What the number actually is

Every trade on an exchange has two sides with different roles. The maker had an order already resting on the book, waiting. The taker came along and accepted that price, crossing the spread to get filled immediately. One side chose patience, the other chose urgency.

The exchange knows which was which, and Binance publishes it. In a kline record, field 7 is the total quote asset volume for the period and field 10 is the portion of that where the buyer was the taker. Divide the second by the first and you have the taker buy share.

Here is a real hourly candle from today, printed exactly as the API returns it:

FieldMeaningValue
7Total quote volume$53,918,317.95
10Taker buy quote volume$20,989,579.23
10 divided by 7Taker buy share38.93%
7 minus 10Implied taker sell volume$32,928,738.72

So in that hour, of the $53.9 million that changed hands, $21.0 million came from buyers lifting offers and $32.9 million came from sellers hitting bids. The reading was 38.93%.

One technical note for anyone reproducing this. Binance also publishes the same split in base currency terms, fields 9 and 5. The two give almost identical answers, differing by around 0.0004 percentage points on the candle above, because price moves slightly within the period. We use the quote version throughout, because dollars are what readers think in.

The first thing to unlearn: 50% is not neutral

The intuitive reading is that above 50% means buyers are winning and below 50% means sellers are. That is close enough to be useful and wrong enough to mislead, because the typical reading is not 50%, and it changes with the timeframe you are looking at.

Timeframe10th pct25th pctMedian75th pct90th pct
5 minute31.6%41.9%50.9%60.6%68.0%
1 hour39.4%44.8%50.5%55.9%60.8%
1 day45.7%47.4%49.0%50.8%52.4%

This single table is the most useful thing in the guide, and it resolves most confusion about the metric.

On five minute candles a genuinely strong buy reading is around 68% and a genuinely weak one is around 32%. On daily candles the same percentiles are 52.4% and 45.7%. The daily median is 49.0%, which is below the halfway line.

So a reader who learned the metric watching a fast chart, where 68% happens routinely, will look at a daily reading of 53% and conclude nothing is happening. In fact 53% on a daily candle is above the 90th percentile. It is one of the strongest daily readings the market produces.

The reason for the narrowing is arithmetic rather than anything mysterious. A five minute window can be dominated by a single participant. A whole day contains thousands of independent decisions in both directions, and averaging them pulls the result toward the middle. The longer the window, the more it averages, and the tighter the range gets.

Does it actually track price?

Yes, and by a consistent amount. We correlated the buy share against the return of the same candle, across 999 candles at each timeframe.

TimeframeCorrelation, same candleVariance explainedCorrelation with the next candle
5 minute0.50025.0%0.037 lower
15 minute0.39315.4%0.008 lower
1 hour0.35512.6%0.029 lower
4 hour0.46321.4%0.046 lower
1 day0.49224.2%0.051

Read the two right hand columns against each other, because the contrast is the entire practical lesson of this guide.

Within the period it measures, the buy share explains somewhere between 12% and 25% of the variation in returns. That is a real relationship. It is not noise, and it is not a rule either: three quarters of what moves price in a given candle is something else.

Against the next period, the correlation is indistinguishable from zero at every timeframe we tested. Four of the five are slightly negative. This is the measurement behind a sentence this site has published many times without evidence: the metric describes the window it measures, and it does not forecast the next one. Now it is a number rather than a claim.

If you take one thing from this guide, take this. A high reading tells you something true about what just happened. It tells you close to nothing about what happens next.

Why price rises on a low reading, and how often

The question we are asked most often is some version of: the price went up, so why is the number below 50? The answer is that this is not the exception. It is close to half of all up days.

TimeframeUp candles in sampleOf those, buy share below 50%Share
5 minute49313527.4%
15 minute50014328.6%
1 hour50816532.5%
4 hour49815130.3%
1 day50923145.4%

On daily candles, almost half of all up days had more aggressive selling than aggressive buying. Taken across hourly candles, the buy share agrees with the direction of price 65.6% of the time. Better than a coin flip, nowhere near a rule.

The mechanism is the one this site spent a week arriving at. Price is set by where resting orders sit, and the taker ratio measures only who crossed the spread to reach them. If patient buyers keep raising their bids while absorbing aggressive selling, the price climbs on a low reading. Nothing is broken. The two things are simply not the same thing.

Worth noting the mirror case is rarer on daily candles: only 15.7% of down days had a buy share above 50%, against 45.4% the other way. That asymmetry follows from the daily median sitting at 49.0%, below the halfway line.

It works better on some assets than others

Before you carry the numbers above onto another chart, check that they transfer. They mostly do, with one clear exception.

