How to Read the Bid-Ask Spread

Every trading screen shows two prices. The bid is the most anyone will pay right now; the ask is the least anyone will sell for. The gap between them is the spread, and it's the first thing a market order pays — before the fee, before any slippage, before the trade has had a chance to go right or wrong. It's also the number people quote least carefully. "Tight" and "wide" get said a lot; how tight, for how long, and compared with what, gets said less.

So this week the spread is measured instead of described. Binance publishes a record of every change to the best bid and ask on its perpetual contracts — the bookTicker feed — for 16 May 2023 to 27 March 2024. That's several million updates a day on bitcoin alone, so the study reads every fourth day plus the days something happened, weights each quote by how long it stood, and reports the spread in ticks and in basis points of the price. Then, for one morning, it reads the live quote on every USDT pair the venue lists.

What the number measures

An order book is two lists. On one side, everyone who has said "I'll buy at this price," sorted from the highest price down; on the other, everyone who has said "I'll sell at this price," sorted from the lowest price up. The top of each list is the best bid and the best ask. Nobody has agreed to trade yet — if they had, they'd have traded — so there's always a gap. That gap is the spread. It's quoted in the instrument's price ($0.10 on the bitcoin perpetual), in ticks (the smallest price step the exchange allows, so one tick is the narrowest it can be), or in basis points of the mid-price, which is the only way to compare an $82,000 coin with a $0.09 one.

Who sets it? Market makers — firms and bots whose business is standing on both sides of the book, buying at the bid, selling at the ask, and keeping the difference. They quote as tight as they can afford to, because a tighter quote gets filled first. What they can afford depends on how likely the price is to move against them before they can get out. When nothing's happening, that risk is small and the quote sits on the tick. When price is moving fast, they pull back — or get run over — and the gap opens. So the spread is a live reading of how nervous the people supplying liquidity are, updated thousands of times a minute.

Finding 1: the spread is one tick, almost all the time

The first thing the data says is that on a book this deep, the spread isn't really a market price at all. It's the exchange's rule. Binance lets the bitcoin perpetual move in steps of $0.10, and for 99.4% of the sampled time — weighted by how long each quote stood — the best bid and best ask were exactly one step apart. There was no month under 98.3%. Ether, with a one-cent tick, was pinned 99.5% of the time. Solana, the thinnest of the three, still spent 96.5% of its time on the tick.

The basis-point columns say the second thing. Bitcoin's tick was $0.10 in May 2023 when the coin was around $26,965, and still $0.10 in Mar 2024 at $67,500. So the spread as a share of the price fell from 0.038 bp to 0.020 bp without the book getting any tighter — the price went up under a fixed floor. Solana makes the point louder: 0.50 bp in May 2023 at $20.25, 0.10 bp in Mar 2024 at $149.76. A 80% "improvement" in the spread that was entirely the price rising past a tick that stayed at a tenth of a cent. When you read a spread in basis points, you're mostly reading the tick divided by the price.

Sitting at that best bid and best ask, on bitcoin, was a median of $259,124 of orders — the amount a market order could take before it had to reach into the next level. On solana, $6,292. A tick-wide spread is a fact about the first $259,124; what a bigger order pays is a fact about the depth behind it, which is a different measurement and a different article.

Finding 2: it follows the clock

Crypto trades around the clock, but the spread doesn't. Bitcoin's quietest hour was 06:00 UTC, at 1.02 ticks on average and one tick 99.8% of the time. Its widest was 17:00 UTC, at 1.24 ticks and 99.0%. The shape is the same on all three contracts: a floor through the Asian and European morning, a bump when New York opens at 13:30 UTC and the US data prints land, and a second bump into the US afternoon. Solana's widest hour, 15:00 UTC, averaged 1.32 ticks.

Weekends are the tightest time to trade bitcoin, by this measure: at least 99.86% of Saturday and Sunday at one tick, against 98.93% on Tuesday, the widest weekday. That's the opposite of the folk rule that weekends are thin and dangerous. Both are true — weekend depth is lower and weekend moves are sharper when they come — but the spread measures how nervous the makers are minute to minute, and with no US session, no data and no equity open, there's less to be nervous about. The spread is narrow when nothing is happening. That's not the same as safe.

