Tweet. Like. Follow. Flag.
These are some of the newest terms in traders lexicons as social media has gone from just a fun pastime of online jokes and photo postings to integral part of execution strategy with data and charts. While social media remains a hotspot for compliance departments to contend with, just looking at Twitter one can see professional traders, news agencies, market pundits and others openly voice their opinions on the market, market structure and just about everything else. But has social media had a definitive impact on trading? Have traders replaced the traditional newswire or newspaper with tweets, Facebook posts or LinkedIn features?
Ian Domowitz, Vice Chairman, ITG told Traders Magazine that Twitter has changed the information used for trading by the retail community, as opposed to professional (e.g. hedge fund and institutional) trading).The phenomenon of trump tweets is a singularity, and should not be confused with general information. The real point: Twitter is public information, and public information is not typically a source of excess returns.
I dont believe that public social media sites affect trading in any material fashion, Domowitz said. Once again, public information. There are Twitter aggregators (like Stocktwits and others) that provide a filtered version of general twitter traffic to eliminate noise relative to the type of news to be disseminated on the site. Perhaps useful to retail, but our own research on the FX market, for example, shows limited predictability, and no impact on general market structure related to trading.
But one new entrant in this space, Pluribus Labs, an unstructured data analytics firm led by former Instinet COO Frank Freitas, sees things differently.
Twitter has definitely enhanced the trading landscape, said Wachi Bandara, Chief Research Officer at Pluribus Labs. One could even make the argument that Twitter constitutes a new form of market data that traders must incorporate into their information gathering process if they want to understand all facets of information that potentially move stocks.
Bandara explained that Twitter usage can be broken down into two distinct categories – event detection and systematic predictive information.
On the event detection side, there have been several well-publicized examples of traders profiting from the early capture of market moving information. Our view is that Twitter is increasingly important from a systematic perspective as well, Bandara said. There is immense value in aggregating the opinions of thousands of traders. We have shown that the volume of messages alone has explanatory power over stock performance. Using a proprietary language engine, we can show that Twitter data can inform profitable trading strategies.
Spencer Mindlin of Aite Group added that firms which desire retail trading business realize that this group can have an outsized ability to create strong short-term movement in some stocks. That being said, quantitative funds have been heavily investing in ways to detect and model retail order flow, and part of their success is contingent on their ability to gauge virality and significance of real-time information in order to predict how the retail crowd will react and a stock will move.
This information is found in the analysis of large data sets of information, Mindlin began. Its unlikely that social media will ever be a standalone predictor of stock movement, but big data sets of information found online and in social media networks will definitely play a significant role in an automated traders ability to detect how retail order flow will impact stock pricing.
Mindlin added the demand for big data and new platforms creating new datasets and distributing them over a variety of financial analytics platforms for investment firms is big business. He cited firms such as Dataminr, StockTwits, Estimize, iSentium, Kensho, and Thinknum are being met with strong demand because they are either creating, aggregating, and/or analyzing alternative sets and making it easy for firms to access this information via portals, terminals, or APIs.
Social media just an early type of big data set, Mindlin continued. Traditional financial statement analysis will become less important and give way to real-time analysis of supply chains and consumer behaviors. And technologies like Hadoop and Spark and whatever the next generations of these analytics platforms will be even more powerful.
This following article originally appeared in the August 2014 issue of Traders Magazine
Twitter Trading Looks for Action
By Renee Caruthers
Talk about a social media addiction. Every day, Cayman Atlantic Investment Management’s systems scan 400 to 500 million tweets looking for a breaking news event. “If the words oil pipeline’ and explosion’ popped up in a tweet, that would ping up an alert and the computer would try to find the geographic area where the event occurred,” said Paul Hawtin, founder and CEO of the managed trading account firm based in the Grand Caymans. “Event detection is a growing area and we are finding it to be very lucrative, but it’s a little bit like finding a needle in a haystack,” he told Traders.
