Why the sentiment gauges disagree — and what each one actually measures

The disagreement is the mechanism, not a malfunction

Here is a familiar afternoon. You open a survey and it says investors are frightened. You open a composite index and it says the crowd is greedy. You look at the volatility index and it looks calm. You conclude that sentiment is noise and close the tabs.

That conclusion is understandable and wrong. Those four readings are not four attempts at the same measurement. They are four different measurements: one of stated opinion, one of market internals, one of a price, and one of recorded positions. They also run on four different clocks, from a weekly poll with a mid-week cutoff to an index that updates every tick. When they disagree, the usual explanation is that people are saying one thing and doing another — which is information, if you know which gauge is listening to the mouth and which is watching the hands.

What follows is what each one reads, mechanically. No current levels appear anywhere on this page, deliberately.

1. AAII: stated opinion, six months out, closed on Wednesday

The AAII Investor Sentiment Survey asks its members one question: whether they feel the direction of the stock market over the next six months will be up (bullish), no change (neutral), or down (bearish). Three buckets. One horizon. That is the whole instrument.

Three properties fall straight out of that design, and they explain most of the confusion around it.

• It is opinion, not position. Nobody is asked what they own. A member can vote bearish and be fully invested, and plenty are.
• The horizon is six months. A reading is not a statement about tomorrow. Treating a six-month opinion poll as a signal for this week's tape is a category error before any question of predictive value arises.
• The clock is weekly and lagged. AAII runs the survey week from Thursday to Wednesday night and publishes on Thursday. So a reading you read on Thursday partly describes how people felt last Friday. If the market moved hard on Thursday morning, the survey has not heard about it yet.

It is also a self-selected poll of one association's members who choose to respond — not a random sample of market participants, and not weighted by anyone's capital. It measures the mood of a particular room. That is worth something. It is not worth what people ask of it.

2. CNN Fear & Greed: seven market internals, not a survey

The Fear & Greed Index is frequently described as a sentiment survey. It is not. Nobody is asked anything. It is a composite of seven measurements of market behaviour, each given equal weight, scaled onto 0–100. CNN lists the seven as market momentum, stock price strength, stock price breadth, put and call options, junk bond demand, market volatility and safe haven demand — the S&P 500 against a long moving average, new highs versus new lows, advancing versus declining volume, the recent trend in the put/call ratio, the yield spread on high-yield debt, market volatility, and the recent return difference between bonds and stocks.

Read that list again and notice what it is. Every input is a price or a volume. The index infers emotion from what markets did; it never asks anyone how they feel. That makes it a coincident description of conditions, and a fast-moving one — several inputs are short-window, so the composite can swing meaningfully in days.

3. The VIX: a priced expectation of width, with no direction in it

The VIX is the most misread of the four, and its own documentation is unusually clear. Cboe's methodology states that the index "is designed to measure the market's expectation of 30-day forward looking volatility of the U.S. equity market, as conveyed by S&P 500 Index option prices."

Three things follow.

It is a price, not a poll. The calculation takes the market prices of SPX and SPXW options as inputs, together with Treasury yield curve rates, and aggregates a strip of out-of-the-money strikes into a variance for each of two expirations, which are then interpolated to a constant 30-day term. Only options with a non-zero bid are included. No option pricing model is used to back out an implied volatility. What you are reading is what someone had to pay.

It has no sign. Both puts and calls enter the calculation. The index is a width, expressed as an annualised percentage — the final step is literally a standard deviation multiplied by 100, with time measured in 365-day years. A wide expected range is consistent with a violent recovery as much as a violent decline. The VIX is called the fear gauge, and the label smuggles in a direction that the arithmetic does not contain.

Its horizon is 30 days. Not today, not this quarter. A calm reading says options are priced for a quiet month; it says nothing about the next hour.

What about "20"?

Twenty appears nowhere in the methodology. It is a convention — a level that sat near the middle of the historical distribution often enough that it acquired the feel of a threshold. Distributions drift. Conventions do not update themselves. We tested the convention directly, and the result is in the next section.

4. Positions and flow: what was actually held

The fourth family does not ask and does not infer. It records. Cboe publishes daily put/call ratios — total, index, equity, and others — measuring contracts traded, puts divided by calls, over a period. That is transacted activity, though it comes with a known ambiguity: a put can be a hedge on a long book or an outright bearish bet, and the tape does not say which.

