Advanced use of Bollinger Bands starts with two distinct quantities: %b describes price location and BandWidth describes relative channel width. They make observations easier to record and test, but do not directly measure buying force or a probability of profit. This lesson uses explicit numerical examples rather than infer trade results from a band tag.

The full anatomy of the bands

Let U be the upper band, L the lower, M the middle and P the selected price. Here P means close. Take all four from the same observation and settings. Traditional bands use an SMA and standard deviation, often with default 20/2 parameters. That construction differs from an ATR-based channel or a highest-high/lowest-low channel, but does not certify three market regimes or a professional-only method.

Full composition of Bollinger Bands (default 20/2 settings)
Middle lineSMA 20 — simple moving average of the last 20 candles
Upper bandSMA 20 plus 2 standard deviations of those same 20 candles
Lower bandSMA 20 minus 2 standard deviations
BandWidth(upper minus lower) divided by the middle line — the channel-width measure
%B(price minus lower band) divided by the channel width — price position inside the channel
Statistical limitationTwo-standard-deviation bands do not guarantee that 95% of actual prices or candles remain inside.

Do not infer fixed coverage probabilities from rolling Bollinger Bands. Their moving center and dispersion describe the supplied price sample; an upper or lower tag alone does not determine the next price. Read %b and BandWidth using their actual numerators and denominators, as the hypothetical diagram shows.

Two formulas and their zero denominators

%b = (P − L) / (U − L). In this decimal convention, 0 is the lower band, 1 the upper and 0.5 the middle of a symmetric channel. Values below 0 and above 1 are allowed. A value 0.75 describes 75% of the distance through the channel, not a 75% probability or price return. Some displays multiply that decimal by 100; record the convention.

BandWidth = (U − L) / M, multiplied by 100 when reported as a percentage. Its denominator is the middle line, not the lower band. If U = L, the %b formula divides by zero; if M = 0, BandWidth does. Record a missing value and define how the calculation or test handles it instead of inventing a zero signal. Positive FX closes make the middle positive, but zero width still needs handling.

Squeeze requires a defined reference

A squeeze denotes relatively narrow bands under a specified reference criterion. Compare BandWidth with the same instrument’s history at the same parameters and frequency. A rule using the lowest width over 120 bars must state whether it includes the current closed bar or only earlier bars. It is a chosen definition; 120 hourly bars are not six calendar months. Narrow width does not determine the timing, direction or distance of a subsequent move.

A hypothetical history-relative width rule

Define position = 100 × (number of earlier widths below the current width + half the number equal to it) / number of earlier widths. Exclude the current reading from history. This tie convention is chosen for the exercise, not a universal indicator standard.

Keep current BandWidth at 0.5%. With nine earlier values 0.6, 0.7, 0.8, 0.9, 1.0, 1.1, 1.2, 1.3 and 1.4%, position is 0%. With nine earlier values 0.1, 0.2, 0.3 and six values of 0.4%, position is 100%. A hypothetical filter requiring position at or below 20% accepts the first history and rejects the second, despite identical current width.

For earlier values [0.5%, 0.5%, 0.6%], the tie result is 100 × (0 + 2/2) / 3 ≈ 33.33%. State history length, ties and threshold before evaluation. With no earlier observations, there is no classification. This describes relative width, without forecasting an executable breakout or its profitability.

A scheduled announcement can be recorded as another condition in a test, but the calendar does not prove that few participants hold positions or that width must compress. Include breakouts that return inside as well as sustained moves. Entry, execution, stop and target rules need their own definitions; the band formula does not select a compulsory stop on the opposite side of the middle line.

Walk the Band — when the band works for the trend

Walking the band describes prices remaining near one band during a trend, as John Bollinger notes. It does not imply that every countertrend method must produce a run of losses. A count such as five or fifteen closes is a possible test parameter, not an established statistical fingerprint or a guarantee of continuation.

A rule may count closes above a selected mean or a chosen %b threshold. State the threshold, number of observations, timeframe and confirmation time before examining results. The calculation records where price stood relative to its then-current bands; it does not make a pullback to SMA 20 an automatically profitable entry or identify the next turning point.

