Mean reversion is one of the oldest ideas in trading: prices deviate from a fair value and eventually snap back. On paper, it works everywhere. In practice, asset selection is everything.
The Altcoin Sweet Spot
We tested our Bollinger Band mean reversion strategy (mean_reversion_bb) across 25 symbols spanning majors, mid-caps, and micro-caps. The results were stark. High-beta altcoins like PEPE, WIF, DOGE, and SUI produced Sharpe ratios between 17 and 19.25. Meanwhile, BTC and ETH delivered negative Sharpe ratios between -12 and -17 during trending regimes.
The reason is structural. Altcoins oscillate around market-maker equilibrium prices with high amplitude and fast mean reversion speed. They overshoot, correct, overshoot again. Bollinger Bands capture these oscillations naturally because the bands expand during volatility and contract during consolidation, providing dynamic entry and exit levels.
Why BTC and ETH Fail
BTC and ETH are too efficient for simple mean reversion. They trend for weeks or months, driven by macro flows, institutional allocation cycles, and ETF inflows. A mean reversion strategy on BTC will fade a trend repeatedly, accumulating losses until the trend exhausts itself. By the time the snapback arrives, drawdown has already consumed most of the capital.
Our validation data confirmed this across five distinct market regime periods from 2021 through 2026. BTC mean reversion was negative in four out of five periods. The single positive period was a sideways consolidation where any strategy would have struggled to lose money.
The bb_period Discovery
Parameter optimization revealed a surprising split. The original six altcoins (SHIB, DOGE, AVAX, SOL, LINK, SUI) performed best with bb_period=30. But when we expanded to seven new symbols (PEPE, WIF, NEAR, ARB, OP, APT, INJ), bb_period=48 beat bb_period=30 on every single one, improving Sharpe by 4.7 to 6.4 points.
The explanation is liquidity depth. The original six are more liquid with tighter spreads and faster mean reversion cycles. The newer seven are thinner, with wider oscillation periods that align better with a 48-bar lookback. One parameter does not fit all, even within the same strategy.
Practical Deployment
We now run 13 paper bots with mean_reversion_bb: six on the original basket at bb_period=30 and seven on the expanded basket at bb_period=48. All use bb_std=2.5 and min_confidence=0.5. Combined, they represent half our deployed capital.
The key takeaway: strategy selection is not about finding the best strategy. It is about finding the best strategy-asset pairing. Mean reversion is mediocre on most assets and exceptional on a specific class of volatile, oscillating altcoins.