Edgewise › Measured

Does the Bollinger Band breakout work?

We tested the Bollinger Band breakout the way it is usually traded, then compared it with random entries on the same crypto coins, taken at times when the market was just as jumpy. Here is what happened.

The short answer

No better than random, wherever we could tell

Where the signal fired often enough to judge, it did no better than random entries. The other time limits have too few trades to say either way.

The rule we tested

The daily close crosses above the upper Bollinger Band (20, 2). Traded long, as a breakout.

What happened

Main setup: stop 2× ATR, target 4× ATR, random entries drawn from the same year. A result only counts if it differs from random in all six test setups (see How we tested it).
Time limit	Result	Differed from random in (needs all 6)	Trades	Hit target: signal	Hit target: random	Hit time limit: signal	Avg result (ATR): signal	Avg result (ATR): random	Smallest gap we could detect (lower is better)
1 day	No better than random	0 of 6	664	1.5%	0.3%	94.7%	+0.12	+0.04	0.11
5 days	No better than random	0 of 6	549	13.3%	5.5%	65.4%	+0.37	+0.14	0.23
10 days	No better than random	0 of 6	458	21.2%	15.3%	40.0%	+0.28	+0.25	0.29
20 days	Not enough data	0 of 6	377	27.6%	24.7%	15.4%	+0.06	+0.15	0.37

ATR (average true range) is how far a coin typically moves in a day, averaged over 14 days. A result of +4 means the trade made four typical days’ moves. Hit target counts trades that reached the target before the stop. Trades that hit the time limit count as not hitting, which is why the one-day rates are so low. Avg result is in ATR: a target is +4, a stop is −2, and a trade that hit the time limit is whatever the close gave. Smallest gap we could detect is how big a real difference from random would have to be, in ATR per trade, before this test would reliably catch it. Lower means a sharper test. Above 0.30 there were too few trades to judge, so the row says “Not enough data”. It is not counted as a fail.

How we tested it

Entry at the close of the day the signal fired. No intraday fills, no judgement calls.
Stop 2× ATR (14) away, target 4× ATR (14) away — the classic 2:1 trade: risk one to make two.
Time limits of 1, 5, 10 and 20 days. A trade that reaches neither level in time closes at that day’s close. A day whose range touched both levels is left out, because daily data cannot say which came first.
The comparison. For every signal, 20 random entries on the same coin, at equally jumpy times (the same volatility band), in the same year, run through the identical stop and target.
Data. Daily bars on 10 crypto perpetual futures (BTC, ETH, SOL, BNB, XRP, ADA, AVAX, LINK, LTC and DOT) through 27 August 2026. Each coin’s first 252 days are skipped so every indicator has history behind it, and signals whose trades would overlap count once.
Six setups. Three ways of drawing the random entries (same year, same half-year, any time) and two stop widths (2× and 1× ATR). A result has to hold in all six.
Many tests at once. 44 signal-and-time-limit combinations were tested together, and every result is corrected for that, so chance alone does not produce a “winner”. The rules were written down before any result was seen.

What this does not show

No fees, spread or funding. Signal and random trades would pay the same costs, so the comparison stays fair, but real averages would be lower than shown.
Only coins that survived. All 10 were still trading in 2026. That does not skew the comparison, since both sides use the same coins, but it flatters the raw averages.
Crypto, daily bars only. Nothing here says how the signal behaves on stocks, forex or intraday charts. The coins also move together, so the sample is smaller than it looks.
“Not enough data” is not “fails”. It means the test could not have told a real edge from noise at that time limit.

If we got the rule wrong

If this is not how you trade the Bollinger Band breakout, tell us how you do — on X at @Edgewise_to — and we will run your version through the same test. Every other signal is on the Measured page, and everything we have checked, including results that went against us, is on Verdicts.

Why Edgewise publishes this

Edgewise shows honest probabilities for live markets and publishes a graded forward record of the direction engine behind them — its next-candle probabilities, which the app itself does not print — misses included. Testing popular signals is the same habit, pointed outward.

See today’s probabilities on Edgewise

See our own graded track record →

The live probabilities are the paid product — have a look if they are useful to you, and ignore them if not. Nothing on this page is gated.