Predicting the stock market with big-data analysis of tweets
Do social media posts about the stock market contain valuable information for investors, or can they be ignored as random noise generated by self-proclaimed financial gurus? A study by Prof. Ting Li and Dr Jan van Dalen of Rotterdam School of Management, Erasmus University (RSM) shows that big data analysis of messages on Twitter can be used to predict developments in the stock market in the short and the long term. This discovery may eventually help investors to make better decisions.
Stock prices always respond quickly to new information about companies, political events or the economy in general, says Professor Ting Li. And for this to happen, the information doesn’t even have to be true. In 2013, the Associated Press Twitter account (@AP) was hacked and the hacked account tweeted that Obama was injured in an explosion. Within seconds, the stock market lost US$136 billion.
This was from just one tweet, so it’s easy to pin down the effect, Ting continues. Together, Twitter users generate a lot of information about stocks, some of it based on valid information and some of it not. This study set out to determine the usefulness of tweets about the stock market for investors. Is there any predictive value hidden in them, and can they be used to improve trading strategies? If so, what’s the best way to use it to make better decisions?
To find out, the researchers analysed the day-to-day movements of over a million Twitter messages mentioning stocks listed in the S&P 100, a US index. With so many Tweets to study, the first problem is to define the words, hashtags and emoticons, Ting says. What do Tweeters actually mean to say about a particular stock?
The researchers developed an algorithm that could extract the ‘buy’, ‘hold’ and ‘sell’ signals embedded in those tweets. They then compared those signals to the actual price fluctuations of the stocks over the following days. The researchers found they could even analyse sentiment about Apple stocks in fifteen minute intervals, simply because there were so many people Tweeting about the company’s stock.
They discovered interesting connections between tweets and stock prices.
First, stocks that are tweeted about using bullish sentiment such as ‘buy!’ experience – on average – higher abnormal returns that day. In other words, they performed better than expected over the period. This could mean more profit or loss then expected, but that in itself is interesting knowledge for investors wanting to make risk assessments.
Interestingly, this relationship between bullish language and increased stock performance was even stronger for influential Twitter users who are frequently retweeted and often mentioned. On average, predictions from power users like these led to higher abnormal returns.
The researchers also found that the number of tweets about a particular stock can predict the trading volume, volatility and follow-up return on a stock within the next 15 minutes and over the next day. Again, this is useful information for investors.
The results often showed that the more that Twitter users disagreed about a particular stock, the higher the trading volumes.
To test if these findings could lead to a profitable trading strategy, the research team ran a simulation. When they used the extra information provided by their model, they discovered it can be used to exploit market inefficiencies profitably, even when transaction costs are taken into account.
This study shows that microblogs such as Twitter can be analysed and read as a proxy for market sentiment, says Ting. It also demonstrates that such an analysis of big data can potentially help individual investors to improve their investment decisions, she concludes.
Check the video here.
Read the paper here: Li, T., van Dalen, J. & van Rees, P.J., More than just noise? Examining the information content of stock microblogs on financial markets, J Inf Technol (2017)