Turn the Candles Off: Reading Volume, Bar by Bar

Pull up any chart and the candles are the first thing you see. They're loud by design: tall bodies, long wicks, red and green fighting for attention, and every one of them a story about where price went. Underneath, in a strip most people never look at twice, sit the volume bars. Try something for a minute. Hide the candles, leave the volume, and look at what's left. It still tells a story, and I'd argue it tells a more honest one, because the volume bars record how many people bothered to go along with the market, whichever way it went.

This post is about reading that strip. It's a preamble, in the same way Read the Spec Sheet was: it won't teach you a trade. It exists because volume runs underneath almost everything in Applied Market Mechanics. Tool 01 picks its climax anchor by volume, the price action and derivatives entries still to come lean on it, and the whole of Wyckoff, which I covered in Part 4 of Understanding Market Mechanics (UMM), is volume dressed up as theatre. What UMM never had was an entry dedicated to the thing itself, and rather than re-explaining it inside every Tool, it gets one page here that the rest of the series can point back to.

We'll go through what a volume bar actually counts, why one bar on its own tells you almost nothing, how a run of them tells you a great deal, and why the units the venue reports in can quietly turn a bull market into a bear one. Then we'll get to the uncomfortable part, which is that volume is the cheapest number in the market to fake, with a worked example from the biggest exchange in crypto, before turning the bars on their side to see what all of it does to volume at price. The last section is a plain checklist of how this series reads volume, so the Tools can link straight to it.

#What a Bar Counts

A volume bar is a count of everything that traded during that bar: shares, contracts, or units of the asset. Every one of those units had a buyer and a seller, which means a volume bar can't lean. If a thousand bitcoin changed hands, a thousand were bought and a thousand were sold, in the same instant, between the same people. When someone tells you a big volume bar means "more buyers than sellers", what they're describing is a bar that can't exist.

What volume measures is participation. It tells you how much business got done while price did whatever the candle above it shows, and so how much the move mattered to the people in the market. A rally on thin volume is a rally few people paid for. A drop on enormous volume is a drop that a lot of people acted on, on both sides of it.

Two things on your screen get in the way of reading it like that. The first is colour. On most charting platforms, TradingView included by default, a volume bar is green or red because the candle above it closed up or down. The colour is the candle's, and it says nothing about who was more aggressive inside the bar. There is a real directional measure, which is whether each trade was a buyer lifting the offer or a seller hitting the bid (the aggressor side), but you need trade-level data for it, and it belongs to the derivatives panel coming as Tool 02. The second is orientation. The bars under a chart are volume over time, one per period. There's also volume at price, a horizontal histogram of how much traded at each level, which is the view behind the observation in Part 3 that markets drift back to where volume built up. We'll start with the vertical bars, the ones every chart already shows you, and turn them on their side once they've taught us how to read them.

I'll write the volume of bar as , in whatever unit the venue reports.

#One Bar, Two Bars, a Run

One bar on its own is close to meaningless. Is 40,000 BTC a lot? On which venue, in which year, on which timeframe? There's no answer, because a single reading has nothing to be large or small against.

Two bars side by side tell you something changed. The second is taller or shorter than the first, which means interest in the market, or at least activity as its proxy, rose or fell from one period to the next. That's the read that matters when price arrives at a level: did more people act, or fewer? The comparison is a ratio,

and I'd sooner you held the ratio in your head than any of the raw numbers. This is the reading that ranges and price action use. In Tool 01 it's how a climax gets told apart from an ordinary week.

A run of bars tells you which way that change is trending: interest building, interest draining, or a sudden climax that breaks the run. Here a single comparison stops being enough, and the natural baseline becomes the average of the run,

That's the reading Wyckoff uses, and it's why Part 4 describes contraction as "volume below its 20- to 50-period average". Part 4's average and the bar-to-bar ratio are the same act of comparison with the window set at different widths, and which width you want depends on what you're asking. Ask whether this test of a level drew more participation than the last one and you want two bars. Ask what phase the market is in and you want a run.

Widen the window far enough and the bars stop being a measurement and start being a story. Which is the experiment I promised at the top.

#Turn the Candles Off

Below is Coinbase's weekly bitcoin volume from the week of the December 2017 top to the week of the November 2022 low, five years and 259 bars, with the candles hidden. Before you read past it, decide what you think happened.

