What Are Bollinger Bands?

Bollinger Bands are three lines on a price chart: a moving average, plus one line above it and one below it placed a chosen multiple of the standard deviation away. John Bollinger, who settled on that formulation in the 1980s, says the three lines exist to answer whether price is "high or low on a relative basis" — a reading of where price sits against the window the band was computed from.

Because the outer lines are scaled by standard deviation, their distance from the average is not fixed. They spread out when the last n bars have been jumpy and close in when those bars have been quiet.

What are Bollinger Bands?

Bollinger Bands are one kind of trading band, and the difference between a band and an envelope is one of construction rather than purpose — John Bollinger writes on his own site that the two serve the same end. A band is built around a measure of central tendency, usually an average; an envelope has no centre line at all and is drawn off highs and lows, or off cycle analysis. Bollinger Bands have a centre by construction, so the middle line carries information of its own.

The stated purpose is narrow. The bands give a relative definition of high and low, which means a tag of the upper line says price is high compared with the window the band was computed from. The definition stops at that comparison. Every upper-band tag his page walks through is then checked against a separate volume indicator before it counts as a signal. The two companions he later built, %b and BandWidth, are readings of the same geometry.

The name was an accident. Bill Griffeth, then an on-air host at the Financial News Network, asked Bollinger live on air what the curves on the chart he had just presented were called, and the answer stuck. The book documenting the method reaches the finance literature's reference lists as Bollinger on Bollinger Bands, dated 2002.

How are Bollinger Bands calculated?

A Bollinger Band calculation takes a price series, a lookback length and a multiplier. Sierra Chart's documentation for its Bollinger Bands study states the three outputs directly: the middle line averages the chosen price series over the chosen length; the upper line is that average plus the multiplier times the standard deviation of the same window; the lower line is the average minus the same quantity. Nothing is defined until the window has filled.

Bollinger's defaults are a 20-period average with the outer lines at two standard deviations, measured on the same data the average was taken from, and his page notes that he computes the standard deviation with the population formula rather than the sample one. He also writes that thirty-five years have not changed them, on a page whose body carries a 2021 copyright line.

Platforms expose the parts as inputs. Sierra Chart's study takes a length, a number of standard deviations, a moving average type and the input data series, so the centre line does not have to be a simple average — exponential, weighted, Wilder's and smoothed are among the options, and each choice changes where the middle line sits and therefore where both outer lines sit. Quantower files the indicator under channels, alongside the Donchian channel, the Keltner channel and a moving average envelope, and its page describes the bands as a way to gauge volatility and to flag overbought or oversold conditions around the average.

Why do Bollinger Bands widen and narrow?

The bands change width because standard deviation changes. The multiplier stays put at two, or wherever the trader set it, and the quantity it multiplies is recomputed on every bar from the most recent n values of the chosen series, so a run of large ranges widens the gap between the lines and a quiet stretch narrows it.

Bollinger describes this as the problem he set out to solve. The common tool when he started was the percentage band — a 7% band sat its top line at 1.07 times the base average and its bottom line at 0.93 times — and those had to be nudged by hand to stay relevant to the price structure. His objection was that the adjustment let the operator's own bias in: a bullish analyst draws bullish bands, and a measure that recalculates itself takes that decision away. Volatility, he adds, was treated then as a fixed property of a security and is now understood to move constantly.

Different bands adapt in different ways, and his page names four families in common use.

What do %b and BandWidth measure?

%b and BandWidth are the two readings Bollinger derived from the bands themselves, and both turn the picture into a number. Sierra Chart documents each as a separate study. %b locates price between the outer lines: subtract the lower band from the input price, divide by the distance between the two bands. With the base band input set to the lower band, that reads zero at the lower band, one at the upper band and one half at the average, and it can run past either end when price trades outside the envelope.

BandWidth measures the envelope itself. Sierra Chart computes it as the distance between the outer bands, divided by the centre average — so it is a ratio rather than a price, and at two standard deviations it works out to four standard deviations expressed as a fraction of the average. That makes it comparable across instruments: a 40-point envelope means one thing on a $4,000 index and another on a $400 one.

Sierra Chart's %b study also draws two horizontal threshold lines wherever the user puts them, so the levels a trader treats as meaningful on a %b plot come from the trader, not from the calculation.

How did John Bollinger say the bands should be read?

Bollinger's own published description of his practice puts confirmation before action. Writing about the percentage-band years, he says he compared where price sat inside the bands against a supply-and-demand measure — David Bostian's Intraday Intensity — and treated an upper-band touch the oscillator failed to confirm as a sell setup, with the mirror case as a buy setup. A chart caption on his page applies the same logic to Bollinger Bands: price tags the upper band on a day when 21-day Intraday Intensity is well below zero, and the entry comes on the first lower close after that.

Two things follow from how he frames it. In his own account of the method, the band tag is the condition that sends him to the volume indicator, and the signal comes after that. And the middle line is doing work: he says Bollinger Bands "work best when the middle band is chosen to reflect the intermediate-term trend".

He names a pattern off the bands as well. The one his page illustrates is a "W bottom", and the example turns on %b reading higher at the second retest than it did at the momentum low that preceded it. That is a convention he defined, and the %b reading behind it describes what the data has already done.

Have Bollinger Bands actually been tested?

Yes, in the foreign exchange market. Coakley, Marzano and Nankervis tested 113,148 technical trading rules on 22 currencies against the US dollar from October 1997 to June 2015, in work published in the International Review of Financial Analysis; 14,112 of those rules were Bollinger Band rules, built by varying an evaluation period, the number of standard deviations between one and four, a crossing threshold, a delay filter and a holding period. The tested signal is mechanical: buy when the rate drops through the lower band by a set percentage, sell when it rises through the upper band by the same amount.

