Twitter Can Predict The Stock Market, If You’re Reading The Right Tweets

In a world where one tweet can send Wall Street into a panic, social analytics company Dataminr tries to be there first, scanning all of Twitter to find individual messages with the right combination of language, context, and location that might end up being breaking–and money-making–news.

Twitter Can Predict The Stock Market, If You’re Reading The Right Tweets
Stock Chart via Shutterstock

Earlier this week, the Associated Press Twitter account posted the following–false–message: “BREAKING: Two Explosions in the White House and Barack Obama is injured.” It didn’t take long for other sources to demonstrate that the president was fine and the AP account hacked, but it was long enough for the stock markets to take a nose dive. The Dow Jones and S&P 500 indexes fropped by close to 1%, the equivalent of hundreds of billions of dollars changing hands.


The incident may prove that Twitter needs better security and algorithmic traders need better quality control, but it’s also evidence of something simpler: News impacts financial markets, and that news is increasingly breaking on Twitter. Dataminr–a social analytics company with clients in finance, government and the larger corporate world–takes this dynamic one step further. They use Twitter to beat the news. “It’s the lack of someone who is a news commentator or a news source saying it,” says Dataminr founder and CEO Peter Ted Bailey. “The point is the things that aren’t there.”

Dataminr can find market-moving information not yet in the news because they aren’t limited to following some manageable group of friends or trusted accounts. They have access to the entire “Firehose” of Twitter’s approximately 200 million active users, and they use it. “We look at every user across Twitter and understand everything that they’ve published and their relative influence on any topic that we know and their local influence,” Bailey says.

The “we” that looks at that deluge of data isn’t Dataminr’s approximately 35 employees; it’s their algorithms. To understand how they work, it helps to have an example.

On March 8, a Royal Caribbean cruise ship arrived in Port Everglades, Florida, with 105 passengers and three crew members sick with norovirus. When that news broke, it sent Royal Carribean Cruises Ltd. Share prices tumbling by 2.9%. But Dataminr clients had the news 48 minutes earlier.

The tweet that tipped them off came at 1:00 p.m., from South Florida news channel WSVN: “Royal Caribbean’s Vision of the Seas cruise ship has pulled into Port Everglades after an outbreak of norovirus on board.” Dataminr’s algorithm found that tweet, and not by searching for “norovirus” or “Royal Caribbean.” ”We detected a slight blip, linguistically,” Bailey tells me, again using “we” to denote the software. “And we saw that the source who published it was one that had local influence.”

The algorithm found that words within the tweet had some resemblance to tweets in the past that had turned out to be newsworthy, and that there was a clear immediate reaction to the tweet, though it had not yet rippled out to national news sources and market commentators.


All of these algorithmic calculations were made with great speed. At 1:02 p.m., only two minutes after the original tweet, relevant Dataminr clients got an email and an alert started flashing in the bottom corner of the screens. It provided not just the WSVN tweet, but an analysis of why it was important. “It’s like, ‘How much context can you possibly put around a tweet?’” said Bailey.

As for what clients did with this early information, Bailey says he is “contractually confined” from giving details. But Dataminr said at least one client told them directly that the alert saved their firm money. Dataminr has always declined to name their clients in the press, but in a presentation at the 2011 Devnest meet-up, they did say that their clients included “three of the top five bulge-bracket investment banks, as well as a leading $15 billion equities hedge fund.”

In the case of Royal Caribbean, traders knowing market-moving news 48 minutes in advance probably meant a big one-time payday. But Dataminr also has government clients–they’re currently hiring three government-focused employees. It’s easy to understand state interest in Twitter intelligence in an age where “Twitter Revolution” recurs in headlines every few months. Just as financial clients use intelligence to further their bottom lines, we can only assume that Dataminr’s clients in the government will be putting the Twitter “Firehose” to work for their own goals, at home and abroad.

About the author

Stan Alcorn is a print, radio and video journalist, regularly reporting for WNYC and NPR. He grew up in New Mexico.