Methodology · Hawkish/Dovish Scoring
Each FOMC statement is fetched from the Federal Reserve monetary policy RSS feed and passed through Claude Haiku 4.5 with a structured prompt. The model returns:
hawkish_score (-100 to +100): the canonical scalar. -100 = aggressive easing language, 0 = neutral hold, +100 = aggressive tightening. This integrates the full statement — current action, forward guidance, balance sheet stance, and qualifying language ("highly attentive," "data-dependent," "for some time").
Forward guidance classification: HAWKISH / NEUTRAL / DOVISH / DATA_DEPENDENT. Concern dimensions: inflation, growth, labor — each scored HIGH/MEDIUM/LOW. Balance sheet: TIGHTENING / STABLE / EXPANDING / NOT_DISCUSSED.
Delta-hawkish is the real alpha. Comparing tone to the prior meeting catches subtle pivots before the consensus rate path moves. Thresholds: ≥+15 HAWKISH_PIVOT · +5..+15 HAWKISH_DRIFT · -5..+5 STABLE · -15..-5 DOVISH_DRIFT · ≤-15 DOVISH_PIVOT.
Regime classification blends absolute score and recent shift:
HAWKISH_STANCE· score ≥+50 — restrictive; equities priced for tight moneyMILD_HAWKISH/TRANSITION_DOVISH· score 20-50 with vs without dovish driftNEUTRAL· score -20 to +20MILD_DOVISH/TRANSITION_HAWKISH· score -50 to -20DOVISH_STANCE· score ≤-50 — aggressive easing; risk-supportive
Data: Federal Reserve press release RSS + statement HTML scrape · refresh every 6h · cached per-statement in S3 so subsequent runs are nearly instant.
ECB, BoE, BoJ coverage coming in v1.1.0.