In an experiment by Alejandro Lopez-Lira, a professor of finance at the University of Florida, a chatbot was fed more than 50,000 news headlines about companies and asked to determine how they might affect stocks.
ChatGPT marked the news as good, bad or irrelevant, after which the researchers calculated a score and analyzed the companies’ stock market performance the next day. As a result, a statistically significant positive correlation was found between scores and stocks – companies with higher scores tended to show better earnings than those with lower scores.
A yet-to-be-peer-reviewed paper by the study’s authors noted that ChatGPT outperformed other “traditional sentiment analysis methods” that also use data from headlines and social media to predict stock movements, although the researchers acknowledge that they did not test each of these methods in an experiment.
In short, our study demonstrates the value of ChatGPT in predicting stock market returns. That is, the inclusion of advanced language models in the process of making investment decisions can provide more accurate forecasts and increase the effectiveness of quantitative trading strategies,” the researchers write.
However, the authors note that investors should not rely solely on ChatGPT — in the future, similar technologies should be used as a tool to “more quickly incorporate news into stock price forecasting.” Lopez-Lira also noted that technology could replace some investment analysts.
Since the launch of ChatGPT in November last year, users have, as soon as they have not tested its capabilities, asked which stocks to invest in or asked for help starting a business. At the same time, its shortcomings were also identified – for example, the fact that it was not connected to the Internet limited its access to current information.
“ChatGPT does not have access to the latest data after the training ends in September 2021. This limitation means that the AI model may not be aware of the latest market trends, news or events that could materially affect stock prices and investment decisions,” said López-Lira.
Note that the study used articles dated last October, while in March 2023 OpenAI added new plugins that allow the chatbot to visit and interact with third-party web pages.
Other limitations include the processing of large texts and numbers. ChatGPT, for example, is unable to handle large amounts of numerical data, such as company accounting data. Addressing these issues, Lopez-Lira said, could “significantly increase the predictive capabilities” of the chatbot.
“The conclusions obtained from this study can help to develop more accurate, efficient and responsible models that will increase the efficiency of financial decision-making processes,” the researchers conclude.
López-Lira said hedge funds have already approached him to learn more about the research. He also said he wouldn’t be surprised if ChatGPT’s ability to predict stock movements declines in the coming months as institutions begin to integrate the technology.
Source: CNBC, Insider