Return Space vs Price Space: What a Band Measures

Take two volatility bands. Give them the same lookback window and the same standard-deviation multiplier, and put them on the same chart. They will still disagree — sometimes dramatically — about where "normal" is. The settings aren't the reason. The reason is a choice made before any setting is applied: what the band measures its statistics on. One band computes its mean and standard deviation on raw price levels; the other computes them on returns — the percentage changes between bars. That choice is called the measurement space, and it shapes everything the band does afterward.

This article is the deep dive on that choice: what each space actually measures, why the difference matters statistically, and what the historical evidence says about it.

What "measurement space" means

Every statistical band does the same three things: collect a window of recent data, compute a center and a dispersion from it, and draw boundaries. The measurement space is the answer to the question, recent data of what?

• Price space: the inputs are the raw closing prices themselves — say, the last 20 closes. The center is an average price; the dispersion is the standard deviation of those price levels. Classic price-space constructions include the moving-average-and-standard-deviation bands most charting platforms ship by default.
• Return space: the inputs are the bar-to-bar percentage changes. The center is an expected move, and the dispersion is the standard deviation of recent moves. To display the result, the band is projected back onto the chart — anchored to the prior close and converted into price levels.

Both end up as lines on a price chart, which is why the distinction is easy to miss. But they are answering different questions. A price-space band asks: how far do closes scatter around their recent average level? A return-space band asks: how large are the moves this market has recently been making?

Why raw prices are a hard thing to average

The statistical machinery inside a band — means, standard deviations — works best on data whose behavior is roughly stable over the window being measured. Statisticians call this stationarity. Returns aren't perfectly stationary (their volatility clusters, which is exactly why a rolling window is used), but they are far closer to it than prices are, and that difference is the crux.

Price levels have two properties that fight the math:

They trend. In a steady uptrend, the last 20 closes aren't 20 draws from one stable distribution — they're a staircase. The average of a staircase sits below its top step, so the band's center lags price. And the standard deviation of a staircase is large not because the market is volatile, but because the staircase climbed. The band widens for the wrong reason. This lag-and-balloon behavior is the drift problem, and it has its own dedicated breakdown in our article on the Bollinger band drift problem.

They aren't comparable across contexts. A $5 standard deviation means something completely different at a price of $50 than at $500, and different again on another symbol or another decade of the same symbol. A band built on price levels inherits that incomparability: its width can't be interpreted without knowing where price happens to sit.

Returns dissolve both problems at once. A 1% move is a 1% move whether the stock trades at $50 or $500, this year or in 1975. Detrending comes free: measuring changes rather than levels removes the staircase, so the dispersion estimate reflects how much the market is moving, not how far it has travelled. This is why quantitative finance — from academic asset-pricing work to institutional risk models — does its statistics on returns nearly universally, and why the volatility being measured is what rolling volatility actually estimates. (For what standard deviation itself is doing as a yardstick, see our primer on judging whether a price move is normal.)

From returns back to price: the projection step

A return-space band still has to live on a price chart, and the way it gets there matters. Each bar, the band takes its expected move and dispersion — both measured in percentages — and anchors them to the prior close. The expected move becomes an expected price; the dispersion becomes boundaries above and below it; the next bar, everything re-anchors to the new close.

That anchoring is the practical payoff. Because the band is rebuilt from wherever price actually is, it stays centered in a trend instead of dragging an average of stale levels behind it. The width reflects current move sizes, scaled to the current price. Oisigma's How It Works page walks through this construction step by step — it's the design choice the whole indicator rests on.

Does the choice show up in the data?

It's a fair question: maybe measurement space is a theoretical nicety that washes out in practice. The historical evidence says otherwise. Oisigma's working paper scored both constructions on the same causal question — how often does the next close land inside the band? — across the same data. At the widely used ±2 standard-deviation width (price-space bands at their standard 20-period settings), the price-space band contained about 83% of next closes against a ~95% nominal target, while the return-space band contained about 94%. The paper's decomposition attributes the gap to the measurement space itself — swapping in fancier variance estimators barely moves the result, but changing the space does. As always: these are historical measurements, and past behavior is not a guarantee of future results.

That containment question — what it is, how to run it on any band — is covered in depth in our guide to comparing band indicators, and what "calibrated" means as a standard is defined in the article on calibrated expected-range indicators.

What return space does not change

Measurement space is a real improvement in what a band describes. It is not a promotion into forecasting. A return-space band still:

• describes, rather than predicts. Its center line is an anchor for the range, not a directional call. Measuring in returns tells you how big recent moves have been; it says nothing about which way the next one goes.
• depends entirely on its window. It reacts at the speed of its lookback, in varying degrees like every rolling construction — when volatility regime-shifts faster than the window can register, any recent-history band runs too narrow for a stretch. We've published that limit and its measured cost in our article on when volatility bands fail.
• carries no trade logic. No measurement space turns a descriptive range into a signal, an edge, or a strategy.

Return space fixes the geometry — where the band sits and how its width scales. The honest limits of describing the future with the recent past remain, and they apply in either space.

The takeaway

When a price-space band and a return-space band disagree on the same chart, the deepest reason is usually not the settings — it's the space. (Two return-space bands can disagree too, but for the ordinary reasons: different windows, different widths.) Price-space bands do statistics on levels, which trend and don't compare across contexts; return-space bands do statistics on moves, which are unit-free, roughly stable, and re-anchor to wherever price actually is. In our published testing, that single construction choice, not the sophistication of the volatility estimate, accounted for most of the difference in how often price actually stayed inside the band.

If you want to see what a return-space band looks like on your own charts, the Behavioral Transform Model — Oisigma's calibrated expected-range indicator for TradingView — is built exactly this way, with the full methodology published. You can start a free 30-day trial and watch the band re-anchor bar by bar on the markets you actually follow.