FedWatcher
From Fed language to a model.
The problem
Federal Reserve communications contain qualitative signals that cannot be read like a standard numerical series. Text needs to be connected to macroeconomic data without losing its source.
The work
A pipeline collects FOMC releases, scores their tone with an LLM and combines it with FRED indicators. A backend and dashboard expose the process and model estimates.
The result
A research application developed in the context of Programming in Finance II at USI: collection, text analysis and modelling in an inspectable workflow. Estimates are experimental model outputs.
Role and tools
Research project and software development
Python, FastAPI, SQLite, LLM, FRED