What Are Bollinger Bands? What They Measure, and What They Don't

A stock climbs for a fortnight, and on your chart the price bumps into a curving line above it. Somewhere on the internet, someone is already typing that this means the stock is overbought and due a fall. It's the most common thing said about Bollinger Bands, and it's the one thing the bands were never built to tell you.

What the lines actually are

Start with the middle line, because the other two are built from it. It's a 20-period simple moving average: add up the last 20 closing prices, divide by 20, plot the answer, repeat tomorrow. If you want the full tour of what an average does to a price chart, we've written that separately — here it's just the spine the bands hang off.

Now the outer two. Take those same 20 closes and work out their standard deviation — a single number summarising how far they scattered from their own average. A calm fortnight gives you a small one. A fortnight with two gap-downs and a recovery gives you a big one. Draw a line two of those above the average, and another two below, and you have the bands.

So the outer lines aren't levels. Nobody chose them, nobody defended them, and no institution is sitting on them. They're the average, plus and minus how far price has been wandering from it. That's the whole construction, and almost everything sensible about the tool follows from it.

The width is the message

Because the outer lines are pinned to how far price has scattered, the distance between them isn't decoration — it's the reading. Bands squeezed into a narrow ribbon mean the last twenty days were quiet. Bands flared wide apart mean they weren't.

This is the part worth internalising, because it flips what the tool is for. Most indicators are trying, in some fashion, to tell you about direction. The bands are telling you about room. They answer "how much has this been moving lately?", which is a question with an actual answer, rather than "what happens tomorrow?", which isn't.

Outside the bands is an ordinary Tuesday

Here's where the two-standard-deviations bit misleads people. In a statistics class, two standard deviations covers about 95% of a normal distribution, so it's tempting to conclude that price should sit inside the bands 95% of the time and that the other 5% is something remarkable.

Bollinger's fourteenth rule tells you not to do this. Prices aren't normally distributed, and twenty readings is a small sample. His own observation is that in practice about 90% of the data sits inside the bands at the default settings — not 95%.

Ninety percent inside means one bar in ten finishes outside. Our chart above, measured rather than assumed, puts 89.6% of its closes inside the bands — the same neighbourhood, arrived at independently. Whichever number you use, an event that turns up every couple of weeks is not an extreme. It's a Tuesday.

The builder above is worth two minutes because it makes the trap obvious. Widen the multiplier and breaches become rare — but by the time price gets out there, the move is well under way. Narrow it and breaches become constant, which tells you nothing at all. There's no setting at which the line becomes a signal. You're only choosing how often it gets touched.

Walking the bands

Now the failure mode that costs people actual money. In a real trend, price doesn't touch the upper band and politely retreat. It leans on it and keeps going — for days, sometimes weeks. Bollinger's seventh rule states it flatly: in trending markets price can, and does, walk up the upper band and down the lower one.

The chart below plays out the naive rule on a trending stretch. Every time price reaches the upper band, sell short; cover six bars later. Six trades. Five of them lose. The net damage is $8.64 a share, on a stock that spent the whole period going up — and at one point price leaned on the upper band for nine bars without a break.

Bollinger goes further than "the touch isn't a sell signal." His eighth rule says closes outside the bands are initially continuation signals, not reversal ones — the opposite of the folk reading. On his reading, strength enough to push price clear of its own recent range is strength — not exhaustion.

The squeeze, and the half of it nobody quotes

There is one band behaviour with a real mechanism behind it, and it's the one where the tool is doing what it's actually built for. When the bands pinch tight, the market has gone quiet. Quiet doesn't last. Sooner or later the range opens back up, and the bands flare.

That isn't folklore — it's the observation that won Robert Engle a Nobel prize. Volatility clusters: calm periods bunch together, violent ones bunch together, and both are time-limited, so a stretch of unusual quiet does tend to be followed by a return to something livelier. On our chart the bands narrow to a width of 0.83 and then open to 6.77 — eight times as wide.

And here's the half that gets left out of every squeeze tutorial. In the same breath as describing volatility clustering, Engle notes that returns themselves are almost unpredictable. Volatility is forecastable. Direction isn't. A squeeze is genuine information that something is coming; it is silent on whether you want to be long or short when it arrives, which is unfortunately the only part you get paid for.

So does trading them work?

The honest answer is that the published evidence disagrees with itself, and you should be suspicious of anyone who tells you otherwise in either direction.

One test, published in Applied Financial Economics Letters in 2007, found that after adjusting for transaction costs Bollinger Band rules were consistently unable to earn profits in excess of simply buying and holding — though a contrarian variant did better. A 2020 study of the Taiwan 50 constituents found close to the opposite: significant positive abnormal returns, and specifically that when price reaches the upper band the profitable move was to go long rather than short. Which, notice, is Bollinger's own eighth rule arriving from a completely different direction.

Two credible tests, two different answers. That disagreement is the finding. If a three-line indicator on a free charting site reliably printed money, the money would have been printed by now and the lines would have stopped working — that's roughly what markets do to simple public rules. What survives is the modest version: the bands measure something real, and the measurement is useful next to everything else you know.

How to actually use them

Bollinger's own first rule frames it neatly: the bands provide a relative definition of high and low. By definition, price is high at the upper band and low at the lower one. Relative — to its own recent behaviour, not to what the company is worth, and not to anything you'd call expensive.

• Read the width first. Pinched bands mean a quiet market, which is a different trading problem from a wide one — and it's the reading the tool does best.
• Treat a tag as a question, not an answer. Price is at the edge of its recent range. Whether that's exhaustion or strength is decided by the trend, the structure and everything else on the chart.
• In a clear trend, expect the walk. Riding the band is the trend working, not a countdown to reversal.
• Don't confuse rare with meaningful. You can widen the bands until touches almost never happen; that doesn't make the touch informative.
• Confirm with something unrelated. Bollinger's own fourth rule warns that two indicators measuring the same thing aren't better than one — pairing the bands with a second momentum gauge tells you less than you think.

And the defaults are defaults. Twenty periods and two standard deviations are Bollinger's starting point, not a law; his eleventh rule even specifies that holding containment steady while lengthening the average means widening the multiplier — 2.1 at 50 periods, 1.9 at 10. If you change one and not the other, you've quietly changed how often price gets outside, which is the number the whole tool rests on.