The theory is perfect
Mean reversion is one of the oldest trading principles: when price deviates too far from its average, it tends to snap back. Buy when RSI is oversold. Sell when RSI is overbought. Enter near the Bollinger Band extremes, target the middle band. In ranging markets, this is a money printer.
In equities, mean reversion strategies have decades of empirical support. SPY, large-cap stocks, even some commodity markets exhibit strong mean-reverting behaviour during certain regimes. The edge is real and well-documented.
So we built one for crypto.
The Ranging strategy: 8 signals, maximum rigour
We didn't build a toy. The Ranging strategy uses an 8-signal confluence system that fuses RSI and Bollinger Bands with multiple confirmation layers:
| Signal | What it detects |
|---|---|
| Bollinger Band revert | Price touched the band and started moving back |
| RSI crossback | RSI crossed back from oversold/overbought zone |
| RSI divergence | Price makes new low but RSI doesn't — hidden momentum shift |
| %B extreme | Bollinger %B at statistical extreme (< 0.35 or > 0.65) |
| MFI exhaustion | Money Flow Index at extreme — buying/selling pressure fading |
| Daily RSI regime | Higher timeframe confirms the range, not trending |
| Volume spike | Above-average volume on the reversal candle |
| Wide bands | Bollinger bandwidth confirms there's room for the revert |
Minimum score: 5 out of 8. Plus hard vetoes: no entry if ADX indicates trending, if bands are too narrow, or if price hasn't actually started reverting. Stop loss at the Bollinger band edge minus an ATR buffer. Take profit at the middle band.
This is arguably the most carefully engineered strategy in our codebase. More signals, more vetoes, more edge cases handled than any other.
The backtest results
| Pair | Return | Profit Factor | Win Rate |
|---|---|---|---|
| BTC/USDT | -9.3% | 0.82 | 52% |
| ETH/USDT | -3.8% | 0.94 | 48% |
| SOL/USDT | -7.2% | 0.90 | 45% |
| BNB/USDT | -1.1% | 1.03 | 50% |
Negative on all four pairs. BNB was essentially flat (1.03 profit factor is break-even after fees). BTC was the worst at -9.3%. An 8-signal, heavily filtered, carefully vetoed mean reversion strategy lost money on every major crypto pair.
Why crypto breaks mean reversion
The answer is structural, not tactical. No amount of signal tuning fixes this:
1. Crypto trends too hard. When BTC decides to move, it doesn't oscillate politely around a mean. It moves 20-40% in one direction over weeks. Mean reversion entries on the way down become underwater positions that never revert. The "mean" itself moves.
2. Ranging windows are too short. True low-ADX ranges in crypto last days, not weeks. By the time your signals confirm a range, accumulate confluence, and trigger an entry, the range is almost over. The breakout catches you in a mean reversion position pointing the wrong way.
3. ADX < 25 still includes mini-trends. Our ranging regime gate is generous. But even within that gate, crypto can trend enough to blow through Bollinger Bands without reverting. A "low ADX" reading in crypto would be a moderate trend in forex.
4. Volatility clusters destroy fixed take-profits. The middle Bollinger Band target assumes price reverts at a predictable pace. In crypto, volatility clusters mean price can blast through the middle band and keep going, or stall short of it for bars before reversing again.
What we learned
Honest backtesting is the only filter that matters. The strategy was theoretically sound. The signals were well-researched. The implementation was clean. But the data said no. We trust the data.
More signals don't overcome a missing edge. We threw 8 signals at the problem. We added vetoes, divergence detection, multi-timeframe confirmation. None of it mattered because the underlying premise — "crypto prices revert to the mean in ranging markets" — is too weak to trade on.
Killing a strategy is a feature, not a failure. We kept the code. The Ranging strategy is available in the backtester so you can verify our results. But we won't deploy it on live accounts until the evidence changes. Having five strategies where two are profitable is worse than having two strategies where both are profitable.
When might it work?
We haven't given up entirely. Mean reversion might work in crypto under specific conditions:
- Stablecoin-adjacent pairs — pairs with very low natural volatility might revert more reliably.
- Very low ADX periods — tightening the gate to ADX < 15 might filter out the mini-trends. But this reduces the number of setups to near zero.
- Higher timeframes — weekly Bollinger Bands revert more reliably than 4-hour. But the trade frequency drops to 2-3 per year.
For now, the Ranging strategy is shelved. We'd rather tell you it doesn't work than let you trade a strategy we know is unprofitable. That's the difference between a platform that sells features and one that trades its own money.
The most profitable thing a strategy can do is tell you when not to trade. The Ranging strategy told us: not here, not now.← Back to blog