Event detection is Cayman Atlantic’s newest trading technique, and it’s just one-third of the company’s three-pronged approach to trading based on Twitter and other social media. It’s a strategy that is not for the faint-hearted. Aside from the time pressures of looking for news before it is pushed out to the world through newswires, to really capture the advantages of an early news you have to be prepared to react to news that can break out about anything, anywhere. “We are learning all the time to be as broad as possible,” Hawtin said. “There are a lot of events that can have some effect on asset prices somewhere, whether it’s an airplane crashing that would affect the company’s stock price, or a flood in a certain region. It’s about getting that information early.”
For Hawtin, being an early adopter of Twitter-based trading strategies is nothing new. In 2010 he launched the Derwent Absolute Return Fund, which became known in the media as the “Twitter hedge fund.” Despite attracting plenty of media attention eager to showcase social media’s new role in business, the fund failed to attract enough investment to remain viable. “At the time, markets were still extremely volatile and the problem with a hedge fund is you need $25 million or more to make it commercially viable,” he said, adding that Derwent had launched with an initial investment of 650,000 British pounds. Cayman Atlantic is the successor to Derwent.
Besides improving its technology and adding a new trading strategy to the mix, Cayman Atlantic has a new business model: Rather than launching as a hedge fund, Cayman Atlantic offers managed trading accounts for high-net-worth individuals. Investors can open an account for a minimum investment of 100,000 pounds or the currency equivalent. One trading strategy covers all accounts, and the managed account model doesn’t require a minimum threshold.
In addition to event-detection-based trading, Cayman Atlantic trades both by monitoring Twitter for “macrosentiment,” meaning indications of the direction of broad indexes like the FTSE 100, the S&P 500, or the Dow, and by tracking social media sentiment on individual companies. The social media sentiment trading on both the macro and individual stock level is based on algorithms that include “large volumes of information” and a basket of keywords that Hawtin won’t reveal, other than to say they “have relevance when measured in terms of economic sentiment.” Hawtin and his team then monitor the economic sentiment scores created by their algorithms to look for changes. “If the score drops overnight or over a few hours, globally people’s confidence in the market is deteriorating,” he said.
Twitter and other forms of social media have risen in acceptance as a market data tool. The sheer number of analytics firms offering social media sentiment analysis is evidence of this. Still, Hawtin is one of the few to publicly embrace social media sentiment analysis as a trading strategy.
“I’m not sure to what extent many algo designers are ready to incorporate social media sentiment as a part of a core algorithmic trading strategy. I’m not sure many are at that level of trust yet,” said Danielle Tierney, Aite group analyst. “What it has been more commonly used for up until now is as a check or a slowdown,” Tierney added, explaining that social media might be used to trigger alerts in existing trading strategies.
Gaining Respect
Perhaps the largest acknowledgment of the growing significance of social media sentiment analysis came earlier this year when leading market data vendors Bloomberg and Thomson Reuters added Twitter-based analytics to their Bloomberg Professional and Thomson Reuters Eikon terminals, respectively. Market data provider Markit also enhanced its Markit Research Signals suite with 22 new social media signals, including some from analytics firm Social Market Analytics as well as others Markit created based partially on SMA signals.
“There are occasions daily, certainly every week, in which important news events in the world break on Twitter from a handle that was not previously on anyone’s list to watch,” said Brian Rooney, Bloomberg’s head of news product. “That said, in the financial sphere, the more common cases and the cases of greater significance of social media moving the markets are from a universe of newsmakers of one type or another who are putting market moving news on Twitter as their broadcast mechanism.”
Bloomberg and Thomson Reuters both view social media sentiment analytics as an extension of news analytics that both firms have been offering for years. Thomson Reuters began offering news analytics back in 2008, three years after it launched machine-readable news, according to James Cantarella, who heads Thomson Reuters’ machine-readable news sector.
