Tweet Retreat: Did high-frequency reading crash the market?
FORTUNE — As Wall Street predictions go, Jamais Cascio had a good one. A little less than a year ago, Cascio, a distinguished fellow at think tank Institute for the Future, in a blog post, predicted that retweeting Twitter bots combined with a fake news story posted by hackers on a major media website would cause a market crash. That’s pretty close to what happened.
By now, you probably know about the so-called Tweet Retreat. On Tuesday, the market dropped 140 points within minutes of a fake tweet about a bombing at the White House. It came from the AP, which said it was hacked. The market recovered just as fast.
Nonetheless, the micro-crash added a new worry to those who are concerned that a glitch in the increasing number of computers trading stocks at lightning-fast speeds will sink the market and erase our retirement accounts: High-frequency reading.
In fact, Wall Street’s computers have been “reading” the news for a while. Dow Jones and Thomson Reuters (TRI) both offer services that convert their news articles into a digitized form that can be fed into computer trading algorithms. They do this by tagging individual words and then assigning a sentiment ranking to those words. The computers add up those sentiment ranks to determine if the news is negative or positive for a particular company or the stock market in general. A sentence with the words bombing, White House, Obama, and injured would presumably get a pretty negative rating.
Back in 2009, Streambase, a company that provides technology to high-speed trading firms, began giving its clients the ability to incorporate tweets into their trading programs that decide when to automatically buy or sell. It’s not clear whether the Dow Jones and Reuters services include twitter in their digitized news feeds. Bloomberg recently began incorporating tweets into its terminals. Last year, at a Wall Street technology conference, Harvard professor Chris Malloy estimated that nearly one-third of all high-frequency trading firms were evaluating whether to include such things as news articles along with traditional market data into their computer trading algorithms.
So it’s likely that the computers of a good number of high-speed trading firms “read” the fake White House bombing tweet.
Still, it’s unlikely that computers were the ones that started Tuesday’s micro-crash. The fake tweet went out 1:07 PM and 50 seconds. According to Eric Hunsader, who runs market research firm Nanex, the first high-frequency trading firms didn’t react to the fake tweet until 15 seconds after that. Most algorithms are written to respond to data within milliseconds. What’s more, most only respond to specific market-moving news events, like the monthly jobs number.
So Hunsader thinks humans made the first trades based on the fake tweet. Nonetheless, computer trading algorithms, trained to follow the market, likely piled into the selling once it got started. Hunsader says that computer trading systems may have incorporated the potentially bad news into their algorithms. When actual selling began to occur, those systems were primed and headed for the exits faster than usual. The result: Hunsader says 20% of the market’s trading volume disappeared in 20 seconds, a faster retreat than what was seen in the flash crash of a few years ago.
“It tells me that the diversity of traders in the market is gone,” says Hunsader. “Just a bunch of machines that act like lemmings.”
But given the fact that the market rebounded, as it had done in the past, it’s hard to tell who got hurt. Mutual funds are trading all the time, slowly working through their large orders. High-frequency traders, Hunsader says, prey on mutual funds and other slow-moving traders. Events like the Twitter crash allow them to really take advantage.
“High-frequency traders are always taking portions of pennies away all the time. In this case they were able to take a whole bunch of whole pennies at once,” says Hunsader. And Hunsader said that’s typical in the market today. “No one really gets killed in these micro-crashes, but there are a lot of paper cuts.”
Worse, Hunsader says high-frequency trading firms are likely to view the Twitter crash as an opportunity. Now that they have seen how fast a tweet, even an erroneous one, can move the market, traders are sure to do what they can to capitalize on it next time.
“If anything, this is likely to accelerate the development cycle of [reading] technology,” says Hunsader. “In the future, the impact of a false tweet could be a lot worse.”