Bollinger Bands are not a buy-at-the-bottom machine. They are a volatility envelope: a moving average with bands whose distance changes with recent dispersion. A trading method appears only when you decide whether a band touch means reversion, a breakout means continuation, or neither, then test that decision with costs and unseen data. John Bollinger's explanation and Fidelity's indicator guide describe the tool as a relative-price and volatility framework, not a standalone answer.
What Bollinger Bands actually measure
Bollinger Bands put a moving average in the middle of a price envelope and set the outer bands from the recent standard deviation of the same price series. The usual construction is:
middle = SMA(close, n)
upper = middle + k × standard_deviation(close, n)
lower = middle - k × standard_deviation(close, n)The familiar starting values are n = 20 and k = 2. They are conventions, not a law of markets. Both John Bollinger's original explanation and Fidelity's calculation guide describe the moving average, standard-deviation width and those conventional settings.
The important part is the moving width. When recent price dispersion rises, the bands spread out; when it falls, they contract. BandWidth makes that change explicit by normalising the distance between the upper and lower bands by the middle band. Fidelity's Band Width guide gives that calculation, while the official cTrader BollingerBands reference describes the same widening and contraction in its indicator interface.
That gives the indicator two kinds of information:
- Relative location: where price sits compared with its recent envelope.
- Recent dispersion: whether that envelope is expanding or shrinking.
It does not give direction by itself. A narrow band says that recent movement has compressed. It does not say whether the next expansion will be higher or lower. Fidelity describes a sharp move in either direction after a tightening, and cTrader likewise treats contraction as a volatility condition rather than a directional forecast.
Why a band touch is easy to misread
The upper and lower bands are often described as relatively high and relatively low prices. That language is about location inside a rolling calculation. It is not the same as a claim that price must reverse there. John Bollinger's own description says the bands combine trend information with relative price level, while Fidelity warns that prices can exceed or hug a band for a prolonged period during a strong trend.
That creates two legitimate but incompatible hypotheses:
| Observation | Mean-reversion reading | Breakout or trend reading |
|---|---|---|
| Price reaches the lower or upper band | The move may be stretched relative to its recent range | The move may be showing persistent pressure |
| Bands contract | The market may be waiting for a return to its recent centre | The market may be storing energy for expansion |
| Price closes outside a band | A fast return inside may invalidate the stretch | A close outside may be the start of continuation |
| Price returns inside after an outside move | The attempted move lost momentum | The breakout failed, at least for that test |
The table is a menu of hypotheses, not a set of signals. Fidelity's guide describes both the tendency for prices to move within the envelope and the possibility of prolonged band-walking or continuation. The cTrader example also shows how a developer can encode a band-cross rule, but code makes a rule reproducible; it does not make the rule profitable.
The practical trap is hindsight. After a chart has printed, a trader can call the same lower-band touch a bargain, a failed breakdown or the first leg of a trend. Those are different strategies with different exits and different risk. If the label changes after the outcome is visible, the method has not yet been specified.
From a chart to a written methodology
Bollinger Bands trading becomes testable when the chart vocabulary is replaced by exact conditions. Two deliberately plain prototypes show the fork:
Mean-reversion prototype, evaluated at the close of bar t:
long when close[t-1] was below lower[t-1] and close[t] returns above lower[t]
exit when price reaches a pre-specified centre or exit rule
short = the exact mirror condition
Breakout prototype, evaluated at the close of bar t:
long when close[t] finishes above upper[t]
exit when price reaches a pre-specified centre, failure or exit rule
short = the exact mirror conditionThese are research starting points, not recommendations. A real specification still has to name the price source, timeframe, lookback, standard-deviation multiplier, entry timing, exit timing, stop treatment, position sizing, and whether a signal is acted on at the close or at the next available price. The cTrader reference exposes the main, top and bottom series for a reproducible implementation, while Fidelity's guide documents the indicator's usual interpretation.
Adding filters can make a chart look more selective. It also adds degrees of freedom. A second timeframe, a momentum oscillator, a bandwidth threshold and a special rule for news may each be sensible, but adding them after seeing the results turns the test into a search for a flattering story. The relevant question is not whether a filter improves one historical curve. It is whether the complete rule was fixed before the unseen sample was opened.
