ChatGPT Can Decode Fed Speak, Predict Stock Moves From Text
The use of artificial intelligence (AI) in finance has been gaining traction, and the recent academic research papers using ChatGPT in market-relevant tasks — one in determining whether Federal Reserve statements were hawkish or dovish, and one in assessing whether headlines were good or bad for a stock — show early promising results. ChatGPT was able to decode Fed speak better than widely used models and even provided explanations similar to the Federal Reserve's own analyst. In the second study, Alejandro Lopez-Lira and Yuehua Tang at the University of Florida found that ChatGPT was able to provide responses which demonstrated a statistical link to the stock's subsequent movements, indicating the technology's ability to correctly parse the implications of news.
Deciphering Federal Reserve Statements
In the first research paper titled "Can ChatGPT Decipher Fedspeak?" by Anne Lundgaard Hansen and Sophia Kazinnik at the Richmond Fed, they showed that ChatGPT performed better than the commonly used model created by Google called BERT and classifications based on dictionaries. ChatGPT also came closest to humans in its ability to figure out if the central bank's statements were dovish or hawkish. Further, ChatGPT explained its classifications of Fed policy statements in a way that resembles the central bank's own analyst.
The paper used a sentence from a May 2013 statement as an example: "Labor market conditions have shown some improvement in recent months, on balance, but the unemployment rate remains elevated." Bryson, described in the paper as "a 24-year-old male, known for his intelligence and curiosity," concluded that the statement is dovish because it implies the economy is not fully recovered yet. ChatGPT's explanation was similar to Bryson's.
Predicting Stock Movements
In the second paper titled "Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models" by Lopez-Lira and Tang, the researchers prompted ChatGPT to act as a financial expert and interpret corporate news headlines. The study used news post-late 2021, an area not covered in ChatGPT's training data. The findings showed that ChatGPT provided responses that showed a statistical link to the stock's subsequent movements.
The paper used the headline "Rimini Street Fined $630,000 in Case Against Oracle" as an example and explained that ChatGPT saw it as positive because the penalty "could potentially boost investor confidence in Oracle’s ability to protect its intellectual property and increase demand for its products and services." This shows ChatGPT's ability to correctly parse news implications.
While using natural language processing in finance is not new, the advances demonstrated by ChatGPT suggest a move toward making the technology more accessible to a broader community of finance professionals. Furthermore, the research suggests that ChatGPT can pull off similar tasks without being specifically trained, and fine-tuning it based on specific examples can make it better.
In conclusion, ChatGPT's early success in deciphering Fed speak and predicting stock movements from text has piqued the interest of the finance community. The technology developed by OpenAI heralds a new level of parsing nuance and context and holds much promise for the future of the industry.
