Analysts fear that their jobs will be lost.

Although the ChatGPT artificial intelligence (AI) chat robot is not allowed to talk about stock investment according to the rules, some people have already discovered the secret technique to break through the restriction. A research by a professor at the University of Florida in the United States found that ChatGPT analyzes whether a certain news headline is good for the stock price or not. Positively, the accuracy is not only better than random prediction, but even better than the traditional data analysis method, namely the human brain. To put it simply, the study found that after examining news headlines, ChatGPT can more accurately predict the rise and fall of related stock prices in the next day. The research results raise concerns about whether stock analysts will lose their jobs. In fact, some hedge funds have expressed their interest in research details to relevant scholars. Goldman Sachs had earlier predicted that as many as 35% of financial practitioners’ jobs may be replaced by artificial intelligence.

ChatGPT can judge the stock price trend of the next day

After the launch of ChatGPT, it has become popular all over the world, and people are curious about the extent of AI’s capabilities, including whether it can create a winning investment method. A recent study by Alejandro Lopez-Lira, a professor of finance at the University of Florida in the United States, found that ChatGPT can predict the stock price trend of the next day by studying news headlines, and its accuracy is better than random prediction. According to “CNBC” reports, although the relevant research papers have not been subject to detailed academic review, Alejandro Lopez-Lira asked ChatGPT to analyze more than 50,000 listings on the New York Stock Exchange, Nasdaq and a small stock exchange during the research. The title of the stock news, and then asked ChatGPT to judge whether the relevant news is beneficial to the stock price. It was found that ChatGPT can judge the stock price trend of the next day more accurately than random prediction. Lopez-Lira pointed out in the paper that ChatGPT’s large-scale language model may be used to predict stock prices on the grounds that it can calmly and accurately analyze whether a news headline is good or bad for a stock.

Lopez-Lira and his partner Yuehua Tang have asked ChatGPT the above 50,000 financial news headlines since October last year. It is worth noting that the news used in these experiments began in October 2022, which was trained on ChatGPT. after the cut-off date for the data. This means that ChatGPT has not “seen” or used relevant news during the training period, which means that there should be no problem of cheating.

ChatGPT’s limitation of not talking about stock prices

But how do researchers break through ChatGPT’s limitation of not talking about stock prices? Scholars revealed that their research method is to add news headlines to ChatGPT 3.5, and attach a command (prompt) to the effect of “assuming you are a financial expert with stock recommendation experience. In the first line (title) In the text, if you think it is good news, please answer yes; if it is bad news, please answer no; if you are not sure, please answer unclear. Then please explain in a short sentence in the next line.” After that , they observed the rise and fall of the corresponding stock in the next trading day, and finally found that ChatGPT’s prediction of the stock’s trend in the next day is much better than random prediction.

In addition, ChatGPT has also defeated traditional business data analysis in the market. As an example in Alejandro Lopez-Lira’s paper, a company settled a lawsuit and paid a fine. This news headline has negative sentiment, which is considered negative in traditional data analysis, but ChatGPT can still correctly infer that the news is actually It was good news for the stock price, and the stock did rise the next day. However, there are still some things that AI cannot do for the time being, such as judging the target price or performing related calculations. Alejandro Lopez-Lira said he was surprised by the results of the study. In addition, some hedge funds have contacted him recently, hoping to learn more about the above research, but he believes that various institutions may integrate technology in the future, so the ability of ChatGPT to predict stock prices will decline accordingly, because if “the more As more and more people use such tools to improve market efficiency, the rate of return will become unpredictable.” He speculated that “the predictability of rewards will return to zero within 5 years of actual use.” What this scholar said simply means that any originally relatively effective method of predicting stock prices, once made public and adopted by too many people, will cause interference and eventually become inaccurate. The report continued that if ChatGPT can further fully understand the content of financial news and analyze its impact on stock prices, some financial industry workers may indeed lose their jobs. Goldman Sachs previously estimated that about 35% of financial industry jobs are at risk of being automated by artificial intelligence.