Modeling Hawkish-Dovish Latent Beliefs in Multi-Agent Debate-Based LLMs for Monetary Policy Decision Classification
Kaito Takano, Masanori Hirano, Kei Nakagawa
TL;DR
This paper tackles predicting FOMC policy rate decisions under uncertainty by modeling the decision process as a structured debate among $n$ LLM-based agents, each carrying a latent hawkish/dovish belief. The approach introduces a Bayesian latent-belief framework in which agents iteratively exchange outputs over $T$ rounds, with predictions drawn from $P(z|x,v,Z^{(t)},\phi)$ and latent stance $\Theta$ guiding interpretation of inputs. Empirically, the full debate-with-beliefs model achieves its best performance (F1 ≈ 0.48) and outperforms ablated variants, with the Beige Book and macro indicators providing the strongest signals; debate dynamics help mitigate a Hold bias seen in non-debated setups. This framework offers interpretable insight into how individual beliefs and social influence shape collective monetary-policy forecasts and can be extended to other central-bank bodies.
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.