AssetHourly correlationDaily correlationDaily 5th to 95th pct
Bitcoin0.3550.49244.4% to 53.2%
Ether0.3760.49745.4% to 53.4%
Solana0.2820.52046.0% to 53.4%
XRP0.1650.22344.8% to 51.8%

Bitcoin, Ether and Solana behave almost identically at the daily scale, clustering around 0.49 to 0.52. XRP is the outlier at 0.223 daily and 0.165 hourly, roughly half the relationship of the others.

This is worth knowing because it explains something we published on August 22. XRP fell 8.11% in a single hour that had a 50.6% buy share, which looked like a spectacular failure of the metric. Given that XRP has the weakest taker to return relationship of the four assets we tested, it was closer to typical behaviour for that particular market than we appreciated at the time.

Practical implication: the percentile tables above are calibrated on Bitcoin. Applying them unchanged to XRP will overstate what a reading means.

The three things it cannot see

A metric built from completed spot trades on one exchange is blind to anything that happens elsewhere or never trades. Over the past two weeks we documented three of these, each with a measurement.

Blind spotWhyMeasured example
ETF creationsSettle through authorised participants, off the public bookA $853 million inflow week with no trace in the ratio
Futures liquidationsExecute on the perpetual book, not spot$36.5 million net on spot against $561.6 million on perps, same 15 minutes
Maker repricingOrders placed and cancelled never trade at allAn 8.11% hourly fall on a 50.6% reading

The first two are venue problems. Money was moving somewhere we were not looking, and the fix in principle is to look there too, which is why our recent articles pull the perpetual book alongside the spot one.

The third is different and cannot be fixed that way. It is on the same book, on the other side, and it consists of orders that never traded. No metric built from trades can see it, by construction.

It does leave one visible trace, though. Across 999 daily candles, 16 Bitcoin highs landed on an exact $1,000 multiple, against a chance expectation near 0.001%. That is the resting book showing up in the price record: large limit orders sit at round numbers, price trades up into them and stops. So the invisible side is not entirely invisible. It just does not appear in the ratio.

A worked example from today

Everything above becomes concrete on a live session. These are Bitcoin candles from this morning, all closed.

Window, UTCMoveBuy shareHow to read it
13:20, five minutes+0.11%27.4%Below the 10th percentile, price up anyway
08:00, one hour+0.85%51.8%Barely above the hourly median of 50.5%
13:00, one hour0.09% lower38.3%Below the 10th percentile of 39.4%, genuinely weak
Session so far+0.24%43.9%Below the daily 10th percentile of 45.7%

The first row is the case that confuses people. Price rose while 72.6% of the value crossing the spread was selling. Using the percentile table, 27.4% on a five minute candle is below the 10th percentile, so the selling really was unusually aggressive. And the price still went up, because patient buyers were holding their bids up underneath it.

The last row shows why the timeframe matters. A session reading of 43.9% sounds mildly bearish if you compare it to 50. Compared to the actual daily distribution, where the 10th percentile is 45.7%, it is a genuinely weak session for aggressive buying, even though the price is up 0.24%. Bitcoin also traded as high as $80,520 during it.

Both facts are true at once, and the guide exists so that they stop looking like a contradiction.

How to actually use it

Five rules, each of which follows from a measurement above rather than from opinion.

One. Match the window to your question. If you want to understand a specific move, read the five minute or hourly figure around it. If you want to characterise a session, read the daily one and judge it against the daily distribution, not against 50.

Two. Compare against the right baseline. A daily 53% is a top decile reading. An hourly 53% is unremarkable. A five minute 53% is barely above the median. The same number means three different things.

Three. Never read it forward. The next period correlation is zero at every timeframe we tested. Whatever the number is telling you, it is telling you about the past.

Four. When price and the ratio disagree, look for the missing venue. A rising price on a weak reading often means the buying happened somewhere the ratio does not cover, or on the resting side. Check open interest for futures activity, and check whether an ETF flow published that day.

Five. Do not use it alone. It explains a quarter of the variance in the best case. Volume tells you whether anyone cared. Open interest tells you whether positions opened or closed. The ratio tells you who was in a hurry. Three different questions.

And one thing to expect: it will regularly disagree with the price on your screen. On 45.4% of up days it already does.

What this guide is not

It is not a strategy. Nothing here has been tested as a trading rule, and the forward correlation of zero is a strong argument that it would not work as one.

It is not calibrated for every market. The percentile tables come from Bitcoin on Binance spot. XRP behaves measurably differently, and we have not tested the long tail of smaller pairs.

It is not a fixed truth. These distributions come from roughly the last three years of daily candles and the last few months of intraday ones, and market structure changes. We will restate the numbers when they move.

What it is: an honest account of one number, including the parts that make it look worse. We publish this metric every minute, and a reader deserves to know that it explains a quarter of the variance rather than all of it, that it says nothing about tomorrow, and that it is blind to three specific kinds of flow. Those facts do not make it useless. They make it usable.