Thesis
The mean-reversion agent runs one of the oldest ideas in systematic trading: prices overshoot and snap back. Concretely, it buys when the 14-period RSI drops below 30 (oversold) and sells when it climbs above 70 (overbought), across a 24-name universe of large-cap US equities spanning tech, financials, healthcare, staples, and energy. It's a clean, interpretable rule — no black box, easy to audit.
Recent Activity
Activity has gone dormant. The last six scheduled runs (2026-07-30 through 2026-08-06) each report 0 executed, 0 rejected — the RSI conditions simply haven't triggered. The paper account sits at roughly $9,913 total on $7,569 cash, essentially flat day to day and marginally below the $10,000 notional start, implying the deployed book is slightly underwater on a live basis. The most recent actual fill was a 21-share WMT buy at $115.75 back on 2026-05-31 — over two months ago. A mean-reversion strategy is expected to sit on its hands when nothing is oversold, so quiet tape isn't a malfunction, but it does mean the live edge is currently untested by fresh signals.
Backtest Performance
On paper the numbers look attractive. Over 451 days the strategy returned 14.73% (final equity $11,473), a 7.98% CAGR, with a 70.59% win rate across 38 trades and just $38 in fees. The catch sits in the risk profile: a 0.58 Sharpe is modest, the 15.64% max drawdown nearly matches the total return, and 879% turnover signals a lot of churn. A high win rate paired with a middling Sharpe usually means the losers are larger than the winners — worth watching.
Validation: The Red Flag
Here the story turns cautionary. Walk-forward validation failed. Across four folds, three were positive — but the most recent and most relevant fold (2025-12-16 to 2026-05-29) delivered -2.84% with a -0.33 Sharpe. That out-of-sample decay is the tell: the strategy's strength was concentrated in fold 2 (+11.1%, 1.32 Sharpe), while folds 1, 3, and 4 were weak or negative.
The deflated statistics reinforce this. With 6 trials, the Probabilistic Sharpe Ratio of 0.785 is borderline, but the Deflated Sharpe Ratio of just 0.304 — which penalizes multiple testing — suggests the observed edge may not survive scrutiny. In plain terms: the full-sample 14.73% looks better than the strategy probably deserves.
Verdict
Mean-reversion is a genuinely defensible thesis with an honest, readable rule set and a respectable headline backtest. But the evidence urges caution, not conviction. The negative most-recent fold, the drawdown roughly equal to total return, and a sub-1 DSR all point to a strategy that may have been favored by a specific regime rather than a durable edge. The current live drought is a natural moment to reassess — perhaps loosening the RSI thresholds, adding a trend filter to avoid catching falling knives, or trimming the universe. Until out-of-sample behavior stabilizes, this one stays firmly in the watch-and-learn column.