Finding 3: it opens for minutes, not hours

On 17 August 2023, bitcoin fell from $27,660 to $24,804 in a quarter of an hour — the sharpest drop in the sample. In the 21:43 UTC minute the average spread was 100 ticks, $10 between the bid and the ask on a coin at $25,478. That's 3.8 bp, a hundred times the normal quote, and the time at one tick fell to 10%. Ether's widest minute came at 21:42, at 5.2 bp. Then it closed: 22 minutes after the peak, bitcoin printed a minute back on one tick. Of the 60 minutes in the window, 11 averaged over five ticks and 2 over fifty. Over the whole day the bitcoin spread was one tick 98.7% of the time.

The widest bitcoin minute in the sample wasn't a crash but a squeeze: 23 October 2023 at 22:42 UTC, 246 ticks — 7.1 bp — as price moved 6.2% inside sixty seconds on the way from $31,926 to $34,690. The direction doesn't matter to the makers; the speed does. Solana's record came on 5 March 2024, the afternoon bitcoin turned down from its old high. The coin fell from $127.37 to $108.28 inside the 19:40–20:10 window; in the 19:57 minute its mid ranged 9.8% and the spread averaged 173 ticks, 15.9 bp, and 6 of those thirty minutes were over fifty ticks. On 10 January 2024, the day the SEC approved the spot bitcoin ETFs, the widest minute was 55 ticks at 20:12 — a big event and a smaller number than either crash, because the tape moved less per minute.

How long does it stay open? Take every minute in the sample that averaged fifty ticks or more and count forward to the next minute back under one and a half. On bitcoin there were 26 such minutes and the longest wait was 36 minutes, on 5 March 2024 from 19:55 UTC. On solana there were 14, and the longest wait was 3.6 hours: on 3 January 2024, the morning a research note predicting the SEC would reject the spot ETFs sent bitcoin from $44,329 to $42,000 inside the 11:50–12:40 window, the solana book never got back to a one-tick minute until the afternoon. The deepest book heals in half an hour; the thinner one can stay wide all session.

That same morning shows something else. Binance's published record of the best bid and ask has holes in it, and they fall where you'd least want them. All three contracts go silent from 12:08 to 12:22 UTC on 3 January 2024 — 14 minutes, starting inside the crash — and again for 43 minutes from 17:32 UTC on 28 February 2024, the minute after bitcoin's second-widest quote of the whole sample (245 ticks at 17:31). Whether the quotes stopped updating or the archive lost them, the effect on a reader is the same: the minutes you'd most want to see are the ones most likely to be missing. The study drops the first minute after each hole rather than credit a stale quote to it.

That's the shape in one table. Across the calmest four fifths of hours — anything up to a 1.0% range — the bitcoin spread averaged between 1.00 ticks and 1.03 ticks. The wildest fifth, with hourly ranges from 1.0% to 13%, averaged 1.43 ticks and was still at one tick 97.7% of the time. The spread doesn't widen gradually as the market gets busier. It sits on the floor until the tape moves faster than the makers can requote, then it jumps, then it's back. Solana's wildest fifth ran 1.68 ticks against 1.00 ticks in its calmest.

Finding 4: the fee is the cost, not the spread

Here's the arithmetic most spread talk skips. A market buy pays the ask and a market sell pays the bid, so a round trip pays the whole spread once. On the bitcoin perpetual at 0.031 bp that's $0.03 per $10,000. The same trade pays the taker fee twice, at 5 bp a side for a regular account: $10.00. The fee is 324 times the spread. Even in the single worst minute of the archive, when the spread was 3.8 bp, the fees were still the bigger number by 2.6 to one. Somebody who worries about the spread on bitcoin and doesn't know their fee tier has the costs the wrong way round.

The maker rate, 2 bp, is what changes the picture. A limit order that rests and gets filled doesn't pay the spread at all — it earns it, or at least half of it — and pays 2 bp instead of 5. On the $10,000 round trip that's $4.00 against $10.00. That's the real cost difference between the two order types on a book like this: not the $0.10 between the bid and the ask, but the 3 bp between the two fee columns.

Finding 5: across the venue, the tick is the floor and volume decides how often it's touched

Read across the whole venue for one morning, the picture has two parts. The busiest fifth of perpetuals — $24m of volume a day and up — quoted a median of 1.7 bp, and 44% of them sat on their tick in nearly every reading. The quietest fifth quoted 6.2 bp and only 10% were pinned. But look at where the busiest pairs land in basis points. Bitcoin at 0.012 bp; ether at 0.037 bp; solana, third-busiest coin on the venue, at 0.89 bp; dogecoin at 1.12 bp. All four were one tick wide in every reading. The difference between them is the tick divided by the price, and nothing else. Binance moved solana's tick to $0.01 — 10 times the archive's — so at $112.32 the floor is 0.89 bp where it was 0.10 bp in Mar 2024. The book didn't get worse. The rule changed.

