Amidst a time of great uncertainty, marked by pandemics, economic downturns, and unprecedented events, AI models may offer a valuable tool for forecasting stock prices. Recently, a finance professor at the University of Florida employed ChatGPT to analyze news headlines to determine whether they had a positive or negative impact on a stock’s value. Although the approach is not yet perfect, the professor’s research suggests that ChatGPT may soon threaten high-paying financial jobs.
As AI technology becomes more advanced and sophisticated through the use of larger computers and better data sets, they may develop “emergent abilities,” skills that were not initially intended when they were created.
Our investigation suggests that ChatGPT sentiment scores exhibit a statistically significant predictive capacity on daily stock market returns, according to a research report that has not yet been peer-reviewed. We discover a significant link between the ChatGPT evaluation and the ensuing daily returns of the stocks in our sample using news headline data and the sentiment scores provided.
Over 50,000 news items regarding publicly traded stocks on the New York Stock Exchange, Nasdaq, and a small-cap exchange were used by Professor Alejandro Lopez-Lira and his colleague Yuehua Tang. The researchers began with the starting date of October 2022 to ensure that the headlines were published after ChatGPT’s learning cutoff date.
The researchers then input the headlines into GPT 3.5 and instructed the AI platform to pretend to be a financial expert and react to the news by answering “YES” if it was good news, “NO” if it was bad news, or “UNKNOWN” if the AI was uncertain. Then, the AI was asked to elaborate on the next line with a short, concise sentence.
When compared to the stock’s returns the next day, the researchers found that ChatGPT performed well in nearly all cases after receiving a news headline. The model outperformed commercial datasets with human sentiment scores, and the researchers found less than a 1% chance that the model would perform equally well by randomly selecting the next day’s move.
If the market doesn’t react exactly, Lopez-Lira claimed that there will be return predictability since “ChatGPT is understanding information meant for humans.”
The researchers came to the conclusion that while basic models like the GPT-1, GPT-2, and BERT could not successfully predict returns, advanced language models would. In the abstract of the research, it is stated that “Our results suggest that incorporating advanced language models into the investment decision-making process can yield more accurate predictions and enhance the performance of quantitative trading strategies.”