Shulou(Shulou.com)11/24 Report--

Alejandro Lopez-Lira, a professor of finance at the University of Florida, said big language models could be useful in predicting stock prices.

In a recent, unpeer-reviewed paper, he wrote that after using ChatGPT to analyze whether news headlines were good or bad for a stock, he found that ChatGPT was far better at predicting the direction of the next day's returns than at random.

The experiment goes to the heart of advanced AI: With more powerful computing power and better data sets, these AI models may reveal entirely new capabilities that were not originally anticipated when the AI models were built.

If ChatGPT has the ability to understand financial news and analyze how it might affect stock prices, it could put some of the high-paying jobs in finance at risk. Goldman Sachs estimated in a March 26 report that about 35 percent of financial industry jobs are at risk of being replaced by AI automation.

"ChatGPT is able to understand information that is meaningful to humans," Lopez-Lira said. This means that ChatGPT can predict returns if the market does not respond perfectly to information. "

But the details of the experiment also suggest that large language models are a long way from being adequate for work in the financial industry. For example, the experiment did not include a target price and left the model to do any calculations. As Microsoft demonstrated publicly earlier this year, technologies like ChatGPT often fabricate numbers. Sentiment analysis of news headlines has long been seen as a viable trading strategy, given the existence of proprietary data sets.

Lopez-Lira said he was surprised by the results. These results suggest that experienced investors have not yet used machine learning techniques like ChatGPT in their trading strategies. "It's definitely going to have an impact on financial analyst employment," he said. The question is, do I want to pay analysts or use AI models? "

For the experiment, Lopez-Lira and its partners used more than 50,000 news headlines from a data provider about companies listed on the New York Stock Exchange, Nasdaq and a small-cap stock exchange. The news started in October 2022, after the cut-off date for ChatGPT training data. This means that ChatGPT did not see or use these news in training.

They then added the news headline to ChatGPT 3.5 with the following prompt: "Forget all previous instructions. Suppose you are a financial expert with experience recommending stocks. In the first line, answer 'yes' if good news,'no' if bad news, and 'unknown' if uncertain. Then explain it in a short sentence in the next line. They then looked at the returns of the respective stocks on the following trading day.

Lopez-Lira found that the model performed better in almost all cases after receiving news headline information. Specifically, the probability that the model randomly selects the next day's trend after receiving news headline information is less than 1%.

ChatGPT also beat commercial datasets in sentiment ratings. In one case in the paper, a company settled a lawsuit and paid a fine, a piece of news that would be considered negative on traditional data analysis, but ChatGPT correctly deduced that it was actually good news.

Lopez-Lira said he has been contacted by hedge funds wanting more information about the study. He also said he wouldn't be surprised if ChatGPT's ability to predict stock market movements declines in the coming months as investment houses begin to integrate the technology. That's because the experiment focused on the next trading day, but most investors believe that the market may have priced in the news within seconds of its announcement.

"As more people use these tools, the market will become more efficient, so one can assume that the predictability of returns will decline," he said. My guess is that if this work continues, the predictability of returns after five years will be zero. "

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.