It should shock no one to learn that these sophisticated models sometimes make mistakes. Recent history is wrought with examples of algorithm-driven trading gone awry, from the shocking collapse of a large hedge fund in the late '90s to several high-profile market failures in more recent years and probably scores of horrifying details that have gone unreleased.
The problems ironically begin with the very human tendency to jump to conclusions. Paul Rowady, senior analyst with consultancy Tabb Group, says that high-frequency trading platforms can use social media data as trading indicators, either supporting or rejecting an existing hypothesis.
However, it would become problematic, he said, for a trading platform to mistake a trading indicator for a trading signal. Differentiating between indicator and signal is important. Indicators are derived from raw market-related data, such as content on social networks, and serve as inputs to models. Signals, usually based on multiple indicators, are the output of those models and eventually become orders to buy or sell financial instruments.
Wall Street calls the mathematicians and engineers who develop these models “quants,” short for quantitative analysts. They program the models that turn multiple indicators into a specific order.
“Social media can play a role in the form of trading indicators, but at this stage, not in the form of pure trading signals," he said. "The signal-to-noise ratio for that dataset is simply way too low."
Figuring out how to determine if a set of social media data is good news or bad news for a particular stock or trade is also problematic. It’s easy enough, said Rowady, to use social networks to capture which stocks will move on abnormally high volume in a given day. Reliably establishing whether they will move higher or lower, however, is another matter entirely.
“You can say, 'There’s a headline on this ticker, so volatility is going to go up.' But as to the bias of that volatility, is [the price] going to go up or down? That’s a much harder problem to solve,” he said.
Rowady is referring to the difficulty of measuring negative sentiment on a social network, a familiar barrier for many other industries grappling with social media analytics. Users don’t frequently un-like something on Facebook, and using Twitter to complain about negative consumer experiences is dependent on that user’s personality, not the platform’s architecture.
Where are these blind spots, and how should high-frequency trading platforms treat them? How do algorithms find patterns in language used across the social media sphere while at the same time controlling for false-positives? Those are a few of the questions that are keeping Wall Street from mastering the use of social media to inform high-frequency trading strategies.
Social Content is Unlike Business News
It is incorrect to assume that social media content is similar to breaking business news, which Wall Street has long depended on for trading-related information. Business news, of course, is generally filtered by editors and varying levels of fact-checking, and its impact on the markets is dampened by news cycles and copy deadlines. Generally, everyone gets access to breaking business news at the same time, which is entirely by design. Business news is also reliable in all but the very occasional instance -– a news outlet may report a hoax as fact once in awhile, but the press gets it right far more often than not.
On social networks, content is oftentimes published with little to no editorial standards, and the flow of content is constant. Social networks can disrupt, complement, reinforce, or inform the traditional news cycle, but they do not replace it. Any trading algorithm that replaces news feeds with social content or co-mingles the two is inherently faulty.
Going a step further, it would be fairly straightforward to use fraudulent content to manipulate high-frequency trading platforms that misuse social media data. That's one reason why the Derwent fund set to launch using Twitter is routinely met with skepticism among some Wall Street insiders.
Realistic Expectations for the Future
For Wall Streeters, however, there are a few reasons to be excited about social media’s potential utility, despite the occasional red flag.
Because social media is a global phenomenon, capturing sentiment in hard-to-reach markets could become easier. Some of these markets don’t have established mainstream press -- or, if they do, the press may be censored -– making the technical difference between "real" news and social content immaterial to the strategy.
Certain algorithms can also analyze words in a language-neutral manner. The contextual advertising industry is currently making strides in using multilingual or language-agnostic technology to deploy paid content, and the financial markets probably won’t lag far behind.
Perhaps the biggest potential lies with social media’s ability to reinforce or refute certain signals that ultimately lead to trading decisions. It’s akin to a friend giving the final arm-twist that drives you to order ice cream for dessert, or a helpful mentor saying you probably shouldn’t make that one decision. Social media wouldn’t be an adequate determinant by itself. But paired with all the other available information, it could nudge a set of observations into an actual decision.
If Wall Street ends up using social networks for this purpose, it would represent a shift in the role of automated and high-frequency trading and potentially earn a role for social media in the market.
