Order book depth: reading a level-2 book

What is the book actually showing you?

An order book is the set of unexecuted limit orders currently resting at a venue, sorted by price. Bids sit below, asks above, and the two best prices define the touch: the best bid, the best ask, the mid halfway between them, and the spread between them. Everything else on the screen is aggregation.

That aggregation has levels, and the vocabulary is worth being precise about. Level 1 is the touch alone — best bid and ask, with their sizes. Level 2 is the depth ladder: total resting size at each price level, usually to some depth limit. Level 3 is the order-by-order view, where each individual order and its queue position is visible; most crypto venues do not publish it, and most public “L3” tooling is reconstructed rather than received.

A live book is maintained, not received. The venue sends a snapshot and then a stream of incremental updates, each carrying a sequence number. Your book is only correct if you have applied every update in order; drop one and your local copy drifts silently, showing size that is no longer there. Any serious consumer of book data checks sequence continuity and re-snapshots on a gap. If you are reading a chart of somebody else’s book, it is worth knowing whether they do.

Hinweis Depth-limited feeds are a common source of confusion. A “20 levels” feed does not show the market’s liquidity; it shows the first 20 price levels, which in a tight book may be a fraction of a percent of the price, and in a stressed one may span a great deal more.

What does the spread tell you?

The spread is what a market maker charges for standing between buyers and sellers. It has to cover three things: the risk of being adversely selected by someone who knows more, the cost of carrying the inventory that accumulates when flow is one-sided, and the plain operational cost of quoting. When any of those rises, the spread widens — no directional opinion required.

That makes spread a useful stress gauge and a poor direction gauge. It widens ahead of scheduled events, in the thin hours, immediately after a large print, and whenever volatility jumps. It narrows when makers compete and inventory turns over easily. A spread that widens while depth stays put is makers pricing uncertainty; a spread that widens while depth collapses is makers leaving.

Compare spreads in basis points of mid, never in ticks or in dollars. A $2 spread on a $100,000 asset is 2 bps and very tight; the same $2 on a $60 asset is over 30 bps and very wide. On a prediction market quoted 0–1, the natural unit is cents, and the same discipline applies: a 0.01 spread is a whole percentage point of implied probability, which is enormous relative to most claimed edges.

Order book depth: what depth data shows, and how deep is “deep”?

Depth is the cumulative size resting within some distance of the mid. The distance is the whole point: “there is $4 million in the book” is meaningless unless you say within what range, because a book always contains a great deal of size far away from the price, where it will never be touched. The convention worth adopting is cumulative notional within a fixed band — 10 bps, 25 bps, 100 bps — measured separately on each side.

Measured that way, depth answers the only question a taker cares about: what will this order cost me? The mechanical exercise is to walk the book — consume each level in turn until your size is filled, then compute the size-weighted average price and compare it to the mid you started from.

Two habits follow. First, size your expectations to the book rather than to the last price — the last trade tells you what someone else’s size cost, not what yours will. Second, watch asymmetry: a book with 5× more depth on the bid than the ask within 10 bps is not a prediction of the next move, but it is a statement about which direction is currently cheaper to push.

Does book imbalance predict the next move?

Order-book imbalance is normally defined as the difference between bid and ask size within a band, divided by their sum, giving a number between −1 and +1. It is one of the most studied quantities in market microstructure, and the literature is consistent about two things: queue imbalance carries information about the next price change, and that information decays extremely fast.

Fast means the horizon over which imbalance says anything is measured in seconds or less on a liquid venue, which is the same horizon over which the imbalance itself changes. This is why the signal is both real and nearly unusable by hand: by the time a human has read it, the configuration that produced it has been replaced, and any edge that remains is competed for by systems that respond in microseconds.

The failure mode to avoid is horizon mismatch — taking a millisecond-scale variable and using it to justify a position you intend to hold for two days. If you want an imbalance measure with a longer half-life, look at flow rather than the resting book: signed trade volume, the ratio of aggressive buys to aggressive sells, and the direction of open-interest change. Those describe what actually happened, and they do not vanish when someone hits cancel.

One more caveat, and it is not a small one: imbalance inverts around large passive orders. A patient institutional buyer resting size on the bid creates a bullish-looking imbalance precisely because it intends to absorb selling, and the price can grind lower against that wall for hours without the wall being wrong.

Are walls real, or are they decoys?

A wall is simply a resting order much larger than its neighbours. It is easy to see and easy to over-interpret, because the one thing it definitely is not is a commitment. Limit orders can be cancelled at no cost and at machine speed, so a visible wall costs its owner nothing until it is hit.

