ChatGPT can decode Fed speak, predict stock moves from headlines

The first wave of academic research applying ChatGPT to the world of finance is arriving – and judging by the early results, the hype of the past few months is justified.

Two new papers have been published this month that deploy artificial intelligence chatbots in market-related tasks – one in understanding whether Federal Reserve statements were hawkish or dovish, and one in determining whether headlines are good for stocks. Were or bad

ChatGPT has passed both tests, suggesting a potentially major step forward in the use of technology to convert tweets and speech from news articles into trading signals.

This process is nothing new on Wall Street, of course, where quants have long used the language model underpinning chatbots to inform many strategies. But the findings point to the technology developed by OpenAI reaching a new level in terms of subtlety and context parsing.

“This is one of those rare cases where the hype is real,” said Slavi Marinov, head of machine learning at the Man AHL, which for years has been using what is known as natural language processing to read texts like earnings tapes and Reddit posts. technology is being used.

In the first paper, Can ChatGPT Decipher Fedspeak? As titled, two researchers at the Fed themselves found that ChatGPT came closest to humans in detecting whether the central bank’s statements were dovish or hawkish. Anne Lundgaard Hansen and Sophia Kaznik at Richmond Fed showed that it beat a commonly used model called BERT from Google and also did classification based on dictionaries.

ChatGPT was also able to explain its classification of Fed policy statements in a way that was similar to the central bank’s own analyst, who also interpreted the language to serve as a human benchmark for the study.

Take this sentence from a May 2013 statement: “Labor market conditions have shown some improvement in recent months, on balance, but the unemployment rate remains elevated.” Robot explained that the line is clear because it shows that the economy is not yet fully recovered. This was similar to the analyst’s conclusion – Bryson, described in the paper as “a 24-year-old male, known for his intelligence and curiosity”.

In another study, can ChatGPT forecast stock price movements? Return Predictability and Large Language Models, Alejandro López-Lira and Yuhua Tang at the University of Florida inspired ChatGPT to pretend to be a financial expert and interpret corporate news headlines. They used news from after 2021, a period that was not included in the chatbot’s training data.

The study found that the answers provided by ChatGPT showed a statistical link to subsequent stock moves, a sign that the technology was able to correctly parse the implications of the news.

In an example of whether the headline “Rimini Street Fine $630,000 in Case Against Oracle” was good or bad for Oracle, ChatGPT explained that it was positive because the fine was “potentially intended to protect its intellectual property and increase demand”. can enhance investor confidence in Oracle’s ability” to deliver on its products and services. ,

Using NLP for most sophisticated quantification is now almost run-of-the-mill, be it finding out how popular a stock is from Twitter or including the latest headlines on a company. But the progress demonstrated by ChatGPT looks set to open up a whole world of new information and make the technology more accessible to a wider community of finance professionals.

To Marinov, while no surprise that machines can now read almost as well as people, ChatGPT could potentially speed up the whole process.

When Mann AHL was first building the model, the quant was manually labeling each sentence for the asset as positive or negative in order to give the hedge fund machines a blueprint for interpreting the language. The London-based firm then turned the whole process into a game, ranking participants and calculating how much they agree on each sentence, so that all employees could get involved.

Two new papers show that ChatGPT can pull off similar tasks without being specially trained. Fed research has shown that this so-called zero-shot learning already outperforms previous techniques, but fine-tuning it based on some specific examples made it even better.

“Earlier you had to label the data yourself,” said Marinov, who also previously co-founded an NLP startup. “Now you can complement it by designing the right signs for ChatGPT.”