The Crowd Shows Up a Week Before Earnings. It Still Does Not Know the Trade.

Social media begins moving before earnings. In aggregate, it still does not hand you the trade.

That is the useful answer.

We aligned social attention around 2,898 US earnings prints from January through June 2026. The crowd begins leaning in roughly seven days before the report. Attention accelerates into the event, peaks on print day, then fades over the following week.

But loud does not mean bullish, bearish, or even important. Some of the loudest conversations barely moved the stock. Some of the largest moves happened in near silence.

Reproduce the public dataset anchors

Download the CSV from LunarCrush Research and run:

import pandas as pd

df = pd.read_csv("lc_earnings_social_events_2026H1.csv")

assert len(df) == 2_898
assert df.shape[1] == 12
assert df["sector"].nunique() == 13

liquid = df[df["baseline_daily_interactions"] >= 500]
share_preprint_hot = (liquid["pre_print_social_z"] > 1).mean()

print(len(liquid))                 # 2,193
print(round(share_preprint_hot, 3))  # 0.284

The 1,665-event curve used in the interactive Atlas applies an internal dollar-volume liquidity cut that is not included in the public file. The public-file anchors above are designed to be independently reproducible.

The crowd leans in a week early

Across 1,665 liquid-name prints, average social attention reached +1.07σ one day before the print and peaked at +1.79σ on print day.

The rise does not prove that the crowd knows the result. It proves that the information environment around a company begins changing before the event.

Loud is not the same as right

The loudest print in the study was Mondelez at +7.7σ. The stock moved only +1.2%. Palo Alto Networks reached +5.4σ and fell 8.7%. UnitedHealth reached +5.2σ and fell 16.4%. Hewlett Packard Enterprise reached +4.2σ and rose 30.4%.

Social reaction measures the size of the conversation, not the direction or magnitude of the stock move.

The blind spots are the better research queue

Quiet shocks are large earnings moves that social attention barely registered. Attention traps are huge conversation spikes attached to small price moves.

Both are useful because they move the question beyond "what is trending?" The better question is "where does attention disagree with the event?"

Read this before quoting the result

At daily granularity, top-decile names moved roughly 1.2 to 1.4 times more than bottom-decile names, depending on the liquidity and season cut. That is a real relationship, but it is not a standalone trading strategy.

Daily bars hide sequencing. They cannot tell you who moved the conversation first, what they said, or whether attention led price inside the day. Treat the published result as a floor. Hourly, point-in-time data is required to test lead-lag structure without look-ahead bias.

Take the data

The complete 2,898-event CSV, Parquet file, and AI companion guide are available from The Earnings Social Atlas, with no registration required.

The GitHub companion includes the data, validation script, and all article graphics: JoeVezzani/earnings-social-atlas.

Institutional teams can request the same study cut to a custom universe with hourly resolution, more than five years of history, and point-in-time membership at [e-mail].

This analysis is for research purposes and does not constitute investment advice.