Above the tick floor, volume is the story. 8% of perpetuals quoted under 1 bp; 43% quoted wider than the 5 bp taker fee, and 5% wider than two fees. The spot market is wider still: 493 USDT spot pairs, median 9.0 bp, with the quietest fifth at 11.3 bp against a 10 bp regular fee. Bitcoin spot, with a one-cent tick on a $81,811 coin, quoted 0.0012 bp — 10 times narrower than the perpetual, because its tick is 10 times finer. Same coin, same venue, same morning, and the number is set by the rulebook.

How to read it yourself

• Count it in ticks first. If the spread is one tick, the book is as tight as the exchange allows and the basis-point number is just the tick divided by the price. A pair at 2 bp with a one-tick spread and a pair at 2 bp that's five ticks wide are not the same market.
• Find the fee before you find the spread. On a big perpetual the taker fee is 324 times the spread; on the quietest fifth of perpetuals it's still around 1.6 times. The spread only becomes the main cost on pairs quoted wider than two fees — 5% of Binance's perpetuals on the study morning, and 45% of its spot pairs.
• Read a wide quote as a clock, not a verdict. It opened because the tape moved faster than the makers could follow. On bitcoin the longest a fifty-tick minute took to get back to a one-tick minute was 36 minutes; on solana it was 3.6 hours. A market order sent into the wide minute pays it; one sent after the book has settled doesn't, and the deeper the book, the sooner that is.
• Don't trade the quiet hours as if they were the deep ones. The 06:00 UTC hour and the weekend have the narrowest spreads in the sample because nothing is happening, not because the book is deep. Depth is a separate reading.
• Match the tick to your order size. The quote covers $259,124 on bitcoin and $6,292 on solana at the touch. Above that, you're paying the next level, and the spread you saw is the least you'll pay, not the most.
• Use a resting order where you can. On this venue it's the difference between 2 bp and 5 bp a side — several times the whole spread on any of the top pairs.

Method, and what this cannot tell you

The archive series is Binance's public bookTicker record for the BTCUSDT, ETHUSDT and SOLUSDT USDT-margined perpetuals: every change to the best bid or best ask, with its quantity and a millisecond stamp. Binance publishes it for 2023-05-16 to 2024-03-30 only, and the files run to 400 MB a day, so the study reads every fourth day plus the days named in the text — a sample, and it says so wherever a figure comes from it. Each quote is weighted by how long it stood before the next update, so a spread that flashed wide for a millisecond counts for a millisecond. A minute's spread is that time-weighted average, in ticks and in basis points of the mid; a day's, a month's or an hour-of-day's is the same average over its minutes. The cross-section is the live best bid and ask for every USDT-quoted perpetual and spot pair on the venue, read forty times at thirty-second intervals on the study date, with the median of those readings as each pair's spread and the exchange's own 24-hour quote volume as its size. Fees are the exchange's published regular-user rates.

• One venue. Binance's perpetuals are the deepest crypto order books in the world, and the spreads here are the narrow end of what exists. The same contracts on a smaller venue, or the same coins on a spot exchange with a thinner book, quote wider — the cross-section's bottom half is closer to what most pairs look like.
• Ten and a half months of history, sampled. The archive starts in May 2023 and stops at the end of March 2024, so it holds one crash day and one all-time-high week and none of 2020-22. Every fourth day is read; a day that is not in the sample is not in the tables.
• The quoted spread is not the spread you pay. It is the gap between the best bid and the best ask for the quantity sitting at each, which on the bitcoin perp is a few hundred thousand dollars. An order larger than that walks into the book and pays the depth, which week 87 measured; an order in a fast market pays the spread that stood when it arrived, not the one on the screen when it was sent.
• The tick sizes are the ones in force. Binance changes a contract's tick from time to time, and a change moves the floor the spread sits on; the archive series uses the tick each file's prices show, and the cross-section reads each pair's current tick from the exchange.
• The published record has holes. All three contracts go silent for three minutes on 21 September 2023, for fourteen from 12:08 UTC on 3 January 2024 inside the flash crash, and for 43 minutes from 17:32 UTC on 28 February 2024, straight after one of the widest minutes in the sample. The study drops the minute after each hole rather than credit the last stale quote to it, and a wide minute that fell inside a hole is not in any table.
• The cross-section is a morning. Forty readings over twenty minutes of a Monday in European hours say what the venue looked like then, not what each pair averages over a week.