Andrew Lo Study Says Twitter Can Help You Trade Fed Meetingsby
It looked at sentiment `polarity' in days before the FOMC met
Researchers found a predictable enhancement of returns
In the social media cacophony, some of the noise rises to the level of stock market signal.
That’s the finding of a working paper overseen by Massachusetts Institute of Technology’s Andrew Lo, which says a trading strategy based on views posted on Twitter prior to Federal Reserve policy meetings regularly turned a profit. A one-standard-deviation increase in tweet sentiment can be exploited to boost Fed-day equity returns by 0.62 percent, it found.
Quant investors have been obsessed for years with pulling signals out of social media, where roughly 1.8 billion active users opine in real time on everything from tech valuations to celebrity breakups. The MIT study, which examined 3.9 million tweets over seven years, joins an expanding pool of research examining topics such as the role of tweets in market volatility and the accuracy of crowd-sourced earnings estimates.
“The key problem with Twitter is separating useful data from the noise,” Pablo Azar, a Ph.D. candidate in the economics department at MIT who co-authored the report, said by phone. “We decided to focus the information from Twitter into one event that we know moves markets, and we found that there’s real economic knowledge to be gathered.”
Social media’s role in the stock market has been expanding since the networks were invented a decade ago. Dozens of companies, including Bloomberg LP, the parent company of Bloomberg News, have products to analyze sentiment expressed on Twitter about markets and economies. ALPS Advisors Inc. just announced the Sprott BUZZ Social Media Insights ETF, which tracks stocks with bullish investor perception based on Twitter and Facebook.
The MIT researchers monitored the tweet flow in the 24 hours before Federal Open Market Committee meetings and developed standards to evaluate how people felt about the outcome. In a model portfolio, they used what they found to adjust exposure to stocks on the day when policy makers met. The system “outperformed on several dimensions a comparably levered benchmark market portfolio.”
One possibility Lo and Azar considered was that Twitter becomes more useful in the days before events with a predictable market impact -- meetings of the Federal Open Market Committee were investigated for this reason. They found that the helpfulness of Twitter is “negligible” on non-Fed days.
The paper, titled “The Wisdom of Twitter Crowds: Predicting Stock Market Reactions to FOMC Meetings via Twitter Feeds,” was co-authored by Azar and Lo, finance professor at MIT’s Sloan School and chairman of AlphaSimplex Group LLC, a quant research firm. They used natural language processing techniques to assign a “polarity” score to each tweet, meant to reflect the emotion in the text. That score was then used as a proxy for sentiment.
It’s worth testing social media to see if they can improve conventional measures of sentiment, things like technical indicators and survey data that are based on a “relatively sparse” population, they wrote. “The rise of social media allows us to overcome these drawbacks and measure the sentiment of a large number of individuals in real time.”
Tweet sentiment as a predictive measure was addressed in a 2015 paper by Johns Hopkins University assistant professor Jim Kyung-Soo Liew, who found that it had power to forecast post-earnings excess returns. Also that year a group of researchers led by University of California-Santa Barbara Ph.D. student Gang Wang found that while social media messages had “minimal correlation” to stock performance in aggregate, a subset of “experts” does in fact contribute more predictive content.
One caveat of the MIT study is that it could only examine a relatively homogeneous segment of the market history, the period after which Twitter emerged, which has been marked mostly by rising stocks and an accommodative Fed.
The next step for the research will be to see if the strategy of calibrating equity exposure based on Twitter sentiment continues to outperform under more hawkish conditions. Regardless, the findings of the study should be viewed as a piece of a bigger puzzle, according to Azar.
“One way people can use our results is as a component in a larger strategy,” he said. “If tweet data can help predict these movements before they happen, then it can help portfolio managers control their exposure to market risk.”