Every sentiment indicator answers the question of how participants feel, and the useful ones answer a different question: what have participants actually done. The gap between those two is the whole subject, and it explains why the indicators people quote are the ones with the least information in them.
The distinction that organises everything here is whether a measurement costs its subject anything. A social post expressing confidence costs nothing to produce, can be produced by somebody with no position at all, and can be produced by somebody with the opposite position who benefits from others believing it. It is a statement, and statements are cheap.
A leveraged position is different in kind. Somebody who is long has committed capital, is paying to hold it, and will lose if they are wrong. Whatever they say publicly, the position is a costly signal, and costly signals carry information that free ones do not. This is not a subtle point and it is the one most sentiment analysis ignores.
So the hierarchy is straightforward. Measurements built from money at risk are worth reading. Measurements built from what people said are worth reading only as a description of what people said, which is occasionally interesting and is not what they are presented as.
Ask what a measurement cost the person being measured. If the answer is nothing, it is a description of talk rather than of the market.
Several are available publicly, and they share the property of being expensive to fake because faking them requires taking real positions.
It shows which side is paying to hold its position and by how much, which is a direct measure of where leveraged demand is concentrated. It is not a forecast of direction and it is a precise reading of positioning, backed by payments actually changing hands. Extreme readings tell you that a large amount of leverage is on one side, which is a statement about fragility rather than about where the price is going.
The total value of outstanding derivative contracts, which tells you how much leveraged capital is committed. Rising open interest with rising price means new positions are being opened; rising price with falling open interest means positions are being closed. The two situations look identical on a chart and describe opposite things.
How far the dated contract trades above or below the current price is a measure of what participants will pay for leveraged exposure over a defined period. It is the cleanest available reading of medium-term positioning appetite, and it is the one input into the sentiment picture that comes with a date attached.
Popular sentiment indices combine social media volume, search interest, surveys and occasionally price-derived inputs into a single number, usually on a scale with labels at each end. They are widely quoted and their construction is worth examining because it explains their behaviour.
The most common problem is circularity. Many of these indices include volatility or recent price change as an input, which means the index moves because the price moved, and the index is then cited as explaining the price movement. An indicator that contains its own subject as a component tells you what already happened in a different font.
The second problem is that the social component measures posting rather than opinion, and posting volume is dominated by whatever is loudest rather than by what most participants think. Automated accounts, coordinated promotion and the general asymmetry between people who post about markets and people who trade them all distort it in ways that have no fixed direction.
The observation that extreme sentiment readings often precede reversals has a real basis and a much narrower one than the way it is used. Where the reading is built from positioning, an extreme means a large amount of leverage sits on one side, which mechanically means a move against that side triggers liquidations that amplify it. That is a statement about how a move would propagate, not about whether one will happen.
Where the reading is built from talk, the same observation is mostly a description of the past. Sentiment surveys are most bearish after prices have fallen, which they always are, and citing that as a contrarian signal is citing the fall itself with extra steps.
The honest version of the contrarian argument survives only in the positioning form: when leverage is heavily one-sided, the market is fragile in that direction, and the size of a move in the opposite one is likely to be larger than the news justifying it. That is useful for sizing and useless for timing, which is the distinction most commentary collapses.
Extreme positioning tells you how a move would propagate. It does not tell you when one will start, and no reading of it ever has.
A single number describing the market conceals the most useful distinction available, which is that different kinds of participant can be measured separately. Retail and professional flow concentrate on different venues and in different instruments, and each leaves its own trace: heavily leveraged perpetual contracts on venues open to everybody read differently from dated futures on venues that require institutional onboarding.
The two frequently diverge, and the divergence carries more information than either reading alone. A market where leveraged retail positioning is heavily one-sided while dated futures show the opposite is a market with a disagreement in it, and that disagreement is a description of who would be forced to act first if the price moved against them. It is also the configuration that precedes the sharpest moves, for exactly that mechanical reason.
