Introduction

Central bank communications heavily influence markets. Fed chair speeches, FOMC statements contain hints about future policy. Dynamic topic modeling tracks how central bank language evolves, revealing policy intent shifts before formal announcements.

Central Bank Language and Market Expectations

When Fed increases mentions of "inflation risk" and "rate hikes," markets anticipate tightening. When focus shifts to "growth concerns" and "employment," markets expect easing. Quantifying this language shift provides signals.

Dynamic Topic Models for Central Bank Speeches

Train Dynamic LDA on all Fed speeches (2000-2024, 200+ speeches). Topics naturally emerge: inflation, employment, financial stability, interest rates, forward guidance. Topic prevalence evolves over time, revealing policy focus shifts.

Key Central Bank Topics

1. Monetary Policy (interest rates, QE, tightening/easing)
2. Inflation (price stability, core inflation, expectations)
3. Employment (labor market, wage growth, unemployment)
4. Financial Stability (systemic risk, credit growth, asset bubbles)
5. International/Geopolitical (trade, exchange rates, global growth)
6. Technical/Forward Guidance (communication strategy, policy framework)

Tone Shifts: Hawkish vs Dovish

Assign hawkish/dovish scores to topics: "rate hike" = hawkish, "employment concerns" = dovish. Track topic sentiment over time. When hawkish topics increase and dovish topics decrease, Fed is tightening policy.

Empirical Results on Fed Communications

Analyze Fed speeches 2015-2024:

  • Hawkish tone increases starting Q4 2021 (inflation concerns emerge)
  • Peak hawkishness Q3 2022 (aggressive tightening)
  • Dovish shift starts Q4 2023 (growth/recession concerns)
  • Shifts precede formal policy changes by 2-6 weeks
A strategy shorting bonds when hawkish tone increases, going long when dovish tone increases achieves 1.1 Sharpe ratio on 10-year Treasury futures.

Forward Guidance Extraction

Extract specific forward guidance: "rates likely to remain at high levels longer than previously expected" = more hawkish than prior guidance. Quantify guidance changes: more explicit commitment to rate cuts → more dovish.

Sentiment vs Topics

Simple sentiment analysis misses nuance. "The economy is strong" (positive sentiment) can be dovish (strong economy doesn't need rate hikes) or hawkish (strong economy justifies rate hikes). Topic models capture this context.

Real-Time Monitoring

Deploy dynamic topic models to monitor recent Fed communications. Track topic evolution: are tone shifts consistent with prior trends or surprising? Surprises (sudden hawkish shift after dovish period) predict market dislocations.

Market Impact Quantification

Regress bond/equity market moves on fed communication tone shifts. Coefficient: 10-percentage-point increase in hawkish topic prevalence causes 5-10 basis point increase in 10-year yields. Use to forecast rates from Fed communication.

Implementation

1. Collect Fed speeches from Federal Reserve website.
2. Preprocess: tokenize, lemmatize, remove stop words.
3. Train Dynamic LDA with 10-15 topics.
4. Track topic evolution over time.
5. Score topics as hawkish/dovish.
6. Monitor recent shifts, generate trading signals.