Every single day I get messages on Twitter asking me if I’m ready to get rich just by tweeting. Believe me, I am, but I’ve yet to see an offer from a spambot that I could put any stock in. The idea that you can make money off Twitter is simply hogwash.Or maybe it isn’t. Researchers at the University of California, Riverside have developed a model for using Twitter to predict financial markets. Unlike tweeting about free iPads, it’s potentially very lucrative.The model, created by UCR associate professor Vagelis Hristidis, a grad student, and a trio of Spanish Yahoo! Researchers, was used to develop a Twitter-based trading strategy. That strategy (which isn’t called "TweetSkrill," but should be) outperformed other "baseline" strategies by between 1.4 and 11 percent. "Baseline" strategies could mean a number of things, but more telling is the fact that the Twitter strategy outpeformed the Dow Jones Industrial Average during a four-month simulation.“These findings have the potential to have a big impact on market investors,” Hristidis, who specializes in data mining research, said. “With so much data available from social media, many investors are looking to sort it out and profit from it.”The goal of the study, which was presented at the Fifth ACM International Conference on Web Search & Data Mining in Seattle, was to try to find a correlation between Twitter activity, stock prices, and trading volume.It’s hardly a crazy idea in concept; the stock market is fueled largely by the perceptions of investors and how those investors react to the news cycle. Theoretically, if you were to know how everyone felt about everything in the business world, you’d have a very reasonable shot at picking up trends as they develop. That’s been tried by combing through financial news for trends, but it’s slow. Twitter, however, is already the platform by which people share exactly those sentiments, aggregating what’s relevant and how they feel in response in real-time.The issue, of course, is that there’s simply an overwhelming mountain of data to sift through to pull out anything relevant, a task not made easy considering the meaning of people’s tweets are much harder to tabulate than simple numbers. (That’s not to mention the insane amount of computing power required to sift through trading data and run the market in the first place.)Hristidis’ group found a way around that. Past studies using Twitter as an investment tool have looked at whether tweets were positive or negative. In this case, the team focused on the volume of tweets pertaining to a specific company, and, more importantly, how those tweets are interconnected and spread throughout the Twittersphere.According to a paper available online (PDF), the team tested the model by first compiling the daily closing price and trading volume for 150 random companies from the S&P 500 for the entire first half of 2010. They then created Twitter filters that sifted out tweets pertaining to those companies during that time period.Comparing the two sets of data, the team expected that the trading volume for a specific company would be correlated with the number of tweets about that company; if a company is trending on Twitter, it’s reasonable to think that more people would be trading in that company’s stock.Instead, they found that trading volume was slightly more correlated with what the team calls "connected components," or in other words, the number of various topics threads being discussed about a single company. So, if Google has people talking about privacy, Android, and ad models one day, trading volume was likely to be higher than a day where Twitter users only focused on one of those topics. In addition to volume, they found that stock price was marginally correlated with the number of connected components.In terms of hard numbers, over the four-month period of the study (March 1, 2010 to June 30, 2010) the DJIA fell 4.2 percent. The team used four trading models to test against their Twitter model. A pair of autoregression models, in which one buys stocks daily based on their performance over the days prior, lost 8.9 and 13.1 percent. A random model, in which a random set of stocks was bought and sold daily, lost 5.5 percent.A fixed model, in which a set of stocks from companies with favorable market cap and balance sheets were bought and held over the course of the study (which is most similar to the strategy employed by your average non-institutional investor) outperformed the market, losing only 3.8 percent. The Twitter model, meanwhile, only lost 2.4 percent.So that means it’s a winner, right? Well, not quite. For example, the model might break down if the market is gaining, a condition it wasn’t tested under. It’s something that requires a lot of testing and refinement before I’d expect to see any major amounts of money being put into Twitter-managed funds. Still, it’s an interesting counterpoint to the increasingly-refined high-speed trading strategies and computer programs of dubious provenance being utilized by the financial industry’s big players.But it’s a truly fascinating concept. It’s easy for people to forget that the stock market is driven so much by perception, and Twitter’s stream of real-time opinion seems like the absolute best source available for trying to predict what people are thinking. The question is, how reliably can one aggregate that data from relevant sources? All I know is that, if the Twitter model catches on, I’ll be the first with a startup named "Bieber" to try to game the discussion volume-weighted system.Follow Derek Mead on Twitter.
Advertisement
Advertisement
Advertisement