CB Policy Analysis

Overview

A research-grade, leak-safe pipeline for Federal Reserve / FOMC policy-text analysis. It scores the monetary-policy stance of every FOMC statement since 1999 using three independent measures — a document-mixed index, a FOMC-RoBERTa transformer, and a transparent hawk/dove lexicon — and tests each one, strictly out-of-sample, against the Bauer-Swanson orthogonalized policy surprise. The result is a rigorously established null: no tone measure adds robust marginal information beyond the surprise already priced in, and the one apparent exception is diagnosed as a policy-regime confound rather than real signal. The same fetch-and-score machinery also powers a live, transparent FOMC Statement Tracker.

What it does

• Leak-safe walk-forward harness — trains only on releases strictly before the one being predicted; a deliberate look-ahead injection must fail the test suite.
• Scores every FOMC statement's stance three ways: a document-mixed index, a FOMC-RoBERTa transformer, and a transparent hawk/dove word-count lexicon.
• Tests each tone measure for marginal out-of-sample explanatory power beyond the Bauer-Swanson orthogonalized policy surprise — the harness's central discipline.
• Stress-tests any apparent signal against regime confounds (rate-level controls, era jackknifing) rather than taking a positive result at face value.
• Reached a rigorous null across all three measures: no tone measure beats the policy surprise out-of-sample; the one apparent exception was diagnosed as a regime confound, not real signal.
• Repackages the same fetch-and-score pipeline into a live, deployed FOMC Statement Tracker: theme intensity, statement change magnitude, communication style, and a word-level redline.

Stack

Python · pandas · statsmodels · pytest · Plotly.

Status

Leak-safe, out-of-sample harness (131 offline tests) reached a rigorous null across three tone measures — no signal beats the policy surprise. The same pipeline runs a live, deployed FOMC Statement Tracker. Live tracker →

GitHub →