“News analytics includes sentiment across all news as opposed to just social media,” Rooney said. “It’s event detection, and analysis of flows and velocity of news around given tickers or entities or topics. It’s readership analytics and the spikes in demand for news around certain issues; and its various forms of market impact analysis around which pieces of news or social media have moved the markets, when and how.”
The emergence of social media sentiment is seen by both firms as a new information medium on which to apply analytics skills they have honed for years.
“We plan to further innovate in both machine-readable news and Twitter social media sentiment and trend detection. We think these techniques will further merge and become consistent on the same analytics platform, which will drive a lot of innovation in 2014,” said Kevin Chen, global head of pricing and text analytics with Thomson Reuters.
Twists and Oddities
Yet this new information medium is not without its quirks. Aside from the challenges of deriving meaning from texts that are 140 characters in length or less, Bloomberg says its analysis shows that Twitter trends positive. “As consumers of social sentiment, it has become clear that the negative sentiment incidents are more rare and might be more telling,” Rooney said.
Meanwhile, Markit found in its research that while social media sentiment has a high degree of accuracy, it has very little correlation with other commonly used indicators. From December 2011 to November 2013, Markit’s analysis found positive social media sentiment stocks had cumulative returns of 76 percent, while negative sentiment stocks had returns of -14 percent. “The most surprising thing that we found is that there is low correlation with short term price movement, with options-implied volatility levels, with equity analyst revisions in the I/B/E/S database, and with short interest levels using market securities finance data to get a gauge on how the short interest market is viewing the stock,” said Chris Hammond, director of the Research Signals team at Markit.
Culling a Vast Universe
Another challenge unique to Twitter and social media sentiment is the unknown nature of who is creating it. Some firms capture a broad cross section of Twitter content, taking the view that an aggregate snapshot offers a better picture of market sentiment.
Others seek to narrow their view of Twitter content to focus on handles that appear to have some knowledge or reason to tweet about the market. But identifying the handles to follow can be more complicated than it seems.
Sandeep Mehta developed algorithms for social media sentiment analysis while working for BrainMatics in Mumbai from 2010 to 2012 (the firm was later acquired by Mzaya). Mehta devised a system for finding handles to follow by starting with a few Twitter handles and using a computerized algorithm to search the initial handles’ “follower” and “followed by” lists. “We would dig continuously through the follower lists using user-supplied keywords,” said Mehta, who is currently enrolled in the Master of Financial Engineering Program at U.C. Berkeley. “You could do it manually, but if you wanted to follow 9,000 links, that would be an almost impossible task,” he said.
Mehta called the method of culling handles “pyramidal navigation technology” because connections between handles branched off in a pyramidal formation like the roots of a tree.
Even Data Gets Emotional
Meanwhile, while much social media sentiment analysis looks for positive and negative sentiment, some algorithms created to measure social media sentiment have other definitions of sentiment. MarketPsychData, founded by board certified psychiatrist Dr. Richard Peterson, first launched as a hedge fund based on economically predictive sentiments before transitioning to an analytics provider.
“We not only measure positive and negative sentiment, but we measure emotions specific to each asset class or category — fear, uncertainty, optimism, anger,” said Changjie Liu, MarketPsych chief of analytics.
MarketPsych measures various keywords and phrases in connection with finance-oriented dictionaries and statistical tools to create scores specific to different emotions. The emotion-based scores predict the market in different ways, according to Liu. “For short-term, some emotions are better and for long-term, other emotions have a greater effect,” Liu said. “For short-term, for example, anger has a bigger effect because I think anger is short-lived. While I think fear, especially for the financial crisis, is something that is always in the back of our minds. Its effect is longer.”
Another indicator, called Market Forecast, monitors predictions of the future by tracking language in which people indicate what they think will happen. Market Forecast has also proved to be a strong indicator, but Liu does not think it’s because predictions on Twitter have insight into the future. “When people make such inferences, others will act on those beliefs, forcing a stock to move in a certain way,” he said. “It’s about other people acting on what they see in social media.”
