What Your Trading Chart Is Really Hiding From You

Picture a security camera pointed at a busy street corner. Now imagine it only takes a photo once every minute, regardless of whether anything is happening. Most of the time, you'd catch nothing — an empty pavement, a parked scooter, a pigeon with opinions. Then a wedding parade would march through in the eleven seconds between frames, and you'd never know it happened.

That, roughly, is the bargain every price chart makes. Markets don't move in neat clockwork; they move in bursts of activity. Yet the default chart on nearly every platform — the one-minute, five-minute or daily candle — chops time into equal slices and treats a frantic earnings-day minute exactly like a quiet lunchtime minute. Understanding that bargain is the difference between reading a chart and being quietly misled by one.

First, what is a bar, actually?

At the bottom of everything is the tick: a single recorded trade, with a timestamp, a price and a quantity. Modern electronic markets generate these in enormous volume — billions of timestamped trades, quotes and cancellations across a trading day, as one practitioner overview describes it. Raw ticks are rich but unwieldy, like trying to drink directly from a waterfall.

So we aggregate. We pick a rule that says "close a bar now," then summarise everything inside that window into four numbers: open, high, low and close — the familiar OHLC candle — plus volume. That act of summarising is called sampling, and it is never neutral. As one market-microstructure text puts it plainly: the representation of market data is not a neutral act.

The clock is a choice, not a law of nature

Time bars are the grandfather clock of market analysis: a new bar every minute, hour or day. They're intuitive, universally supported, and give you a fixed number of bars per session — handy if your model assumes evenly spaced observations.

The trouble is that markets are not chronological in any meaningful sense. Volume clusters at the open and close, spikes around announcements, and thins out to almost nothing overnight. A time bar formed during a market freeze looks structurally identical to one formed during a flash of violent repricing. One commentator compares it to a reporter filing a story every hour — even when nothing has happened for days, or so much has happened that it won't fit.

Statistically, this mismatch leaves fingerprints. Returns sampled on a clock tend to show heteroskedasticity — a mouthful meaning simply that the size of price moves isn't constant, because the information arriving isn't constant — along with serial correlation and fat tails. Several sources make this point; a controlled study of Bitcoin perpetual futures notes that calendar sampling conflates periods of genuine price discovery with noise-driven drift.

Letting activity set the tempo

The alternative is to let the market's own pulse close the bar. Tick bars form after a set number of trades. Volume bars form after a set number of shares change hands. Dollar bars form after a set amount of turnover — price times quantity — has traded. Each tries to give every bar a roughly similar dose of market information rather than a similar dose of clock.

Think of it as the difference between a metronome and a seismograph. One source uses exactly that framing: uniform time sampling is the metronome, event-driven sampling the seismograph. During a quiet spell, activity bars simply take longer to form; during a frenzy, they arrive thick and fast.

There are subtler variants too. VWAP bars use the volume-weighted average price inside the bar as the close. Imbalance bars and run bars close only when buying and selling pressure diverge beyond an expected threshold — an attempt to capture intention, not just quantity. Range and Renko bars close on price displacement instead, filtering out small noise.

Dollar bars deserve a special mention for a practical reason: raw share volume isn't comparable across time for an asset whose price has moved by orders of magnitude. Turnover normalises for both price level and quantity at once.

Does the evidence actually favour them?

Partly — and the honest answer is "it depends," which is less satisfying than a slogan but more useful. One widely cited microstructure text reports dollar bars achieving markedly better return normality than time bars in an equity example, and treats them as a sensible default for many machine-learning workflows. A 2020–2025 study of Bitcoin perpetual futures found that the advantage of tick-level construction was bar-type specific: tick Renko bars showed the smallest deviation from random-walk behaviour, and tick volatility bars cut serial dependence substantially versus a minute-based baseline.

But the same study offers a useful splash of cold water. When tick series were coarsened to match the minute pipeline's bar count, much of the apparent tick underperformance on distributional measures vanished — suggesting some of it was a sampling-frequency artefact rather than a data-quality one. Normality improvements were also regime-dependent, and formal independence tests were rejected across the board at the sample sizes studied. Translation: better-behaved, not magically well-behaved.

Where repainting sneaks in

This is where charts can quietly lie to you. Repainting is when a bar or indicator that looked one way in real time settles into a different shape after the fact. The most mundane version: because a bar's high and low are only final at closure, a chart drawn mid-bar shows a moving target. Watching a live candle is a bit like judging a cricket score mid-over — dramatic, provisional, and not what the record book will say.

A more consequential version comes from resolution. If you build activity bars from pre-aggregated one-minute OHLCV data instead of raw ticks, the accumulation is necessarily coarser. Three intra-minute reversals of 0.05% that cancel out contribute roughly nothing to a close-to-close volatility measure, yet add up on a tick-based one. Minute-based Renko construction can only check its displacement condition at minute boundaries, so it may overshoot the intended brick size — whereas a tick pipeline detects the crossing at the exact trade. The bar you see is a slightly different object from the bar you think you see.

The traps worth knowing

Sampling always loses something: four prices cannot preserve the precise sequence of trades or transient spikes inside a bar. Bar size matters too — too small and you import microstructure noise like bid-ask bounce, inflating measured volatility; too large and you smooth away real moves. Irregular bars also complicate correlation estimates, since standard tools assume evenly spaced observations.

There's a cross-asset wrinkle as well. Volume bars on two correlated instruments will rarely align in time, which can manufacture artificial lead-lag relationships. And tick bars have their own blind spot: they treat every trade equally, so a flurry of one-share orders can trigger more bars than a single large block that actually moved the market. More ticks is not the same as more significance.

Finally, jurisdiction and market structure matter. Continuous crypto markets, with no opening bell, make the calendar-sampling problem especially acute. Equity markets have sessions, auctions and halts that change what a bar even means. Vendor data conventions differ. None of this is portable without checking your own context.

So how should you look at a chart?

Not with suspicion — with literacy. Ask three questions of any chart: What rule closed each bar? What data resolution fed it? And would this pattern survive if I resampled it a different way?

Time bars remain perfectly reasonable for human-readable charts and for models that assume fixed intervals. Activity bars can offer more uniform information content, which is why researchers argue they reduce look-ahead and serial-correlation biases that inflate in-sample results. But every scheme is a lens, and one source says it best: neither view is wrong — they're different perspectives on the same underlying reality.

Technical analysis itself has a long, contested history, and the efficient-market view holds that prices are essentially unpredictable. Research on whether chart-based methods help has produced mixed results. Sampling choices don't settle that argument. They just make sure you're arguing about the right picture.

The candle isn't the market. It's a photograph of the market, taken by a camera you chose to point.