What the bands actually are

John Bollinger developed the bands in the 1980s (Bollinger Bands). The construction is three lines:

• Middle band = an N-period moving average, typically 20 periods, by default a simple moving average

• Upper band = the middle band plus K times an N-period standard deviation, typically K = 2

• Lower band = the middle band minus the same amount

Two derived readings come from that: %b, calculated as (last minus lower band) divided by (upper band minus lower band), which tells you where price sits within the envelope, with 1 at the upper band and 0 at the lower; and Bandwidth, calculated as (upper band minus lower band) divided by the middle band, which measures the width on a normalised basis and is used to identify relative extremes in volatility and for trend identification.

Notice what the middle band is. It is the same 20-period moving average this series covered in Part 14, with all the lag that implies. The bands add a volatility measurement around it. They do not add a new source of information about direction.

The 95% that is actually 88%

Here is the number that should change how you read a band touch.

The reference material is explicit: "instead of finding about 95% of the data inside the bands, studies have found that only about 88% of security prices (85-90%) remain within the bands." The three reasons given are exactly the three assumptions the two-standard-deviation rule depends on, and financial prices break all of them.

This matters because the entire emotional weight of a band touch comes from the assumption that it is rare. At 5%, price reaching the upper band is a genuine outlier worth reacting to. At 12%, it happens roughly every eighth bar, more than twice as often as the 5% assumption implies. That is still a minority event, but it is far too frequent to carry the weight the word "extreme" gives it.

So "price is at the upper band, it is overextended" is not a statistical statement. It is a description of where price sits relative to its own recent average, scaled by its own recent volatility. Useful as context. Not evidence of anything being stretched.

What the testing actually shows

The same 2023 crypto study this series has drawn on tested Bollinger Bands as one of its four rule families, on BTC/USDT and ETH/USDT (Chen, Wang & Yang, 2023). As with the other families, they did not test one setting. They swept the period from 15 to 30 in steps of 3, and the deviation factor across 1.0, 2.0 and 3.0.

On Bitcoin, the best in-sample configuration was a period of 21 with a deviation factor of 3.0, producing a compounded yearly yield of 88.059% with a Sharpe ratio of 1.519 and a maximum loss of 52.234%. That looks like a strategy. Out of sample it returned 12.826%, with p-values that do not reject the null.

Ethereum is the more instructive case, because the best in-sample result was already poor. A period of 21 with a deviation factor of 2.0 produced an annualised return of 7.998%, a Sharpe ratio of 0.12, and a maximum drawdown of 90.552%. Read that combination carefully: this is the configuration chosen with full hindsight, on the data it was fitted to, and it still delivered single-digit returns while giving back over ninety percent at its worst point. Out of sample it returned 14.978%, again failing to reject the null.

The authors' conclusion on this family is blunt. After White's Reality Check and Hansen's Stepwise test, the Bollinger Bands strategies "fail to generate profits in both the in-sample and out-of-sample periods."

Two details deserve attention. First, the winning settings were not the defaults: period 21 rather than 20, and on Bitcoin a deviation factor of 3.0 rather than the conventional 2.0. That is the same pattern Part 14 found with moving averages, where the tested winners were nothing like the famous settings. Second, this is what a sweep across eighteen configurations buys you, six periods from 15 to 30 crossed with three deviation factors, which is exactly the search that the data-snooping correction exists to discount (Hsu & Kuan, 2005).

Reputation versus evidence

Using Bollinger Bands honestly

• Read them as a volatility gauge, not a boundary. When the bands lie close together, low volatility is indicated; as they expand, an increase in price action and volatility is. Bandwidth normalises that width, and is used for spotting relative volatility extremes and for trend identification. The edges are not walls.

• Recalibrate what a band touch means. Roughly one bar in eight sits outside the envelope, more than twice as often as the textbook figure implies.

• Be wary of fading a band touch in a trend. Traders commonly say price can ride along the upper band during a strong advance rather than reverting, but that is trading lore rather than anything the research here establishes. What the sourced numbers do support is narrower and enough on its own: at roughly one bar in eight outside the envelope, a touch by itself is not a high-confidence reversal signal.

