How to Read Order-Book Depth

Every exchange draws it. Two mountains meeting at the current price: green on the left, the bids stacking up as you go lower; red on the right, the asks stacking up as you go higher. Hover over any point and it tells you how much money would have to trade before price reached there. That picture is the depth chart, and the number under it is order-book depth.

It is the only number in this series that describes the future rather than the past. Funding, open interest, the taker ratio and the basis are all records of trades that have already happened or positions already open. Depth is the orders that haven't traded yet. That is why traders read it as support and resistance in dollars, as buy walls and sell walls, and as a warning that a market is thin. The question is whether it deserves to be read that way.

What the number actually measures

Binance publishes a snapshot of its perpetual books every 30 seconds: the dollar value of resting orders within 1%, 2%, 3%, 4% and 5% of the mid, on each side, cumulative. The 1% band is the one this study reads — a $100,000 bitcoin has $1,000 either side of it, and everything queued inside that is what would have to be eaten for price to move a full percent. Bids within 1% below plus asks within 1% above is "depth" everywhere in this article.

Three things it is not. It is not the whole book — orders 5% away are real, they just aren't near. It is not hidden orders, iceberg orders, or the orders a market maker will place when price gets there. And it is not a promise: a resting order can be pulled in the time it takes you to read this sentence. What the number is, exactly, is the money that was showing at the moment of the snapshot.

Finding 1: the book is small next to the tape

On a median day, $268M rested within 1% of the mid on bitcoin's perpetual, $116M on ether's and $29M on solana's. That sounds like a lot until you set it against what trades. A median day's volume was 47× the resting depth on bitcoin, 66× on ether and 86× on solana. The book does not hold the day's business. It is refilled dozens of times over.

The book thickens as you move away from the mid, but not by as much as the mountains suggest. Depth within 2% is about 2.0× the 1% figure on bitcoin, and within 5% about 3.6×. The second percent holds about as much as the first; the next three together hold less than twice that. The mountains are steep near the price and flatten fast.

Growth in dollars hides a shrink in coins. In 2023 the 1% band held a median 7,360 bitcoin; in 2026 it holds 4,650. Market makers size their quotes in dollars of risk, not in coins, so as the price rose the same money bought a thinner book. A 1,000-bitcoin market order eats a bigger share of the band today than it did three years ago.

Finding 2: the clock barely moves it

Traders talk about thin Asian hours and thick London ones. The book agrees, faintly. Bitcoin's depth in each UTC hour, divided by its own day's mean, peaks at 1.04× at 11:00 and bottoms at 0.92× at 22:00. That is a spread of 12 points across the whole day. Ether and solana draw the same curve.

Weekends are the sharper version of the same point. Saturday and Sunday's book is 1.04× its recent median against 1.01× on weekdays — a little fuller — while the weekend range is 0.61× against 1.12×. The book is fullest when nothing is happening. Market makers leave their quotes out when there is nothing to fear, and that is exactly when nobody needs them.

Finding 3: it vanishes in the move

Take each day's thinnest reading — the smallest the bid side and the ask side each got in any hour, added — and set it against the day's range. The rank correlation is −0.62 on bitcoin, −0.56 on ether and −0.53 on solana. On the busiest 5% of bitcoin's days the book's thinnest reading was 33% of its recent median; on the quietest 5% it was 80%.

That is the shape of the thing. The mountains on the depth chart are drawn by people who will pull them the moment they are about to be hit. When price moves fast the bids that were "support" are cancelled and re-placed lower, the asks that were "resistance" are cancelled and re-placed higher, and for a few minutes the book near the price is mostly empty. The day you most want the book to be there is the day it is not.

17 August 2023 is the thinnest hour in the record. At 21:00 UTC bitcoin fell from about $27,700 to a low of $24,581 — an 11% range in one hour — and the bid and ask within 1% of the mid thinned to $17M against a trailing-week average of $317M. That is 5% of the book. It did not come back the next day, either: the 18th ran at 47% of the 16th.

Finding 4: a thin book says "bigger range tomorrow", a little

Now the forecasting question. Sort bitcoin's days into five bins by the depth ratio and look at the next day. The thinnest fifth — a book at 0.52× to 0.88× its recent median — was followed by a day whose range was 1.23× its own recent median. The thickest fifth was followed by 0.83×. The rank correlation between today's depth and tomorrow's range is −0.23.

Part of that is volatility persistence: a big day is usually followed by a big day (today's range against tomorrow's: +0.28), and big days thin the book, so a thin book can just be a busy day in disguise. To check, split the days by whether today was busy or quiet, and sort each half by depth. Among busy days the thinnest third saw a next-day range of 1.34× and the thickest third 0.92×; among quiet days, 0.96× against 0.81×. The book adds something on its own. Not much, but something.