Read straight, it's a market that got steadily busier. A quiet 2018 and 2019, a burst in March 2020, a livelier 2021, and then 2022 busiest of all, finishing on the tallest bar in the whole window. The averages agree with your eye: 94,670 BTC a week in 2018, 132,933 in 2021, and 156,544 in 2022. If you'd been told nothing else, you'd say interest peaked at the end, in 2022.

Now the same bars, converted into dollars.

A completely different market. 2018 to 2020 flatten into a floor, and 2021 towers over everything: about 6.07 billion dollars a week on average, against 0.74 billion in 2018, roughly eight times as much, with 2022 well below it at 4.19 billion. Same exchange, same weeks, same trades. The only thing that changed is the ruler.

And here's what actually happened.

The tallest bitcoin bars all sit on weeks when price fell hard: the March 2020 crash, the May 2021 sell-off, the June 2022 collapse, and the FTX week in November 2022, which is the tallest of the lot at 432,341 BTC. They're capitulations, and they're exactly what the climax anchor in Tool 01 goes looking for. The bars were telling the truth about those weeks. Where they misled you was the long comparison, 2018 against 2021 against 2022, because a bitcoin in 2018 and a bitcoin in 2021 are not the same amount of anything.

Here's the part I want to be careful with, because it's easy to walk away with the wrong lesson. Dollars aren't simply right and bitcoin simply wrong. Look for March 2020 in the dollar chart and it has almost vanished. In bitcoin it's the second-heaviest week of the 259. In dollars it ranks 110th, because bitcoin's typical price that week was under 6,000, and the dollar view can't help but shrink anything that happened cheaply. Yet against its own neighbourhood it stands out in both: 5.2 times the median of the eight weeks before it in bitcoin, 3.2 times in dollars. Neither chart lied about that.

So the rule is narrower, and more useful, than "always convert". Over a short window, where price hasn't moved by much, reading bar to bar works in any unit, because every bar is measured on roughly the same scale. Across eras, where price has moved by multiples, the raw count stops being comparable and you need a common scale before the bars can be set against each other. Neither unit is the truth. Each answers a different question, and the mistake is asking one of them the other's question.

#The Same Scale

The fix is the one I described in Tool 01's failure modes: convert each bar into dollars before comparing them. To make the conversion, I multiply each bar's volume by its typical price,

with , and the bar's high, low and close. It's an approximation, since the real figure would weight every trade by the price it actually happened at, but it's close enough to compare bars by eye, and it's what every dollar figure in this series uses, Tool 01's included. If it helps to think of it in physics terms, it's normalisation. Each bar gets rescaled by its own factor so they can be laid side by side, the same way you'd normalise a set of wave functions before comparing them, and that comparison is the whole reason you're reading the bars in the first place.

Bitcoin is just the example. The same thing happens wherever volume is reported in units of the thing traded and the price of that thing moves a long way. A share that's gone up twentyfold trades a twentieth as many shares for the same money. A stock split does the opposite overnight: a 10-for-1 split multiplies the share count by ten without anyone trading more, which is why charting platforms quietly adjust split history for you. Futures report contracts, and contract sizes differ wildly between venues and products, which Read the Spec Sheet walks through for CME against the perps. In every case the question is the same: is this bar counting the same amount of anything as the bar I'm comparing it with?

#Where It's Measured

Equities have something like a consolidated tape, a single record that pulls together trades from across the venues. Crypto has nothing of the kind. Every exchange reports only its own trades, so the volume under your chart is the volume of the venue you happen to be looking at, and an index ticker that blends several venues' prices usually carries little or no volume at all, which is why Tool 01 tells you to anchor on a venue ticker.

That matters because moves don't happen everywhere at once. They start on whichever venue the size is sitting on, and the others follow because arbitrage drags their prices into line, the effect behind The Long Shadow of Markets. (I learned this the practical way before I knew the word for it: my first real trading win was a cross-exchange arbitrage at the December 2017 peak, and the gaps between venues that made it possible are the same gaps that make one venue's volume a partial record.) The follower's price moves with the leader, but its volume doesn't have to. Tool 01's March 2020 example is the clearest case I know, where BitMEX's derivatives led the collapse and the spot venues followed with very different volume.

Three more splits are worth keeping in mind. Spot and derivatives volume are different populations doing different things, and comparing them is the derivatives panel's job, so it waits for Tool 02. For equities, a large share of trading happens off-exchange in venues your chart may or may not include, which is the subject of Part 7. And even on a single venue, the same asset trades against several quote currencies, so activity can move from one pair to another for reasons that have nothing to do with the market. Which brings us to the part of this post I care most about.

