To understand the economy, researchers are now using AI to listen to what the Fed *didn't* say.
April 16, 2026
Original Paper
LLM-Based Expectations and Fed Communication
SSRN · 6580319
The Takeaway
When the Fed Chair speaks, the market reacts, but it's hard to tell if the reaction is to the news itself or how it was delivered. This paper uses LLMs to generate 'counterfactual' speeches—essentially what the market *expected* to hear—to isolate the pure surprise in the actual speech. It turns out that the 'vibe' and the unexpected phrasing are often more powerful than the hard data being reported. This allows economists to measure the 'market shock' of a single sentence with surgical precision. For you, it means that your mortgage rate or 401(k) is being moved by subtle nuances in language that only an AI can now fully quantify.
From the abstract
We introduce a new approach to measuring monetary policy surprises in FOMC press conferences. Our method uses LLMs to generate counterfactual press-conference responses by the Federal Reserve Chair and measures policy surprises as deviations between realized and simulated answers. Using high-frequency intraday data from 2011 to 2025, we show that equity prices respond strongly to these surprises: hawkish surprises lead to significant declines in stock prices, while dovish surprises increase pric