Order Book Imbalance: What Depth Says That the Last Price Does Not

One trade is a point, the book is a distribution

The last price is the price of the most recent match. It is a single number produced by a single event, and it is the number most charts are built from. It does not say how much was traded, how much is available one tick away, or whether the next trade will happen near it.

The order book answers a different question. It lists the resting orders on each side, by price and size, and therefore describes the queue that a market order would have to walk through. Depth is that list. Imbalance is one way of summarizing it.

The distinction matters because the two numbers can disagree. A market can print a new low on a tiny trade while the bid side is deep, or print a new high while the offers are thin. Neither number is wrong; they describe different parts of the same market. The spread, depth and slippage explainer covers the cost side of that difference; this article covers the information side.

What depth actually reports

Binance’s GET /api/v3/depth returns two arrays, bids and asks, each entry a [price, quantity] pair. The default limit is 100 levels and the maximum is 5,000. The request weight rises with the window: 5 at 1–100 levels, 25 at 101–500, 50 at 501–1,000 and 250 at 1,001–5,000. The response carries a lastUpdateId, which is the identifier a client uses to stitch snapshots together without silently mixing two book states.

OKX’s GET /api/v5/market/books returns up to 400 levels per side. Each level is [price, size, deprecated, number of orders]; the third field is documented as always 0, and the fourth is the count of orders at that price. The endpoint is documented as updating once every 50 milliseconds and as returning the latest data once the server-side cache has been updated, rather than recomputing on demand. Its rate limit is 40 requests per 2 seconds, by IP.

Those details are not trivia. The number of orders at a price tells you whether a level is one large order or many small ones, which changes how likely it is to disappear at once. The update cadence tells you how stale a snapshot can be. The level cap tells you the maximum window any ratio can use.

Building an imbalance ratio

The common construction is a normalized difference between bid depth and ask depth over the same number of levels:

imbalance = (bid depth − ask depth) / (bid depth + ask depth)

The result runs from −1 (all bid, no ask) to +1 (all ask, no bid), with 0 meaning balanced. Depth can be summed in base units or in notional; notional is usually the better choice when comparing across pairs, because one BTC and one DOGE are not the same amount of market.

The table below is illustrative. It is not a market snapshot, not data from Binance or OKX, and not a claim about any instrument.

Bid size totals 60 and ask size totals 30, so the five-level imbalance is (60 − 30) / (60 + 30) = +0.33. In notional terms the totals are roughly 5,998 versus 3,001. The ratio stays near +0.33 because the prices are close, but on a pair whose bid and ask levels sit at different distances from the mid, the base-unit and notional versions can diverge.

Three choices are embedded in that one number: the number of levels, whether size is measured in base units or notional, and whether the touch is included. Change any of them and the number changes. Two analysts can report opposite imbalances on the same pair and both be arithmetic-correct.

The last price is a single point

A last price can be set by a trade of any size the venue allows. In a thin moment, a small print at an extreme can become the reference price for every chart, every alert and every stop order that watches the last trade. The book may still hold most of its size several ticks away.

OKX makes this explicit in its trigger documentation. It describes last as triggering when any trade occurs at or beyond the trigger price, calls it the most responsive option but vulnerable to brief price wicks in thin markets, and recommends mark for derivatives; last is the only option for spot instruments. That is a venue telling you, in its own documentation, that the last price is not the same object as the market.

Depth is not a correction to that problem. It is a different measurement. It tells you what is quoted around the price, not what has traded. A market can have a deep book and a last price that is stale, or a thin book and a last price that is current.

What depth cannot tell you: spoofing and layering

The CFTC’s interpretive guidance on disruptive trading practices quotes the statutory definition: spoofing is “bidding or offering with the intent to cancel the bid or offer before execution.” The definition is about intent, and intent is not in the feed.

This is the central limitation of every imbalance ratio. A large bid may be a genuine hedge, a market-maker’s inventory quote, or an order placed to make the book look deeper than it is. Layering — placing several orders at different prices to create the appearance of a wall — produces exactly the pattern a naive imbalance calculation rewards. The orders can be canceled before anyone trades against them.

The venues give partial help. OKX’s order count field separates one large order from many small ones at the same price. Persistence over repeated snapshots separates a level that stays from a level that flickers. Neither is proof. An order can rest for minutes and still be canceled before it trades, and a market maker can also pull a genuine quote in milliseconds when the market moves.

What the ratio measures is the state of the queue at one instant. What it does not measure is whether the queue is committed.

Reading depth without over-reading it

A few habits keep depth useful without turning it into a signal it cannot be:

• Sample repeatedly. One snapshot is a frame; the question is whether the level survives. Compare the same window across several snapshots and note how much of the depth is replaced between them.
• Match the window to your size. A ratio built on five levels is about the touch. If your order is large enough to walk ten levels, a five-level ratio describes a market you are not trading in.
• Use the same construction everywhere. Fix the level count, the notional convention and the treatment of the touch before comparing pairs or venues. Otherwise the differences you find are differences in methodology.
• Read the spread and the tick first. A wide spread makes the first levels less meaningful, and a one-cent tick near a price of 100 is a different distance than a one-cent tick near 80,000.
• Confirm with trades, not with more depth. If the imbalance persists and the trade flow leans the same way, the book and the tape agree. If they disagree, the tape is the record of what actually happened. The maker and taker explainer covers why the distinction between resting and taking matters for what the tape shows.

What the number is good for

Imbalance is a compact description of the queue. It is useful for comparing the same pair to itself over time, for noticing when the book has thinned before a size-sensitive order, and for explaining why a market order cost what it did. It is not a forecast, and it is not evidence of intent.

The honest summary is that depth tells you what is advertised, the last price tells you what last traded, and neither tells you what will happen next. A trader who reads the book, the tape and the last price, and knows which question each one answers, is harder to fool than one who treats any single number as the market.

Risk note. The endpoint behavior, level limits and update cadence in this article come from Binance’s Spot API documentation and OKX’s API documentation as read on 2026-10-25. The depth table is explicitly hypothetical. The spoofing definition is quoted from the CFTC’s published guidance on the Commodity Exchange Act’s disruptive trading provisions. Nothing here is investment advice.