It adds nothing about direction. The next day's average return and the share of up days are flat across the five bins — the up-share spans 6% from the highest bin to the lowest, which is noise. A thin book is a reason to expect a bigger move and to size for it. It is not a reason to expect the move to go one way.

Finding 5: a buy wall is yesterday's candle

This is the one that matters most, because it is the way most people read the depth chart. Bids outweighing asks near the price reads as demand — a floor, a wall, support with a number on it. Sort bitcoin's days by the bid share of the 1% band. In the most bid-heavy fifth (bids above 54% of the band) bitcoin was DOWN that day 83% of the time. In the most ask-heavy fifth it was up 85% of the time. The correlation between imbalance and the same day's return is −0.55.

Read the left half of that table and the mechanism is obvious. When price falls, it falls into the bids: the asks that were within 1% are now 3% away and no longer count, and the bids that were 3% below are now within 1% and do. The book leans against the move that just happened, automatically, because the band moved and the orders didn't. A bid-heavy book is what a down day leaves behind. It is the same thing week 85 found in the taker ratio — the number is the candle restated — from the other side of the trade.

Now read the right half. The next day's return is flat across the five bins, and the correlation between today's imbalance and tomorrow's return is +0.01. Ether's is −0.02, solana's +0.01. And the same-day link holds in every year of the sample, between −0.53 and −0.58 — it is not a bull-market artefact. The wall tells you where price has been. It does not tell you where price is going.

How to read it yourself

• Know the scale. On bitcoin's biggest perpetual, about $268M rests within 1% of the mid on an ordinary day, and a day trades 47× that. On ether it is $116M; on solana $29M. A market order the size of the band moves price a full percent — and that is before anyone pulls a quote.
• Read it as a ratio, not a dollar figure. The book has grown with the price, so compare today's depth with the last few weeks', not with a number you remember.
• Discount the clock. The difference between the fullest and emptiest hour of the day is about 12 points, and weekends are fuller than weekdays. The time of day is not why the book is thin.
• Expect it to vanish when you need it. On the busiest days the thinnest reading is about a third of normal. Size a position for the book you'll find at the low, not the one you see now.
• A thin book means a wider range tomorrow, a little — 1.23× for the thinnest fifth of days against 0.83× for the thickest. Use it for stop placement and size, not for direction.
• A buy wall is not support. When bids outweigh asks, bitcoin fell that day 83% of the time and did whatever it liked the next. The wall shows you the candle that already printed.
• Watch the thinnest reading, not the average. A day's mean depth barely moves; the minute-by-minute low is what tells you the market makers left.

Method, and what this cannot tell you

Each hour's depth is the mean of Binance's 30-second snapshots of the USDT notional resting within 1% of the mid on the bid side plus the ask side; each UTC day is the snapshot-weighted mean of its hours, and its thinnest reading is the smallest hourly bid-side minimum plus the smallest hourly ask-side minimum. A day's depth ratio is its depth over the median of the trailing 28 days. Imbalance is bid over bid-plus-ask within 1%. Depth, its thinnest reading and imbalance are compared with the same day's range (high minus low over open), return, and the next day's range and return, and — sorted into quintiles of depth ratio and of imbalance — scored on both. The hour-of-day profile is each hour's depth over its own day's mean; the weekday profile is the depth ratio by day of the week. The event table ranks every hour in the sample by its thinnest reading over the trailing seven days' mean.

• One venue, one contract. Binance's perpetual book only. Spot books, Coinbase, OKX and Bybit hold their own depth, and a bitcoin market maker quotes across all of them at once — so the total resting within 1% of the mid is several times what this one book shows.
• The archive freezes. On every symbol the feed printed an identical bid-side notional for a whole hour or longer in 16 April to 19 May 2025, 10 to 15 October 2025 and 3 to 7 September 2026 (ether also 30 October to 1 December 2025). Those hours were dropped, and the 10 October 2025 cascade — the one day a reader would most want to see — is not in the sample because the bid side froze at 17:00 UTC that day.
• Notional within 1% counts resting limit orders. It does not count iceberg orders, orders that will be placed when price gets there, or orders that will be pulled when it does. What a market maker shows and what it will fill are different numbers, and this study only sees the first.
• Each side's thinnest reading is its own minimum within the hour; the two are added, so the 'thinnest' figure can be slightly lower than any single snapshot showed.
• Daily readings, no leverage on intraday timing. A book that thins for eight minutes at 21:04 UTC and refills is one number in this study. The hourly and 30-second data exist and would show more; this article reads the day.
• Correlations near −0.2 are real and small. A depth-ratio quintile moves the next day's range by about a fifth either way; it does not tell you which day will be the one that gaps.