Paper: arXiv 2511.02469
Authors: Kaito Takano, Masanori Hirano, Kei Nakagawa
Abstract
Accurately forecasting central bank policy decisions, particularly those of the Federal Open Market Committee(FOMC) has become increasingly important amid heightened economic uncertainty. While prior studies have used monetary policy texts to predict rate changes, most rely on static classification models that overlook the deliberative nature of policymaking. This study proposes a novel framework that structurally imitates the FOMC’s collective decision-making process by modeling multiple large language models(LLMs) as interacting agents. Each agent begins with a distinct initial belief and produces a prediction based on both qualitative policy texts and quantitative macroeconomic indicators. Through iterative rounds, agents revise their predictions by observing the outputs of others, simulating deliberation and consensus formation. To enhance interpretability, we introduce a latent variable representing each agent’s underlying belief(e.g., hawkish or dovish), and we theoretically demonstrate how this belief mediates the perception of input information and interaction dynamics. Empirical results show that this debate-based approach significantly outperforms standard LLMs-based baselines in prediction accuracy. Furthermore, the explicit modeling of beliefs provides insights into how individual perspectives and social influence shape collective policy forecasts.
Complexity vs Empirical Score
- Math Complexity: 3.5/10
- Empirical Rigor: 7.0/10
- Quadrant: Street Traders — empirical and applied, lower math complexity
Why this score: The paper employs complex LLM-based multi-agent architectures and latent variable modeling, but the mathematical framework is primarily conceptual and probabilistic rather than dense with advanced derivations. Empirical evaluation is robust, involving backtested prediction accuracy against baselines and ablation studies on real FOMC/Beige Book data, making it highly implementation-ready.
Research Flowchart
flowchart TD
A["Research Goal<br>Predict FOMC Monetary Policy<br>via Simulated Debate"] --> B["Inputs<br>Economic Reports & Policy Texts"]
B --> C["Agent Initialization<br>Assign Hawkish/Dovish Latent Beliefs"]
C --> D{"Multi-Agent Debate Loop<br>Iterative Prediction & Consensus"}
D -- Rounds of Interaction --> D
D --> E["Decision Classification<br>Final Rate Change Prediction"]
E --> F["Outcomes<br>Higher Accuracy<br>Interpretable Belief Dynamics"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a debate-based multi‑agent LLM framework that imitates FOMC decision‑making and introduces a latent hawk‑dove belief variable to enhance interpretability.
Source sentence
This study proposes a novel framework that structurally imitates the FOMC’s collective decision-making process by modeling multiple large language models (LLMs) as interacting agents.
Assumes 1
- Conditional independence of an agent’s output from inputs once its latent belief is known. statistical
Source sentence
Assumption 1 (Conditional Independence Given Latent Belief) For any round t and any belief θ ∈Θ, the output z(t+1)i by agent i is conditionally independent of x, v, and Z(t) once θ is known: Pz(t+1)i|θ,x,v,Z(t),ϕi = Pz(t+1)i|θ,ϕi (∀θ ∈Θ, t ≥0).
Data
- Policy Rate macro commercial
Source sentence
We obtain this data from the Bloomberg terminal6.
- Macroeconomic Indicators macro public
Source sentence
We obtain these data from the Federal Reserve Economic Data (FRED) database7.
- Beige Book corpus text-news public
Source sentence
In our research, we work with the Beige Book corpus created by [33].
Costs
Not applicable
Finds 3
- The full model (debate + latent beliefs) achieved an F1 score of 0.48, outperforming versions that remove debate, beliefs, the Beige Book, or macro indicators. predictive-accuracy
Source sentence
The full model that includes both debate and beliefs reached an F1 score of 0.48, outperforming versions that remove the debate, the beliefs, the Beige Book, or the macroeconomic indicators.
- Ablation shows the Beige Book is especially important for accuracy. performance
Source sentence
The ablation study showed that the Beige Book is especially important for accuracy.
- Debate rounds reduce the strong Hold bias observed when agents do not interact. performance
Source sentence
the debate rounds lessen the strong Hold bias seen when the agents do not interact.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 16 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (38)
- Multi-LLM debate: Framework, principals, and interventions (2024)
- Federal Reserve information and the behavior of interest rates (2000)
- High-frequency identification of monetary non-neutrality: the information effect (2018)
- The macroeconomic impact of news about policy and news about the economy in ECB announcements (2018)
- Treasury yield spread prediction with sentiments of Beige Book and macroeconomic data (2023)
- International transmission of US monetary policy shocks: Evidence from VAR’s (2001)
- Capital flows and the risk-taking channel of monetary policy (2015)
- Communication by central bank committee members: different strategies, same effectiveness? (2007)
and 30 more
- Does central bank tone move asset prices? (2025)
- Monetary Policy with Uncertain Inflation Persistence (2024)
- Forecasting US inflation in real time (2021)
- Inflation rate tracking portfolio optimization method: Evidence from Japan (2022)
- Artificial intelligence and central bank communication: the case of the ECB (2024)
- Artificial intelligence in the financial system: implications and progress from a central bank perspective (2024)
- Transparency and deliberation within the FOMC: A computational linguistics approach (2018)
- Machine learning and asset allocation (2019)
- Financial market effects of FOMC projections (2021)
- Sequential deliberation in collective decision-making: The case of the FOMC (2016)
- Deliberating American monetary policy: A textual analysis (2013)
- ChatEval: Towards Better LLM -based Evaluators through Multi-Agent Debate (2024)
- Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate (2024)
- Construction of the FRB Beige Book Corpus and Analysis~(Japanese) (2023)
- Improving factuality and reasoning in language models through multiagent debate (2023)
- The dissemination of culture: A model with local convergence and global polarization (1997)
- Generative Agents: Interactive Simulacra of Human Behavior (2023)
- ResearchTown: Simulator of Human Research Community (2024)
- Spontaneous Emergence of Agent Individuality Through Social Interactions in Large Language Model-Based Communities (2024)
- Scope of Large Language Models for Mining Emerging Opinions in Online Health Discourse (2024) — link
- ChatGPT (2023)
- GPT-4 Technical Report (2023) — link
- Scaling and Criticality in a Stochastic Multi-agent Model of a Financial Market (1999)
- Attention is All you Need
- BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding (2019)
- Improving Language Understanding by Generative Pre-Training (2018)
- Language Models are Unsupervised Multitask Learners (2019)
- Language Models are Few-Shot Learners (2020)
- Performance and biases of Large Language Models in public opinion simulation (2024)
- Models of Segregation (1969)
0 of 38 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.