Tweet Investing

By Jonathan Yates

John Burbank III, founder of Passport Capital, once observed that, “There is no way you can make 30 percent a year going long on the S&P 500.  If you’re going to hit that kind of return, you have to go to places where it is going to happen.”   This focus on intensive research has taken analysts from Passport Capital into all corners of the world, seeking nuggets of information on companies to be mined for profit.  For other hedge managers in pursuit of a competitive advantage, having to “…go to places where it is going to happen” now entails mining millions of real time tweets and online posts that are a mother lode of information from social media about the behavior of the masses and then determining how it will move the financial markets.

The goal in computer driven analysis of tweets is certainly nothing new: collecting market intelligence from unique sources that will result in actionable investing.   It has become more intense since a study was released last fall by Johan Bollen, a professor of informatics at Indiana University linking the collective mood of millions, as determined by their interactions on social media sites, with the way the Dow Jones Industrial Average moved in direction.  Bollen’s analysis of tweets, according to his study, “Twitter Mood Predicts the Stock Market,” claims an 87 percent chance of successfully predicting stock prices.

Information of this sort can hardly be expected to be ignored for long by the denizens of Wall Street or across The Pond in The City.  And it has not: Derwent Capital Markets, an upstart hedge fund based in London, announced it will soon launch a fund based on the methodology from Professor Bollen’s study.   For this exclusive right, Derwent is paying $40.7 million.  Professor Bollen and Huina Mao, co researcher of the study, are being hired as private consultants.  Bollen and Mao will also receive more than one third of the proceeds from the licensing agreement with Indiana University.

This is hardly the first effort to tap into the mood of the masses for investing profits.  Financiers from Baron von Rothschild to Warren Buffet have long noted the wisdom in buying when others are selling, and selling when others are buying.  But this is guidance for long term investing.  Derwent Capital Markets, headed by 28 year old Paul Hawttin is not looking to buy a railroad.

Hawttin is investing (or betting, perhaps) that information gleaned from social-networking sites, known as unstructured data, will yield short term trading gains.  The competitive advantage here for Derwent Capital Markets, explains Adam Honore, research director for Alte Group’s capital markets group, is that unstructured data is generally not utilized at present for the placing of buy and sell orders in current trading patterns (this can also include news, blogs, and regulatory filings). Quantitative market data, or structured data, such as prices of trades executed at exchanges and government statistics does not result in 30 percent gains as it is as widely followed by investors as the S&P 500.

Investing and Social Media

While Hawttin’s sizeable purchase would seem to indicate otherwise, deployment of unstructured data in investing is hardly unique in the financial marketplace.  According to research from the Alte Group, investment firms utilizing unstructured data has burgeoned from 2 percent in 2008 to a current rate of 35 percent.  “We’re talking about competitive advantage,” declares Honore. “It’s all about the machine. The smallest advantage can mean substantial returns on investment.”

Even with this growth, Honore questions the utility of unstructured data as the foundation for trades.   “You can use it for some trading decision support,” he says. “But would I run a whole strategy off of it? No.”  Eric Davidson, Vice President at Titan Trading Analytics, a behavioral finance research firm that just started  using social-media data for trading, agrees for now, noting, “This is just the tip of the iceberg.  But it’s too early to build strategies solely off of social media. I don’t think anyone has nailed this concept yet.”

Professor Bollen’s research disputes those who claim that analyzing and quantifying millions of 140-characters-or-less tweets cannot serve as the basis for a profitable trading strategy now.  In 2008, Bollen analyzed the text of daily Twitter feeds (9.6 million in total) over a nine-month period with two mood-tracking tools. The mood of daily tweets (positive or negative) was measured by the first.  The second categorized the tweets into six segments associated with different moods: calm, alert, sure, vital, kind and happy.  From this came the claims of an 87 percent success rate for predicting short term movements in the Dow Jones Industrial Average.  “If 100 million people are getting nervous, it could affect the market” observes Bollen.

Data such as this would obviously lead to selling and/or going short on a variety of financial instruments.  Trades could also be based on how it might affect investing based on structured data.  If the mood was negative from millions of Tweets, the University of Michigan’s Consumer Sentiment Index would be impacted.  Published monthly, trades could be based around the unstructured data and its affect here on the financial markets.

If Tweets revealed that many were going out of town for vacations, it could yield information into higher gasoline buys for the summer months.  Or if holiday shopping was going to be strong, retail stocks would be expected to rise.  Individual stock prices for retailers or oil companies would move based on this. How movies were received based on Tweets from those leaving the theater could give an immediate edge to investors mining this data.

Jeff Catlin, CEO of Lexalytics, a text analysis firm that also monitors social media, observes that, “Because Twitter tweets are such an early stream of information, if you can measure it well, then you have an earlier signal. You can make an investment decision.  Hawtin, unsurprisingly agrees, “There’s more than 100 million tweets a day, and if you compile them all together they give you a general sense of how people are feeling.”

Hawtin will not be the first with Derwent Capital Markets, however, to trade based on tweets.  Richard Peterson, a managing partner at MarketPsych, ran a hedge fund from late 2008 through 2010 that traded from what Tweeter mood measurements and other social-media sentiment were indicating. The fund rose 5.4 per cent in the final four months of 2008, vs. a 30 per cent loss for the S&P 500. It beat the S&P again by 7 percentage points in 2009.  In 2010, however, the S&P topped it by 21 percent.  Since then, the firm has shifted its focus and now sells sentiment data feeds from its proprietary software to money managers.

MarketPsych’s mood analyzing tools now being sold can detect changes about the feelings for a stock.   It can also detect rumors and leaks of negative information before this becomes public information via the news. There are 400 different sentiments, such as optimism or anger, and tones, such as uncertainty or confidence, in the software package.

Peterson reflects on how this unstructured data resulted in profits from the trading of an individual stock.  In 2009, public anxiety about the swine flu epidemic was detected on the message boards of AMR, parent of American Airlines.  Specifically. on April 25, 2009, MarketPsych’s software detected a rise in anxiety on AMR stock message boards after a “public health emergency” was declared. On April 30, three days after the flu pandemic alert was raised to Phase 4, the fund bought AMR at $4.81 a share. It sold the shares at $5.95 a week later for a gain of 24 per cent.  As there was plenty of public information about the swine flu epidemic, the value of any data coming after the institution of a public health emergency is suspect.  Moreover, it would seem that positions should have been taken in the stock of every passenger air carrier.

Peterson does claim that trading based on Tweets “…was a viable strategy for us.”    While Peterson claims it was a “viable strategy,” the basic economic principle of revealed preference evinces that his firm abandoned it.  No one does this because they are making too much money.

Imitation is the sincerest form of flattery on Wall Street.  If Derwent Capital Markets posts returns reflecting anything near an 87 percent success rate from Tweet induced trades, it will be replicated by other hedge funds and money management groups as expeditiously as possible.  Probably the greatest benefit for investors, however, will be more emphasis on analyzing unique social media content and then trading based on it.  With social media evolving in China and in other countries, there will be tremendous opportunities in emerging markets.

Foreign exchange trading would seem to have huge potential as a result, as would individual stocks on overseas bourses.   An 87 percent success rate is obviously a tremendous competitive advantage.  But margins like that do not last in the world of finance as imitators immediately pile in.  Taking trading of this type, as Burbank counseled, “…to where it is going to happen” is what will result in the outsized returns.

, , , , ,

Trading involves substantial risk of loss and is not suitable for all individuals. Past Performance is not indicative of future results.