Twitter Data and the Financial Markets: Sentiment Analysis, Volatility & Other Indicators

This series looks at the many ways that financial market participants are using Twitter data to inform their decisions.

Financial analysts, traders and market professionals globally are increasingly using Twitter to stay abreast of the market and make critical decisions. This is the second in a series of blog posts in which we aim to cover some of the ways that Twitter data is being used by a variety of financial market participants. In the first post we discussed fundamental and consumer analysis. Now let’s look at how market participants are using sentiment analysis and other indicators from Twitter data to inform their decisions.

Sentiment analytics in the financial markets

Sentiment analysis can use natural language processing, artificial intelligence, text analysis and computational linguistics to identify the attitude of a writer with respect to a topic. It’s an important cornerstone of behavioral finance, where theorists believe that markets are irrational and that asset prices are driven by human emotion (e.g., fear, greed, hope and overconfidence, among others). With the growth in global conversation on social media - Twitter in particular - where a vast amount of real-time market conversation occurs on a daily basis, academics and practitioners have been studying and measuring the global conversation to understand if it can meaningfully impact markets. Most concur that Twitter sentiment is correlated to asset price moves, but the debate has been about the predictive nature of Tweets on price. Well, the results are in and the early movers in this space are seeing success.

A number of firms are active in this area, including Bloomberg, that has integrated company-based sentiment, as well as Tweet velocity (an indication of volatility), into their social analytics solution on the Terminal. For this post, we’ll dig into three firms who focus on analyzing sentiment at the ticker level - PsychSignal, iSentium and Social Market Analytics.

Next generation squawk box

PsychSignal describes its business as the “next generation squawk box.” Squawk boxes, introduced to Wall Street in the late 70’s, are intercom-like systems that connect the various trading floors and management offices, allowing traders to listen in on the action and mood on the trading floor. By listening to Twitter’s global conversation, PsychSignal’s hedge fund customers stay abreast of market mood with new indicators that enable them to execute investment strategies and to generate additional alpha. PsychSignal processes millions of cashtagged ($ticker) Tweets for more than 10,000 stocks and ETFs, applying sophisticated filtering and natural language processing to quantify the meaning of each word and each Tweet. This entails understanding nuances of Twitter, trading slang, and market abbreviations. Sentiment score is measured as both bullishness and bearishness (e.g. 10% bulls, 90% bears); further each is scored on “intensity” on a scale of 0 (no intensity) to 4.0 (extreme; i.e., shouting on the squawk box) to reflect the real nature of markets. In addition to sentiment, PsychSignal recently launched its HIVE-MIND product designed to be an early warning system of risk and high volatility by uncovering nonlinear patterns in trader mood. The firm has been working with University College London which published a recent white paper titled “A nonlinear impact: evidences of causal effects of social media on market prices” that concludes that social media conversations can reliably anticipate market movement.

Wisdom of the crowd

iSentium is a believer in the “wisdom of the crowd,” not just the opinion of experts and traders. They analyze millions of ticker-based Tweets and apply light filtering to keep a very broad cross section of some 350,000 users. iSentium provides its hedge fund clients a range of products, including a Daily Directional Indicator (DDI), which is a straight-forward Buy or Sell indicator sent right before the market opens and available for both stocks and ETFs. Sentiment coverage is on all US equities and ETFs, with the DDI product available for 1500 of the most liquid. iSentium reports an open-to-close strategy based on the $SPY DDI that has significantly outperformed the S&P and it’s a strategy they’ve been investing in since October 2014. Off the back of these results, the firm has recently partnered with JP Morgan Chase to launch an investable index <JPUSISENT Index>, available for institutional clients of JP Morgan.

iSentium also markets a DDI iSentium Fear Gauge product. Similar to the VIX, it’s bound between 0 and 100 and is derived using sentiment from a cross section of the broader market. Typically, when its value crosses a threshold of 90, it is predictive of extreme fear in the market.

iSentium’s work is backed by large amounts of in-house industry and academic research in linguistics and computer science, with a patent granted for sentiment calculus for a method and system using social media for event-driven trading.

It’s who’s Tweeting

Social Market Analytics (“SMA”) believes it’s not just what is being said that matters, it’s who is saying it. A part of their proprietary approach includes filtering Tweets to identify accounts of professional investors. SMA is looking for future-oriented Tweets, so that “bought $AAPL last week” is not relevant and discarded from analytics, while “…raising price target”, “…flying off the shelves” and “buying calls” are all relevant statements that are future-oriented in nature. SMA publishes a total of 7 quantitative metrics around sentiment. The “S-Score” is a statistical Z-score that quantifies when Twitter conversations are significantly more positive, or more negative, than normal. When an S-Score reaches positive (or negative) 2.0 it indicates the current conversation is 2 standard deviations (95%) more positive (or negative). Several of the other metrics include: S-Mean, a smoothed weighted average; S-Volatility, a percent measurement of the variability; S-Delta, a change in S-Score from one period to the next; and S-Dispersion, a measurement of Tweet source concentration factor (i.e., is it one person Tweeting like crazy, or a broad based signal from many?). Time periods for SMA signals range from intraday to longer term periods of 90 days. SMA provides scores on 3200 securities including US equities, ETFs, FX pairs, as well as commodities where signals also appear strong. SMA’s work is backed by extensive research and they have collaborated with the University of Illinois on a number of papers, including a recent study that looked at enhanced beta strategies of three ETFs (SPY (SPDR S&P 500), XLV (Health Care Select Sector SPDR Fund) and XLY (Consumer Discretionary Select Sector SPDR Fund)) and finds that Twitter sentiment data really can enhance ETF portfolio returns.

Looking ahead

Hedge funds and other market participants are actively leveraging the predictive nature of sentiment derived from Tweets to invest and trade. Leaders in the space are taking a diligent, scientific approach, backed by academic study and implemented successfully in the market. Participants are also marrying the knowledge of $cashtag based conversation with a broader based consumer sentiment of products and brands as an indicator of asset price moves over short and longer time periods. We expect to see increased work on sentiment, volatility and other indicators from Twitter data across asset classes, with FX, commodities and non-US equities increasingly a focus, as academic research continues to validate such methods.

For more information about how you can use Twitter data to better understand the markets, check out our website or email us at