Trading bots create extreme events faster than humans can react Our algorithms now show collective behavior that we do not control.
by John Timmer - Sept 16 2013, 9:30am +0950
66 Flickr user: Andreas Poike
High-frequency trading is the practice where automated systems search for minor differences in price of stocks that can be exploited for small financial gains. Executed often enough and with a high enough investment, they can lead to serious profits for the investment firms that have the wherewithal to run these systems. The systems trade with minimal human supervision, however, and have been blamed for a number of unusually violent swings that have taken place in the stock market.
A new paper has gone searching through historic trading for these sorts of glitches and ended up finding a lot of—over 18,000, all of which took place too fast for human intervention to have driven them. When they generated a mathematical model of this trading, they found that they showed indications of many traders executing a similar strategy, exactly as you'd expect from automated trading systems. The rise in this style of trading appears to be an emergent property of computerized trading, and it seems to have reached an inflection point near the start of the financial crisis.
The primary victims of these glitches? The stocks of the investment banks themselves.
The primary feature of these systems is speed, since the market conditions they seek to take advantage of are ephemeral. The authors of the new paper (a mix of academics and private sector employees) note that a new fiber optic cable is being laid across the Atlantic just to shave five milliseconds off the network transit times. Some companies are offering processors customized to execute trades even faster. This all ensures that the trading bots are acting far too fast to have direct human supervision. Fully attentive individuals will, at best, take close to a full second before they can react to changing circumstances.
To identify activities that might be triggered by automated systems, the authors defined something called an ultrafast extreme event (UEE). These are cases where a stock price moved at least 10 consecutive times in the same direction, all within 1,500 milliseconds. The total magnitude of these mini-crashes and rises had to be at least 0.8 percent. That may not seem like much, but it represents over 30 standard deviations from the normal run of trading.
Searching through the data, the authors found over 18,500 of these UEEs. As the authors narrowed the time window to below one second (the limit of human reaction time), the number of spikes and crashes increases rapidly, indicating that the behavior is likely to be triggered by something that can act faster—the automated systems. As a result, the paper's title makes reference to a "new machine ecology" represented by the behavior they trigger in the stock market.
The first UEEs in the data set appeared in late 2006, and they grew rapidly and steadily from there. Right as the financial crisis started in 2008, however, they experienced a very significant growth spurt. The other notable thing linking them to the financial crisis is that the 10 stocks that experienced the most UEEs were all banks, including some that would later go bankrupt or be sold.
The authors next built a mathematical model in which a large number of traders were able to employ an equally large number of trading strategies. As they steadily decreased the number of strategies, their market model showed a threshold effect—events like UEEs appeared suddenly once the threshold was crossed. They also referenced a separate paper where another team showed a similar threshold appearing as the reaction times of computerized traders dropped below one second.
Overall, the authors make a pretty compelling case that fast trading systems, coupled to a limited number of trading strategies, have caused a fundamental change in the behavior of the stock market. This shift leads to sudden changes in the value of stocks that aren't linked to any underlying financial factors. (In most cases, the change seems to be transient, and stocks return to their former value rapidly.) What's not at all clear is what triggers these UEEs, and whether changes in market regulations or trading strategies could eliminate them.
