Two recently published academic papers have utilized the artificial intelligence chatbot ChatGPT to perform market-relevant tasks, with promising results. The papers explore ChatGPT’s ability to decipher Federal Reserve statements as hawkish or dovish and to determine whether headlines are positive or negative for a stock. The success of ChatGPT in both tasks indicates a significant advancement in the use of technology to extract trading signals from vast amounts of textual data, including news articles and social media posts. These early results support the recent hype surrounding ChatGPT’s potential applications in the field of finance.
The first academic paper, titled “Can ChatGPT Decipher Fedspeak?” has found that ChatGPT, an artificial intelligence chatbot, is highly effective in determining whether Federal Reserve statements are hawkish or dovish. The study was conducted by two researchers, Anne Lundgaard Hansen, and Sophia Kazinnik, from the Federal Reserve Bank of Richmond. They compared the performance of ChatGPT to that of Google’s commonly used model called BERT and classifications based on dictionaries.
The results showed that ChatGPT performed better than both BERT and the dictionary-based classifications. Moreover, ChatGPT could even explain its classifications in a way that closely resembled how the central bank’s own analysts interpreted the language. This makes ChatGPT a valuable tool for turning vast amounts of news articles, tweets, and speeches into actionable trading signals.