Job Market Practice Seminar

Title: Monetary Policy Surprises: A New Hope – LLM agents and Central Bank Communication

Abstract: I develop a novel framework for measuring monetary policy surprises using Large Language Models to systematically process Federal Reserve communications. My  multi-agent system analyzes the complete institutional communication cycle—Beige Books, FOMC Minutes, and policy Statements—as an integrated narrative rather than isolated documents. Specialized LLMs extract economic assessments, analyze internal deliberations, and generate genuine surprises by comparing ex-ante expectations with realized decisions using only information available before each meeting's blackout period. The resulting narrative surprises explain 61% of  policy variation compared to 15-17\% for market-based measures, while remaining equally unpredictable (8-12% R² on standard predictors). Impulse response analysis reveals that narrative surprises generate theoretically consistent contractionary effects and coherent yield curve dynamics, while market measures produce puzzling expansionary responses at every horizon. A duration-hedged trading strategy validates the economic significance, generating significant positive per-trade returns. These findings demonstrate that LLMs can systematically extract policy-relevant information from central bank communications, providing a complementary approach to high-frequency identification that captures the deliberative evolution of Fed thinking.