Bollinger Bands Guide: Volatility, Squeezes, and the Mean-Reversion Myth

John Bollinger, a Chartered Market Technician and Chartered Financial Analyst, developed the bands in the early 1980s while working as a market analyst and floor technician. His motivation was simple: existing "trading bands" of the era used fixed percentage widths (e.g., ±3% around a moving average), but markets have different volatility at different times. A 3% envelope around a calm utility stock is far too wide; the same envelope around a tech growth stock in a volatility spike is far too narrow.

Bollinger's solution – published in full in Bollinger on Bollinger Bands (McGraw-Hill, 2001) – was to make the envelope adaptive by using standard deviation of the closing price over a lookback window. When volatility rises, the bands widen. When it falls, they contract. This simple change turned fixed-width envelopes into a self-adjusting volatility tool used on every charting platform in the world.

The default parameters Bollinger settled on – 20-period SMA, 2 standard deviations – have a statistical intuition: if returns were normally distributed (they are not), ~95% of closes should fall within 2σ of the mean. In practice, because returns have fat tails, 80–90% of closes fall inside the bands depending on the market. The overshoot into the tails is where most of the interesting action happens.

The three lines on a standard Bollinger Band plot are:

Middle band = 20-period SMA of closing prices.

Upper band = 20-SMA + (2 × 20-period standard deviation of closing prices).

Lower band = 20-SMA − (2 × 20-period standard deviation of closing prices).

The standard deviation at each bar is computed from the last 20 closes, so the bands reflect recent volatility, not long-run average volatility. This is why Bollinger Bands react quickly to regime changes – a calm market punctuated by a news shock will see the bands fan out within 2–3 bars.

The choice of 20 periods comes from the original work: it corresponds to one trading month on daily bars, 20 hours on an hourly chart, 20 minutes on a 1-minute chart. Bollinger explicitly says in his book that traders should adjust the parameters to fit their timeframe and market. A common short-term setting is 10-period/1.9σ; a common long-term setting is 50-period/2.1σ. The standard deviation multiplier rarely drops below 1.8 or exceeds 2.5 without the bands becoming pathologically narrow or wide.

The most common and costly misunderstanding of Bollinger Bands is treating a touch of the upper band as an automatic sell signal (and a touch of the lower band as an automatic buy). Bollinger himself has stated repeatedly that band touches are not trade signals. They are volatility references. In a strong trend, price spends long stretches pressed against one band – a phenomenon known as "walking the band".

Consider the empirical reality: in the SPY 2017 bull run, the daily close touched or closed above the upper Bollinger Band on roughly 30% of days and the market still gained ~20% for the year. Every one of those touches, as a short signal, would have bled the account. The correct reading of a band walk is "trend is strong, stay in the trend", not "trend is over".

The actionable version of the upper-band signal requires a failure: price pushes above the upper band, then closes back inside, then makes a lower high without retesting the band. This is the classic M-top pattern – confirmation that the momentum that pushed price into the tail has failed. Bollinger described M-tops and their mirror image W-bottoms as his primary reversal patterns.

A squeeze occurs when bandwidth – the distance between upper and lower bands – contracts to a multi-month low. Low volatility persists, then typically resolves with a volatility expansion in one direction or the other. This is the most empirically robust setup that Bollinger Bands produce.

Bandwidth = (upper band − lower band) ÷ middle band. Expressed as a percentage, a reading of 3% means the bands are 3% apart; a reading of 10% means much wider. A common squeeze definition: bandwidth is in the lowest 10% of its last 125-day range (roughly six months).

The mechanism is not magical. Volatility is mean-reverting – periods of low vol tend to be followed by periods of higher vol (Engle's ARCH, Bollerslev's GARCH; Nobel Prize 2003). So the squeeze works as a timing signal for when the next move will happen, not which direction. Most traders running a simulated evaluation combine the squeeze with a directional trigger – a break above recent highs, a volume surge, a trend filter from the higher timeframe.

The chart below shows simulated bandwidth over 200 trading days with two squeeze events. Each squeeze – where bandwidth bottoms out – is followed by a volatility expansion (in this case, a directional breakout). Not every squeeze produces a clean breakout, but the expansion always follows; volatility cannot stay compressed indefinitely.

TradingView, MT5, and most modern platforms plot bandwidth as a separate sub-indicator. Setting an alert for "bandwidth below Nth percentile of last 125 bars" gives you a systematic squeeze-detection workflow instead of eyeballing chart after chart.

