Two research papers have been released this month that utilize ChatGPT, OpenAI’s Artificial Intelligence chatbot, with market-relevant tasks. One paper tested the ability of ChatGPT to interpret Federal Reserve statements and the other tested the ability of ChatGPT to interpret stock prices based on corporate news headlines.
Remarkably, ChatGPT provided successful results in both tests, leading to speculation that the technology is ready to be used in transformation of large texts from news articles, to tweets and speeches, into trading signals. This is an advancement from traditional use of the technology, as “Quants” have always harnessed the underlying language models of ChatGPT to inform their strategies.
In the paper “Can ChatGPT Decipher Fed Speak?”, two researchers from the Federal Reserve, Anne Lundgaard Hansen and Sophia Kazinnik from the Richmond Fed, found that ChatGPT was closest to humans in determining whether Federal Reserve Statements were hawkish or dovish. It performed better than a common model from Google called BERT, and also outperformed models based on dictionaries.
In the paper “Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models”, Alejandro Lopez-Lira and Yuehua Tang from the University of Florida prompted ChatGPT to function like a financial expert and interpret news headlines. They examined news over the period of late 2021, which was not present in ChatGPT’s training data. Results showed that ChatGPT’s responses were statistically linked to subsequent stock moves, indicating that the technology accurately understands the implications of the news.
Slavi Marinov, head of machine learning at Man AHL, which has previously used natural language processing to read texts such as earnings transcripts, Reddit posts and more, believes that the hype around ChatGPT is deserved. He noted that ChatGPT has the potential to speed up the process of interpretation, as Man AHL had originally had to label every sentence as either positive or negative for an asset to provide a guideline to machines.
Bloomberg LP, the parent of Bloomberg News, also released a language model for finance last month. With this AI technology, the process of parsing nuance and context has advanced to a new level, ensuring that financial professionals have in their hands an accessible, efficient and powerful tool to interpret texts and potentially capture the right signals in order to make the right moves in the stock market.
Oracle, one of the companies mentioned in this article, is a multinational computer technology company. It develops and markets computer hardware systems, enterprise software products and other related services. Oracle is a well-known acquirer of smaller companies, the most notable being its acquisition of the enterprise resource planning software provider, PeopleSoft.
Bryson, a 24-year-old human analyst mentioned in the article mentioned above, is known for his intelligence and curiosity. As a human benchmark, he is in charge of accurately interpreting and classifying Federal Reserve statements. Bryson’s results were also successfully compared to ChatGPT’s, showing that the chatbot is capable of replicating the accuracy of and providing an insight much like any expert analyst.
