SPECIMEN 005 / STRATEGY AUTOPSYCRYPTO · OUT-OF-SAMPLE TESTED
Bollinger Band Bounce
Price hit the lower band, so it's due a bounce, right? Buy the band on crypto and you buy the knife.
Mean reversion
Spot · daily · Bollinger(20,2)
10 coins · out-of-sample
Failed out-of-sample
The claimIt touched the lower band, so it's due a bounce.
The realityBeat holding on only 1 of 10 coins, and mostly lost money doing it. Worse than nothing.
Cause of death Bought the knife
01 The seduction
Price touches the lower Bollinger band = two standard deviations cheap = oversold = due a bounce back to the middle. It's the textbook mean-reversion trade, and on a range-bound chart it prints.
Bollinger Band Bounce · BTC
+7%
$1,000 → $1,072 vs $7,766 holding
Beat buy & hold
1 / 10
almost never
Out-of-sample
4 / 10
faded out-of-sample
Simulated on real historical prices, fees included, and past results are not the future. See how we grade →
02 The one line that ends it
Failed
Bollinger Band Bounce vs. doing nothinglog scale
The strategyBuy & hold (doing nothing)
Bollinger Band Bounce beat holding on 1 of 10 majors — and 4 of 10 out-of-sample. Same money, same window, fees included — measured against simply buying Bitcoin once and never touching it.
03 The evidence
| Coin | Trades | Win % | Strategy | Buy & hold | Verdict |
|---|
| BTC | 41 | 59 | +7% | +677% | lost |
| ETH | 38 | 50 | −22% | +1236% | lost |
| BNB | 39 | 67 | +50% | +2761% | lost |
| SOL | 33 | 70 | −4% | +2766% | lost |
| XRP | 42 | 76 | +84% | +365% | lost |
| LINK | 35 | 63 | +1% | +140% | lost |
| LTC | 38 | 68 | +23% | −36% | beat |
| DOGE | 33 | 48 | −69% | +2138% | lost |
| ADA | 36 | 53 | −82% | +173% | lost |
| AVAX | 29 | 45 | −96% | +60% | lost |
Beat holding on 1 of 10 coins — and on 4 of 10 in the recent out-of-sample period (the honest test that separates a real edge from a bull-market coincidence).
04 The other direction — what if you also shorted?
The headline test sits in cash when it is not long. The fair question is whether trading both ways helps — so we ran the identical rule with the flat side flipped to a 1x short (still no leverage), same fees, same ten coins.
| View | BTC $1k→ | Median return | Median maxDD | Beat hold | Verdict |
|---|
| Long / flat · headline | $1,072 | +7% | 66% | 1/10 | Failed out-of-sample |
| Long / short 1x · illustrative | $155 | −85% | 100% | 1/10 | Failed out-of-sample |
Trading it both ways was worse on the headline test — median return −2% → −99%, worst drawdown 66% → 100%. And that is before funding: a short is not spot, it needs a derivative that pays borrow/funding (~10–30%/yr) this omits — add a realistic 20%/yr and the median return drops to −100%. Shorting a market that mostly went up is a wrecking ball, not a rescue.
Why the short side is illustrative
You cannot short on spot — it needs a derivative (a perpetual or margin), which is why the headline test is long/flat. These short figures are shown at 1x with no funding or borrow cost, so the real-world version is worse than what you see here, not better. The window is also net-bullish; the short side only pays in a sustained bear, and betting on that is a different strategy entirely.
05 Verdict
The worst in the set. It beat simply holding on one coin out of ten. Mean reversion is a bet that price comes back; crypto majors trend, so mostly it doesn't, and you're left holding the fall.
Method — the headline test is long/flat spot, no leverage (the long/short view above is a 1x illustration, funding excluded) · a signal read at the daily close is filled at the NEXT close (a full bar of lag — no look-ahead) · 0.06% fee each side, buy-and-hold charged one entry fee too · each coin over its available Binance daily history with warm-up · out-of-sample = held-back most-recent ~45% of each coin's data (a recency holdout, not rolling walk-forward). Honest limits: the ten coins are today's majors (survivorship bias) and are correlated, so “beat on N of 10” is not N independent tests. Reproducible.
The honest other side
There isn't much of an other side to this one. It assumes the range holds. When the range breaks — which on crypto is most of the time — the band just gets ridden all the way down and you're long the entire way.
06 Your turn
That was the textbook version.
We tested it exactly as it's usually described. Yours is probably a little different — a tweak to the settings, an added filter, a different exit. You don't need to code any of it: just describe your version to Vera, our assistant, in plain English, and she builds it and runs it through this same honest test. Most don't survive. A few do — we've proven that. Worth knowing which yours is.
$1.99per strategy
full report + code
Coming soon
✓ The price is shown up front, right here. You will never build a test and then get hit with a paywall to see the result.