Can ChatGPT predict future interest rate decisions?

The public interest in Large Language Models (LLMs) and generative artificial intelligence (AI) has gained significant traction, with academic studies attempting to assess their application to a range of fields. In particular, the rise in Generative Pre-trained Transformer (GPT) models, in the form of ChatGPT has prompted a range of calls to examine beyond its use in writing and language generation; to test its predictive qualities and assess its abilities to process complex textual information. Our paper extends upon the examination of GPT model use in the monetary policy context. We set out to test the hypothesis that LLMs and GPT models can offer predictive qualities of future interest rate decisions through their textual processing and sentiment capabilities. Specifically, we use GPT-3.5 to evaluate and label the speech of every Bank of England Monetary Policy Committee (MPC) member based on linguistic expectations. We then model a preferred policy rate vote equation for each speech giving committee member (i) in each of their future rate decision windows (t+n). We find that ChatGPT can predict future interest rate decisions. Our results provide evidence for the potential of LLMs to help us better process latent human beliefs, make out-of-sample predictions and navigate possible models of rational expectations.