ML-QuantSubscribe

arXivMacro-Finance & Rates

Jawboning the Fed: Do Markets Listen When Administrations Talk?

Analysis of four decades of news finds dovish political pressure lowers near-term rate expectations by roughly one basis point per event, while hawkish pressure shows no effect.

Featured in No. 137 on 30 Oct 2026 · 4 days after release

Three-Stage LLM Classification Pipeline
Figure 1 : Three-Stage LLM Classification Pipeline
Released
26 Oct 2026
First featured
No. 137 · 30 Oct 2026
Published in
Not yet, as far as Semantic Scholar knows
Fanfare
3 of 5
Identifier
arXiv:2610.32300
Authors
Michela Baretti et al.

Abstract

From arXiv (CC0).

We provide the first systematic evidence on whether financial markets price political pressure on the Federal Reserve as a signal of future monetary policy. Using approximately four decades of Reuters newswire articles from 1988 to 2025, classified by an open-weight large language model ensemble into dovish and hawkish pressure, we estimate the daily response of interest rate expectations across the term structure, the US dollar, inflation compensation, and market volatility. Dovish pressure lowers near-term rate expectations by roughly one basis point per pressure event---equivalent to a four-percent increase in the implied probability of a rate cut at the next FOMC meeting---with the effect decaying monotonically and vanishing beyond the three-month horizon. Hawkish pressure has no detectable effect. Furthermore, excluding the Trump presidencies does not weaken the estimate, ruling out the possibility that the result is confined to an idiosyncratic recent episode. The response is accompanied by dollar depreciation, with stable inflation compensation and market volatility. We interpret the pattern as markets pricing anticipated monetary accommodation, rather than a credibility tax on the nominal anchor.

Citations and venue from Semantic Scholar (ODC-BY), refreshed weekly. Summary: Quant Letter (CC BY 4.0).

    Type to search. Try rough volatility, LLM agents or FinGPT.

    ↑↓ move↵ openesc closeFull search page