Point and Figure Chart: The Column Rule Our Engine Runs

A point and figure chart is the oldest way of drawing price without drawing time. There are no bars and no sessions — only columns of X for advances and O for declines, each one stack of fixed-height boxes, each new column meaning the market changed its mind by enough to be worth recording.

Our chart engine ships that transform, and this page is the rule it runs rather than the rule textbooks describe. Three things decide what you see: how tall a box is, how many boxes it takes to start a new column, and where the grid lines fall. All three are quantified below on 347,171 real daily bars, and one of them is not what the classic literature would predict.

What a point and figure chart is

Columns replace candles. A rising column keeps growing while price keeps clearing whole boxes upward, and it stops the moment price falls far enough to justify a fresh column beside it.

Because the column only advances in whole boxes, everything smaller than a box is invisible. Sixty daily candles can collapse into a single column, and below you can watch exactly that happen: one stack twelve boxes tall carrying a 143.2% advance, with the clock nowhere on the picture.

The x-axis is ordinal — column number, not date. Columns keep the timestamp of the bar that last extended them, so approximate labels are possible, but two neighbouring columns can be a week or a year apart. Across the whole tape our engine averaged 52.6 daily bars per column. That is the compression on offer, and it is also the information being discarded.

The exact column rule our engine runs

The transform is one grid, one direction flag, and four comparisons.

How Vike computes it. point_and_figure in vike-chart opens an up column at floor(close / box) * box, then walks the closes. While the column is rising, a close one full box above its top raises the top to the close's own grid line; a close three boxes below the top ends the column and opens a falling one whose top starts a single box below where the last one finished. Falling columns mirror that exactly. The box comes from auto_box — the mean true range of the series — floored at one fiftieth of the total high-to-low range, and the output caps at 20,000 columns.

I transcribed that rule from the crate, transcribed its named Python twin beside it, then wrote a third version from the description above alone, structured differently on purpose. Across 243 Binance pairs the transcription and the twin agreed on 13,431 values — both edges of every column plus each box — with a worst absolute difference of 1.164e-10, and the from-scratch version reproduced every column on every symbol exactly. A port agreeing with the file it was copied from proves the copy. Only the third one proves the reading.

Two artefacts fall out of the loop that nobody writes down. The first column is always opened as a rising one, whatever price does next: on all 243 pairs the chart began with an X column because the constructor never asks. And a reversal starts one box inside the column it left, so consecutive columns overlap by exactly one row — 6,351 of 6,351 column pairs in the sample, without exception.

The box is coarser than renko's on purpose

Here is the detail that separates our point and figure charts from our renko bricks, which share the same auto_box helper.

Renko takes the average true range and stops. Point and figure takes the larger of that range and one fiftieth of the whole span the series covered, because the crate's own comment says a tiny box "smears into thousands of 1-px columns". The floor binds far more often than I expected: on 165 of 243 pairs — 67.9% — the range term won outright, so the volatility measurement never got a vote at all.

Median ratio between the two boxes is 1.27x, and the extreme is 28.7x. Feed both transforms the identical bars and the difference shows up as unit count: 88,989 renko bricks against 6,594 point and figure columns, thirteen and a half times as many drawing units from one tape.

Bitcoin's box works out at $2,467.65, or 3.88% of the last close, and its whole 3,287-bar history fits in 59 columns. Dogecoin's lands at 13.54% of current price, which is what happens when a box is sized against a decade that included a far wilder era than today.

Three boxes to turn the chart

Reversal count is the second parameter, and it is doing more work than the box.

Our engine uses three, the classic convention. Price must travel three whole boxes against the live column before a new one opens, so two boxes of counter-move draw nothing whatsoever and leave the chart looking untroubled. Below, one POND window is rendered three times over identical candles and an identical box: five columns at three, eight columns at one, and a single unbroken column at five.

That spread is the honest answer to what these charts show. They show whatever the reversal count admits. When I see a point and figure setup quoted without both numbers beside it, I treat the picture as decoration.

What the columns throw away

Time goes first, and it goes completely.

Highs and lows go with it. Only closes are read, so an intraday collapse that recovered by the bell contributes nothing at all — the same blindness renko has, for the same reason, and the reason our candlestick detectors never run on either output.

Precision goes third, and this one is subtler. Every column edge is snapped down to a grid line with floor, so the printed extreme sits a median 0.56 boxes inside the close that set it. On Bitcoin's box that is roughly $1,380 of real move the chart declines to draw.

Watch the IQ window below. Sixty sessions, a 2.3% net drift, one unfinished column — and now compare it with the LUMIA stack further up, also a single column, which covered 143.2%. Same drawing, wildly different market. The count of columns tells you nothing whatever about magnitude, and reading it as though it does is the mistake I made when I first put one of these on screen.

Where point and figure charts fail

Lateness is structural, and it is worse here than on any other chart type we ship.

A reversal needs three boxes of travel before it prints anything. At Bitcoin's box that is nearly 12% of price committed before the chart will concede that the trend changed at all. You pay for the smoothing every time, and the bill arrives at the turn.

Then there is the trap of a column that looks finished and is not. The rightmost column on any such chart is still forming — our transform emits it unclosed, deliberately, because hiding it would be worse. The PYR window below closed on a rising stack eight boxes tall, which is exactly the picture people describe as strength. Price fell 50.0% over the next twenty sessions. Nothing in the construction was wrong; the column simply had not finished, and a tall X column is a record of what already happened rather than a forecast.

And the parameters are the strategy. Change the reversal from three to one and POND's window went from five columns to eight; change it to five and the whole window became one. Any result quoted off this chart is a result about its box and its reversal count first, and about the market second. I skip any backtest that does not print both.

Point and figure vs renko

Both delete time, and they delete it differently.

Renko emits one brick per box of travel and turns after a single box, which makes it granular and jumpy — 88,989 bricks where point and figure drew 6,594 columns on the same bars. Point and figure quantises to an absolute grid instead of stepping from wherever the last brick ended, so its levels are the same for everyone with the same box, and it demands three times the counter-move before conceding a turn.

Practically, bricks read better for following an established move and columns read better for spotting that a range has finally broken. Neither belongs anywhere near execution, since fills happen at prices both transforms rounded away. The heikin ashi page covers the third member of that family, and the trading indicators overview sets out the lag every summary tool shares.

Frequently asked questions

What box size should point and figure charting use? Ours is automatic: the greater of the series' average true range and a fiftieth of its total range. The floor won on 67.9% of the pairs tested, so on most charts the box is a range fraction rather than a volatility measure — worth knowing before comparing our grid to anybody else's.

What is the 3-box reversal? Price has to clear three complete boxes in the opposite direction first. It is our default, and one POND window drew five columns under it, eight at a one-box setting and one at five.

Do point and figure charts repaint? Closed columns never change. The final column is live and can keep growing or be replaced, and because auto_box reads the entire loaded series, two charts with different history lengths can differ everywhere.

Can you use candlestick patterns on point and figure columns? No. Columns have no open, no shadow and no session, so every premise behind candlestick shapes is gone by construction.

Is point and figure better than a bar chart? It answers a narrower question more cleanly. Bar charts record everything and leave the filtering to you; this one filters first, at a cost in precision that the 0.56-box grid snap above puts a number on.

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

See what renko bricks throw away