Large language models (LLMs) like ChatGPT and BERT can predict far more than just stock prices. A recent research paper by University of Florida professors Alejandro Lopez-Lira and Yuehua Tang revealed that ChatGPT, an OpenAI-developed model, can make reliable predictions with regards to stock market returns based on sentiment analysis.
Sentiment analysis involves analyzing text, such as news headlines, for the sentiment expressed about a certain subject or company. If done accurately, it can make quantitative algorithms used by traders more accurate. Through their analysis, the two professors discovered ChatGPT generated more reliable sentiment scores from news headlines and better market predictions than the other LLM models – GPT-1, GPT-2, and BERT.
However, ChatGPT should not be looked at as a standalone solution for sentiment analysis. Alejandro Lopez-Lira, assistant professor of finance at University of Florida and one of the paper’s co-authors, believes ChatGPT has potential for improvement, especially with regards to being up-to-date on current events. He suggested adding more context when inputting prompts to the model as well as fine-tuning the model to make predictions more accurate.
In a separate paper, titled, “Language Models Trained on Media Diets Can Predict Public Opinion,” MIT researchers Eric Chu, Jacob Andreas, and Deb Roy, along with Harvard researcher Stephen Ansolabehere, explored how large language models trained on particular media could predict public opinion. Through experiments with BERT, they concluded that media diet models can generate accurate predictions of how a group of media consumers will answer poll questions.
Unfortunately, the use of media diet models could also be abused by media manipulators to assess the effectiveness of their disinformation campaigns. The authors of this research urge for more in-depth analysis on how media affects people and shapes public opinion through the effects of selective exposure, echo chambers, and filter bubble.
Alejandro Lopez-Lira is an assistant professor of finance at the University of Florida who had co-authored the paper, “Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models.” Eric Chu is a Google research scientist who was a MIT doctoral candidate at the time of his research project “Language Models Trained on Media Diets Can Predict Public Opinion.” Both have provided groundbreaking research into the efficacy of large language models when it comes to prediction and forecasting.
