Why Closing Prices Matter: Close vs High/Low Volatility

Every bar on a chart carries four prices: open, high, low and close. Most volatility measures, and most of the bands built from them, use only the last of the four. That looks like a waste of information, and traders reasonably ask why a wick that reached well past a band should count for nothing while a close a fraction inside it counts as "contained."

The answer is not that the close is the most important price. It is that the close is the one price a series can be built from, and a volatility measure is a statement about a series. Which price feeds the series decides what the measure is measuring — and what any test of it can honestly claim.

A bar has four prices; a return series uses one

Volatility, in the sense a band uses it, is the dispersion of returns: the percentage change from one observation to the next. A return needs two observations of the same kind in sequence. Yesterday's close to today's close is one such pair. Yesterday's high to today's high is another in principle — but a bar's high is the single most extreme tick in the period, not a price the market settled at, and two consecutive extremes do not describe a move from one state to another. A chain of closes is the only chain in which each link is a settlement rather than an excursion.

That is why the standard deviation of close-to-close returns is the default input for a rolling volatility estimate, and why the working paper builds its range from exactly that: a rolling window of recent close-to-close returns, projected forward from the prior close. The mechanics of the window are the subject of what is rolling volatility; the yardstick itself — one standard deviation as the unit of "normal" — is walked through in is this price move normal.

This is a different choice from whether a band is built on returns or on raw price levels, which is the subject of return space vs price space. This article is about the other axis: given that the input is a series, which of the bar's four prices supplies it.

What a close-scored test counts, and what it ignores

A band built from closes makes a claim about closes: that a stated fraction of next closes should land inside the range drawn from the prior close. The test reads each bar's close and classifies it as inside or outside. The high and the low never enter — not the construction, and not the scoring.

Two consequences follow. First, the published figures are figures about closes. Historically, across the instruments tested in the working paper, the next close landed inside the inner band about 71% of the time and inside the outer band about 94%; the tables are on the Proof page. Past behavior is not a guarantee of future results. Those numbers say nothing about how often a bar's high or low crossed the band, because that was not the quantity scored.

Second, a wick beyond the band with a close back inside it was, in that test, an inside observation. This is not a loophole; it is what "expected range for the next close" means. A close-scored test is deliberately blind to the path a bar took to reach its close, and a band that reported wick crossings as failures would be answering a different question than the one it was built to answer.

Two questions, two kinds of measurement

There are two families of quantity a trader might want a volatility measure for.

The first is close-to-close: where did the market settle relative to where it settled last time? Returns, drift, the dispersion of daily moves, and the expected range for the next close are all in this family.

The second is intraday: how far did price travel within the bar, how deep did it push past a prior high or low, how wide was the session? These are questions about highs and lows, and a measure built from closes does not contain the information to answer them. Average True Range does — it takes each bar's high-low span, extended to the prior close so that a gap counts as travel, and averages it — which is why ATR is the natural width for a session-range tool and why our expected-range guide for TradingView describes ATR-based ranges as a legitimate family with a different purpose, not a mistake.

We have measured this boundary rather than assumed it. In a study of order blocks and liquidity sweeps, an ATR-normalised measure calibrated excursion depth past a prior extreme — an intraday quantity — more consistently across markets than our own standard-deviation measure did. The rule that result established is the one this article rests on: standard deviation calibrates close-to-close quantities; ATR calibrates intraday ones. Neither is better in general, and a claim that a close-based measure calibrates intraday quantities better would be wrong as stated.

Why more information did not make a better band

The natural next thought is to have both: keep the close as the scored quantity but let the highs and lows improve the volatility estimate feeding the band. Range-based variance estimators do exactly that, and in the textbook they are several times more efficient at estimating a single bar's variance than a close-to-close standard deviation.

When the working paper rebuilt the band on three of them, inner-band containment fell by roughly seven points and its consistency across markets got substantially worse; the result and the estimators involved are set out in what we found testing the popular bands. The lesson was narrower than "closes are better": a more efficient estimate of one bar's variance is not a better-calibrated range for the next close, because the extra information was about something other than the quantity being scored.

What this means on a chart

A candle whose wick leaves the range and whose body returns inside describes a bar that travelled outside the expected range intraday and settled inside it. The excursion is real information about the bar's path, but not information the band's calibration record speaks to. A candle that closes outside the range is the event the record was measured on, and the Proof page framing applies to it: an expected range, not a claim that price reverses at its edges.

Some traders use that separation deliberately — reading the close-scored range as the answer to "was this settlement ordinary?" and a range-based measure as the answer to "how much ground did the session cover?" — and keep both on the chart because the questions differ. That is a description of use, drawn from the use list on the How It Works page, not a recommendation. The working paper validates the range's calibration on closes; it does not validate the profitability of any use built on that reading, and whether any such use delivers value after costs is an open question.

One further consequence: a chart type that changes what a "close" is changes the series. Non-standard charts that synthesise their closes feed the band a different input than the one the published record was measured on; the Behavioral Transform Model explainer covers which chart types those are.

The takeaway

The close matters to a volatility band not because it is a privileged price but because it is the price a return series can be built from, one settlement to the next. A band built from closes makes a claim about closes, is scored on closes, and is silent on wicks by design. Highs and lows belong to a different family of question — intraday travel and excursion — where a range-based measure such as ATR is the matched tool, and we have published the result that says so against our own measure. Knowing which of the four prices feeds an indicator is the fastest way to know which question it answers.

Frequently asked questions

Which closing price does the indicator use on a market that trades 24 hours? The close of whatever bar the chart is showing, as defined by TradingView's session and timezone settings for that symbol. On a 24-hour market such as crypto or spot forex, a "daily close" is a convention set by the data feed rather than a bell, so two charts of the same symbol with different session settings can produce slightly different close series and therefore slightly different bands. The containment figures in the working paper were measured on the daily series from the data sources itemised on the FAQ page.

Can the source be changed from close to HLC3 or typical price? On TradingView many indicators expose a source input, and a user can point it at an average of the bar's prices instead of the close. Doing so changes the series the band is computed on and the series it would be scored against, so the published statistics — which were measured on closes — do not carry over to that configuration. The How It Works page lists source alongside symbol, timeframe and window as the settings that change the structure drawn.

Does the indicator work on a line chart? Yes. A line chart plots closes only, which is precisely the series a close-based band uses, so nothing the model needs is missing; the FAQ page lists standard candle, OHLC bar, line and area charts as supported. What a line chart hides is the wick — the intraday excursion — which, as described above, the band's calibration record does not speak to in any case.

Which of Oisigma's published results use highs and lows at all? Only the comparison rows, never the headline figures. In the working paper, bands built from the Parkinson, Garman–Klass and Rogers–Satchell range estimators, and the ATR-width Keltner-style constructions, were scored on close containment and did worse than the close-to-close construction; in the order-block and liquidity-sweep study, excursion depth past a prior extreme was measured from highs and lows and an ATR-normalised measure calibrated it better. Every headline containment figure on the Proof page is closes only.

The quickest way to see the distinction is on a chart where wicks and closes disagree. BTM draws a close-based expected range on any TradingView symbol, and you can start a free 30-day trial to watch the settlement, rather than the excursion, against the range on the markets you already follow.