Changing %b while price remains unchanged

Separate price movement from movement of its reference. Three model observations keep P = 100 while the specified bands change.

Scroll sideways to see the full diagram

Changing %b while price remains unchangedIn A, L=90, M=100, U=110 and P=100, so %b=0.50. In B, the bands shift by 5: L=95, M=105, U=115; P stays 100, so %b=0.25. The third observation returns to A. Lines connect supplied observations, not bands calculated from an unspecified history.Price stands still; the bands shift its reference90100110ABAP = 100 in every observationA: (100 − 90)/20 = 0.50B: (100 − 95)/20 = 0.25Lower %b does not require falling price.

Hypothetical model: specified symmetrical bands, arbitrary units and constant P = 100. We do not reconstruct the indicator from a full closing-price series. BandWidth also changes: 20/100 = 20%, while 20/105 ≈ 19.05%, despite identical absolute width. Meaning of %b and the BandWidth denominator: John Bollinger's rules.

%b describes price relative to the bands, not price movement alone.

Worked location and width examples

Use the following hypothetical completed band and price values. Each row is an independent observation, not consecutive bars of a reconstructed series. The bands are symmetric around M. No unseen history is assumed to have produced them, and these price-location calculations do not measure the force of buyers.

Hypothetical EUR/USD: P / L / M / U
A: 1.1020 / 1.0960 / 1.1000 / 1.1040%b = 0.75; BandWidth ≈ 0.7273%
B: 1.1050 / 1.0940 / 1.1020 / 1.1100%b = 0.6875; BandWidth ≈ 1.4519%
C: 1.1120 / 1.0940 / 1.1020 / 1.1100%b = 1.125; BandWidth ≈ 1.4519%

In A, width is 0.0080: %b = 0.0060 / 0.0080 = 0.75 and percentage BandWidth = 100 × 0.0080 / 1.1000. B has a higher price but lower %b because the bands also changed. In C, price exceeds U, so %b exceeds 1. None supplies a future return probability.

A lower %b at a later price high means a different normalized position relative to the then-current bands. It does not by itself measure buying force: the boundaries and denominator can change. Check the underlying values and treat any divergence rule as a hypothesis to test rather than proof of reversal.

The most common mistakes with Bollinger Bands

  • Confusing location with outcome. A band can be a selected chart reference, but a tag does not guarantee support, resistance, reversal or continuation.
  • Ignoring moving inputs. %b can fall at an unchanged or higher price because L, U and the denominator change.
  • Equating narrow width with no opportunity. BandWidth describes dispersion under a window, not whether every trend or mean-reversion rule makes money.
  • Assuming default parameters are optimal. Bollinger calls 20/2 defaults. A different multiplier changes width, not the underlying observations or their future outcomes.
  • Retuning after seeing results. Keep parameter development separate from evaluation on unused data and retain false breakouts and costs.

Practical takeaway: A band describes relative price location under a chosen calculation; touching it does not establish a trading signal.

Dispersion of closes differs from True Range

Traditional bands use standard deviation of a selected price-level window. ATR averages True Range from high, low and the previous close; long wicks or gaps can therefore affect the two measures differently. Substituting ATR for σ changes the construction, not merely the label. Also distinguish live values from confirmed readings: a signal available only at close cannot be tested at an earlier price from the same bar.

What to do tomorrow

  1. Record a complete observation. Save P, L, M, U, source, timeframe and settings from the same closed bar.
  2. Calculate both measures. Keep the %b denominator U−L separate from BandWidth’s M and record decimal or percentage scaling.
  3. Define the historical comparison. State the prior-window length, whether the current observation is excluded, tie handling and threshold. Missing inputs or zero denominators must have an explicit rule.
  4. Test a complete method. Specify entry, exit, size, costs and fills and assess later unused data. Compare the whole sample, including failed breakouts, rather than only successful chart examples.

Related reading: Bollinger Bands from the ground up, ATR as a volatility measure when scaling stops and position sizes, and three volatility indicators in one decision.