The thesis
Mean-reversion is one of the oldest ideas in systematic trading, and this strategy states it plainly: buy oversold names (RSI below 30) and sell overbought ones (RSI above 70). It trades a concentrated universe of 24 US large-caps — mega-cap tech, payments, healthcare, staples, and energy — the kind of liquid, well-covered stocks where fewer structural surprises get in the way of a clean signal.
Backtest headline
On paper the numbers are inviting. Over a 451-day window the strategy returned 14.73% (7.98% CAGR), finishing on $11,473 of equity with a 70.59% win rate across 38 trades. Drawdown topped out at 15.64% and the Sharpe ratio landed at 0.58. Costs were trivial — $38 in fees, no FX drag. The high win rate is characteristic of mean-reversion: many small, frequent wins. But note the 879% turnover: this strategy churns hard, and in a higher-friction environment those thin margins would compress fast.
The validation problem
Here is where enthusiasm should cool. Walk-forward validation failed. Across four out-of-sample folds, three were positive — folds 2 and 3 posted 11.1% (Sharpe 1.32) and 2.21% respectively — but the most recent fold (Dec 2025 to May 2026) lost -2.84% at a Sharpe of -0.33 and a 14.96% drawdown. The aggregate out-of-sample return is negative, and the deflated Sharpe ratio (DSR) of just 0.304 against 6 trials tells the real story: once you adjust for how many variants were tried, the edge is barely distinguishable from luck. The probabilistic Sharpe (0.785) is healthier, but DSR is the harsher, more honest judge, and it says the recent-period edge has thinned.
Live activity: mostly silence
Despite being marked live, the strategy has essentially gone quiet. The only executed trade on record is a single WMT buy — 21 shares at $115.75 on 31 May 2026. Every scheduled run since mid-August reports 0 executed, 0 rejected, with cash parked at $7,569.25 and total account value drifting between roughly $9,730 and $9,998. That total sits below the $10,000 baseline, so the live book is modestly underwater. The zero-execution streak is consistent with the thesis, not a bug: if nothing in the universe is genuinely oversold or overbought, the strategy correctly does nothing.
Verdict
Mean-reversion has a coherent thesis, a clean cost profile, and a genuinely strong hit rate. But the deteriorating final fold, the negative out-of-sample aggregate, and a DSR barely above 0.3 mean it has not earned unconditional capital. The current live posture — mostly in cash, waiting for a real signal — is arguably the right one. We would keep it on the bench, watch the next validation refresh closely, and treat any live gains until then as unproven.