Thesis
Mean-reversion is about as classic as systematic trading gets: buy the oversold (RSI below 30), sell the overbought (RSI above 70). It trades a 24-name universe of US large caps spanning tech (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare, staples, and industrials. The bet is simple — short-term price dislocations snap back — and the broad, liquid universe is well suited to it.
Recent Activity
The strategy is live and running on schedule. In its 2026-10-30 run it executed 2 trades with 0 rejections, and on 2026-10-29 it executed 1. Recent fills include buys of 4 CAT at $622.54, 8 GOOGL at $280.28, and 48 BAC at $51.15 — all firmly on the "buy the dip" side of the book, with no sells, suggesting the model currently reads several names as oversold rather than stretched.
After those runs, total equity sits at $9,876.81 against $2,742.49 in cash — a touch below the $10k starting line. The Reviewer signed off at risk 0, flagging it as a built-in, hand-written strategy rather than machine-generated code.
Backtest vs. Validation
Headline numbers look appealing. Over 451 days the backtest returned 14.73% (7.98% CAGR), with a 70.59% win rate across 38 trades, a Sharpe of 0.58, and a 15.64% max drawdown. Costs were negligible — $38 in fees, no FX.
The catch is that it fails validation. Walk-forward testing split the history into 4 folds; 3 were positive, but the out-of-sample aggregate returned -2.84% with a Sharpe of -0.33. The most recent fold (2025-12-16 to 2026-05-29) is precisely that losing stretch: -2.84% return, negative Sharpe, and a 14.96% drawdown — the deepest of the four. Earlier folds carried the record, especially fold 2's 11.1% at a 1.32 Sharpe.
The deflated metrics tell the same story. A probabilistic Sharpe of 0.785 is middling, but the deflated Sharpe ratio of 0.304 — which penalises for the 6 trials run — is weak, and the negative out-of-sample Sharpe is the real red flag.
Strengths and Risks
Strengths: a high win rate and a genuinely low-cost, interpretable rule set. Mean-reversion is cheap to reason about, and the drawdowns, while non-trivial, stayed contained in three of four folds.
Risks: the 70% win rate is the classic mean-reversion mirage — many small wins masking the occasional large loss when a "dip" is actually a trend. Turnover is a punishing 879%, so any slippage assumptions matter enormously. Most importantly, performance has decayed: the strategy worked in earlier regimes and has not in the latest one.
Takeaway
Mean-reversion earns its place as a live experiment, but its validation failure is not noise — recent out-of-sample behaviour is genuinely negative. Treat the backtest as the optimistic ceiling, not the expectation, and watch whether it can sell the overbought as reliably as it buys the oversold.