Further along the same axis, the CFTC's Commitments of Traders report breaks down each Tuesday's open interest in futures and options markets by trader category. It is the cleanest positioning data in the list and the slowest: the data is from Tuesday and is generally released the following Friday at 3:30 p.m. Eastern. By the time you read it, it is three days old by construction.

This is the standing gap the whole subject turns on. A survey records what investors say. Flow and positioning record what they did. Those two things routinely diverge, and neither is lying — people are quite capable of describing themselves as bearish while holding a full book, because opinion is cheap and repositioning is not.

What we found when we used a sentiment level as a switch

We ran the obvious experiment: take the most popular sentiment threshold and use it as permission. Block all new entries while the VIX sits above 20, and see what it costs.

In a research replay on real historical rules over 2016–2026, against a benchmark version of the same strategy with no gate at all, blocking new entries above VIX 20 lowered compound annual return from 12.0% to 10.1%, lowered the Sharpe ratio from 0.99 to 0.92, and made the worst drawdown deeper — −22.6% against the ungated −20.1%. It paid for nothing. On a separate panel of leader setups from 2019 to 2026, entries taken with the VIX between 20 and 30 went on to beat entries taken below 20 over the following month, by a small margin that survived a conservative test on non-overlapping windows. Above 35 the sample thins out badly and the effect vanishes into the noise; we do not claim anything there.

One variant did buy something. Force-selling the book when the VIX crossed 20 improved the worst drawdown to −18.3%, and cost 2.2 points of annual return to do it. That is a real trade-off rather than a free lunch, and it is exactly the shape of result that argues for using sentiment to size rather than to veto.

Where sentiment actually belongs

Sentiment is real information. It describes how crowded a trade is, how much protection has been bought, and how far opinion has travelled from positioning. What it does not do is tell you when. Crowds get frightened well before bottoms and well after them, and stay complacent through quiet years and the beginnings of bad ones alike.

The failure mode is treating a sentiment reading as permission — as a gate that opens and closes. We tested that directly and it was worse than not doing it, which is consistent with the broader finding that market weakness is not a veto. The same trap catches people who try to classify how bad things are first; the vocabulary question of correction versus bear market is answered in hindsight and does not help you in the moment either.

Where these readings earn their place is in risk sizing. A high priced expectation of volatility is a statement that the range of outcomes over the next month is wide. That is a good reason to hold a smaller position, to set a wider stop so ordinary noise does not remove you, and to be slower to add leverage. It is a poor reason to refuse to participate. One adjusts your exposure to conditions; the other hands a single number a veto over your whole process.

To disclose the bias: this is how our own engine treats it. Coil's scanner lets volatility affect how large a position is and how wide its stop sits, and never lets it decide whether a qualifying setup is allowed to exist — a design choice the replay above is the reason for, not a slogan.

Read the gauges as four instruments answering four questions on four clocks, and the contradictions stop being frustrating. Then put the answer where it works: in how much, not in whether.

FAQ

What does the AAII sentiment survey actually ask?

It asks members whether they feel the direction of the stock market over the next six months will be up (bullish), no change (neutral), or down (bearish). It is a self-selected weekly poll of opinion about a six-month horizon, collected over a survey week that closes Wednesday night and published Thursday. It does not record what anyone owns.

What does a high VIX mean?

That S&P 500 index options are priced for a wider range of outcomes over the next 30 days. It is stated as an annualised percentage and carries no direction — the same reading is consistent with a violent recovery or a violent decline. It is a width, not an arrow.

Is 20 a meaningful threshold for the VIX?

Twenty appears nowhere in Cboe's methodology. It is a convention that grew from the level sitting near the middle of the historical distribution. In our research replay, using it as a permission switch to block new entries produced lower return, lower Sharpe and a deeper drawdown than not gating at all.

Is everyone bearish a buy signal?

Not on its own, and not on a schedule. Extremes in stated sentiment have historically clustered near turning points, but they also persist for long stretches without one, and the surveys that record them are lagged by design. Crowd positioning is a reason to think about how much risk you are carrying. It is not a reason to time an entry, and no gauge on this page has ever told anyone what happens next.