#When Volume Lies

A trade gets recorded whether or not it moves anything. Put a resting order on the book and send a market order into it from another account you control, and a trade prints, volume goes up, and price doesn't have to move at all, because you were on both sides. That's a wash trade. The only thing stopping anyone from doing it all day is the trading fee, which makes every fake trade cost real money.

So what happens when the fee goes to zero?

On 8 July 2022, Binance removed trading fees on thirteen bitcoin spot pairs, BTC/USDT and BTC/BUSD among them. On 22 March 2023 it brought them back on all of them except BTC/TUSD, which stayed free. Nothing about bitcoin changed on either date. Here is what the volume did.

Over the 26 weeks before the change, Binance's BTC/USDT averaged 410,655 BTC a week, about 3.1 times Coinbase. Through the zero-fee weeks it averaged 1,830,253, about 10.6 times Coinbase. In the 26 weeks after, it averaged 282,492, back to about 3.6 times. The number of trades went from roughly 8.7 million a week to 43.7 million, while the average trade barely changed size, about 0.047 BTC before and 0.042 during. The extra volume came from five times as many trades of much the same size.

By the day, the edges are almost comically sharp.

On 7 July 2022, BTC/USDT traded 85,015 BTC and closed at 21,625. On 8 July it traded 403,082 and closed at 21,595. Nearly five times the activity, and price didn't notice. At the other end, it traded 420,930 BTC on 21 March 2023, 224,113 on the 22nd, and 50,844 by the 25th, while BTC/TUSD, the one pair still free, went from 185 BTC on the 21st to 4,108 on the 22nd and 6,499 on the 23rd. (BTC/TUSD hadn't traded at all for five months before 11 March, which is its own small footnote.) Coinbase did rise on the 22nd as well, but that was a Federal Reserve rate decision, the kind of event that's supposed to move volume, and it rose on a venue with no fee change.

I can't tell you how much of Binance's block was wash trading. Some of it surely was, and some of it was real trading that only made sense at zero cost: arbitrage bots and market makers churning back and forth because it was free to. The data can't separate the two, and I'm not going to pretend it can. It doesn't matter for reading the chart, though, because either way those bars aren't comparable with the bars around them. Describe the footprint, and treat intent as inference.

That footprint gives you a diagnostic you can carry anywhere. Genuine climaxes are spikes. They cluster around price events, they're steep on both sides, and they show up across venues. Policy distortions are blocks. They have flat tops, and their edges fall on calendar dates rather than on anything price did. When volume changes at a moment when price didn't, go looking for a reason that isn't on the chart.

There's one more twist worth spelling out. Turn the candles off on Binance's chart across late 2022 and you'd read the biggest accumulation in its history, right at the cycle low. The cycle low was real. Coinbase confirms it with the FTX spike, an honest capitulation on an honest venue. So the distorted reading pointed the right way, and was wrong about the size by close to an order of magnitude. That's what makes it dangerous. A reading that's obviously wrong gets thrown out. One that's right for the wrong reason gets trusted next time.

Part 4 already listed the general checks: compare across venues, be suspicious of prints with no price impact, and be suspicious of volume that ramps too neatly. This is what those checks look like in the wild.

#Volume at Price

Everything so far has read volume along the time axis, one bar per period. Turn the same trades on their side and add them up by price instead, and you get a volume profile: a horizontal histogram of how much traded at each level over some stretch of time. On TradingView it's the visible-range volume profile, the VPVR most traders know it as, and it's the picture behind Part 3's observation that markets drift back to where volume built up.

A profile has three landmarks. The point of control is the single price with the most volume. The value area is the band around it holding most of the volume, conventionally 70%. And the high- and low-volume nodes are the bulges and gaps in between, prices where the market spent a long time doing business and prices it passed through in a hurry.

Three things to know before reading one. The first is that a profile is an estimate. Without trade-by-trade data, a platform spreads each bar's volume across the bar's range, so a day that ran from 20,000 to 22,000 gets its volume shared out evenly between them, however it actually traded. Lower-timeframe bars make the estimate sharper, which is why TradingView builds its profiles from them, one-minute bars for a daily session, and every profile in this series is built from daily bars spread evenly across each day's range, and says so. The second is colour. TradingView splits each row into two shades, yellow and blue by default, for up and down volume, and they carry the same caveat as the bar colours earlier: a bar's volume counts as up when that bar closed at or above its open, so the split follows the candles' direction and says nothing about who was more aggressive. The third is the window. The VPVR recalculates for whatever happens to be on your screen, so scroll or zoom and the profile changes under you, and two people looking at the same chart can come away with two different points of control. That's fine for a quick look and useless for a record, so this series reads profiles over a fixed window with its dates written down.

