Automated, high-speed stock trading on the rise
One of the more interesting trends in stock market trading these last few years has been so-called “high-frequency trading,” which uses the combination of sophisticated high-speed computers and a firehose of information about the market to execute trading strategies that would be completely impossible by hand.
Using a preferred connection, complicated algorithms can detect subtle trends in the market and do things like try to lead the curve in real-time, or subtly probe the other traders to see what their limits were. It’s the latter trick that has led some traders to cry foul, as the algorithms propose trades and cancel them as fast as possible, just to try and figure out the behavior of other individual traders. It’s a pattern of behavior specifically designed to stretch prices as far as possible, at the expense of liquidity and other attributes that are healthy for the stock market to possess.
And there’s a further concern: as an anonymous hedge fund manager notes in a sprawling and fascinating interview, most of these automated traders are running algorithms based off the same theories about the stock market; if one hits a blind-spot, they all hit a blind-spot. He tells a story from August of 2007:
The problem is that the DNA of a lot of these models is very, very similar, it’s like an ecosystem with no biodiversity because most of the people who do [statistical arbitrage] can trace their lineage, their intellectual lineage, back to four or five guys who really started the whole black box trading discipline in the ’70s and ’80s. And what happened is, in August, a few of these funds that have big black box trading books suffered losses in other businesses and they decided to reduce risk, so they basically dialed down the black box system. So the black box system started unwinding its positions, and every black box is so similar that everybody was kind of long the same stocks and short the same stocks. So when one fund starts selling off its longs and buying back its shorts, that causes losses for the next black box and the people who run that black box say, “Oh gosh! I’m losing a lot more money than I thought I could. My risk model is no longer relevant; let me turn down my black box.” And basically what you had was an avalanche where everybody’s black box is being shut off, causing incredibly bizarre behavior in the market.
We had a loss over the course of like three days that was like a ten-sigma event, meaning, you know, it should never happen based on the statistical models that underlie it. Why? Because the model doesn’t assume that everybody else is trading the same model as you are. So that’s sort of like a meta-model factor. The model doesn’t know that there are other black boxes out there.
