Volume-price analysis

Price tells you where the market went. Volume tells you how hard it had to work to get there. Comparing the two is one of the oldest ideas in technical analysis — and one of the few where we can now show you exactly which parts survive measurement and which parts do not.

Effort versus result

Effort is volume: how many contracts or shares changed hands. Result is how far price moved while that was happening. The entire method is reading the ratio between them.

• Heavy volume, big move. Effort matched by result. Unremarkable — the move cost what it should.
• Heavy volume, small move. Effort without result. A lot of trading happened and price barely budged, which means someone large was taking the other side of everything. The classic reading: absorption. Somebody wants that price and is willing to soak up supply to keep it.
• Light volume, big move. Result without effort. Price slid a long way because nobody was there — thin books, few participants. Easy to reverse, because whatever moved it was not size.
• Light volume on a pullback. The one people misread most often. A pullback that gives back very little on very little volume is not a lack of buying. It is a lack of selling. Nobody is trying to get out.

That is the whole conceptual apparatus. Everything below is a way of making it precise enough to count.

What Wyckoff and Weis were actually claiming

Richard Wyckoff was a stock operator writing in the 1900s–1930s. His central claim was about who, not what: markets are moved by large, well-informed participants who have to accumulate and distribute positions gradually, because doing it all at once would move the price against them. That gradual process leaves a footprint in the relationship between volume and price, and a careful reader can see it.

Note that this is a mechanical argument, not a mystical one, and it does not require anyone to be clairvoyant. A fund that needs to buy a large position cannot simply take every offer — it would pay far more than it wanted to. So it buys into weakness, absorbs supply, waits. That behaviour is different from what a crowd of small participants does, and the difference shows up in the data. Whether it shows up strongly enough to be worth anything is an empirical question, and that is what this page is about.

David Weis spent decades developing a specific presentation of these ideas. His insight was about visualisation: bar-by-bar volume is close to unreadable, because each bar's volume depends on time of day, news, expiries and noise. Group the bars into waves instead, sum the volume within each wave, and the comparison becomes obvious to the eye.

Both men were explicit that this was interpretive, not mechanical — a way of forming a view, not a rule that fires. That framing turns out to be well supported by what we measured. Quant Data is independent research and is not affiliated with or endorsed by David Weis or any Wyckoff organisation.

What a wave is

A wave is a run in one direction that continues until price reverses by more than a threshold. Reverse by less and you are still in the same wave; reverse by more and a new wave begins in the other direction. Then you total the volume that traded inside each wave.

A schematic. Wave 1 travels a long way on heavy volume. Wave 2 gives back a fraction of it on very light volume — few people selling into the pullback. Wave 3 pushes to a new high on heavy volume again. This is the shape the tradition reads as buyers in control.

The threshold is the only parameter that matters, and picking it badly ruins everything. Set it too tight and every wiggle becomes a wave; too loose and a whole session is one wave. Quant Data sets the reversal threshold at 0.3937 × ATR(60) on closing prices — a coefficient calibrated on S&P 500 E-mini futures. Because it is expressed as a multiple of recent average range rather than in points, the same setting behaves consistently on a $30 stock and a 6,000-point index. Nothing here is re-tuned per instrument, which is deliberate: re-tuning per asset is how you manufacture results that do not replicate.

The five events

Once you have waves, you can define specific configurations the tradition considers meaningful. Quant Data detects five. Descriptions first — the scoreboard comes after, and the order matters.

Read those descriptions on their own terms first. Every one of them is a reasonable story about market behaviour, and each has been taught in books and courses for decades.

The honest scoreboard

Now the measurement. Pre-registered tests on S&P 500 E-mini futures, 5-minute bars, 2010–2026, using symmetric ±3×ATR barriers — the outcome is whichever barrier price hits first, so there is no bias from asymmetric targets. Corrected for the number of tests run. Required to hold with the same sign in both halves of the period.

The baseline is 49.3% — the win rate of any wave flip at all, with no event condition. That is the number every row below has to beat.

Two confirmed. Two reversed.

The bullish half of the tradition holds up. Both confirmed events are small — 51.8% and 51.4% against a 49.3% baseline is roughly a two-point edge — but they are measured on thirteen and twenty-three thousand samples, which is enough to see two points.

