web_istock-458351045.jpg

You can’t beat tweets for predicting corporate earnings

From celebrity gossip to political messaging to natural disaster updates, Twitter provides real-time news, data, and opinions. Associate Professor of Accountancy Lucile Faurel finds that the social media platform can do much more in her study.

By Jenn Woolson

With more than 325 million users worldwide, Twitter is one of the most immediate and widespread information sources. From celebrity gossip to political messaging to natural disaster updates, tweets provide real-time news, data, and opinions.

Associate Professor of Accountancy Lucile Faurel finds that the social media platform can do much more in her study called, “Can Twitter Help Predict Firm-Level Earnings and Stock Returns?" The research, published in 2018 in The Accounting Review, was co-authored with Eli Bartov of New York University’s Stern School of Business and Partha Mohanram of the University of Toronto’s Rotman School of Management. To answer the study’s title, the researchers find that, yes, Twitter is indeed a valuable tool.

While Faurel doesn’t use Twitter personally, she still finds the medium intriguing from a research point of view. “I'm not a Twitter user and I most probably never will be,” she says. “What led me to look at Twitter was the fact that it's so widely used. In our study, we focus on what investors are tweeting when they are exchanging information or opinions about stocks.”

Predicting earnings

The study employs a sample of around 900,000 tweets from 2009 through 2012. The researchers selected more than 3,600 companies and gathered tweets that included stock symbols of those specific firms.

Then, they selected the tweets written during the nine trading days preceding each quarterly earnings release date. Analysis of those tweets helps to answer three firm-level questions:

  1. Can tweets help predict upcoming quarterly earnings news?
  2. Can they help predict the stock returns around those earnings news?
  3. Are the results more pronounced for firms with better or worse information environments?

What they find is that Twitter successfully predicts the firms’ earnings and the related stock price reaction. The study’s results hold for tweets that convey original information, as well as tweets that disseminate existing data, and are stronger for tweets providing information directly related to firm fundamentals and stock trading. Importantly, the results hold even after controlling for concurrent information or opinion from traditional media sources.

Moreover, the study finds these results are more pronounced for firms in weak information environments. “In stronger information environments, as an investor, you could get information about a firm from analyst reports or forecasts, even articles in the local or national news. However, if we are talking about much smaller firms, ones that have far less visibility, we find in our results that Twitter then plays a greater role,” Faurel explains.

Twitter wisdom

Rather than looking for general economic or industry-level sentiment as several past studies have done, Faurel and her co-authors take a unique firm-level view of investors’ tweets. That's a crucial distinction, she explains.

Other studies look at companies’ use of social media to disseminate information to investors. This is the first study that looks at individual tweets and predictions of firm-level earnings and stock returns. It is also the first study that looks at individual investors tweeting among themselves or tweeting to others about their thoughts on specific firms.

This is not the company tweeting about their own company, Faurel says. “This is a very new area of looking at the demand side of financial information and what investors are doing to acquire and share information.”

What they find is that these individuals’ tweets were more accurate at predicting earnings than those of analysts. One potential element at play is “the wisdom of crowds,” which states that if you have a set of people — not necessarily experts in an area — more often than not the group’s opinion will beat one expert’s opinion. The other element Faurel cites is diversity.

Twitter users are very diverse, more so than even a group of analysts, in their background and knowledge. Related theories state that diversity provides benefits.

While there are many diverse Twitter users in their sample, they find that not all are equal. Just as with your personal social media activity, there are some “super” users. In the study, 838 individuals accounted for the top 1 percent of users. Each of them tweeted at least 159 times, and an average of 647 times, compared to an average of 10 tweets per user overall.

“You see huge variations in how many tweets each user sends,” Faurel says. “We saw that these top users accounted for a very significant portion of our sample of tweets. It's consistent with what happens in social media regardless of the platform or the discussion topic.”

Weighting words

To make their earnings and returns predictions, the researchers use textual analysis of the tweets to capture the opinion or sentiment conveyed and classify them as negative, positive, or neutral. To do this, the researchers employ four approaches. Three of the approaches use dictionaries or word lists — the Loughran-McDonald sentiment word lists, the Harvard Psychological Dictionary, and the Hu and Liu opinion lexicon — to analyze the sentiment of specific words in each tweet. Each of these tools contains thousands of words and phrases grouped by their negative or positive connotations.

In the fourth approach, they use a naive Bayes classifier, which was developed by a computer scientist to evaluate movie reviews. That algorithm analyzes the entire message and outputs whether the information is positive, neutral, or negative, and with what confidence level, Faurel explains. For example, it would say “positive with an 80 percent confidence level.” This analysis allows them to put more weight on the messages that appear to be more strongly positive or more strongly negative.

We used different methods to capture the sentiment or opinion of these tweets. We were pleased to see that our results were holding with all of the different methods.

Their findings also hold whether the tweet is original or disseminating already existing information — the two primary roles of social media.

“When you think about in our personal lives, social media serves these two purposes almost equally,” Faurel says. For example, you can send out original information, such as the news that your daughter made the spelling bee final. Alternatively, you can disseminate already existing data, such as a link to the video of the spelling bee or a retweet of the school’s announcement.

The same is true for financial tweets. One investor could share an opinion about a company while another might link to a story or press release about that same firm.

“You couldn’t say that one does more than the other. And that's exactly what our results show,” she says. “Both are informative.”

The role of regulation

With more than 500 million tweets per day, Twitter has come to the forefront as one of the most popular ways people receive and spread information. That ubiquity is part of what attracted Faurel to the research.

People are allowed to tweet anything they want. There are no rules or guidelines to follow. For better or worse, this information is — with no filter — what comes out of people.

There have been a handful of rogue tweets over the years — containing misleading information — that contributed to specific stocks tanking. This has led to discussions about whether Twitter needs more regulations around who can say what about particular topics.

However, the study’s findings show that the information conveyed by tweets is by and large informative and can be used to help predict earnings and returns around earnings.

With this study under her belt, Faurel is already diving into additional financial research involving social media.

“This was step one,” she says. She thinks their findings have opened the door to look at many other variables that Twitter could help predict in capital markets. While she’s excited about the prospect for future research, she’s not likely to tweet about it any time soon.