Do liquidity walls predict price? We measured 2,146 walls

Open any guide to reading an order book and you will find some version of this claim: a large resting sell order is resistance, a large resting buy order is support. Size on the ask pushes price down; size on the bid holds it up.

We are in an unusual position to check that. Our detectors watch the order books of 16 exchanges continuously, and a separate scoring job records what the price did afterwards — at 15 minutes, an hour, four hours, a day. So rather than argue about it, we measured it.

The answer is not the one the guides imply.

The short answer

Density on its own does not predict direction. It shows where activity, volume or liquidity concentrates — which helps locate zones where the market is more likely to react at all. What happens inside that zone is decided by price behaviour, not by the presence of density.

Our measurements bear that out. Across 2,146 large walls, the side a wall sat on made almost no difference to what price did next: walls on the ask were followed by a higher price an hour later 46.6% of the time, walls on the bid 47.3%. A gap of 0.7 percentage points — far smaller than the difference between two coins picked at random.

A wall shows where visible liquidity is sitting. It does not tell you which way price is about to go.

Four different things called "density"

Worth separating these before the data, because they get conflated constantly and the confusion starts there.

Price density. How long price spent inside a given range. An area where price lingered is a zone of agreement about value. A historical property, unrelated to the current book.

Volume density. How much volume traded at each price level. Shows where trading actually happened — a fact about completed trades.

Liquidity density (order-book density). How visible liquidity currently placed in the book is distributed across price levels. This is about intent, not execution: an order can be cancelled and never fill.

A large wall. A special case of the last one — a single unusually large order at one level. This is what our alerts detect, and what we measured.

The first two describe what already happened. The last two describe what is standing in the book now. This study covers only the fourth, and its conclusions should not be transplanted onto the others.

What we measured

Every wall in this study is a real detection from our production system, not a backtest over historical snapshots.

Two details matter for interpreting this.

Entry price is a traded price, not a quote. We take the first actual trade at or after the moment the wall was detected. Using the wall's own quoted price would flatter the result by measuring against a level that may never have traded.

Only large walls. The smallest wall that reached the scorer was $200,007. This is a study of significant resting liquidity, not of every small order. Whether the same holds for a $5,000 wall on a thin altcoin is a question this data cannot answer.

The result

If sell walls suppressed price and buy walls supported it, these two rows should look clearly different. They do not. At one hour the gap is under a point, and the direction of the small difference is not even consistent across the two horizons.

Both sides sit slightly below 50%, which is a separate and unsurprising fact: over a three-day window in this sample, prices drifted down marginally more often than up. That drift applies to both rows equally, which is exactly why comparing the two rows to each other is the meaningful test, not comparing either to 50%.

The mistake we nearly made

The first cut of this analysis looked much more exciting.

Broken down by exchange, the numbers ranged from 41.7% to 67.7%. That looks like a real finding — as though walls behave differently on Kraken than on Gate, which would be a genuinely interesting result about venue microstructure.

Then we looked at what those numbers were made of. The OKX figure came from four coins. The Kraken Futures figure came from eleven. Over a three-day window, a couple of trending coins dominate a venue's entire sample, and the "venue effect" is really those coins' price action wearing a venue label.

So we ran the test that removes the confound: compare ASK against BID within the same coin. If the wall side matters, the gap should be consistent in sign and meaningful in size across coins.

The gaps swing in both directions and do not settle on a sign. On MU the sell walls were followed by more upside; on XMR and SKHYNIX the buy walls were. That pattern — similar magnitudes, inconsistent direction — is what noise looks like.

Meanwhile the coin explains a great deal. Per-coin outcomes in the full sample range from about 24% to 77%. Which asset you are looking at swamps which side the wall is on.

Which coin you are in influences the next hour far more than which side the wall sits on.

Why this does not mean walls are useless

It would be easy to read the above as "the order book is noise." That is the wrong conclusion, and the distinction matters.

A wall is a fact about the present: someone has committed real capital at a specific price, and you can see it before price gets there. What the data says is that this fact does not resolve into a reliable directional forecast an hour out. Those are different claims.

Three things a wall genuinely tells you:

Where the fight is. A large resting order may indicate a price level with a notable concentration of visible liquidity. That is useful for placing stops and targets, whichever way you think price is going.

What happens when it is tested. A wall absorbed by aggressive trades means something different from one cancelled a second before price arrives. The behaviour is the information, not the existence of the order — which is why we stamp the moment a wall leaves the book, not just its appearance. Which of the two occurred requires confirmation from price behaviour. More on that in what liquidity walls are.

