Would You like a feature Interview?
All Interviews are 100% FREE of Charge
Alejandro Lopez-Lira, a professor of finance at the University of Florida, says large language models can be useful when: Stock price forecast.
He uses ChatGPT to analyze news headlines to determine whether they are good or bad for stocks, and finds that ChatGPT’s ability to predict the direction of next-day returns is far superior to random. discovered. Recent unpeer-reviewed papers.
This experiment gets to the heart of the promise of state-of-the-art artificial intelligence. With bigger computers and better datasets to power ChatGPT, these AI models could look like this:Emergent ability”, or features that were not originally planned at the time of construction.
If ChatGPT can demonstrate a new ability to understand financial news headlines and their impact on stock prices, high-paying jobs in the financial industry could be in jeopardy. About 35% of financial operations are at risk of being automated by his AI, Goldman Sachs estimates in his March 26 memo.
“The fact that ChatGPT understands information for humans almost guarantees predictability of returns even if the market doesn’t fully react,” said Lopez-Lira.
Traders work on the floor of the New York Stock Exchange.
Jason DeCrow
But experimental details also show whether so-called “large scale language models” can perform many financial tasks.
For example, the experiment did not include a target price or let the model do any calculations. In fact, ChatGPT-style technology often makes numbers, as Microsoft learned in a public demo earlier this year. Headline sentiment analysis is also a well-understood trading strategy and already has its own data set.
Lopez-Lira said she was surprised by the results, adding that it suggests sophisticated investors are not yet using ChatGPT-style machine learning in their trading strategies.
“On the regulatory side, headlines become more important when you have computers that just read them, and you know if everyone has access to machines like GPT,” Lopez-Lira said. Second, it will certainly have some impact on employment in the world of financial analysts. The question is, do you want to pay analysts?
How the experiment works
In the experiment, López Lira and his partner Yue Hua Tang examined over 50,000 headlines from data vendors on public stocks on the New York Stock Exchange, Nasdaq, and small-cap exchanges. They started in October 2022 — after ChatGPT’s data cutoff date, meaning the engine wasn’t seeing or using these headlines in training.
Then I sent the headline to ChatGPT 3.5 with the following prompt:
“Forget all previous instructions. Assume you are a financial professional. You are a financial professional with experience in stock recommendations. Answer ‘NO’ if you are not sure, or ‘UNKNOWN’ if you are unsure about the first line. Then I’ll elaborate on one short, concise sentence on the next line. ”
We then looked at the stock’s return on the next trading day.
Ultimately, Lopez-Lira found that the model performed better in almost all cases when informed by news headlines. Specifically, she found that there was less than a 1% chance that a model would randomly choose a move for the next day compared to being informed by a news headline.
ChatGPT also beat commercial datasets on human sentiment scores. According to the researcher, his one instance in the paper had a headline about the company settling lawsuits and paying fines, and there was some negative sentiment, but ChatGPT’s response suggests it’s actually good news. I have correctly inferred that
Lopez-Lira told CNBC that hedge funds have reached out to learn more about his research. He also said he wouldn’t be surprised if ChatGPT’s ability to predict share price movements declines in the coming months as institutions begin to integrate the technology.
This is because the experiment only looked at the stock price for the next trading day, and most people expected the market to have already set the price a few seconds after the news was published. It’s because
“As more people use these types of tools, we expect markets to become more efficient and returns less predictable,” said Lopez-Lira. “My guess is that if you do this exercise, the next five years, by year five, there will be zero predictability of returns.”