US Rates StrategySim FOMC

中文标题

美国利率策略 · 模拟联邦公开市场委员会（FOMC）

SMBC09/01/2026updated 10/17/2026

Citable research brief

Core conclusion · Tlines extraction

SMBC's US Rates Strategy report explores the use of AI and large language models in analyzing and simulating FOMC meetings. It finds that while AI like ChatGPT can accurately classify the hawkishness of Fed commentary, simulating FOMC deliberations is less accurate than mechanical models. The report also discusses FOMC dissents, political pressure effects, and the challenges of simulating individual members like Chair Warsh.

Institution: SMBC
Published: 09/01/2026
Research topic: Political pressure effect · FED · Chair proposal dynamics
Time horizon: not_applicable

FED · FED

Key arguments

ChatGPT performs better than dictionary-based models in grading the hawkishness of Fed commentary, closely mimicking human reasoning.
AI-based FOMC simulations have higher mean absolute forecast errors (8bp) compared to mechanical Monte Carlo models.
Political pressure on FOMC outcomes is more effective when applied publicly rather than privately to the Chair.
FOMC dissents have become rarer over time, with a shift from 6% in 1957-2013 to 3.7% since 2013.
Simulating FOMC members like Chair Warsh is difficult due to lack of information about his reaction function.

Key numbers

Mean absolute forecast error: 8bp
Dissent share 1957-2013: 6%
Dissent share since 2013: 3.7%
Public pressure effect: 10% closer to administration view
Number of simulated meetings: 218

Main risks

AI simulations may suffer from false memory effects, leading to overfitting in-sample.
Simulated FOMC outcomes may not accurately reflect real deliberations due to complexity.
Political pressure could distort FOMC independence if public criticism becomes prevalent.

Corresponding source evidence

Kazinnik and Sinclair find that public pressure moves Committee voting 10% closer to the administration’s view. · not_applicable
The AI sim results suggest that if the Chair’s proposal is too far outside the consensus, the other FOMC members will ignore it . · not_applicable
Creating a simWarsh is complicated because he provides little information about his reaction function and his commentary is terse. · not_applicable

Conditions / invalidation

Public pressure applied to all members
Chair's proposal is too far outside consensus