Macroeconomic Regime Detection and Narrative Dynamics in U.S. Central Bank Speeches

This repository contains a small, self-contained research project exploring the relationship between the U.S. macroeconomic environment and communication patterns in Federal Reserve speeches. The project combines quantitative macroeconomic signal extraction with text-based analysis of central banker speech transcripts. A composite U.S. macroeconomic indicator is constructed from standard macroeconomic variables retrieved from FRED, using rolling standardization and Principal Component Analysis to summarize joint dynamics in inflation, growth, labor market conditions, and monetary policy. Structural changes in the macroeconomic environment are identified using unsupervised change point detection methods. On the textual side, the project analyzes Worldwide Central Bankers speech transcripts from the BIS Gigando dataset. Named Entity Recognition (NER) is applied to identify institutions of policy-relevant entities, region and hawkish/dovish classification. The analysis is designed to support downstream narrative modeling by providing structured representations of speech content aligned with detected macroeconomic regimes. The notebooks included in this repository focus on methodological transparency and reproducibility rather than finalized empirical claims. They are intended as building blocks for larger research projects on macroeconomic narratives, central bank communication, and the interaction between economic regimes and policy discourse.