The thesis
Mean-reversion is one of the oldest ideas in systematic trading, and this strategy states it plainly: buy the oversold (RSI below 30), sell the overbought (RSI above 70). It runs over a 24-name universe of US large caps — the mega-cap tech cohort (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare, staples, and industrials. The bet is that short-term price extremes in liquid, well-covered names tend to snap back. It is currently live.
Backtest performance
The headline numbers are respectable. Over 451 days the strategy returned 14.73% (a 7.98% CAGR), turning a notional $10,000 into $11,473. The win rate is a flattering 70.59% across 38 trades, and max drawdown stayed at a tolerable 15.64%.
The texture underneath is less comfortable. The Sharpe ratio is 0.58 — modest for a strategy winning seven trades in ten, which tells you the losers are considerably larger than the winners, exactly the asymmetry mean-reversion is prone to when a "cheap" name keeps falling. Turnover of 879% is also high, though at $38 total fees the frictional cost was negligible in the sim.
Validation: the warning light
This is where enthusiasm should cool. Walk-forward validation failed. Of four out-of-sample folds, three were positive — but the most recent (Dec 2025 → May 2026) returned −2.84% with a −0.33 Sharpe and a 14.96% drawdown. The deterioration is directional, not random: the strategy worked best in the 2025 H1 fold (+11.1%, Sharpe 1.32) and has weakened since.
The probabilistic metrics echo this. The Probabilistic Sharpe Ratio of 0.785 is borderline, but the Deflated Sharpe Ratio is just 0.304 — once you penalise for the six trials run, the edge is no longer convincing. In plain terms: the full-sample 14.73% is partly a product of fitting, and recent out-of-sample behaviour does not support it.
Recent activity
Live trading has been quiet. The last six scheduled runs mostly executed zero trades — unsurprising, since RSI extremes are infrequent in a calm market. The exception was 23 Oct, when it sold 8 HD at $318.76 (bought 21 Sep at $299.12, a clean winner) and bought 3 CAT at $688.79. The book sits around $9,999 total with roughly $3,293 in cash, meaning it is only partially invested and waiting for signals. That patience is a feature, not a bug, for this style.
Verdict
Strengths: a simple, explainable thesis, a high hit rate, and contained drawdowns. Risks: a low Sharpe that betrays fat-tailed losers, high turnover, and — most importantly — a failed validation driven by a loss-making recent fold and a weak deflated Sharpe.
The honest read is that mean-reversion looks good in aggregate but is decaying at the edge. Keeping it live in a small, paper-sized allocation is defensible as a live-data probe; sizing it up on the backtest number would be a mistake. Watch the next out-of-sample window closely — if fold-4 weakness persists, this is a retirement candidate.