New AI Reads Fed Speeches, May Predict Your Loan Rates

Central bank press conferences are watched by millions of people, mostly for one reason: to guess where interest rates go next. Economists Martin Feldkircher, Márton Kardos and Kristoffer Laigaard Nielbo tried a new approach. They fed transcripts from the US Federal Reserve and the European Central Bank into an AI system and measured sentiment on three things: how strict policy sounds, how optimistic the economy sounds, and how uncertain officials sound. Crucially, they didn't just average those feelings across the whole speech. They charted them as a curve — a path that rises and dips from the opening statement to the final question.

What they found is striking. The shape of that curve predicted actual rate decisions better than older tools that just count 'hawkish' versus 'dovish' words. In plain terms, it matters not only what officials say, but the order in which they say it and how much they emphasize it. The researchers also found a second effect: the curve seemed to influence how professional forecasters updated their inflation expectations, and how much those forecasters disagreed with each other.

If you've ever watched your mortgage rate, savings account yield or credit card APR move after a Fed meeting, this is your world. A tool that reads the way language is structured could, in theory, give a slightly earlier and sharper read on where rates are heading — useful for anyone with a loan, a retirement account, or a business that borrows money. The researchers argue that how a central bank designs its message is not a minor stylistic choice but a core part of the signal itself.

The honest catch: this is a 39-page academic paper, not a product you can download, and it covers only two central banks using past speeches. The AI may be picking up patterns that markets already partly anticipate. And if officials learn that the sequence of their words moves markets, they may simply change how they speak.