There are at least four different objects that look identical on a depth chart:

• A genuine passive order from someone who wants that size at that price and will let it be filled.
• The visible tip of an iceberg, where the displayed quantity is a fraction of the true order and refreshes as it is consumed — here the real liquidity is larger than the screen says.
• A hedge that exists only while some other position exists, and that will be pulled the instant the underlying moves.
• A decoy: size posted to be seen and cancelled, intended to move other participants’ pricing. Layering and spoofing are illegal in regulated markets and are prosecuted; on venues without that regime, the practice exists and you should assume it does.

You cannot distinguish them from the depth chart alone, but you can from behaviour. Watch the tape against the book: does the wall print trades — does size decrement while executions arrive at that price — or does it step away as price approaches? Absorption is a real order doing its job. Retreat, repeatedly, at the same distance, is a decoy or a hedge. A wall that survives being hit is the only kind that has told you anything.

Hinweis The same logic applies to sudden disappearances. A wall that vanishes without a single trade against it never provided liquidity; it provided information to whoever was watching, and possibly the wrong information deliberately.

Top of book vs depth of book: why displayed liquidity is not available liquidity

This is the single most expensive misunderstanding in book reading, and it cuts in both directions.

Why less is available than you see

• Cancellation beats you. Between the moment the book reaches your screen and the moment your order reaches the matching engine, resting orders can be withdrawn — and the fastest participants withdraw exactly when an informed order is arriving.
• The same inventory is quoted in several places. A market maker running a book across multiple venues may show size on each while intending to fill it once; a fill on one venue triggers cancellation on the others.
• Depth is state, not a promise. Nothing obliges anyone to be there a moment later, and no venue penalises a cancelled order.

Why more is available than you see

• Hidden and iceberg orders do not display their true size. A level that keeps refilling as it is consumed holds far more than the ladder claims.
• Replenishment is behavioural. In a calm market, makers re-quote within milliseconds of being filled, so the effective depth over a few seconds is a multiple of the instantaneous snapshot.
• Liquidity that is not resting can still be summoned. Size arrives when the price becomes interesting; a thin book at a level nobody cares about says nothing about the book that will exist when the price gets there.

Net of both effects: treat displayed depth as an upper bound on what a fast, informed market will let you take, and as a lower bound on what a calm market will provide to a patient order. Those two statements are not contradictory — they describe different regimes, and knowing which one you are in is most of the skill.

What can a book never tell you?

It is worth being explicit about the limits, because a rich data source invites over-reading. A level-2 book contains no identity: you cannot tell a hedge from a directional bet, a market maker from a fund, or one participant from ten. It contains no intent: an order is a price and a size, not a plan.

And critically, it contains no latent orders. Stop-losses are not resting orders — they are instructions held elsewhere that become market orders when a price prints. Neither are liquidation levels, which live in the venue’s risk engine, not its book. Both represent flow that is invisible until it arrives, arrives all at once, and arrives in the direction that hurts. That is the mechanism behind most violent moves, and it has its own guide: liquidations and cascades.

Nor does one venue’s book describe the market. Depth on the venue you are watching can be shallow while the same asset is deep elsewhere, and cross-venue arbitrage will import that depth with a lag. For anything above small size, the relevant book is the aggregate — with the caveat, again, that shared inventory means the aggregate overstates the true total.

How do you read a book in practice?

A short sequence, in this order, takes about fifteen seconds and avoids most of the errors above:

1. 01Spread in basis points of mid. Where is it relative to its own recent range? That is your regime check.
2. 02Depth within 10 and 25 bps, on each side separately, in notional. That is your cost estimate, and it is the number that decides your size.
3. 03Imbalance, and its path over the last minute — the change matters more than the level.
4. 04The tape against the book: are prints hitting the bid or lifting the offer, and is resting size decrementing or disappearing?
5. 05Refresh behaviour after the last sweep. A book that rebuilds in seconds is a different market from one that stays hollow.
6. 06The same five readings on a second venue. Divergence is information; agreement is confirmation.

Finally, match the horizon. Everything on this list decays in seconds to minutes. If your decision has a horizon of days, the book tells you how to execute it, not whether to take it — and for the “whether”, funding and positioning are the better instruments. That is the funding rates guide.

Häufige Fragen

Is a large bid wall bullish?

Not by itself. It is a statement that someone wants that size at that price, and it can be cancelled for free. It becomes evidence only when trades hit it and it stays — absorption, not appearance, is the signal.

What is the difference between L2 and L3 data?

L2 aggregates all orders at a price into a single size per level. L3 shows each order individually, which lets you infer queue position and watch specific orders being placed and cancelled. Most public crypto feeds are L2; treat “L3” dashboards as reconstructions unless the venue documents otherwise.

Why did my market order fill worse than the book showed?

Some combination of three things: resting orders were cancelled between your snapshot and your arrival, your size consumed more levels than you accounted for, or the book you were looking at had drifted from a dropped update. The first is the most common and the least visible.