The Thesis
Mean-reversion is one of the oldest ideas in systematic trading, and this agent runs a textbook version of it: buy when a name is oversold (RSI below 30), sell when it's overbought (RSI above 70). It operates over a 24-name large-cap universe spanning tech, financials, healthcare, staples, and energy — AAPL, MSFT, NVDA, JPM, V, JNJ, WMT, XOM, and peers. The bet is simple: short-term price extremes in liquid blue chips tend to snap back.
Backtest: Encouraging on the Surface
Over 451 days the strategy compounded to a final equity of $11,473, a total return of +14.73% (roughly 7.98% CAGR). The win rate is a headline-grabbing 70.59% across 38 trades, with a Sharpe of 0.58 and a max drawdown of 15.64%. Fees were negligible at $38.
Two caveats temper the enthusiasm. First, turnover ran to 879% — this is an active strategy, and in a live setting slippage would eat into that clean fee number. Second, a Sharpe under 0.6 alongside a 15.6% drawdown means the ride is bumpier than the return alone suggests.
Validation: The Part That Matters
Here the story turns. Walk-forward validation did not pass. Across four folds, three were positive — folds one through three returned +2.06%, +11.10%, and +2.21% — but the most recent fold (Dec 2025 to May 2026) came in at -2.84% with a Sharpe of -0.33. That out-of-sample weakness is the crux: the strategy's edge is concentrated in fold two (Sharpe 1.32) and thins out badly toward the present.
The deflated statistics confirm the caution. The Probabilistic Sharpe Ratio sits at a respectable 0.785, but the Deflated Sharpe Ratio — which penalizes for the six trials run — drops to 0.304, well short of the confidence you'd want before trusting the edge is real rather than a product of search.
Live Activity: Quiet, and Drifting Down
The live paper account tells its own story. The last executed trade was a single buy of 21 WMT shares at $115.75 on May 31. Since then, six consecutive scheduled runs (July 17–24) have executed nothing and rejected nothing — no name has hit an RSI extreme. Meanwhile total account value has slipped from $9,975 to $9,867, leaving the portfolio below its $10,000 starting line with $7,569 parked in cash.
That idleness is itself informative: in a trending, non-extreme market, a mean-reversion agent simply has nothing to do.
Verdict
Mean-reversion has a genuinely attractive full-sample profile — high win rate, positive return, low fees. But the negative most-recent fold, the sub-threshold DSR, and the flat, cash-heavy live book all point the same direction: the edge is unconfirmed out-of-sample. Keep it live for observation, but treat the 14.73% as a hypothesis awaiting proof, not a track record.