Monetary Policy Tracking and Central Bank Sentiment Analysis
Last Updated: 19 Sept 2025
Objective:
Analyze and quantify the progression of sentiment (dovish to hawkish) in successive central bank statements. Extract associated information regarding central bank projections and compare comments to market expectations for rates.
Explanation:
Rate decisions move markets - equities, currencies, bonds and even crypto. We can get to the heart of the signals and block any noise from Central Bank and FOMC statements, minutes and transcripts by employing AI sentiment analysis.
AI models are incredibly proficient at extracting quantifiable sentiment scores from documents given the way they vectorize language into mathematical representations with embedded contextual meaning. This means that sentiment scores produced by AI models are not subjective assessments like you or I might make when reading a statement, but a bias-free, calculation based on relative language representations.
That makes this analysis valuable for investors. It can be performed within seconds of a central bank release, like FOMC statement, minutes and transcripts being published. Quantitative sentiment scores allow you monitor the progression of central bank thinking through time, comparing sequential meetings so you capture not just the statement itself but the change in thinking and nuance.
As always, be aware that models can make mistakes. At each step, examine the response and challenge information or conclusions that appear erroneous before proceeding to any subsequent steps. If in doubt use a second model with the same prompt to verify the information and generate challenge questions and answers (CoVe process) to correct interpretations of data.
Link to blog post explanation:
Preferred Model(s):
ChatGPT-5+ is preferred for sentiment related analysis tasks based on performance
Important Execution Notes:
It is strongly recommended to download and attach (or past in the URLs of) the central bank statements from its website to the prompt in order to avoid the model referring to lagging memory prior to its training cut-off.
The prompt is presented below with a Lite version and a Professional version depending on the level of detail in response and verifications the investor requires.
Note that sentiment scores are best interpreted as relative to each other, rather than absolute in magnitude.
Sample Output
Copy/Paste Prompt Set:
Important note: Subscribers can use this prompt set for their own analysis. However, the prompt is copyrighted by The Inferential Investor, paywalled, and must not be shared without permission.
Analysis Lite/Quick Prompt:
ROLE
You are a market strategist producing a quick investor readout of recent FOMC statements.
INPUTS
- Last three FOMC statements (PDFs or URLs), most recent first: {LINK1} {LINK2} {LINK3}
OBJECTIVE
Deliver a clean, investor-ready brief: (1) redline changes, (2) sentiment score 0–20 (0=dovish, 20=hawkish), (3) compact table, (4) one-paragraph read-through vs market pricing into year-end and end-of-next-year.
METHOD (QUICK)
1) Parse each statement; normalize whitespace.
2) Extract four blocks: Growth, Inflation, Labor/Unemployment, Policy/Guidance (incl. QT).
3) Redline T-2→T-1 and T-1→T. Output: HTML diffs + a 5–8 bullet “material changes” list, tag each bullet as Growth / Inflation / Labor / Guidance / QT.
4) Sentiment 0–20:
- Lexicon count: hawkish (“inflation has moved up”, “remains elevated”, “strong”, “tight”) vs dovish (“growth moderated”, “job gains slowed”, “unemployment edged up”, “downside risks to employment”).
- Action tilt: hike +2, hold 0, cut –2.
- Score = map(net_hawk - net_dove + action_tilt) to 0–20; clamp.
- Show hawk/dove counts, action tilt, final score.
5) Summary table: date, decision & range, one-line Growth / Inflation / Labor, risk-management phrase, QT note, vote/dissents, sentiment.
6) Market check:
- Pull headline market-implied path for year-end and end-of-next-year (Fed funds futures or a reliable aggregator).
- One paragraph: “Market pricing is {aligned / too dovish / too hawkish} vs statement tone,” citing which phrases drive the judgment.
7) Export: redlines (HTML), summary CSV, and a tiny sentiment trend line (3 points).
OUTPUT FORMAT
1) Executive Summary (5 bullets max)
2) Redline Highlights (two lists; each 5–8 bullets; tagged by theme) + links to HTML diffs
3) Sentiment (0–20): mini table with hawk/dove counts + action tilt; tiny 3-point trend
4) Summary Table (inline + CSV link)
5) One-Paragraph Interpretation vs Market Pricing (year-end and end-of-next-year)
6) Files: redlines.html, summary.csv, sentiment.png
SCORING RUBRIC (LITE)
- Start baseline 10.
- Add +1 per unique hawkish phrase; +2 if phrase includes “has moved up”, “higher for longer”, or explicit inflation “elevated”.
- Subtract –1 per unique dovish phrase; –2 if includes “job gains have slowed”, “unemployment has edged up”, or “downside risks to employment have risen”.
- Policy action tilt: hike +2, hold 0, cut –2.
- Clamp to 0–20. Report the components.
SUMMARY TABLE COLUMNS
[Date] | [Decision & Target Range] | [Growth (1 line)] | [Inflation (1 line)] | [Labor (1 line)] | [Risk/Guidance (1 line)] | [QT] | [Vote/Dissents] | [Sentiment 0–20]
VERIFICATION (MIN)
- Check decision and range numbers against the implementation note.
- Confirm dissenter names and preference direction.
- Verify any quoted phrases (“has moved up”, “moderated”, “edged up”) verbatim.
STYLE
- Tight bullets, no jargon, no recommendations. Bold the decision, numbers, and final sentiment per meeting. Provide 3 download links at the end.