What the evidence and the critics say
The research does not support one portable Bollinger Bands edge. Two studies reach different, tightly scoped conclusions. Fang, Jacobsen and Qin's work on the relationship between popularity and profitability reports that the historical predictive power of Bollinger Band rules weakened as the method became more widely known (the Journal of Portfolio Management record, the authors' SSRN version).
Ni, Day, Huang and Yu report a different pattern in their study of constituent stocks from Taiwan 50: the result depends on which band is reached and on the rule used around that event (the paper's DOI record, the RePEc record). The point is not that one paper wins. The point is that market, sample, implementation and period change the question being answered.
The cautious inference from those studies is that Bollinger Bands are a useful way to state a volatility and relative-price hypothesis, but their portability is unproven. A result in one market does not become a result in FX, futures or a different timeframe by changing the label on the chart. The two studies above are evidence for that limitation, not evidence of a universal failure or a universal edge.
There is a simpler criticism. The envelope is objective once its inputs are fixed, but the interpretation around it is not. “Touch,” “close outside,” “walk,” “re-enter” and “confirm” can all be defined several ways. Fidelity presents several legitimate readings, and John Bollinger's own overview includes additional tools such as %b and BandWidth. That flexibility helps a discretionary analyst describe a chart. It also makes it easy to choose the version that looks best after the fact.
How you'd actually test it
The fair test is not “did price bounce from the band?” It is “which written rule, in which market state, survives execution costs and data it never saw?” A clean protocol would look like this:
- Freeze the question. Choose the instrument universe, timeframe, session treatment, data source and execution convention before reviewing the final results. State whether the test is for a mean-reversion hypothesis, a breakout hypothesis or two separate experiments.
- Freeze the indicator. Start with the documented
20and2settings as a baseline. If other lookbacks or multipliers are allowed, list the candidate set before the comparison and report all candidates, not only the winner. - Write the signal in bar-time. Decide whether a touch uses a high or low, whether a breakout requires a close, and whether the order is placed at the close or the next available price. Do not let the chart renderer decide this implicitly.
- Cost the path. Include spread, slippage, commission and swap where they apply. The real-cost backtesting explainer shows why a clean zero-cost curve is not a fair test of a trading method.
- Keep data unseen. Hold out a period that did not participate in choosing the rules. If the method is allowed to change through time, use a time-ordered design and document when a refit occurs. The out-of-sample testing guide explains why the split matters.
- Measure the path, not just the total. Record expectancy, max drawdown, turnover, exposure, time in the market, the contribution of costs and any loss-limit breaches. Compare the result with a simple benchmark appropriate to the instrument. A positive total can still hide a path that a prop account cannot tolerate.
- Probe sensitivity. Re-run the frozen rule across nearby parameters, symbols and volatility conditions without quietly selecting the most flattering slice. Parameter sensitivity in prop backtesting is the useful companion here.
- Reproduce the implementation. If the intended platform is cTrader, rebuild the signal from closed bars and verify that the top, main and bottom series are indexed as the test assumes. realbacktesting's verification standard is the broader rule: the result should be something another trader can reconstruct, not a screenshot to admire.
That is the test-it-don't-trust-it ethos in practical form. The band is easy to draw. The hard part is proving that the interpretation survives after the chart has stopped helping you.
Frequently asked
Are Bollinger Bands support and resistance?
Not in the fixed-level sense. They are rolling boundaries built from a moving average and recent standard deviation, so they move as the data changes. Fidelity and John Bollinger's explanation frame them as relative-price references rather than permanent market levels.
Do narrow bands predict an upward move?
No. Narrow bands describe compressed recent dispersion; the next expansion can be in either direction. That two-sided interpretation is stated by Fidelity and reflected in the cTrader indicator reference.
Is 20, 2 the correct setting?
It is the conventional starting point, not a universal optimum. John Bollinger and Fidelity both document those defaults while allowing the inputs to vary.
Does a touch of the outer band mean price will reverse?
No. A strong trend can exceed or hug a band for a prolonged period. Fidelity and the Wiley discussion of Bollinger Bands both describe that band-walking problem.
Takeaway
Bollinger Bands tell you how far price has moved from a rolling reference. They do not tell you what that distance means until a written, costed test does.