The bearish half does not hold up. It does something more interesting than failing: it points the other way. When no_demand_short fires — a weak rally on light volume, the textbook picture of a move with nothing behind it — the market went up 54.4% of the time across 22,262 instances. Betting against that rally lost.

Why? Because of what each kind of result costs you:

• A confirmed result tells you a thing you already believed is worth about two points. Nice. Modest.
• A reversed result tells you that a thing you believed — and would have acted on — was costing you money in a systematic, repeatable way. Removing a negative-edge habit is worth more than adding a small positive one, and it is available immediately at no cost.

And there is a coherent explanation, which matters because a reversal with no mechanism is usually just noise. Every bearish volume-price read tested below 50% on this market. Index products have long-run upward drift and a structure that punishes crowded short positions. In that environment, "the rally looks weak" is simply not a reason to be short — a weak-looking rally in a drifting-up market is what most rallies look like. The asymmetry is a property of the instrument, not a flaw in the detection.

Which also means: do not assume it holds elsewhere. The asymmetry is an explanation, not a law, and the next section is about exactly where it stops.

The strongest result: gating by day type

A volume event on its own is a weak trigger. The same event filtered by the character of the session it happens in is considerably stronger.

Take cib_long and only count the ones that fire on a day the day-type model reads as TrendUp:

Zero-shot means the day-type model had never seen a single bar of NQ or QQQ during training. It learned the structure of a trending session on one instrument and that structure transferred to two others it had never met. That is the closest thing here to evidence of a general property rather than a curve fit.

Two details are worth more than the headline. First, adding a third filtering layer on top made results worse — it shrank the sample without adding information. More conditions is not more edge. Second, look at the TrendDown column: 44–45%. The most valuable use of the gate is often not taking the long trigger when the session reads as a downtrend. Avoiding a negative-edge situation is cheaper and more reliable than finding a positive-edge one.

Where this stops working

Publishing this next part is the point of the exercise. If it were missing you would have no way to judge the rest.

The conclusion is uncomfortable and worth stating flatly: these win rates are a property of index-type assets, not a law of markets. Anyone showing you Weis statistics on crypto without having tested them there is extrapolating, and in this specific case the extrapolation is measurably backwards.

What this is actually good for

A two-point edge sounds like something you could trade. Do the arithmetic and it is not. Gross expectation on these events is roughly the same size as transaction costs — you would be trading to pay your broker.

The honest uses are narrower and they are real:

• A confirmation layer. You already have a reason to be interested in a market. The wave structure either agrees or does not, and you weight accordingly.
• A filter that tells you when to stand aside. 44% under a TrendDown read is the most actionable number in this whole page, and acting on it means doing nothing.
• A description that trains your eye. Reading effort against result on actual waves, with the measured statistics next to them, is how you build the intuition — and unlike most chart education, here you get told when your intuition is backwards.

Which is what Weis said all along. He framed the method as interpretive rather than mechanical, and 16 years of counting agrees with him.

How to practise

1. Look at waves, not bars. Pick a liquid instrument and mark the swings by hand. Write down each wave's rough volume and how far it travelled. Do it for a week before reading a single event name.
2. Call the pullbacks before you check. When price pulls back, decide whether it looks like light-volume drift or genuine selling. Then check.
3. Log the bearish setups especially. Every time you see something that looks like "no demand", write down what you expected and what happened. This is where the measured reversal will show up in your own data, and seeing it yourself is worth more than reading it here.
4. Always pair it with the day type. A volume event without session context is the ungated 51.8%. With context it is 56.5%. That gap is the single most reliable thing on this page.

Try it

The wave engine is the Weis Wave API — $30/month. It returns the wave in progress with its volume relative to the last twenty, the eight most recent completed waves, and any events that fired, each carrying its measured win rate and sample size so nothing gets quoted from folklore. Pair it with Brooks Daily Bias for the gate; both together are $50/month.

The quantdata-weis-wave Skill teaches an AI agent to read the output correctly — including refusing to call a no_demand_short bearish, which is exactly what an agent working from its training data would otherwise do.

Email for a key How to read these numbers

Next

Educational analytics. Nothing on this page is investment advice or a recommendation to buy or sell anything.