The thesis
Mean-reversion runs one of the oldest ideas in technical trading: buy weakness, sell strength. Concretely, it accumulates names when the 14-day RSI dips below 30 (oversold) and exits when RSI pushes above 70 (overbought). The universe is 24 large-cap US blue chips — AAPL, MSFT, NVDA, JPM, JNJ, WMT and peers — a deliberately liquid, low-idiosyncratic-risk basket that keeps the signal, not single-name blowups, in the driver's seat.
Backtest scorecard
On paper the strategy looks appealing. Over 451 trading days it returned 14.73% (final equity $11,473 on a $10k base), a 7.98% CAGR, with a 70.59% win rate across 38 trades. That hit rate is the headline strength: mean-reversion systems tend to win often and small, and this one fits the pattern.
The caveats sit right next to the wins. The Sharpe ratio is a modest 0.58, and the max drawdown of 15.64% is roughly equal to the entire annual return — a thin margin of reward over risk. Turnover of 879% is also high, meaning performance leans on frequent round-trips; the $38 in modelled fees is trivial here, but slippage in a live book would bite harder than a backtest suggests.
Where validation pushes back
This is the part that matters, and it is not flattering. Walk-forward validation failed. Across four out-of-sample folds, three were positive, but the most recent one (2025-12-16 → 2026-05-29) returned -2.84% with a -0.33 Sharpe and a 14.96% drawdown. So the newest, most decision-relevant slice of history is exactly where the edge disappeared.
The robustness statistics tell the same story with more nuance. The Probabilistic Sharpe Ratio of 0.785 is reasonable, but the Deflated Sharpe Ratio is just 0.304 — once you penalise for the 6 trials run in search of this configuration, the confidence that the true Sharpe exceeds zero is weak. A strong full-sample number (14.73%) alongside a negative out-of-sample tail is the classic fingerprint of a strategy that fit the past better than it will fit the future.
Recent activity: mostly waiting
Live paper trading has been quiet to the point of dormant. The last executed trade was a buy of 21 WMT at $115.75 on 2026-05-31. Every scheduled run since — July 7 through 14 — reports 0 executed, 0 rejected: no names have hit the RSI<30 or >70 thresholds. The book holds $7,569 cash against a total value hovering around $9,957, meaning the live account is fractionally below its $10k start. In a market with no oversold extremes, a mean-reversion strategy simply has nothing to do, and that inactivity is currently a small drag rather than a source of returns.
Verdict
Mean-reversion is a coherent, high-win-rate strategy with a respectable historical return — but its risk-adjusted profile is thin, its deflated Sharpe is unconvincing, and it flunked out-of-sample validation on the freshest data. The recent stretch of do-nothing runs is a feature of the design, not a bug, yet it underscores the real question: when signals do fire, will the edge still be there? Until the validation gate clears, this one belongs on a short leash.