Where liquidity is available. If you need to fill size, resting depth helps estimate what it will cost. The estimate is approximate: some of that depth may disappear before you reach it.

What a wall does not do is tell you to buy or sell. We publish walls because they are real and visible, not because they are predictions — and this study is the honest version of that claim, with numbers attached.

Why a wall alert doesn't contradict density analysis

A fair question: if a wall doesn't predict direction, why alert on it at all?

Because a wall alert is a detection notice, not a recommendation. It says one thing: unusually large visible liquidity has appeared at this level. It is a special case of detecting concentrated liquidity — the most visible kind, and the fastest to verify.

The value is in timing, not forecasting. A large resting order is visible before price reaches it. Spotting one by hand across ~7,400 instruments is not possible, and knowing about a potential reaction zone in advance is useful whichever way you happen to be leaning.

What follows is the trader's work. A wall marks a potential reaction zone, and what actually happens there is shown by price behaviour:

All four begin identically, with the same large wall. Only the price reaction separates them, and watching that is something you do yourself.

What the system records, and what it doesn't — plainly. We stamp the moment a wall leaves the book, so the map shows only standing walls. But the published alert does not say whether the wall was absorbed by trades or cancelled — those are different events, and our signal cannot distinguish them for you. Internal machinery uses that distinction to filter market makers, but it is not surfaced in the alert. Liquidity being repositioned to a new level is a separate signal (pushing), not a property of a wall alert.

A visible wall is not executed volume

Worth stating separately, because it is easy to be fooled here.

The order book shows intent, not trades. A $500,000 order in the book does not mean anyone bought or sold $500,000 — it means someone has declared willingness. A limit order can be pulled at any moment, instantly and at no cost.

So "wall size" and "volume that actually traded through the level" are quantities that must not be conflated. The first is visible in advance and can vanish; the second is only visible afterwards and is irreversible.

The practical consequence: large visible liquidity is a reason to look at a level, not proof that money is genuinely standing there ready to fill.

Limits of this study

Stating these plainly, because content that hides its limits is not worth citing.

Three days is a short window. Market regime dominates short samples. A different three days could produce different absolute numbers. What is more robust is the comparison between ASK and BID, since both sides experienced the same regime.

The sample is concentrated. The twelve most-detected coins account for about 35% of all observations, and several are tokenised equities rather than crypto natives. A sample dominated by a dozen assets is not a clean cross-section of the market.

Large walls only. Nothing here speaks to small walls on thin books, where the dynamics may be entirely different.

We measured direction, not magnitude. "Did price end higher" ignores how far it travelled and what happened in between. A wall that was defended for fifty minutes and then broke registers the same as one that never mattered.

We would rather publish this with the caveats visible than round it into a cleaner story.

FAQ

Does this mean order book walls are fake? No. The walls are real resting orders, and most of them are genuine. The finding is narrower: the side a wall sits on does not predict which way price moves next. Existence and prediction are different things.

Should I ignore walls when trading? Not necessarily — but treat a wall as information about market structure rather than a directional signal. Where liquidity sits is useful for stop placement and execution cost. Whether price will go up or down is a separate question this data says walls do not answer.

Why are both sides slightly below 50%? Because prices in this sample drifted down marginally more often than up over the three-day window. That baseline applies to both rows equally, which is why the meaningful comparison is ASK against BID, not either against 50%.

How is this different from studies showing order book imbalance predicts price? Most of that research measures imbalance over very short horizons — seconds to a minute — using full order book depth on a single venue. We measured a different thing: individual large walls, across 16 venues, at 15-minute and 1-hour horizons. Both can be true. Short-horizon microstructure effects do not necessarily survive to an hour.

Can I see the underlying data? The walls themselves are published live in our terminal as they are detected, with exact sizes and prices. The outcome scoring is internal, but every alert it scores appeared publicly first.

---

Conclusion

Density and wall alerts are not standalone forecasts. They are contextual analysis tools: they help you notice important zones and anomalies quickly, across thousands of instruments where doing so by hand is impossible.

The decision stays yours. Density shows you where to look; direction is suggested by what price then does — whether the wall held, was absorbed by trades, simply left the book, or liquidity reappeared at another level. An order book is a map of commitments, not a forecast: it shows where money is currently standing, which is genuinely more than a chart gives you. It does not tell you what happens next, and anyone selling you a wall as a buy signal is selling you something the data does not support.

Don't watch 10,000 order books. Let the system tell you which order book deserves your attention. What you conclude when you get there is your call.