The thesis
The mean-reversion agent runs one of the oldest ideas in systematic trading: buy oversold, sell overbought. Concretely, it enters when RSI drops below 30 and exits when RSI climbs above 70, across a 24-name universe of large-cap US equities (AAPL, MSFT, NVDA, JPM, V, JNJ, WMT, XOM and peers). It's currently flagged live.
The appeal is simplicity: no regime forecasting, no earnings models — just a bet that stretched prices snap back toward their average. On the right names in the right conditions, that bet pays off often.
In-sample performance
The backtest reads well at first glance. Over 451 days the strategy returned 14.73% (7.98% CAGR), finishing at $11,473 on a $10,000 book, with a 70.59% win rate across 38 trades. A win rate that high is the signature of a working mean-reversion system — small, frequent gains as prices revert.
But the supporting stats temper the enthusiasm. The Sharpe ratio is a modest 0.58, max drawdown reached 15.64%, and turnover was a heavy 879% — a lot of churn for the return, even if fees stayed low at $38 total. High hit rate plus middling Sharpe usually means the losers are bigger than the winners.
Validation: it didn't pass
This is where the story turns. Walk-forward validation failed. Across four out-of-sample folds, three were positive — folds 1–3 returned +2.06%, +11.10% (Sharpe 1.32), and +2.21% — but the most recent fold (Dec 2025–May 2026) lost -2.84% at a -0.33 Sharpe with its deepest drawdown yet (14.96%).
The aggregate out-of-sample numbers are the red flag: -2.84% OOS return and -0.33 OOS Sharpe. The Probabilistic Sharpe Ratio sits at a respectable 0.785, but the Deflated Sharpe Ratio — which penalizes for the 6 trials run — collapses to 0.304. In plain terms, once you account for how many variants were tested, the edge looks a lot more like luck, and the deterioration is concentrated in the newest data.
Recent activity: gone quiet
Live behavior echoes the concern. The last executed trade was a 21-share WMT buy at $115.75 on May 31. Since then, six consecutive scheduled runs (July 28 – Aug 4) logged 0 executed, 0 rejected — no name in the universe hit the RSI thresholds. Cash has been parked at $7,569.25, and the paper book's total value hovers around $9,900, below where it started.
The verdict
Mean-reversion is a coherent, cheap-to-run strategy with a genuinely strong hit rate, and it performed honestly in three of four folds. But the failed validation, negative out-of-sample Sharpe, deflated DSR, and a live book that is both underwater and idle all point the same direction: the edge is fragile and regime-dependent. It's a reasonable research candidate — not a strategy to size up. Watch whether it re-engages when volatility returns; a strategy that only wins when it trades needs signals to fire.