PREDICTING THE EFFECT OF FOMC WORDING ON THE U.S. TREASURY YIELDS: A MACHINE LEARNING ANALYSIS

Central banks have increasingly relied on their post-meeting statements to shape market expectations around future monetary policy. This paper explores how the wording of the Federal Open Market Committee (FOMC) statements impacts the U.S. Treasury markets over the period from 2000 to 2023. Using a state-of-the-art natural language processing (NLP) approach, I develop a custom large language model that captures the effect of the statement’s wording on the change in Treasury yields following its release. To train the model, I build a novel dataset of analysts’ reactions to the FOMC policy statements. These, along with the change in Treasury Yields, serve as a proxy for the change in market expectations. I apply the trained model to analyze the effects across different maturities. The model is able to explain 53%, 58%, and 63% of the variance in 5, 10, and 30-year treasury yields post-FOMC announcement time, which is significantly higher than the variance explained by other econometric approaches. Additionally, I explore the potential impacts of the alternative FOMC statements found in the Tealbook. This paper advances macroeconomic research by providing a modern forecasting methodology technique, while also serving as a tool for policymakers in understanding the formation of monetary policy expectations.