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 falls below 30 (oversold) and sells when RSI climbs above 70 (overbought). The strategy is live, trading a 24-name universe of US large caps spanning tech (AAPL, MSFT, NVDA), financials (JPM, BAC, V, MA), healthcare (JNJ, UNH, PFE), and staples (PG, KO, WMT, COST). It is a mechanical, single-signal system — easy to reason about, and easy to overfit.
What the backtest says
The headline numbers are attractive. Over 451 days the strategy returned 14.73% (7.98% CAGR), turning a notional $10,000 into $11,473. Its win rate is a striking 70.59% across 38 trades, and fees were negligible at $38 with zero FX cost. The catch is quality-of-return: a Sharpe of 0.58 against a 15.64% max drawdown means the equity curve paid for its gains with real volatility. Turnover of 879% also signals a strategy that trades a lot to earn that Sharpe.
Why validation fails
This is where the story turns. Walk-forward validation split the history into four folds. Three were positive — +2.06%, +11.1%, +2.21% — but the most recent fold (Dec 2025 to May 2026) returned -2.84% with a -0.33 Sharpe and a 14.96% drawdown. That single out-of-sample window is the tell.
The deflated statistics confirm the concern. The Probabilistic Sharpe Ratio is a respectable 0.785, but the Deflated Sharpe Ratio is just 0.304 — once you account for the 6 trials behind this configuration, the edge is no longer statistically convincing. The verdict: validation did not pass. The high win rate looks less like durable skill and more like a strategy that books many small wins and occasionally gives them back in a drawdown.
Live behaviour
Recent activity is quiet. Scheduled runs from Sep 14–18 mostly executed nothing — the RSI thresholds simply weren't tripped. The exceptions were a 42-share BAC buy at $58.21 (Sep 17) and, further back, a 21-share WMT buy at $115.75 in May. More sobering: the live paper account currently sits around $9,808 total against its $10,000 start. Despite the flattering backtest, the live book is modestly underwater — a reminder that in-sample returns and forward results are different things.
The balance sheet
Strengths: a transparent, low-cost, high-win-rate rule set that trades sparingly when signals are absent, and three of four positive folds.
Risks: a failed validation gate, a negative and most-recent out-of-sample fold, a deflated Sharpe that undercuts the edge, a 15%+ drawdown profile, and live performance that has yet to match the backtest.
Verdict
Mean-reversion is a clean idea with a seductive win rate, but the numbers that matter most for deployment — out-of-sample return and deflated Sharpe — argue for caution. We'd treat it as a research candidate, not a capital-allocation decision, until it demonstrates a stable positive out-of-sample window. The backtest earned attention; validation earned skepticism.