Can Twitter beat the stock market?

By Phil Pearlman
June 14, 2011

By Phil Pearlman

The views expressed are his own.

Can you really use what shows up on Twitter to beat the stock market? The case for being able to do so rests largely on this October 2010 study, published by Bollen, Mao and Zeng, which measured the correlation between “mood states” derived from largescale Twitter feeds and the Dow Jones Industrial Average over time.

Results of the study suggested to the researchers a correlation and even predictive relationship between Twitter sentiment and directional biases in the large cap stock index and perhaps the broader market as well. The rationale behind the study implies–correctly, I think–that Twitter has scaled enough and tapped into a real-time stream of consciousness massive enough that it might reflect collective near-term sentiment, assuming it can be measured correctly.

In May of this year, the London-based hedge fund Derwent Capital raised $40 million and launched a hedge fund which incorporates a sentiment analysis component, based on the measurements in the study and licensed from Bollen and his colleagues.

Paul Hawtin, manager of the new fund, is taking the right approach by integrating Bollen’s Twitter sentiment analysis with existing trading algorithms. This will allow him to utilize market signals he has experience with and confidence in, while at the same time beginning to incorporate and test the Twitter sentiment-analysis data with real capital at risk. Nevertheless, we will not know how this integration is going until he has established a track record over 1-3 years.

Presumably, there are other funds as well with assets under management much larger than $40 million and with significantly greater resources at their disposal that are not publicizing their Twitter-based sentiment analysis but who are more quietly conducting their own analyses on large scale Twitter feeds. The public will not hear about the results of such research or even that it is occurring, especially if the findings prove of value to these large quantitatively based funds. In addition, these larger funds should obtain meaningful results much more quickly than academia, which by tradition tends to plod along as other researchers converge to retest, refine methodologies and then publish. (At StockTwits, we are currently formulating sentiment measurement models based on our own rich market-focused data which we may choose to iterate at some point in the future.)

Correlating Twitter sentiment and market activity is all well and good. To sophisticated market participants, however, the only variable that really matters is “alpha”–that is, can the Twitter sentiment-analysis algorithm increase excess market return?

Late last year, I predicted that hedge funds would waste no time pouring over Bollen et al’s study.  The question I raised, though, still applies to Derwent’s new fund as well as others who are likely applying sentiment analysis of the Twitter API to a market timing algorithm.  I wrote:

There is a big difference between discovering a method to measure close to real time collective sentiment and its relationship to market behavior and exploiting it for market gains.  For one thing, the methodology described in the study above, while provocative, will require scrutiny from the scientific community, reproduction and refinement.

But, and perhaps more importantly, the individuals who will attempt to trade off of such data will be subject to not only market structural limitations but their own emotions as well.

Aside from the development of the scientific methodology, which I believe is in its early days, the question then becomes how managers control for their own emotions, which inevitably exist within the same sentiment reservoir that is being measured.

It’s an age-old question: can you completely stand apart from the crowd when you are part of it? Even if Twitter turns out to be the best market sentiment tool ever, it doesn’t necessarily provide a new answer.


We welcome comments that advance the story through relevant opinion, anecdotes, links and data. If you see a comment that you believe is irrelevant or inappropriate, you can flag it to our editors by using the report abuse links. Views expressed in the comments do not represent those of Reuters. For more information on our comment policy, see

Let me first state, I really admire the initiative of the underlying study. However, I just do not see how this fund will differentiate itself from the countless other optimized quantitative strategies that ultimately fail.

The basis of this study was to use a lag of 3 days on sentiment readings as one parameter in a neural network which then learns over the historical period (in this case Feb 28 to Nov 28, 2008) to predict over the next 15 trading days.

“February 28, 2008 to November 28, 2008 is chosen as the longest possible training period while Dec 1 to Dec 19, 2008 was chosen as the test period because it was characterized by stabilization of DJIA values after considerable volatility in previous months and the absence of any unusual or significant socio-cultural events.”
They handpicked a test period where nothing happened to generate their findings and under these ideal conditions, the prediction exhibited minimal statistical significance.

“The cases in which the t -3 mood time series fails to track changes in the DJIA are nearly equally informative as where it doesn’t. In particular we point to a significant deviation between the two graphs on October 13th where the DJIA surges by more than 3 standard deviations trough-to-peak. The Calm curve however remains relatively flat at that time after which it starts to again track changes in the DJIA again. This discrepancy may be the result of the the Federal Reserve’s announcement on October 13th of a major bank bailout initiative which unexpectedly increase DJIA values that day. The deviation between Calm values and the DJIA on that day illustrates that unexpected news is not anticipated by the public mood yet remains a significant factor in modeling the stock market.”
This particular variance would represent a noticeable drawdown to the fund, exacerbated by the leverage used in implementing the strategy.

Like any other optimized quantitative strategy, recalibration is a necessary requirement. However frequent this calibration is, it does not ensure survivability of the fund. A recent Bloomberg op-ed touches on this.
“The models aren’t capable of any great level of precision. In finance, when models are calibrated, they always have to be recalibrated a week later. But those parameters are supposed to remain fixed for evermore. If they have to be changed, then the model either was wrong before, is wrong now, or more likely both.”  /bankers-can-t-avoid-risk-by-hiding-it. html

Again, I don’t wish investors to lose capital in this investment and perhaps the final strategy is more loosely interpreted from the study with extremely favorable risk management / leverage considerations. But, how can this work be directly translated into an investment with survivability if it relies on predicting a lagged outcome with 80% accuracy using optimized historical data? Regardless of its behavioral representations (which I admire), the data appear to be victims of a suspect methodology.

Posted by choral | Report as abusive

[…] Can Twitter beat the stock market? — Reuters […]

Posted by Train Reading: The Internet and Ugly Husbands | co-fx | Report as abusive

[…] Can Twitter beat the stock market? — Reuters […]

Posted by Invest in Stocks | Train Reading: The Internet and Ugly Husbands | Report as abusive

[…] Can Twitter beat the stock market? — Reuters […]

Posted by Train Reading: The Internet and Ugly Husbands | Hot Otc Stock Alerts | Report as abusive

Yes, but don’t forget one of the basic principles of Psychohistory (wee Hari Seldon): Predictions of behavior are valid only if the subjects (in this case, those who buy and trade in the markets) are unaware of the predictions and the analysis.

Posted by jane33w | Report as abusive

I suspect that this methodology about as much validity as moon cycles/gravitational pull and Elliot Wave theory – ok, maybe not as much as the ridiculously complex to the point of unusable “Elliot Wave” theory.

Open up a ThinkorSwim trading account and you can have free access to more indicators and forecasting tools than you’ll ever need. Unless you have a robust trading methodology that has an edge, they won’t work. In fact, you’ll probably blow your money.

Posted by MyNameIsObama | Report as abusive