The Monetary Policy Statement Database: An LLM Application to Global Financial Conditions

This study introduces the Monetary Policy Statement Database (MPSD), comprising 6,693 statements from 51 central banks worldwide (1990–2024). We develop a reproducible pipeline combining standard natural language preprocessing with large language model (LLM) tools for cross-country analysis. Four key findings emerge. First, statements lengthened substantially after the Global Financial Crisis while readability improved modestly. Second, inflation references comove across countries during global inflation episodes. Third, LLM-based question answering and aspect-based sentiment reveal that central banks attribute global financial conditions primarily to broad U.S. macroeconomic developments rather than to Federal Reserve policy actions specifically. Fourth, using a benchmark dictionary-based sentiment index and LLM-derived aspect-based sentiment indicators, Granger causality tests suggest that statement sentiment predicts the Global Financial Cycle rather than merely responding to it. The MPSD and accompanying codebase support reproducible research on monetary policy communication and international transmission.