• Remember the middle band is a moving average. Everything Part 14 said about lag applies to the spine of this indicator.

• Distrust tuned settings. The best crypto configurations came out of a parameter sweep and still did not clear the significance bar.

Common mistakes

• Treating the bands as a 95% confidence interval - the real figure is about 88%, and the assumptions behind the maths do not hold for price.

• Shorting the upper band automatically - at roughly one bar in eight outside the envelope, a touch alone is not a high-confidence reversal signal.

• Reading narrow bands as a directional signal - a contraction says volatility is low, not which way the expansion will go.

• Assuming 20 and 2 are optimal - they are conventions, and the tested winners used different values.

• Judging a strategy on return alone - the best in-sample Ethereum configuration returned 8% a year with a 90% drawdown.

Bollinger Bands on ApeX Omni

ApeX Omni's charts carry Bollinger Bands alongside the price action this series has taught you to read. The defensible use on a perpetuals venue is as a volatility read: let bandwidth tell you whether the market is compressing or expanding, then size accordingly and take the actual decision from structure, levels and volume. Volatility sizing matters more here than on a spot chart, because the same leverage behaves very differently in a quiet market than in an expanding one, and your liquidation price does not move just because volatility doubled, as Part 9 covered. If you do trade a band touch, price it as considerably more frequent than the textbook figure implies, put the stop where the structure says, and trigger it on the mark price as Part 12 argued. Trade the chart on ApeX Omni, and let the bands describe conditions rather than dictate entries.

The bottom line

Bollinger Bands are a good volatility instrument wearing the costume of a statistical one. The envelope widens and narrows with volatility, and bandwidth normalises that width, which is used to identify relative extremes in volatility and for trend identification. What the bands do not do is mark a boundary that price has statistically overshot, because the two-standard-deviation intuition depends on assumptions that financial prices comprehensively break. Around 88% of prices sit inside, not 95%, which makes a band touch more than twice as frequent as the textbook figure implies. Add a crypto test where the tuned configurations failed to clear significance in the very sample they were fitted to, and the honest position is clear: use the bands to see volatility, and take your entries from something that is not derived from the same twenty-period average.

Next in this series: the Stochastic Oscillator, the last of the big five, and the question of whether comparing a close to its recent range tells you anything the other four have not already said.

Frequently asked questions

What are Bollinger Bands and who created them? John Bollinger developed them in the 1980s. They consist of a moving average, typically 20 periods, with an upper and lower band placed a number of standard deviations away, typically two.

Do 95% of prices really stay inside the bands? No. Studies find about 88%, in a range of roughly 85 to 90%. The 95% figure assumes a normal distribution, independent observations and a large sample, and price data violates all three.

Does a touch of the upper band mean sell? Not on its own. Because roughly 12% of prices fall outside the bands, a touch is more than twice as frequent as the textbook 5% figure implies, which is too frequent to treat as a high-confidence reversal signal by itself.

What are %b and Bandwidth? %b measures where price sits within the envelope, reading 1 at the upper band and 0 at the lower. Bandwidth is the distance between the bands divided by the middle band, measuring width on a normalised basis, and it is used to identify relative extremes in volatility and for trend identification.

Are Bollinger Bands profitable to trade? In a 2023 crypto study that swept periods from 15 to 30 and deviation factors of 1.0, 2.0 and 3.0, Bollinger Bands strategies failed to generate profits in both the in-sample and out-of-sample periods after correcting for data-snooping.

Should I change the default settings? The tested winners in that study used a period of 21, and a deviation factor of 3.0 on Bitcoin, rather than the conventional 20 and 2. That is not a recommendation to use those numbers, since they came from a hindsight sweep that still failed the significance test.

Explore more from this series: Part 14: Moving Averages | Part 15: RSI Explained | Part 16: MACD Explained | Part 18: Stochastic Oscillator

This article is for educational and informational purposes only and is not financial, investment, or legal advice. Do your own research before making any trading decision.