Alejandro Lopez-Lira, a finance professor at the University of Florida, claims that large language models can be useful in predicting stock prices. He used ChatGPT to analyze news headlines for whether they were positive or negative for stocks, and found that ChatGPT’s ability to predict the direction of returns the following day was much better than random, according to a recent unpublished study.

This experiment hits at the heart of promises around cutting-edge artificial intelligence: With bigger computers and better data sets — like those powering ChatGPT — these AI models can exhibit “emerging capabilities,” or abilities that weren’t originally planned when they were created.

If ChatGPT can exhibit an emerging capability to understand headlines from financial news and how they might impact stock prices, it could threaten high-paying jobs in finance.

About 35% of finance jobs are at risk of automation through AI, according to a note from Goldman Sachs on March 26.

“The fact that ChatGPT understands information meant for humans almost guarantees that if the market doesn’t react perfectly, there will be predictability of returns,” Lopez-Lira said.

But the specifics of the experiment also illustrate how far so-called “large language models” are from being able to perform many financial tasks.

For instance, the experiment didn’t include target prices or require mathematical operations from the model. In fact, ChatGPT-type technology often creates numbers, as Microsoft learned in a public unveiling earlier this year.

Sentiment analysis of headlines is also a well-known trading strategy, with existing proprietary data sets.

Lopez-Lira said the results surprised him and suggest that sophisticated investors aren’t yet using ChatGPT-style machine learning in their trading strategies.

Source: CNBC