The thesis
Mean-reversion is one of the oldest ideas in systematic trading, and this agent runs the textbook version: buy when the 14-period RSI falls below 30 (oversold), sell when it climbs above 70 (overbought). It operates across a 24-name universe of large-cap US equities — megacap tech (AAPL, MSFT, NVDA), financials (JPM, V, MA), staples (PG, KO, WMT, COST) and healthcare (JNJ, UNH). The premise is that liquid blue chips overshoot in both directions and snap back toward a mean, so the edge comes from disciplined entry timing rather than directional conviction.
Backtest performance
Over a 451-day window the strategy returned 14.73% (final equity $11,473 on a $10k base), a 7.98% CAGR, with a Sharpe of 0.58 and a maximum drawdown of 15.64%. The headline number that flatters it is a 70.6% win rate across 38 trades — most positions closed green. But two caveats stand out. First, turnover ran to 879%, an aggressive churn that would be far more punishing under real-world slippage than the modest $38 of modelled fees suggests. Second, a Sharpe below 0.6 with a drawdown near 16% is a thin risk-adjusted profile; the high win rate implies many small wins offset by fewer, larger losers — a payout shape that punishes any regime shift.
The validation problem
Here the story turns. The four-fold walk-forward test did not pass. Three of four folds were positive, and the strong middle fold (Jan–Jul 2025, +11.1%, Sharpe 1.32) does most of the heavy lifting. The crucial, most-recent fold — Dec 2025 to May 2026 — lost -2.84% with a -0.33 Sharpe and a 14.96% drawdown. The aggregate out-of-sample return of -2.84% and OOS Sharpe of -0.33 are the honest measure of forward expectancy, and both are negative.
The deflated statistics tell the same tale: a Probabilistic Sharpe Ratio of 0.785 is decent, but the Deflated Sharpe Ratio of 0.304 — which penalises the six trials run during selection — sits well below the 0.5 confidence threshold. In plain terms, the in-sample edge may be an artefact of selection rather than a durable signal.
Recent live activity
The live book reflects this hesitancy. The last actual fill was a WMT buy (21 shares @ $115.75) on 31 May; since then the scheduled runs from 20–27 August have all reported 0 executed, 0 rejected — no name has been oversold or overbought enough to trigger. Cash has sat flat at $7,569.25 while total equity drifted between roughly $9,724 and $9,795, leaving the paper account modestly below its starting line.
Verdict
The strengths are real: a coherent thesis, a clean rule set, and a high hit rate. The risks are equally real: negative out-of-sample results, a failing deflated Sharpe, punishing turnover, and a live book that is idle and slightly underwater. This is a strategy worth watching, not scaling — the next few active trades will tell us whether the recent losing fold was noise or the start of a broken regime.