Point and figure chart backtests: box size, reversal and the breakout you missed

Contents

1. What is a point and figure chart, and why do its backtests overstate breakouts?
2. How is a point and figure column built from real bars?
3. Where does a point and figure backtest see the future?
4. The box price is not a tradeable price
5. The same-bar ordering guess
6. Stops on a chart with no lows
7. How does AlphaPrime backtest a point and figure strategy?
8. How do you choose box size and reversal for a backtest?
9. Which markets and strategies suit point and figure?
10. Frequently asked questions

What is a point and figure chart, and why do its backtests overstate breakouts?

A point and figure chart records price in columns of X boxes (rising) and O boxes (falling), adding a box only when price moves a full box size and starting a new column only after a reversal of several boxes. Its backtests overstate breakouts because they fill at the box price, a level the market has already moved through.

Point and figure, often written P&F, is the oldest of the synthetic chart types and the only one that most traders learned from a book rather than a platform. A box size is the fixed price step one X or O represents. A reversal amount is the number of boxes price must move against the current column before a new column starts; three is the traditional setting. Time does not appear on the chart at all. A column can last an afternoon or a quarter.

P&F signals are patterns of columns: a double top breakout when an X column rises one box above the previous X column's top, a double bottom breakdown when an O column falls one box below the previous O column's bottom, and their triple and catapult variants. The patterns are clear and the chart is quiet. The way most engines backtest it leans on two assumptions that do not hold live. The rules shared by every synthetic chart type are on the overview of why synthetic-bar backtests cheat; this page is the P&F-specific part.

How is a point and figure column built from real bars?

A column is extended when a completed underlying bar's price reaches the next box level, and it reverses when a completed bar's price has moved the reversal amount of boxes against the column. The chart is built from bars, not from a continuous line, and that detail decides whether the backtest is honest.

Two construction methods exist. The close method looks only at each bar's close, so a bar can either extend the column or reverse it, never both. The high-low method uses the bar's high to extend an X column and its low to test for reversal (and the opposite in an O column). The high-low method draws more boxes and more signals, and it has a built-in problem: when a single bar's range is large enough both to extend the column and to reverse it, the method needs to know which happened first, and the bar does not say.

AlphaPrime (Valynox) builds P&F columns from completed underlying bars, in either the close or the high-low method, and applies two rules that most chart engines do not. Rule one: every new box hands the strategy a bar, visible at the close of the underlying bar that reached that box, and nothing is drawn before that close. Why: a box, once added, can never be removed, so it is safe to act on; a box that the current bar might still reach is not. What it avoids: breakout signals that were there on the chart at one moment and gone the next, and fills at a box the bar had not yet reached. Rule two, for the high-low method: when one bar could both extend the column and reverse it, the reversal counts only if it would happen under both possible orderings; otherwise the column simply extends. Why: inside one bar the sequence of high and low is unknown, and an engine that picks the order that completes the pattern is choosing the outcome. What it avoids: reversals and breakouts manufactured by the engine's guess, at the price of recognising some real reversals one column late.

Where does a point and figure backtest see the future?

In three places: the fill at the box level, the same-bar ordering, and the stop check against a chart that has no intrabar prices at all.

The box price is not a tradeable price

A breakout box at 130 means price reached at least 130 at some point in the bar. It does not mean you could buy at 130. On a daily-built P&F chart the real entry is the next day's open, which can be several boxes away. Filling at the box turns every breakout into a perfect entry, and P&F breakout strategies backtested that way show win rates that live trading never reproduces.

The same-bar ordering guess

Described above, and worth repeating because it is unique to P&F and to a lesser degree Kagi and Line Break: when an engine must choose between extend-then-reverse and reverse-then-extend inside one bar, it is choosing the trade. Any method that picks the order that favours the pattern is manufacturing signals.

Stops on a chart with no lows

A P&F chart shows boxes, not highs and lows. If a stop is judged only on whether a box was printed against you, the stop is hit only when price has moved a full box past it, and sometimes only when a full reversal column has formed. A stop that is checked on the real underlying bar's low will be hit much more often, and that is the number you will see in your account.

P&F feature	Box-price backtest	AlphaPrime
Box confirmation	Drawn the moment price touches the level, inside the bar	Each new box becomes a bar at the close of the underlying bar that reached it; never undrawn
Same bar can extend and reverse	Engine picks an order, usually the one that completes the pattern	Reversal only if both orderings reverse; otherwise the column extends
Entry fill	The breakout box price	Next real price plus doubled slippage
Stop check	Only when a box prints against the position	Real highs and lows of the underlying bars, stop first

How does AlphaPrime backtest a point and figure strategy?

