Project Description
Communication by the Federal Reserve (Fed) plays a crucial role in shaping financial markets and regulatory expectations. This research project uses Large Language Models (LLMs) to systematically extract whether approximately 11,400 Fed speeches (1993-2025) signal a restrictive ("tight") or accommodative ("loose") stance on financial stability-following the methodology of Bybee et al. (2023).
Based on these classifications, a monthly Belief Index is constructed, capturing the average tone of Fed communication regarding financial stability. To enable causal inference, the index is purged of predictable components: a predictability regression following Swanson & Bauer (2023) removes the influence of prior financial market and macroeconomic variables (including stock returns, interest rate spreads, VIX, and NFCI). The resulting Belief Surprise Index captures the genuine informational content of the speeches.
The research addresses the following questions:
(1) How reliably can LLMs classify the stance of Fed speeches on financial stability?
(2) What are the causal effects of unexpected changes in Fed communication on bank stocks, bond yields, and financial spreads?
(3) How do short-term market reactions differ from longer-term effects?
The findings provide insights into the effectiveness of central bank communication as a tool for financial stability and are relevant for regulators, market participants, and monetary policy research.
