What it measures

Bollinger Bands draw a moving average with an envelope above and below it whose width is set by recent volatility. When price is quiet the bands pinch in; when it moves the bands open. John Bollinger's point was that "high" and "low" are relative to volatility, not fixed distances.

The calculation

middle = SMA(close, 20) upper = middle + 2 × standard deviation of close over 20 bars lower = middle − 2 × standard deviation of close over 20 bars

Band width (upper − lower, divided by the middle) is the volatility reading; percent B, the position of the close between the bands, is the oscillator reading.

Inputs and defaults

  • Length: 20.
  • Multiplier: 2 standard deviations. With normally distributed returns about 95 in 100 closes would sit inside; returns are not normal and more fall outside than that.
  • Source: close.

How it is read

Two readings that contradict each other. Reversion: a close at the upper band is stretched, sell toward the middle. Breakout: a close outside the band after a squeeze is the start of a move, buy with it. Both are in the screened universe with the same default settings. The squeeze itself (band width at a multi-month low) is read as a warning that a move is coming, without saying which way.

Where it fails

The standard deviation is computed on price, not returns, so a trending market widens the bands and the reversion reading sells into every leg of the trend. The breakout reading, meanwhile, buys every false break in a range. The bands say that something unusual is happening; the trade depends on a regime call the bands cannot make.

How the desk reads it

The desk reads a Bollinger strategy first for which of the two readings it codes, reversion or breakout, and then for what it did in the year the market trended without looking back: the reversion reading loses there and the breakout reading loses in the ranging year, and a backtest that shows neither has a short window.

In the desk's screen

Of the 614 screened strategies in the library, 39 (6.4%) use this indicator somewhere in their logic.

14 of those were gross-profitable on enough cells to reach the after-fee replay.

The count is read from the compiled code of each strategy in the library; a strategy that computes the indicator by hand under another name is not counted. The desk's own rules are not part of the count.