The thesis
The mean-reversion strategy runs a classic, easy-to-audit rule across a 24-name large-cap universe (AAPL, MSFT, NVDA, JPM, JNJ, WMT and peers): buy when RSI falls below 30, sell when it climbs above 70. The bet is simple — oversold names snap back, overbought names cool off. Simplicity is a genuine strength here: there is little room for hidden curve-fitting, and every trade is explainable.
Backtest performance
On paper the numbers are attractive. Over 451 days the strategy returned 14.73% (final equity $11,473 on a $10k base), a 7.98% CAGR, with a 70.59% win rate across 38 trades. Costs were negligible — $38 in fees, no FX drag.
But the quality metrics temper the enthusiasm. The Sharpe of 0.58 is modest for a strategy leaning on a 70% win rate, which tells us the average loser is meaningfully larger than the average winner — reversion strategies tend to "pick up pennies" and occasionally cop a full drawdown. The 15.64% max drawdown and an 879% turnover confirm this profile: frequent trading, thin per-trade edge, and real tail risk.
The validation gate says no
This is the headline caveat. Walk-forward validation failed. Three of four out-of-sample folds were positive, but the most recent fold (Dec 2025 → May 2026) returned −2.84% with a −0.33 Sharpe and a 14.96% drawdown — the worst of the four. Deflated Sharpe (DSR) sits at just 0.304 against a probabilistic Sharpe (PSR) of 0.785 across 6 trials. In plain terms: once we account for how many variants were tried, the evidence that this edge is real — rather than lucky — is weak. The strategy's own gate correctly refuses to certify it.
Live activity: quiet, and slightly underwater
The live paper sleeve reflects this ambivalence. The last executed trade was a WMT buy (21 shares @ $115.75) on 31 May. Every scheduled run since — six consecutive sessions through 29 July — logged 0 executed, 0 rejected, with cash pinned at $7,569.25. Total account value has drifted between roughly $9,837 and $9,971, i.e. still marginally below the $10k line, carried by that single Walmart position.
The dormancy isn't a bug; it's the rule working as designed. With most of the universe trading in RSI's neutral middle band, there is simply nothing to do. That discipline is admirable, but it also means the strategy is currently a one-stock bet, not a diversified reversion engine.
Verdict
Mean-reversion is transparent, cheap to run, and boasts an eye-catching win rate — but the failing validation gate and the negative most-recent fold are exactly the warning signs our process exists to catch. Keep it live for observation; do not scale capital into it until an out-of-sample fold recovers and the deflated Sharpe clears the bar.