Patterns are read from the columns; fills, stops and targets come from the underlying bars. Each rule, its reason, and the error it removes:

• Every new box is a bar, confirmed at the close of the underlying bar that reached it. Why: a box can never be removed once added, and the forming bar's close is unknown until it prints. Avoids: signals from boxes that were drawn and then undrawn, and signals that change when you reload the chart.
• No intrabar ordering guess. Why: the bar does not record which came first; a reversal counts only if both orderings produce it. Avoids: patterns completed by the engine rather than by the market.
• Market orders fill at the open of the next underlying bar, plus slippage. Why: that is the first price available after the signal exists. Avoids: fills at box levels the market had already left.
• Stops and targets are tested against the real highs and lows of the underlying bars, stop before target. Why: the real bars contain the prices the chart threw away. Avoids: stops that the chart says were never touched and the broker says were.
• Slippage starts at twice the time-bar figure. Why: a P&F signal arrives after a multi-box move, when the market is already running. Avoids: pricing a breakout at a quiet-market spread.

The slippage multiplier for P&F is two and can only be raised, because one underlying bar can draw several boxes and even a whole column, the same clustering that range and Renko bars show. On exits there is a larger cost that is not slippage at all: a reversal is known only after price has moved the reversal amount of boxes against the column, so the exit the chart draws at the column's top is really taken three boxes lower, plus the gap to the next real open. The honest fill shows that distance; the box-price fill hides it. The general method for weighing costs against the average trade is on the page about commission, slippage and the average trade.

How do you choose box size and reversal for a backtest?

Choose a box that your underlying bars can resolve and a reversal that fits the holding period you want; the traditional three-box reversal is a reasonable default, and the box size is where the real decision sits.

1. Box size relative to the underlying bar. A box much smaller than the typical bar range means most bars draw several boxes and the same-bar ambiguity appears on most bars. A box of roughly one average bar range or more keeps the chart honest.
2. Fixed versus percentage boxes. Fixed boxes suit a single instrument over a short history. Percentage boxes keep the chart comparable across years of a rising market, and the percentage must be applied to a price known at the time, not to the current price.
3. Reversal amount. Three boxes filters most noise; one box gives a chart that reacts to everything and is rarely worth backtesting; five boxes suits long holds. Change the reversal only together with the holding period you are testing.

Keep box and reversal fixed across in-sample and out-of-sample. The right pair is the one where the rule still works when you move the box up or down by a step. If it only works at one box size, the pattern is fitted to the chart. Kagi and Line Break make similar decisions with different names; their pages are Kagi chart backtests and Line Break chart backtests.

Which markets and strategies suit point and figure?

P&F suits slow trend and breakout rules on liquid markets with long histories: index futures, major currency pairs, metals, and the larger cryptocurrencies on daily or hourly underlying bars. It is a poor fit for intraday scalping, where the box-to-fill gap dominates, and for thin markets where a single print can draw a column.

Before trusting a P&F result, compare the close-method and high-low-method versions of the same rule. If the high-low version is much better, the difference is mostly the same-bar ordering guess, and it will not survive real fills. Then run the rule with the box one step larger and one step smaller. A pattern that holds across three box sizes has something in it. A pattern that needs one exact box is a coincidence drawn in X's and O's. If it holds, the next step is walk-forward analysis, exactly as for a time-bar strategy.

What this means if you use AlphaPrime: choose point and figure as the chart type in strategy generation, set the box size, the reversal and the underlying resolution, and run the search. Every candidate is tested with boxes confirmed at the underlying close, no intrabar guessing, next-real-price fills, doubled slippage and real-bar stop checks, in Classic and in Reserve. The breakouts that remain are the ones you could actually have bought.

Frequently asked questions

How do you read a point and figure chart?

Each column is a run in one direction: X boxes stack while price rises by at least one box size per box, O boxes stack while it falls. A new column starts only when price moves the reversal amount of boxes the other way. A breakout is an X column that rises above the previous X column's top; a breakdown is an O column that falls below the previous O column's bottom. Time is not shown.

What box size should I use for point and figure?

One that your underlying bars can resolve: roughly the average range of the bar you build from, or larger. Smaller boxes draw more signals and more same-bar ambiguity. Test one step larger and one step smaller; keep the setting fixed across in-sample and out-of-sample.

Can point and figure charts be backtested at all?

Yes, if the engine confirms each box at the close of the underlying bar that reached it, fills at the next real price rather than the box level, and does not guess the order of highs and lows inside a bar. Without those three rules the backtest is a drawing exercise. With them it is a test like any other.

Does point and figure work for forex and crypto?

The chart works on any price series. For forex, set the box in pips; for crypto, use a percentage box, because prices move across orders of magnitude over a few years and a fixed box becomes meaningless. The same fill and stop rules apply.

Why does my P&F backtest show fewer signals in AlphaPrime?

Because boxes are confirmed at the close of the underlying bar that reached them, and because the engine does not pick the friendlier order when one bar could both extend and reverse the column. Signals that depended on a mid-bar box or on that guess disappear. The signals that remain are the ones that were there, and final, in real time.