HOSI tracks whether roughly 50 oil-market accounts are becoming more
bullish or bearish than their own normal stance. It is designed to measure capitulation, conviction,
and narrative shifts — not to predict oil prices.
Where is sentiment now?
Oil Sentiment vs. WTI
Front-month WTI and HOSI are each standardized against their own
full-period average and volatility, allowing both to share one axis. The chart shows when
sentiment moved with, ahead of, or behind oil prices.
WTI priceHOSI sentiment
Daily Sentiment Trend
The smoothed daily reading shows the direction of sentiment across the
cohort. The faint line is the unsmoothed daily result; thinner weekend activity can make
individual readings noisier.
Smoothed HOSIDaily reading
Who is driving sentiment?
Bullish vs. Bearish Breadth
The share of active accounts expressing a bullish outlook minus the share
expressing a bearish outlook. Positive readings indicate bullish consensus; negative readings
indicate bearish consensus.
Who Is Driving Sentiment?
Weekly sentiment by account. Blue = bullish, orange = bearish, neutral gray
in between; dashed = no rated posts. Click any cell to see the daily posts
and model-generated rationale behind that reading.
bearish −2+2 bullish
Major Sentiment Events
Largest Sentiment Swings
The largest changes in smoothed hourly sentiment, measured from one
turning point to the next. Fast shocks lasted three days or less; slower repricings developed
over longer periods.
Fast shocks — 3 days or less
Slow repricings — more than 3 days
Conversation Volume
Daily number of posts expressing a directional oil view, split between
bullish (up, blue) and bearish (down, orange) outlooks. Spikes indicate unusually intense
attention; volume alone does not indicate the direction of sentiment.
Sentiment Shock Score — anomaly detail, click to expand
Measures how unusual each sentiment move was across periods
ranging from 12 to 96 hours. Bars show each day's most anomalous hour; readings beyond ±3 standard
deviations (solid bars) are classified as major sentiment events. Thin-volume hours are excluded.
Does Oil Twitter Lead WTI?
Verdict: HOSI is
primarily a fast mirror of oil prices, not a reliable leading indicator. Sentiment usually moves during
or shortly after the price move, and forward correlations are generally weak.
Did Sentiment Predict the Next Move?
Correlation between hourly sentiment and cumulative WTI returns over the
following several hours and days. Results near zero indicate little evidence that sentiment
predicted the subsequent price move.
Across most periods and time horizons, forward correlations remain
within the statistical noise band. During the war, extremely bullish hourly readings were followed by
somewhat weaker 24-hour WTI returns, but the sample is too limited to treat this as a reliable
contrarian signal.
WTI and Sentiment on Their Original Scales — raw panels, click to expand
The same timeline shown separately in dollars per barrel and
HOSI stance units. Hovering links both panels.
Average cohort stance is the equal-weight mean outlook (−100…+100).
Relative to normal adjusts for each account's own habitual bias, so a permanent bull only
moves it when becoming more or less bullish than usual.
Average cohort stanceRelative to normalhover for values
Completed Monday–Sunday readings for the full HOSI
history. The current partial week is excluded.
How HOSI Works
Cohort. HOSI follows a fixed group of roughly 50
oil-focused accounts on X/Twitter.
Scoring. Posts are evaluated for their implied crude-price
outlook on a five-level bearish-to-bullish scale. Posts without a directional oil view are excluded.
Baseline adjustment. Each account is evaluated against its
own historical behavior. Consistently bullish accounts only move the adjusted index when they become
more or less bullish than usual.
Aggregation. Individual post readings are combined into
hourly, daily, and completed weekly results. Active-account weighting prevents low-participation periods
from carrying the same confidence as high-participation periods.
Auditability. Heatmap cells can be opened to inspect the daily
posts and the model's one-sentence rationale.
Technical methodology
Each (account, week) batch of tweets is rated
per-day on a 5-level bull/bear scale by Claude Opus 4.8 with cross-account context in the same prompt;
no-signal days are excluded. Weekly readings cover completed Monday–Sunday weeks only, so the newest
weekly reading is the most recent closed week, not today. The hourly signal study labels each tweet
individually and aligns to front-month WTI; "relative to normal" is a per-account z-composite; swings
use an ε-drawdown decomposition (ε = 0.15 stance units) of the smoothed hourly series; the shock score
scales sentiment change by each window's own volatility across 12–96h; forward correlation uses
|r|·√n ≥ ~2 as a nominal significance gate before any multiple-testing discount. Data is served live
from the oil-sentiment index; every heatmap cell drills to the model's
per-day rationale. Cohort registry v1.
Hyperforage is an AI research tool, not investment advice.