Potential of ChatGPT in predicting stock market trends based on Twitter Sentiment Analysis
This study demonstrates that ChatGPT can effectively predict short-term stock market trends for Microsoft and Google by analyzing Twitter sentiment, revealing a positive correlation between its evaluations and subsequent stock performance.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the modern financial world, the pulse of the market is often felt not just in boardrooms or trading floors, but in the digital chatter of millions of people online. For decades, experts have tried to predict whether stock prices will rise or fall by studying company reports and economic numbers. However, a newer idea has taken hold: that the collective mood of the public, expressed in real-time on social media, might actually move the needle on stock values. This concept relies on "sentiment analysis," a method of using computers to read text and determine whether the feelings behind the words are positive, negative, or neutral. The question researchers are now asking is whether the newest generation of artificial intelligence, specifically a powerful language model capable of understanding human conversation, can read this digital mood and use it to forecast the future of the stock market without needing to be taught the rules of finance first.
Two researchers, Ummara Mumtaz and Summaya Mumtaz, set out to test this idea using a tool called ChatGPT. They focused on two of the world's largest technology companies, Microsoft and Google, and asked a simple question: could this artificial intelligence look at a day's worth of tweets about these companies and guess whether their stock prices would go up or down the next day? The researchers did not train the computer to be a stock expert. Instead, they used a technique known as zero-shot learning, which allows a model to perform a task it has never seen before by relying on its general understanding of language. They gathered a massive collection of half a million tweets from early 2023 that were extracted by searching for the term "gpt." From this original dataset, they specifically sought out and filtered for tweets that mentioned or related to three distinct terms: 'ChatGPT', 'Microsoft', and 'Google.' To keep the process fair, they cleaned the data only slightly, removing web links but leaving the raw, unfiltered opinions, hashtags, and emojis exactly as people had posted them, preserving the authentic voice of the internet.
The method was straightforward. For each day in their study period, the researchers fed a sample of about 15,000 words worth of tweets into ChatGPT. They gave the computer a simple instruction: read these posts and predict the stock market trend for Microsoft and Google for the following day. The model was not given any historical stock data to study; it had to rely entirely on the tone and content of the tweets. The computer would then output a prediction, such as "upward," "downward," or "stable," often adding a brief explanation of why it thought that, such as noting that a company was being challenged by a new competitor or that a specific product launch was causing excitement. The researchers then compared these computer-generated guesses against the actual percentage change in the stock prices reported by the stock exchange the next day.
The results showed a surprising level of success. For Microsoft, the artificial intelligence correctly predicted the direction of the stock market trend on 26 out of 37 days, achieving an accuracy rate of 70 percent. For Google, the model was accurate on 23 out of 36 days, resulting in an accuracy of roughly 63.88 percent. The researchers noted that while these numbers are not perfect, they are significantly better than what one would expect from a model that is simply guessing at random. Beyond just getting the direction right, the study highlighted that the computer was able to identify specific reasons behind the trends, such as mentioning the impact of a new rival product or the potential threat of the technology itself. This suggests that the model was not just matching patterns but was actually interpreting the context and nuance of the public conversation.
The study concludes that this approach, which uses untrained artificial intelligence to read social media sentiment, holds real promise for understanding market movements. The researchers emphasize that they did not need to spend months cleaning the data or teaching the computer specific financial formulas. By simply letting the model read the raw opinions of the public, they were able to capture a signal that correlated with real-world financial outcomes. While the study was limited to a specific set of tweets and a short timeframe, the findings suggest that the ability of these advanced language models to understand human emotion and context could become a valuable tool for financial forecasting, offering a new way to listen to the market's voice.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.