The thesis
Mean-reversion runs one of the oldest ideas in technical trading: prices overshoot and then snap back. Concretely, it buys names whose 14-day RSI drops below 30 (oversold) and sells when RSI climbs above 70 (overbought). It trades a 24-name large-cap universe spanning tech, financials, healthcare, staples and energy — AAPL and NVDA alongside JNJ, KO and XOM. The strategy is currently live on paper capital.
Recent activity
The last two weeks have been quiet. Of the six scheduled runs between 14 and 21 October, five executed nothing at all — unsurprising, since a mean-reversion engine only acts when something is genuinely stretched, and little was. The one exception was 16 October, when it sold 21 shares of WMT at $118.83, closing a position opened at $115.75 back in May for a thin gain. The earlier buys — MCD, HD and BAC through September — reflect the oversold side of the book being put to work. As of 21 October the account holds $2,819.82 in cash against a total of $9,915.80, which means the live sleeve is currently a hair underwater versus its $10k start, even as the backtest shows a profit.
Backtest and validation
On paper the headline numbers flatter the strategy. Over 451 days it returned 14.73% (7.98% CAGR), with a 70.59% win rate across 38 trades and a max drawdown of 15.64%. That win rate is the strongest argument in its favour — it is right far more often than not.
The caveats arrive quickly. The Sharpe ratio is only 0.58, which tells you the losers are meaningfully larger than the frequent winners — a classic mean-reversion asymmetry where the occasional trade that keeps falling erases many small wins. Turnover of 879% also signals a lot of churn for that return.
Most important, validation did not pass. Across four walk-forward folds, three were positive — including a strong fold 2 (+11.1%, Sharpe 1.32) — but the most recent fold (Dec 2025–May 2026) lost 2.84% with a negative Sharpe of −0.33 and a 14.96% drawdown. The out-of-sample window is exactly that losing stretch. After adjusting for the six trials run, the deflated Sharpe ratio is 0.30 — below the 0.5 threshold we'd want before trusting the edge is real rather than a product of selection.
Verdict
Mean-reversion is a coherent, well-behaved strategy with a genuinely high hit rate, and its quiet recent tape is consistent with its discipline. But the failed validation and negative out-of-sample fold are real warnings: the most recent regime did not reward it, and the deflated statistics suggest its edge is fragile. Worth watching live — not yet worth trusting with size.