Then the lessons from the bars come back, louder. A profile adds bars together, so whatever distorted the bars distorts the profile, except that now the distortion is baked into a level you might act on.

This is the window from the candles-off experiment, every Coinbase day from the December 2017 top to the November 2022 low, added up by price. In bitcoin, the point of control sits at about 19,300, with a second heavy shelf down around 9,000 to 10,000, where the market spent much of 2019 and 2020. Count the same trades in dollars and the point of control climbs to about 47,200, in the middle of the 2021 range. Same trades, and a level two and a half times higher. If you were drawing support and resistance off that profile, the unit you happened to have selected would decide where you drew it.

And here's the zero-fee block seen from the side. Over 2021 to 2024, with both venues counted in dollars so the units effect hits them equally, Coinbase's point of control sits at about 61,500, in the range where the most money actually changed hands. Binance's sits at about 23,500, inside the price zone of its zero-fee months, and a third of all its dollar volume over the four years sits below 25,000, against 13% on Coinbase. That heavy shelf is what free trading looks like once you add it up by price. Anyone reading Binance's profile on its own would see the strongest support of the cycle at a level built largely out of churn.

So a profile is only as good as the bars it's made from. The venue question, the units question and the fake-volume question all carry over, and the window is a new one on top. What a profile is for, anchoring one to a range and reading how price behaves at its nodes, is procedure, and procedure belongs in a Tool. All this post needs is for you to know what the picture is counting.

#Traps

A few smaller ones, all of which I've walked into at some point.

• The live bar is partial. The current bar is still being filled, so comparing it with closed bars always makes it look quiet early in the period. Compare closed bars only, which is the same rule Tool 01 applies to closes.
• Compare like with like. Crypto trades through weekends, when volume thins out, and equities close for holidays. A Saturday against a Wednesday, or a holiday week against a full one, is a comparison of calendars. Daily vs Weekly Market Closes covers where those bars start and end.
• Continuous futures roll. A continuous futures series like COMEX:HG1! stitches one contract onto the next, and volume migrates from the expiring contract to the new one around each roll, so the series shows drops and jumps that are scheduling. It's why Setup 01's copper figures hide volume entirely.
• Thin markets. On a small-cap or a quiet pair, a single large print can make a bar that looks like a climax and is really one order. Check the bar on a lower timeframe before treating it as a crowd.
• Footprint and intent. Volume tells you that people acted. It never tells you why, and every story about who was buying and what they knew is inference layered on top. Keep the two apart in your notes, or you'll end up defending the story instead of reading the bars.

#How This Series Reads Volume

The working rules, written plainly, for the Tools to link to.

1. Read volume relatively. A single bar means nothing on its own. Compare a bar with the one before it () to see whether participation rose or fell, and against the average of a run () to judge a phase or call a climax.
2. Use the venue's own units by default. Figures show volume as the venue reports it, because that's what you'll see on your own chart. Over short windows those units are fine.
3. Normalise across eras. When price has moved by multiples across the window you're comparing, convert to dollars first: , labelled as approximate.
4. Know what one unit is. Shares, coins or contracts, and for contracts, what one contract is worth on that venue (see Read the Spec Sheet).
5. Use the venue where the move happened. Anchor on a venue ticker with real volume, never an index, and when venues disagree, find which one led.
6. Read profiles over a fixed window. Volume at price is read over stated dates, never the visible range, and it's an estimate built from the bars, so the venue, units and fake-volume checks apply to it too.
7. Distrust blocks. A step in volume whose edges fall on calendar dates is a policy, a promotion or a listing until proven otherwise. Check it against a venue that didn't change anything.
8. Compare closed bars, and like with like. No live bar, no weekend against weekday, no roll week against a normal one.
9. Climax checks go top down. On a higher-timeframe range, look for the climax on the weekly first, then the daily, then the 4h, and never lower, as Tool 01 sets out.
10. Record the footprint and label the inference. What traded is data. Why it traded is a hypothesis.

Turn the candles back on whenever you like. They're useful, and Tool 01 can't draw a range without them. But every so often, hide them and look at what's underneath, because the bars that look quietest sometimes carried the most money, and the loudest ones are sometimes the cheapest thing on the screen.