The study, carried out by researchersin Sheffield Business School at Sheffield Hallam University, analysed speeches given by Bank of England Monetary Policy Committee members prior to interest rate voting .
Using ChatGPT, the researchers classified each speech as dovish, neutral or hawkish based on the tone and content. This classification was then used in an econometric model which could successfully predict how each member would vote at the next one or two policy meetings.
The results showed ChatGPT sentiment analysis of the speeches was a statistically significant determinant of future voting behaviour. Committee members who gave more neutrspeeches were more likely to vote for interest rate hikes at subsequent meetings.
Dr Drew Woodhouse, in Sheffield Business School and lead author of the research, said: "Our findings highlight the predictive potential of tools like ChatGPT for processing human beliefs and expectations. This has major implications for forecasting policy decisions and modelling economic expectations.
“As generative AI capabilities expand, these tools hold promise for gaining insights from complex communications and texts across many fields. Our findings reveal new opportunities at the intersection of technology, language and economics."
By leveraging natural language processing and textual analysis, ChatGPT demonstrated proficiency in classifying the nuanced language of central bankers. The technology was able to grasp the intricacies of "Fed speak" and relate speech content to eventual policy actions.
The researchers suggest this approach could be extended to study other aspects of central bank communications, like forward guidance. It also illustrates how publicly available AI like ChatGPT can empower economic analysis and financial decision-making.
The full paper can be read here.
