ROC-12 Mean Reversion in Low-ADX Regimes

We tested a mean reversion strategy that buys when the 12-period rate of change (ROC) falls below -2%, filtered to trade only when ADX stays below 35. It passed all six gates and graduated to the survivor library with 0.103% average profit per trade and a 76.1% win rate in both walk-forward training and validation.

The strategy hunts for oversold bounces but adds a crucial regime filter: it only enters when ADX indicates weak trend strength. This keeps it out of strong directional moves where mean reversion tends to fail. Trades are limited to a 12:00 to 22:00 UTC session window, and exits trigger on a net profit target of 1.5% (decoded from the tp1.5 parameter in the gene name). We tested it on BTC, ETH, XRP, and BNB using 5-minute bars.

The validation metrics are clean but modest. A deflated Sharpe ratio of 0.47 suggests the edge is real but thin after accounting for multiple testing across the full search space. The 76% win rate is encouraging for a mean reversion approach, though the small 0.103% per-trade profit means transaction costs and slippage will matter in live execution.

Here's where you should stay skeptical. The strategy has zero forward test trades so far, which means we have no live paper trading data to confirm the backtest held up out of sample. The probability of being a fluke is listed at 100.0%, which appears to be a data artifact or placeholder value and does not align with the deflated Sharpe or gate passage. We recommend treating this number as missing data until the research pipeline is updated. The four-coin coverage is narrow, and we don't yet know how this performs on altcoins outside the top tier or in extended high-volatility regimes.

Despite these caveats, the strategy passed our gates honestly and offers a starting template for regime-aware mean reversion. Check the full survivor registry at stratproof.com/survivors to compare this approach against other graduates, or head to stratproof.com/prove to backtest your own ROC or ADX variants with custom parameters.