Interpreting Fedspeak with Confidence: A LLM-Based Uncertainty-Aware Framework Guided by Monetary Policy Transmission Paths

TL;DR

This work tackles the challenge of interpreting Fedspeak by introducing an uncertainty-aware LLM framework augmented with domain-driven reasoning anchored in monetary policy transmission paths. It combines data augmentation via financial entity relations with a formal transmission-path model and a dynamic uncertainty decoding module, enabling accurate and reliable policy stance classification. Empirical results on the FOMC dataset show state-of-the-art performance and reveal a strong link between perceptual uncertainty and prediction errors, highlighting PU as a diagnostic tool for reliability. The approach enhances economic interpretability and supports human analysts in robust policy stance analysis, with practical implications for financial forecasting and risk management.

Abstract

"Fedspeak", the stylized and often nuanced language used by the U.S. Federal Reserve, encodes implicit policy signals and strategic stances. The Federal Open Market Committee strategically employs Fedspeak as a communication tool to shape market expectations and influence both domestic and global economic conditions. As such, automatically parsing and interpreting Fedspeak presents a high-impact challenge, with significant implications for financial forecasting, algorithmic trading, and data-driven policy analysis. In this paper, we propose an LLM-based, uncertainty-aware framework for deciphering Fedspeak and classifying its underlying monetary policy stance. Technically, to enrich the semantic and contextual representation of Fedspeak texts, we incorporate domain-specific reasoning grounded in the monetary policy transmission mechanism. We further introduce a dynamic uncertainty decoding module to assess the confidence of model predictions, thereby enhancing both classification accuracy and model reliability. Experimental results demonstrate that our framework achieves state-of-the-art performance on the policy stance analysis task. Moreover, statistical analysis reveals a significant positive correlation between perceptual uncertainty and model error rates, validating the effectiveness of perceptual uncertainty as a diagnostic signal.

Figures (5)

• Figure 1: Comparison between (a) the traditional method and (b) our proposed method. Our method introduces a domain-specific reasoning approach grounded in the monetary policy transmission mechanism to emulate how human experts analyze policy stances, providing the model with relevant domain knowledge. In addition, we employ a dynamic uncertainty decoding module to capture perceptual uncertainty and estimate the model's confidence in its predictions, thereby improving overall reliability.
• Figure 2: The workflow of data augmentation.We extract economic entity relations from Fedspeak, and then perform reasoning grounded in the monetary policy transmission mechanism using structured templates to derive policy advice.
• Figure 3: Overview of Dynamic Uncertainty Decoding module.When the LLM is about to generate a prediction token, we obtain the corresponding logits over the vocabulary. Dynamic uncertainty decoding module quantifies the model's PU via estimating CR and EA. The decoding strategy for the current token is selected based on whether the PU exceeds the threshold.
• Figure 4: P-values of T-test, Mann-Whitney U test and logistic regression for different $K$ values on FOMC dataset.
• Figure 5: Comparison of F1 Performance across Temperature, Threshold Percentile and K value.