Reading it requires no subscription and does require choosing your instruments deliberately. Compare the funding rate on a widely accessible perpetual against the annualised premium on a dated contract from a venue with higher barriers, over the same period. When those two point the same way, positioning is uniform and the market is less fragile. When they oppose each other, one group is going to be wrong expensively, and the mechanism will amplify whichever way it resolves.
Any indicator that fires often enough to be interesting will precede a reversal sometimes, because reversals happen. Establishing that it has predictive value requires comparing how often it fired before a reversal against how often it fired at all, and that comparison is almost never presented alongside the claim.
This is the mechanism behind most sentiment analysis that appears to work. Somebody notices that an indicator was extreme before three large turns, which is a real observation about three occasions, and does not count the forty other occasions it was equally extreme and nothing happened. The three are memorable and the forty are not.
The test that settles it is short and rarely run: take every occasion the indicator crossed the threshold, in order, without selecting, and record what followed each one. Most indicators do not survive that, and the ones that do survive it weakly. Running it yourself on the indicator you actually use is more informative than any amount of reading about it.
Two uses hold up, and neither is prediction. The first is sizing: when positioning is heavily concentrated, moves in the opposite direction are amplified by forced closures, so the same position carries more risk than usual. That adjustment requires no view on direction and follows directly from the mechanism.
The second is as a description of who else is in the trade. If you have reached a conclusion and every measure of positioning shows that most leveraged participants have reached the same one, you are not early, and the move you are anticipating has to be paid for by somebody who has not yet acted. That is a reason to expect a smaller move rather than a reason to expect the opposite one, and it is a different adjustment from reversing your view.
What sentiment cannot do is tell you when. Positioning can stay extreme for months, and the observation that it is extreme has been made continuously throughout every period in which it stayed extreme. Anybody using it as a timing signal is using a measurement that has no time dimension in it.
A useful picture needs three numbers and no subscription. The current funding rate on the instrument you trade, compared with its own recent range rather than against zero, since what matters is whether it is unusual for this market. Open interest, and whether it rose or fell alongside the last significant price move. And the spread between spot and the nearest dated contract, annualised so it can be compared over time.
Read together, those three describe the positioning honestly: how much leverage is committed, which side is paying for it, and what participants will pay for exposure over a defined horizon. That is the whole of what a sentiment reading can legitimately contain, and it takes ten minutes to assemble from public data.
What it will not give you is a single number with a label, which is precisely why the published indices exist and why they are quoted more than the underlying data. A number with a word attached is easier to repeat than three figures that require interpretation, and the ease of repetition is the reason for its popularity rather than evidence of its usefulness.
The ones built from social posts and search interest measure talk, which costs nothing to produce and can be produced by anybody including people with the opposite position. Many also include recent price change as an input, which makes them partly a restatement of the price. Positioning data is a different and better category.
Anything built from money at risk, because it is expensive to fake. The funding rate on perpetual contracts shows which side is paying to hold its position, open interest shows how much leveraged capital is committed, and the spot to futures spread shows what participants will pay for exposure over a defined period.
Extreme positioning tells you a large amount of leverage sits on one side, which means a move against that side would be amplified by liquidations. That is a statement about how a move would propagate rather than about whether or when one will start, and the distinction is the one most commentary drops.
Because it is backed by payments actually changing hands between participants who have committed capital and will lose if they are wrong. A social post expressing the same opinion costs nothing and carries no such commitment. Costly signals contain information that free ones do not.
That new leveraged positions are being opened. Rising price with rising open interest means new money is entering; rising price with falling open interest means positions are being closed, which is a very different situation. The two look identical on a price chart alone.
Take every occasion it crossed its threshold, in order, without selecting, and record what followed each one. Most indicators that appear predictive are being judged on the memorable occasions rather than all of them. That test is short and almost nobody runs it.
No. Positioning can stay extreme for months, and it has been observed as extreme continuously throughout every such period. The measurement has no time dimension, so using it as a timing signal is asking a question the data cannot answer.
Adjust size rather than direction. When leverage is concentrated on one side, moves in the opposite direction are amplified by forced closures, so the same position carries more risk than usual. That adjustment requires no view on where the price is going.