The thesis
Mean-reversion is one of the oldest ideas in systematic trading, and this strategy states it plainly: buy when a name is oversold (RSI below 30) and sell when it is overbought (RSI above 70). It runs across a 24-name universe of US large caps — the usual megacap tech (AAPL, MSFT, NVDA), financials (JPM, V, MA), staples (PG, KO, WMT, COST) and a few industrials and energy names. The premise is that short-term price extremes in liquid, well-covered stocks tend to snap back toward a mean.
Backtest: strong on the surface
The headline numbers are attractive. Over 451 days the strategy returned 14.73% (7.98% CAGR) with a 70.59% win rate across 38 trades. A seven-in-ten hit rate is exactly what you would hope for from a mean-reversion engine — it wins often and small.
But two figures temper the enthusiasm. The Sharpe ratio is only 0.58, meaning those wins are not translating into strong risk-adjusted returns, and the max drawdown of 15.64% is roughly the size of the total return itself. Turnover of 879% is also high, though fees came in at a modest $38.
Validation: the strategy fails the gate
This is where the picture darkens. Walk-forward validation across four folds did not pass. Three of four folds were positive, but the pattern matters: folds two and three were healthy (11.1% and 2.2%), while the most recent fold — the truest out-of-sample window (Dec 2025 to May 2026) — returned -2.84% with a Sharpe of -0.33 and a 14.96% drawdown.
The robustness statistics reinforce the caution. The Deflated Sharpe Ratio is just 0.304, well short of confidence after accounting for the six trials run. In plain terms, once you adjust for how many variants were tested, the edge may be closer to noise than signal. Mean-reversion is also structurally fragile: it works until a trend or a shock keeps pushing an oversold name lower, and that is precisely the risk the final fold hints at.
Live: quiet and slightly underwater
Recently the strategy has done almost nothing. Every scheduled run from 19–26 August executed zero trades, sitting on $7,569 in cash with total equity drifting between roughly $9,730 and $9,990 — below the $10k line. Its last actual fill was a WMT buy back on 31 May. With RSI signals not firing, the book is mostly idle cash, which is defensible discipline but also means no edge is being captured.
Verdict
The high win rate is real and the logic is sound, but the evidence does not yet justify conviction. A failed validation gate, a negative most-recent fold, a sub-0.35 deflated Sharpe, and a live account that is flat-to-down all point the same way: treat this as a candidate under observation, not a proven edge. The next meaningful out-of-sample stretch — ideally one that includes a sharp drawdown — will tell us far more than the backtest already has.