Have you ever watched a tweet go viral and wondered if someone is quietly making money from it? That idea is at the core of “Twitter trading”: turning tweets into data that guide trades.
In one study, Catherine Xiao and Wanfeng Chen collected tweets about Tesla and Ford between 2015 and 2017 and linked them to daily stock returns. They cleaned the tweets, removed spam and non-English posts, and used a language tool to rate each tweet as negative, neutral, or positive. From this they built simple indicators such as average daily sentiment and how much this sentiment changed from one day to the next. These indicators were then combined with classic price-based variables in trading rules that try to predict whether Tesla’s return, relative to its sector ETF, will be higher or lower the next day. For Tesla, the rules that used Twitter sentiment , especially tweets mentioning products or news, like “Model S” or “Elon Musk” did better than rules using only prices. The improvement was much weaker for Ford, suggesting that Twitter is most helpful for “story stocks” whose prices are driven by expectations, hype, and news rather than slow-moving fundamentals.The authors also went beyond simple rules and trained a reinforcement-learning trader. This agent learned, by trial and error on historical data, when to buy, hold, or sell based on a mix of price information and Twitter indicators. When Twitter features were included, the agent generated higher risk-adjusted returns and adapted better across different market conditions. In plain language: having a live feed of crowd mood and attention helped the model decide when a move in Tesla’s price was just noise and when it was part of a bigger story building up on social media.
In another paper on Bitcoin, researchers used millions of tweets to forecast how volatile the price would be. They again created sentiment scores and tweet counts, but they also added information about the users themselves: follower numbers, verification badges, how often they tweet, and how central they are in the network. These features were fed into deep learning models that tried to predict the next day’s realised volatility, not just the direction of price. They found that information about the users themselves was often more useful than the exact wording of the tweets, suggesting that “who” talks about an asset can matter more than “what” they say. When influential accounts became more active, the models expected larger swings in Bitcoin prices, even if the overall sentiment was neutral.
Taken together, these two studies sketch a practical picture of how Twitter trading works in the real world. For growth stocks like Tesla, sentiment and attention can contain clues about future excess returns, especially around product announcements or controversy. For Bitcoin, Twitter looks more like a risk measure, signalling when the market is about to become turbulent. In both cases, the data are messy and the edge is small, but they show that social media is not just background noise – it is part of the information set that prices react to.
Of course, there are serious drawbacks. Twitter is full of bots, coordinated campaigns, and plain old sarcasm that sentiment tools can misread. Once a strategy becomes popular, its edge may disappear. And none of these models replace basic risk management or an understanding of fundamentals. Instead, the lesson from Twitter trading is more modest but still powerful: if markets are shaped by the beliefs and emotions of thousands of human traders, then watching what they say online can help us understand, and sometimes anticipate, what they will do with their money.