ROC-12 mean reversion killed by negative validation Sharpe

We killed this mean-reversion strategy before it ever saw forward testing. The idea was simple: buy when 12-period rate-of-change drops below -2%, targeting a quick bounce in low-ADX (ranging) markets. The backtest ran 231 out-of-sample trades across BTC, ETH, XRP, and BNB on 5-minute bars during the 12:00 to 22:00 UTC session window, and the numbers told us everything we needed to know.

Average P&L per trade in validation was -0.003%. Win rate was 82.3%, which sounds impressive until you realize the wins were smaller than the losses. The deflated Sharpe ratio, which adjusts for multiple testing and the risk we cherry-picked a fluke, came in at -0.03. The probability this strategy has real edge after accounting for how many candidates we tested hit 100%, meaning the backtest itself flagged this as statistically indistinguishable from noise.

The strategy used a net profit exit, meaning it closed positions once they hit a small gain or a stop. That 82% win rate came from capturing tiny mean-reversion bounces, but the 18% of trades that lost apparently gave back more than the wins collected. The ADX filter was supposed to keep us out of strong trends where mean reversion fails, but it didn't save the P&L.

We soft-killed this gene under the strict audit gate on May 2, 2026. The kill is reversible if we refine the gate logic, but right now the backtest's own validation metrics say there's insufficient evidence of edge. No forward trades were ever placed.

The lesson here is that high win rate alone means nothing. A strategy that wins 8 out of 10 times but bleeds on the losses is just expensive false confidence. Mean reversion on rate-of-change can work, but this parameter set and exit logic didn't survive contact with out-of-sample data.

Check the full killed registry at stratproof.com/survivors to see what did pass, or test your own mean-reversion idea at stratproof.com/prove with stricter exit rules baked in from the start.