The thesis
Mean-reversion makes a simple, old bet: prices that stretch too far snap back. Operationally it buys names whose 14-day RSI falls below 30 (oversold) and sells those pushing above 70 (overbought), across a 24-stock large-cap universe spanning tech (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare, staples and energy. It's a counter-trend, mechanical rule — no discretion, no narrative. That discipline is the appeal; it's also the limitation, because the strategy only acts when the oscillator hits an extreme.
Recent activity
That limitation is visible right now. Since the 29 October reset the strategy opened three positions: BAC (48 shares @ $51.15) on the 29th, then CAT (4 @ $622.54) and GOOGL (8 @ $280.28) on the 30th. Every scheduled run since — the 2nd through the 5th of November — reports the same line: 0 executed, 0 rejected. No name has been oversold or overbought enough to trip the rule. Meanwhile total equity has drifted from $9,976 down to $9,675, leaving the book roughly 3% underwater with $2,742 in cash sitting idle. Four quiet sessions in a row is the strategy working as designed, not a malfunction — but it does mean capital is doing little while it waits.
Backtest performance
The historical simulation is the strong side of the story. Over 451 days it returned +14.73% (CAGR 7.98%), with a 70.59% win rate across 38 trades, a Sharpe of 0.58 and a max drawdown of 15.64%. Fees were negligible ($38 total, no FX cost). A high hit-rate with a middling Sharpe is characteristic of mean-reversion: many small wins, punctuated by sharp losses when a falling knife keeps falling. Turnover of 879% confirms it trades actively when signals fire.
Validation and risks
Here the picture turns cautionary. The walk-forward test ran four folds; three were positive — but the fourth and most recent (Dec 2025–May 2026) lost -2.84% with a -0.33 Sharpe. That final fold is the out-of-sample window, so the strategy's freshest, least-fitted evidence is negative. The probabilistic Sharpe (PSR 0.785) is respectable, but the deflated Sharpe (DSR 0.304), which penalises the 6 trials behind this configuration, is weak. The validation gate's verdict: failed.
Translation: the in-sample returns are real but may be partly a product of selection across trials, and the regime that worked in 2024–25 stopped paying in the latest window. The 70% win rate is genuine edge; the failed out-of-sample and low DSR are genuine overfitting risk. Treat this one as live-but-on-probation — worth watching for whether the recent drought breaks into profitable signals, not yet worth sizing up.