October 19, 1987 has become known as one of the world's stock markets Black Mondays. Part of the reason blamed for the rapid market fall-offs was automated trading.
By 1989, there were calls by Congress and others to stop the use of automated trading; one firm, Dean Witter Reynolds said they were going to stop using it as it "threatened the integrity of the market in customers' minds."
This attitude didn't last long. Program trading volume now runs about 30 - 35% of the New York Stock Exchange on a weekly basis, and occasionally as high as 50%.
Well, here we are twenty-years later, and program trading surfaces as a cause again in the current world's stock markets' wild gyrations, which as of today, has wiped out all this year's market gains. This time, it appears that the algorithms used by the trading programs underestimated (i.e., missed) the risks that the sub-prime mortgage meltdown implied.
In an interesting article in the Wall Street Journal (subscription required), there were many excuses given for why the trading programs failed this time: "A unique combination of factors," " A perfect negative storm," etc., etc.
For more on the problem, just Google the word "quant," which is the modern slang for quantitative automated trading programs.
Using computer models for market prediction is great as long as the current reality meets the model reality. Once they diverge, then they don't work very well if at all. This has been known and warned about for over forty years - yet it is a lesson that people just keep insisting they want to painfully relearn, which is why I call it the déjà vu to the nth power problem.