The headline numbers look striking. Averaged across currencies, 42% of the Bollinger rules counted as significant on their excess-return measure before any snooping correction, against 34% to 36% for the three moving average classes and 12% for a channel breakout rule. The single best Bollinger rule in each currency averaged 20.6% a year across the 22, topping out at 29.7% on the South African rand. Across the same currencies the best of the traditional rules averaged 6% to 7%.

Two caveats travel with those figures and the authors state both. Returns are excess returns that account for interest rate differentials, and the study applies no transaction costs at all — they judged the available indicative bid-ask quotes too wide to use and noted that spreads are narrow on large FX trades. And the best rule in each currency is the best of thousands, selected after seeing the results.

What survives a data-snooping correction?

Very little of the raw count, which is the most useful result in the study. Picking the best of 14,112 rules produces a good-looking number whether or not the method has any content, so Coakley, Marzano and Nankervis ran a stepwise test designed to control for exactly that. After it, an average of 112 Bollinger rules per currency remained significant — about 0.8% of the Bollinger rules tested, against 9.1% of the relative strength index rules.

The survivors are also unevenly spread: 2 robust rules on the Hong Kong dollar, 12 on the Canadian dollar, 186 on the rand, 1,322 on the Indian rupee. The two currencies with the largest robust-rule counts overall are the rupee and the Hungarian forint, and the authors note that both are managed against the dollar directly or indirectly by their central banks; their reading is that a rule can clear a statistical bar inside that kind of controlled market while the returns stay small in economic terms. Their conclusion is that the indicator rules, Bollinger's among them, stay robustly profitable once interest differentials and snooping are allowed for, while virtually no traditional rule stayed significant in their 2006 to 2015 sub-period.

The best-performing rule is worth a closer look. Before the correction it was near-identical in every currency: the authors report an evaluation period of five and one standard deviation, with the other three parameters at zero. One standard deviation is half the width of Bollinger's own default. A rule found by searching a grid lands where the data in that grid put it, which is the whole difficulty with curve fitting and the reason the stepwise correction exists.

What does the broader evidence on technical trading say?

Mixed results and a warning about method. Park and Irwin's review for the University of Illinois AgMAS project surveyed 92 modern studies of technical trading and counted 58 with positive results, 24 with negative results and 10 mixed. Their verdict on the field is that most of those studies suffer from data snooping, from rules selected after searching for them, or from the difficulty of estimating risk and costs honestly.

Two further findings matter. The survey evidence they collect shows technical analysis in wide use in both futures and currency markets, with roughly 30% to 40% of practitioners treating it as an important factor in price at horizons out to six months. And on futures specifically, they found technical trading looked profitable through the ten years from the mid-1970s, with nothing comprehensive documented after the mid-1980s.

The name Bollinger does not appear anywhere in that review — a text search of the whole report returns nothing — so the band evidence in this post rests on the currency study above, run on spot rates against the dollar.

What do Bollinger Bands mean on a funded account?

On a funded or evaluation account, the gap between the bands moves and the line that can close the account does not. A Lumen account is a simulated one that pays out real profits, and its drawdown line is set at a fixed dollar gap under the best daily close the account has ever recorded, and never moves down; see how the drawdown line works for the mechanics. That makes your room a dollar figure. BandWidth is a ratio. When the standard deviation of the last 20 bars doubles, a trade sized from the lower band to the upper one risks roughly twice the dollars it did a week earlier, while the line beneath you has not moved an inch.

That makes the bands a sizing input before they are anything else. If you place a stop off the envelope, the stop distance in ticks is a different number every session, and holding contract count constant through a volatility expansion quietly multiplies the risk per trade. How many stops of a given width fit inside the room you have is the arithmetic in our guide to position sizing on a funded account.

One more detail matters on an intraday chart: the bands are computed from whatever bars the chart holds, and they do not know where one session ended and the next began. A 20-period band on a 15-minute chart reaches back five hours, so the first bands after a reopen are built largely from the previous session's dispersion.

FAQ

What are the default Bollinger Band settings?

A 20-period moving average with the outer bands at two standard deviations of the same 20 values. Bollinger's page gives those defaults, says thirty-five years have left them unchanged, and specifies the population formula for the standard deviation rather than the sample one. Platforms expose every part of it: Sierra Chart's study takes a length, a number of standard deviations, a moving average type and an input series.

Does a touch of the upper Bollinger Band mean price will fall?

Bollinger's own page does not say so. He describes a band tag as a relative high — price is high compared with the window used to build the band — and in his published account of the method, an upper-band touch that the volume indicator did not confirm was a sell setup, with the entry taken on a later bar. Quantower's page does describe the bands as a way to flag overbought or oversold conditions around the average; that is the vendor's description of the tool, not Bollinger's. The band itself is a measurement of dispersion that has already happened.

Are Bollinger Bands the same as Keltner or Donchian channels?

They share a shape and differ in the mechanism. Bollinger's page sets out the distinction: Keltner bands adapt through average true range, Donchian bands take the prior n days' highest high and lowest low, percentage bands are fixed until someone moves them, and Bollinger Bands use standard deviation. Quantower lists Bollinger Bands, the Keltner channel and the Donchian channel together in the channels section of its indicator list.

Have Bollinger Bands been tested on futures?

The test cited here is a foreign exchange study: 22 currencies against the dollar, October 1997 to June 2015, with no transaction costs applied. Park and Irwin's review of 92 technical trading studies covers futures markets among others and records profits there in the ten years to the mid-1980s, with nothing comprehensive after; the name Bollinger does not appear in that review's text. No test of the bands on a futures contract turned up in the sources read for this post.