A STOCK market that鈥檚 not only run by computers but designed by them too would generate hundreds of billions of dollars鈥� worth of extra profit for their human controllers. So says artificial intelligence expert David Cliff, who applied the principles of evolution to allow his software to develop the ideal marketplace.
Cliff鈥檚 software created a super-auctioneer that can orchestrate buying and selling in a new type of stock market: one where the traders are all software-based agents. The idea could even introduce an ethical dimension to trading. Cliff found that in his marketplace, fewer people paid inflated prices and everyone was able to maximise profits.
Last year, 快猫短视频 reported on research from IBM which showed that software agents trading commodities such as precious metals or coffee beans outperformed their human counterparts, earning on average 7 per cent more (11 August 2001, p 21). One of the software agents used in that experiment was developed by Cliff, who works at Hewlett-Packard鈥檚 research lab in Bristol.
Advertisement
But Cliff found it odd that these software agents were being shoehorned into computer models of marketplaces designed by people for people. So he set about designing a marketplace that is better suited to artificial buying and selling agents.
To do this, he designed a range of possible markets using software incorporating a genetic algorithm, a type of program inspired by evolution. The GA uses a process similar to natural selection to 鈥渆volve鈥� continually until it reaches the best solution to a problem. The idea is that for every set of trading conditions, there is an ideal marketplace, and it鈥檚 this the GA seeks. It evolves by putting a range of parameters through a series of mutations until it maximises profit. For example, Cliff鈥檚 GA found that one of the most critical factors is the number of 鈥渢raders鈥� allowed to bid at any one time.
Today鈥檚 markets use a system known as a continuous double auction, where buyers and sellers simultaneously yell offers at each other, with the sellers slowly lowering their asking prices and the buyers gradually increasing their offers.
The GA found that the fittest market of this sort was one that allowed three buyers to make bids for every sell offer, generating 2 per cent more profit than a typical market. In the real world, with trillions of dollars percolating through the markets, this small gain might translate into hundreds of billions of dollars, says Cliff.
One crucial difference between the evolved marketplace and a human one is that traders have to reveal their maximum and minimum prices to the auctioneer program, though not to each other, before the market can be set up for trading. This gives it key data on which to base its evolution.
But this strategy would only work if the information were kept confidential and was known only to the marketplace itself, says Lawrence Barwick, head of proprietary trading at the Bank of America in London. 鈥淚f it鈥檚 closed like that and everybody has to play by the same rules then it could work.鈥