The crowd leans in a week early

Average social attention (z-score vs each name's own 60-day baseline) across all 1,665 liquid-name prints. The build starts ~7 days out, doubles the day before, peaks at +1.8σ on the print, and takes a week to decay. This is daily data; the intraday lead-lag lives in our hourly feed.

02 — THE LEAGUE TABLE

The loudest prints of the season

Mega-caps ranked by social reaction. Note what tops it: not the biggest moves, the biggest conversations. UNH's collapse hit 5.2σ. HPE moved +30% at 4.2σ.

#	Ticker	Company	Print date	Social reaction	2-day move	Crowd sentiment

03 — THE BLIND SPOTS

Where the crowd wasn't looking

Both edges are screens. Left: double-digit moves the crowd barely reacted to (under-covered names your competitors saw late). Right: multi-sigma conversation storms that priced almost nothing in.

😴 Quiet shocks · big move, no crowd

Ticker	Date	Move	Social

📣 All hype · big crowd, no move

Ticker	Date	Move	Social

04 — SECTOR PROFILES

Who gets talked about on print day

Median print-window social reaction by sector. Materials, tech and staples run hot; utilities report in silence.

05 — WHERE THE SIGNAL ACTUALLY LIVES

The daily number, straight up — and why it is the floor

We tested the obvious hypothesis ourselves: does pre-print social momentum predict print-day move size? At daily granularity, top-decile names moved ~1.2–1.4× more than bottom-decile. We publish that number because you would find it yourself in an hour with the file below — and because it is the signal measured the crudest possible way: daily bars, no timing, no sequencing, no constituent history. That is the floor, not the ceiling.

29.7% of liquid prints build more than 1σ of attention before the number is public. The question that decides whether that is tradeable is sequencing: who moved the conversation first, at what hour, and whether attention led or chased price inside the day. Daily bars cannot answer it by construction. The hourly feed with point-in-time constituent history exists for exactly that question. Teams that license it are not betting that attention matters — they are measuring when it arrives, hours before the tape shows it.

06 — THE DATASET

Don't take our word for it

Every event in this study: symbol, report date, pre/post social z-scores, crowd sentiment, print move, 5-day drift, EPS surprise, sector. No registration.

⬇ Parquet (2,898 events) ⬇ CSV ⬇ AI guide (.md for your LLM)

The .md file is a companion brief for AI assistants: schema, methodology, validation anchors, and starter analyses. Drop it into Claude or ChatGPT next to the data and your model starts from the right place.

Methodology: attention = log1p(daily interactions); z vs each name's own t−67…t−8 baseline (min 40 obs). Pre-window t−7…t−1, post-window t…t+2. Moves = close-to-close straddling the print (2-day, timing-agnostic), |move|>60% excluded. Earnings dates + surprises: Nasdaq calendar. Universe: 3,048 US equities continuously tracked by LunarCrush.

Talk to Joe

Two-week institutional trials include the full hourly feed, 5+ years of history, point-in-time constituent membership, and this exact panel cut to your tickers. Most teams have it in their backtester the same week.

Joe Vezzani · Co-founder & CEO, LunarCrush · joe@lunarcrush.com

Email Joe