large to remember, said Wakeling.

Another possibility is that because fluctuations in complex markets are generally very small, it's difficult to try out ideas without actually changing the market situation, Wakeling said. In this case, "an attempt to exploit a pattern can actually destroy it," he said.

Repetitive behavior may outperform random behavior for a similar reason. "Because the market fluctuations are so small, if you change your position, this means that your action decides what the market outcome is," said Wakeling. "So by changing often you can put yourself at a disadvantage."

It could also be that players are picking up a different pattern than the one they use in simple markets. Over any given time period in a market, "there will be a slight bias in one direction -- the market is rising overall, or falling overall," said Wakeling. "If you can work out what the long-term trend of the market is, by repeating the same action throughout that period you can exploit that slight imbalance," he said. The process is probably not conscious, but instinctive, he added.

The results also suggest that there is a real limit on the human ability to spot useful information in the markets, said Wakeling. If this is true, "contrary to the propositions of neo-classical economics, there will always be some inefficiencies left behind in the market," he said.

Today's relatively fast Internet connections made the experiment possible, said Wakeling. "Our experiment was able to take place because we now have fast Web browsers which can transmit dynamically-changing graphics at high-speed," he said. This allowed for a graphical interface without users having to download a program, which meant more subjects and thus quicker data for the researchers. "You simply log onto the Web site and you can play -- it's all there in your Web browser," said Wakeling.

The researchers used a Web-based C program to do the number crunching and used Flash to construct the graphical interface.

The next step is to do more testing to find out why the transition between deductive and repetitive behavior exists, and why players choose the repetitive strategy rather than something else, said Wakeling.

The researchers' long-term goal "is to have a proper theoretical understanding of how humans make economic decisions, and how those individual decisions add up to the macroscopic behavior we see around us every day," said Wakeling.

A system to train financial traders that is based on the interactive minority game could be developed within three or four years, said Wakeling.

Wakeling's research colleagues were Paolo Laureti, Peter Ruch and Yi-Cheng Zhang. The work is slated for publication in Physica A. The research was funded by the Swiss National Science Foundation.

Timeline:   3-4 years
Funding:   Government
TRN Categories:  Applied Technology
Story Type:   News
Related Elements:  Technical paper, "The Interactive Minority Game: a Web-Based Investigation of Human Market Interactions," slated for publication in Physica A and posted at arxiv.org/abs/nlin.AO/0309033


November 5/12, 2003

Page One

Crystal bends light backwards

Micro waterflows make power

Web game reveals market sense

Crystal fiber goes distance

Stored data continues to swell
Electrons spin magnetic fields
Textbook queries video
Rig fires more photon pairs
Process prints silicon circuits
Paired molecules store data


Research News Roundup
Research Watch blog

View from the High Ground Q&A
How It Works

RSS Feeds:
News  | Blog  | Books 

Ad links:
Buy an ad link


Ad links: Clear History

Buy an ad link

Home     Archive     Resources    Feeds     Offline Publications     Glossary
TRN Finder     Research Dir.    Events Dir.      Researchers     Bookshelf
   Contribute      Under Development     T-shirts etc.     Classifieds
Forum    Comments    Feedback     About TRN

© Copyright Technology Research News, LLC 2000-2006. All rights reserved.