Professional Prompt: (longer execution time, additional analysis)
ROLE
You are a senior market strategist preparing an investor-ready readout of recent Federal Reserve policy communications.
INPUTS
- Primary: The last three FOMC policy statements (PDFs or URLs) including the most recent meeting. {ATTACH_OR_LINK_HERE}
- Optional but preferred: implementation notes, Chair press-conference transcript, and SEP/dots if in scope. {OPTIONAL_URL_LINKS}
- If browsing is available, you may use official sources and top wires (FOMC site, BLS, BEA, CME FedWatch/FF futures, Bloomberg/Reuters summaries) to confirm meeting dates and market-implied paths.
OBJECTIVE
Produce an investor-focused analysis that (1) redlines the last three statements, (2) quantifies hawkish/dovish sentiment on a 0–20 scale, (3) summarizes key language and vote/dissents, and (4) interprets the evolving reaction function vs current market pricing for the policy path into year-end and end-of-next-year.
METHOD & REQUIREMENTS
A) TEXT NORMALIZATION & PARSE
- Extract text from each statement, normalize spacing, and isolate these blocks: growth, inflation, labor market/unemployment, balance-sheet/QT, guidance/risks, decision paragraph, and voting/dissents.
- Validate meeting dates.
B) REDLINE (3-WAY)
- Produce two redlines: T-2→T-1 and T-1→T (word-level). Export (i) HTML diffs and (ii) a concise “material changes” bullet list per pair.
- Tag each change as primarily related to Growth / Inflation / Employment / Policy Guidance / Balance Sheet / Risk Management.
C) SENTIMENT (0–20 HAWK/DOVE)
- Compute a 0–20 sentiment score per statement where 0 = unanimously dovish, 20 = unanimously hawkish.
- Approach:
1) Lexicon pass: count hawkish vs dovish phrases (e.g., “inflation has moved up,” “remains elevated,” “higher for longer,” versus “growth moderated,” “job gains slowed,” “downside risks to employment”).
2) Action tilt: hike/hold/cut adjustments (+2/0/–2).
3) LLM cross-check: classify tone paragraph-by-paragraph; resolve conflicts conservatively.
- Report: show hawk/dove term counts, action adjustment, raw net, and final mapped score. Plot a mini trend chart across the three meetings.
D) SUMMARY TABLE (INVESTOR-READY)
Provide a single compact table with columns:
- Meeting date
- Decision & target range (bps change)
- One-line growth language (concise)
- One-line inflation language (concise)
- One-line labor/unemployment language (concise)
- Guidance/risk-management phrase (concise)
- Balance sheet/QT note
- Vote: for/against; list dissenters and their rationale if stated
- Sentiment (0–20)
Export CSV.
E) INTERPRETATION & REACTION FUNCTION
- Synthesize how the Fed’s view of growth, inflation, and unemployment has shifted across the three meetings.
- Make explicit the reaction-function evolution (e.g., “greater weight on labor slack / risk-management tilt / inflation persistence still cited”).
- Quote the 3–5 most decision-salient lines verbatim with citations (page/paragraph).
F) MARKET PRICING VS FED SIGNALS
- Pull current market-implied path:
- Table of each remaining meeting this year and next: probability distribution over hike/hold/cut; implied target range (midpoint) at year-end and end-of-next-year.
- Compare to your interpretation of the Fed stance. Conclude: “Market pricing is broadly appropriate / too dovish / too hawkish,” with the key conditional drivers (upcoming data that could swing pricing).
G) SCENARIO PATHS (BASE/UPSIDE/DOWNSIDE)
- For each scenario provide:
- Qualitative trigger set (inflation prints, labor trend, financial conditions).
- Policy path (per-meeting bps changes, year-end and next-year-end levels).
- Directional macro implications (growth/inflation/unemployment).
- Confidence level and top “change-my-mind” indicators.
H) CROSS-ASSET REACTION (OPTIONAL, IF DATA AVAILABLE)
- 1-hour and 24-hour post-statement moves: 2y/10y UST yields, FF/OIS, S&P 500, USD (DXY), gold, IG/HY spreads. One-line interpretation.
I) PRESENTATION SPEC
- Start with a 6–8 bullet Executive Summary (no acronyms; bold the decision, the key new phrase, the sentiment score, the bottom-line on market pricing).
- Use clean Markdown headings, tight bullets, and one mini chart (sentiment trend).
- Provide download links for: redlines (HTML), summary CSV, and the chart image.
- Keep tone neutral, avoid recommendations.
J) VERIFICATION CHECKLIST
- Confirm meeting dates and the stated target range changes.
- Confirm dissenter names and stated direction.
- Confirm any “has moved up / remains elevated / moderated / slowed / edged up” phrases before quoting.
OUTPUT STRUCTURE
1) Executive Summary (6–8 bullets)
2) Redline Highlights (per pair) + links to HTML diffs
3) Sentiment Scores (0–20) + method + mini line chart
4) Summary Table (shown inline + CSV link)
5) Interpretation of the Outlook & Reaction Function (with 3–5 verbatim quotes + citations)
6) Market Pricing vs Fed Signals (meeting-by-meeting probability table; year-end & next-year-end implied rates; conclusion on appropriateness)
7) Scenario Paths (base/upside/downside) with triggers and watchlist
8) (Optional) Cross-Asset Reaction
9) Appendix: Sources and methodology notes