Renko Charts: How Bricks Are Built, and What They Throw Away

A renko chart throws the clock away. Instead of drawing one candle per session, it draws one brick per fixed amount of price travel, so a trending week becomes a staircase and a flat month can vanish into nothing at all. The chart looks decisive because everything indecisive has been deleted before you see it.

Deleting is the right word, and this page measures what gets deleted. Below is the exact brick rule our chart engine runs, what the box size does to the same tape when you move it, and three things a renko chart cannot show you — each quantified on real Binance daily bars rather than asserted.

What a renko chart is

One brick means the close travelled one full box beyond the last brick's edge. Nothing else draws anything.

The x-axis is ordinal — brick number, not time. Each brick keeps the timestamp of the daily bar that completed it, so the chart can still print approximate dates, but equal widths on screen can be a day apart or a quarter apart. In the example below, forty ATOM candles covering a 31.7% decline compress into twelve red bricks: the chop is gone, the staircase is what remains.

The exact renko rule our engine runs

The whole transform is a starting point, a box, and two while-loops.

How Vike computes it. renko in vike-chart anchors at the first bar's close, then walks the series reading closes only: while a close sits a full box above the last brick edge it emits an up brick [last, last + box], while a full box below, a down brick [last, last − box], stepping the edge each time and capping the output at 20,000 bricks. The box comes from auto_box: the mean true range of the entire series, falling back to 0.5% of the last close if that degenerates. Bricks carry no wicks by construction — an up brick's high is its close, its low is its open.

Notice what never appears in that loop: the bar's high and low. A $20,550 intraday round trip — the widest bar Bitcoin printed in the sample, on 2025-10-10 — contributes nothing unless the close moved, because the excursion happened between prices the loop never reads.

I transcribed that rule from the crate source, then transcribed its named Python twin, and ran both over fourteen symbols with a thousand daily bars each: 12,938 compared values across every brick edge and the box itself, worst absolute difference 1.819e-12. The residue has a cause worth naming — CPython's built-in sum() compensates floating-point error (Neumaier summation, since Python 3.12) while the Rust source accumulates in a plain running total, so the two boxes differ by one unit in the last place and roughly two hundred chained brick edges amplify that ulp into the twelfth decimal. Same rule, one rounding philosophy apart.

The box size decides the chart

Every property a renko chart appears to have is really a property of its box.

On Bitcoin's last thousand daily bars, auto_box works out to $2,841.56 — 4.49% of the final close. Shrink that box to a quarter and the same tape draws 1,460 bricks with 261 direction changes; double it and you get 73 bricks with 30 changes; quadruple it, 20 bricks and 7 changes. Between the extremes the chart tells five different stories about one price history, and none of them is the neutral one, because there is no neutral box.

Below, the same forty ATOM candles are drawn at half the box and at double it. Twenty-six bricks with two reversals versus five bricks with none — a scalper's tape and a serene trend, from identical data.

There is a second subtlety I nearly missed when I first measured this: the automatic box averages true range across everything on the chart. Load more history and the box changes, so today's bricks are partly sized by volatility from years ago. Dogecoin makes the failure vivid — its box lands at 18.03% of the current price, because the average still remembers a far wilder era, and bricks that tall barely move at today's volatility.

What the bricks throw away

Time, first. Across all fourteen tapes, 11,469 of 14,000 daily bars produced no brick whatsoever — just under 82% of all sessions left no mark on the chart. The longest silence ran 120 days, and the opposite extreme is just as distorting: a single ADA session in March 2025 emitted twelve bricks at once. On the renko axis those twelve bricks occupy the same visual space as twelve quiet months.

Wicks, second. Bricks have none — a rising brick's top is its close, and the mirror holds on the way down, always. Intraday rejection, the raw material of every candlestick pattern, does not survive the transform, which is why our pattern detectors run on raw candles only and a hammer "found" on bricks is fiction.

Watch the DOGE window below: forty candles of churn, three bricks — deletion at its most extreme.

One box reverses it — and that is unusual

Here is the detail to check before comparing our bricks against another platform's.

Classic renko convention demands a two-box counter-move before drawing the first brick of a reversal, which filters one-box noise at the cost of even later turns. Our source reverses on a single box. Once the close travels a full box past the last edge in the opposite direction, the brick prints. Measured on Bitcoin at the automatic box, the one-box rule drew 217 bricks with 67 direction changes; the classic two-box variant drew 150 with 44. Roughly half again as many reversals — same tape, same box, different convention, visibly different chart.

NEAR, below, is the rule at its best: a 74.7% advance rendered as ten bricks, nine of them green.

Where renko charts fail

Lateness is structural. A brick prints only after a full box of travel has already happened, so at Bitcoin's automatic box every fresh signal starts about 4.5% late by definition — that is the entry cost of the smoothing, paid every time.

Risk is invisible. Because wicks are deleted and quiet bars vanish, a renko chart cannot tell you what the excursion inside a move looked like, and stop placement made off bricks alone inherits that blindness. I learned to keep the raw chart open beside the bricks for exactly this reason: one shows the story, the other shows what it cost to sit through.

And the parameter is the strategy. Any backtest on renko is mostly a backtest of its box; shift the box and the 261-versus-7 reversal spread above says the signal count can move by 30x. When I see renko results quoted without the box size and the rule for choosing it, I stop reading.

Renko vs candlestick charts

Candles record everything and ask you to filter; renko filters first and shows you the remainder. That makes bricks a genuinely good display for trend inspection — and a poor sole basis for execution, since fills, stops and patterns all live on prices the bricks discarded.

One workable division of labour: bricks for structure, a close-based tool like the MACD for momentum, the raw tape for execution. The heikin ashi page makes the same argument for a different transform — smoothing and deletion are cousins, and both charge for their clarity in information.

Frequently asked questions

What is the best brick size for a renko chart? There is no neutral answer — that is the finding, not an evasion. Our engine defaults to the loaded series' average true range (4.49% of price for Bitcoin daily, measured above). Whatever you choose, publish it next to any result; the same tape produced between 20 and 1,460 bricks depending on that one number.

Does a renko chart repaint? The current brick's edge is fixed, and closed bricks never change. But two renko charts with different history lengths can differ everywhere, because auto_box reads the whole series — more history, different box, different bricks throughout.

Why does my renko chart differ from another platform's? Three suspects, in order: the box size, the reversal convention (ours draws a reversal after one box; the classic rule waits for two), and tick-based versus close-based construction — ours reads daily closes only.

Do candlestick patterns work on renko bricks? No. Bricks have no wicks and no session boundaries, so the premises behind candlestick patterns — rejection at real intraday extremes — are deleted by construction. Trend tools like the ichimoku cloud also assume one bar per period and misbehave on an ordinal axis.

This is educational material, not financial advice. A chart type changes what you see, never what the market does — every renko figure here was measured on past data, and trading carries real risk of loss.

See how heikin ashi smooths the same tape