Dear clients,
The first wave of academic research on the application of ChatGPT in the world of finance is approaching – and judging by the preliminary results, the hype of the past few months was not without reason.
Two new articles were published this month that used an AI-powered chatbot to solve market-relevant problems — one to decipher whether the Federal Reserve’s statements were hawkish or dovish, and one to determine whether the headlines were good or bad for stocks.
ChatGPT passed both tests, suggesting a potentially important step forward in using the technology to convert text from news articles to tweets and speech into trading signals.
Of course, this process is nothing new on Wall Street, where quants have long used the language models underlying chatbots to inform many strategies. But the results indicate that the technology developed by OpenAI is reaching a new level in terms of parsing nuance and context.
In the first article, titled „Can ChatGPT Decipher Fedspeak?“, two researchers from the Fed itself found that ChatGPT came closest to people in figuring out whether central bank statements were dovish or hawkish. ChatGPT was even able to explain its categorization of the Fed’s policy statements in a way that would be done by a central bank analyst’s own, interpreting language as a human benchmark for research.
In the second study, „Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models“, scientists at the University of Florida prompted ChatGPT to pretend to be a financial expert and interpret corporate news headlines. The study found that the responses given by ChatGPT had a statistical relationship with subsequent stock movements, an indication that the technology was able to correctly analyze the impact of the news.
Two new papers suggest that ChatGPT can perform similar tasks even without special training. The FRS study showed that this so-called zero-shot learning is already superior to previous technologies, but fine-tuning it based on some specific examples has made it even better.
