How Your Tweets Could Crash The Global Economy

A new study outlines the frightening ways financial algorithms harvest social media data — including yours.

Two years ago, a hacker took control of the Associated Press Twitter account and tweeted, "Breaking: Two Explosions in the White House and Barack Obama is injured." Within two minutes, the Dow Jones dropped nearly 150 points and the S&P 500 had lost nearly $150 billion in value. Within five minutes, the market had recovered.

The whiplash caused by the tweet was held up by many in the press as an example of the power — and ultimately the correctness — of the proprietary, lightning-fast algorithms that automatically execute financial trades based on reams of incoming data. Though these algorithms are closely held secrets, it is widely understood that they take into account social media data, including tweets. That a tweet from the AP could cause this kind of market fluctuation was seen as a kind of double proof of the sophistication of these formulas, which recognized and heavily weighted a tweet from a world-renowned news organization, then recognized it as a fake and course-corrected.

For Tero Karppi, a professor of media theory at the University at Buffalo, the drama raised much more fundamental questions about the way online speech affects the global financial market.

In a new case study, "Social Media, Financial Algorithms and the Hack Crash," Karppi, along with co-author Kate Crawford of Microsoft Research, mapped the ways that social media data intersects with financial algorithms, and the potential consequences of that integration.

"The biggest surprise was just how interconnected the systems are," Karppi told BuzzFeed News. "There are systems that buy access to social media and mine that data. Social media plays a significant role in financial markets."

Indeed, much of the study focuses on services like Dataminr (which played a role in the 2013 crash and recovery) that harvest data from social networks and turn it into "actionable signals" for financial companies. As the study puts it, these companies "assess emotion, importance and social meaning in order to 'predict the present' and thus transform social media signals into economic information and value." (Through a spokesperson, Dataminr declined to comment for this article.)

The problem with converting social media speech to algorithmic data, according to Karppi, is that this speech is not necessarily accurate or truthful. Indeed, the 2013 crash was precipitated by a trusted Twitter account being amplified by thousands of people at once who had no idea they were spreading a lie. "People don't necessarily represent their actual being while they are on social media," Karppi said. "There seems to be this neo-positivistic epistemology where we believe the data we gather from social media actually represents reality in some way. I think we need to be critical towards that."

Software like Dataminr performs so-called "sentiment analysis": measuring how people feel from their online speech. It's not hard to imagine a series of performatively negative tweets about a marketing campaign, or a global news event, snowballing into a speech trend that sentiment analysis converts into data points for trading algorithms. And that's where the trouble could begin.

The algorithms into which firms funnel social media content perform so-called high-frequency trading, and they value speed above all, a fact that can lead to "weird and scary consequences," said Karppi. As he and Crawford write in the study, "algorithms and other actors respond to sudden changes in financial markets which are then imitated and repeated; when someone or something begins to sell in earnest, other entities follow." In other words, our online speech can indirectly lead to to huge self-fulfilling prophecies that shake financial markets, as algorithms follow other algorithms that are following faulty social media data.

It's a bracing idea, that our stray thoughts — and more importantly, the digital speech of influential people, brands, and organizations, many of which hardly pay a pittance to security — are imperfectly integrated into a vast layering of formulas that play a dominant role in determining the economic health of the world. Even more frightening is the fact that we have no idea how our speech is being weighted, because the math is all kept secret.

And because all of these processes happen faster than human cognition, the only solution, according to Karppi, is building better automated systems to govern them — another layer of abstraction.

"No one has the time to check if its true or not," Karppi said.

In the shadow of a global economic collapse that was characterized by the massive abstraction of financial products, it's yet another example of the sheer complexity of global financial system, and our complicity in it, whether we like it or not.

Skip to footer