The first wave of academic study using ChatGPT in the finance industry is now underway, and preliminary findings show that the recent excitement was well-founded. This month, two more studies were released that used the artificial intelligence chatbot to do tasks that were relevant to the market, such as identifying whether Federal Reserve remarks were hawkish or dovish and whether stock-related stories were favourable or unfavourable.
Both tests were successfully completed by ChatGPT, indicating a possible significant advancement in the use of technology. Now ChatGPT has touched another height by changing text into strategies. Of course, that procedure is nothing new on Wall Street, where quants have long utilised the same language models that the chatbot is built upon to inform several strategies. However, the results suggest that OpenAI’s technology has advanced to a new level in terms of context and nuance processing.
Slavi Marinov, head of machine learning at Man AHL said, “It’s one of the rare cases where the hype is real.” Two researchers from the Federal Reserve discovered that ChatGPT was most accurate in predicting whether the central bank’s pronouncements were dovish or hawkish in the first publication, Can ChatGPT Decipher Fedspeak? At the Richmond Fed, Anne Lundgaard Hansen and Sophia Kazinnik demonstrated how it outperformed classifications based on dictionaries as well as the widely used BERT model from Google. Even more impressively, ChatGPT was able to classify Fed policy statements in a manner that mirrored those of the central bank’s own analyst, who also used the language to serve as a human standard for the study.
The second study was about whether ChatGPT can Predict Changes in Stock Prices or not. Alejandro Lopez-Lira and Yuehua Tang of the University of Florida, using Return Predictability and Large Language Models, trained ChatGPT to pretend to be a financial expert and analyse corporate news headlines. They used news from a time frame that the chatbot’s training data didn’t cover: late 2021.
The study discovered a statistical relationship between the stock’s future movements and the replies provided by ChatGPT, indicating that the technology was able to correctly comprehend the significance of the news.
