in part of this how to make money in microseconds article (found via the discovery blog) donald mackenzie tells a frightening story about automated trading programmes and the algorithm they use.
it happened on 6th may 2010 on the amercian stock market. for a mere 20 minutes starting around 2.40pm the overall prices of US shares, and futures contracts (which are bets on those prices) fell by some 6% in around 5 minutes then recovered almost as quickly.
as 2% is the usual maximum fall for a day this was an enormous drop. indeed some gigantic price fluctuations took place in some individual shares (accenture shares - trading at around $40.50 dropped to a single cent. sotheby’s - trading at around $34 jumped to $99,999.99.
After 5 months of investigation it was found that this "flash crash" was triggered by an algorithm used in an automated trading programme. fortunately the electronic platform on which these trades were executed had a "stop logic" functionality designed to detect and interrupt such self-feeding crashes by giving human traders time to assess what was happening, step in and pick up bargains.
somebody called perrow argues that of course the worry is that the combinations of all these complex algorithm in automated trading systems creates tightly coupled, decentralised, highly complex systems. if something goes wrong it takes time to work out what. but tight coupling means we don't have the time.
postscript - this study from the uk may also be useful