What Is Sentiment Divergence?

The idea

Sentiment and market data usually move together — communities get positive when prices rise and negative when they fall. Divergence is the case where they come apart, and the interest in it rests on a simple thought: if the two normally agree, disagreement means one of them is early.

That thought is reasonable and incomplete. Disagreement can equally mean the measurement is wrong.

The usual forms

Pattern	Observation	Common reading
Positive divergence	Sentiment strong, price weak or flat	Conversation ahead of the market
Negative divergence	Price strong, sentiment weak	Rally without conviction behind it
Extreme agreement	Both at an extreme together	Crowded positioning

The "common reading" column is how these are usually described. Each is a hypothesis about what a mismatch implies, not something the measurement establishes.

What Moonboard actually computes

This is where the general concept and the specific implementation part company, and knowing the difference prevents a real misreading.

SDE compares sentiment against a stability term derived from realised volatility, not against price direction:

stability = (1 − min(volatility, 1)) × 100 divergence = sentiment − stability SDE = percentile of divergence across the universe

Volatility has no sign. It cannot distinguish a 20 % rally from a 20 % fall. So SDE measures enthusiasm relative to turbulence, and a high SDE requires both high sentiment and high volatility together.

sentiment 85, volatility 0.09 → stability 91 → divergence −6 sentiment 85, volatility 0.40 → stability 60 → divergence +25 Same sentiment. Very different SDE — purely from volatility.

If you expected two assets with identical sentiment to rank together, that expectation describes a sentiment metric. SDE is not one, and the sentiment field is available on its own for that purpose.

Why divergence is an observation, not a signal

Divergence is often presented as actionable — fade the euphoria, buy the despair. Three reasons to hold that loosely:

Sentiment error is largest at the extremes. Very high and very low readings come disproportionately from crisis language, sarcasm and coordinated promotion — exactly the text classifiers handle worst. A large divergence is more likely than average to be partly measurement error.
The bias runs the wrong way. Classifiers tend to read defiant community language during drawdowns as positive, which manufactures "positive divergence" during falls — the pattern a contrarian reading would treat as a buying opportunity. See sentiment research.
Divergences persist. Mood and prices can disagree for a long time. Identifying a state says nothing about when it resolves, or which side moves.

What Moonboard claims

That an asset currently exhibits a measurable gap between its sentiment reading and its volatility reading, relative to other assets. Not that the gap will close, not in which direction, and not that acting on it produces any particular outcome.

The other divergence metric

WSD applies the same idea to a different pair: trading activity against social attention. It compares two percentile ranks rather than subtracting two raw values, which avoids the problem of subtracting quantities with no shared unit — SDE's structural weakness.

SDE methodologyThe full derivation
Market sentimentHow the sentiment field is produced
VolatilityWhy the second term has no direction
Sentiment researchClassification failure modes