%B = (price − lower band) ÷ (upper band − lower band). A %B of 0 means price is at the lower band, 1 means at the upper band, 0.5 at the middle. Values above 1 or below 0 indicate closes outside the bands – the fat-tail events.

%B converts a visual signal into a number, which is essential for backtesting. Instead of "price touched the lower band", you can write rules like "long when %B crosses above 0.05 after being below 0 for at least 3 bars". Such precision lets you measure the real edge of a band-based rule over thousands of trades.

In Bollinger's own writing, %B is used to identify W-bottoms (a classic reversal pattern): price makes a low with %B below 0, rallies to the middle band, then makes a higher low with %B above 0. The internal strength has improved even though the absolute price low is similar or lower – a bullish divergence expressed in volatility-normalised coordinates.

M-top construction: (1) price makes a high outside the upper band, %B above 1; (2) price pulls back below the middle band; (3) price rallies to a lower high with %B inside the bands – no new tag of the upper band; (4) the second high fails and price breaks the pullback low. The failure to reach the band the second time signals momentum exhaustion even as the second peak may be nominally close to the first.

W-bottom is the mirror: low outside lower band with %B negative, rally to middle, second low with %B less negative (higher low in volatility-normalised terms), followed by upside break of the intermediate high. Bollinger treats the pattern as triggered when the "neckline" between the two lows is broken on the upside with confirming volume.

These patterns work because they combine structure (double top/bottom, which represents seller/buyer exhaustion at a level) with volatility context (the band test and %B divergence add statistical weight). Like all classical patterns, the edge is modest (win rate roughly 52–58% on major instruments per Bulkowski's Encyclopedia of Chart Patterns), but the risk-reward profile when stop is below the pattern low is often favourable.

Mean-reversion traders default to "buy the lower band, sell the upper band". In a ranging market this is profitable – the bands are basically the range walls. In a trending market it is ruinous. The fix is a trend filter that switches the rules on and off:

Regime-based rules:

• If price is above 200-SMA on daily: only take long setups – W-bottoms, lower-band bounces with confirmation, squeeze-then-upside-break.
• If price is below 200-SMA on daily: only take short setups – M-tops, upper-band failures, squeeze-then-downside-break.
• If ADX is below 20 on the trading timeframe: mean-reversion rules (both sides) are valid.
• If ADX is above 25: only trade with the trend; treat opposite-band touches as trend-continuation pauses, not reversals.

Tests of such regime-aware rules (Colby 2002; Pardo 2008 in Evaluation and Optimization of Trading Strategies) show meaningful Sharpe improvements over naïve band-touch rules. The indicator is the same; the filter is what turns a random walk into a system.

Traders routinely change the 20/2 default. Some common alternatives, all of which Bollinger has endorsed in his writings or interviews:

10 / 1.9 – short-term, more signals. Useful on 5-minute or 15-minute charts where default 20 lags too much. More false positives.

50 / 2.1 – position trading. Smoother bands, fewer but higher-quality signals, better for weekly chart analysis.

20 / 1.0 – inner envelope for partial profit-taking. Sometimes plotted together with 20/2 to show multiple levels of volatility extension.

Rule of thumb from Bollinger: if you shorten the period, lower the multiplier slightly (otherwise the bands hug price too tight); if you lengthen the period, raise the multiplier slightly. Always backtest changes on out-of-sample data – the temptation to curve-fit parameters to a single chart is the fastest way to manufacture a fictional edge (see backtesting methodology).

Step 1 – weekly regime read. Pull up the weekly chart of your instrument with 20/2 bands. Is price above or below the middle band? Walking the upper band? In a wide range? This is your bias for the week.

Step 2 – daily squeeze scan. Screen for bandwidth in the lowest 10% of its 125-day range. Compressed markets are setting up for something; identify which.

Step 3 – 4-hour setup identification. Within the weekly bias, look for M-tops or W-bottoms on the 4-hour chart. These are your trade structures.

Step 4 – 15-minute execution. Time entries on the 15-minute chart: %B reversal from extreme, retest of breakout level, volume confirmation. Stop below the pattern extreme.

Step 5 – management. If price walks the opposite band (i.e. trend picks up strongly in your favour), trail stop behind the middle band instead of beneath static swing levels. If price fails to break your entry bar's low/high after 10 bars, close and reassess – the setup is stale.

This multi-timeframe flow is covered in more depth in session-timing guides and structure-based playbooks, both of which pair naturally with Bollinger Bands.