The thesis
Mean-reversion is one of the oldest ideas in systematic trading: prices that stretch too far from their recent range tend to snap back. This strategy implements it with a single, transparent rule — buy when the 14-day RSI drops below 30 (oversold), sell when it climbs above 70 (overbought). It runs live against a universe of 24 US large caps spanning tech, financials, healthcare, staples and energy — names like AAPL, NVDA, JPM, JNJ, XOM and CAT. The appeal is simplicity: no regime models, no machine learning, just a well-understood oscillator.
Recent activity
The strategy trades patiently. Over the past week (2026-10-21 through 2026-10-28) most scheduled runs executed nothing — a sign that few names in the universe reached RSI extremes. The exceptions: a 7-share GOOGL buy at $322.21 on 2026-10-27, and on 2026-10-23 a two-sided day that bought 3 CAT at $688.79 while selling 8 HD at $318.76. That HD exit is instructive — the shares were bought at $299.12 on 2026-09-21, so the overbought signal closed a position in the green. A WMT round-trip (bought $115.75 in May, sold $118.83 on 2026-10-16) tells the same modest-but-positive story. The paper book stands near $10,096 total with roughly $1,032 in cash.
Backtest performance
Over 451 days the backtest returned 14.73% (7.98% CAGR), with a striking 70.59% win rate across 38 trades and total fees of just $38. That hit rate is the headline strength — most individual round-trips make money. But the quality metrics temper the enthusiasm: a Sharpe of 0.58 is pedestrian, and a 15.64% max drawdown is meaningful for a strategy earning single-digit annualised returns. Turnover of 879% confirms the churn inherent to oscillator trading — a high win rate on many small trades, not a few large winners.
The validation problem
Here is where caution is warranted. Walk-forward validation failed. Across four folds, three were positive, but the most recent — 2025-12-16 to 2026-05-29 — returned −2.84% with a Sharpe of −0.33 and a 14.96% drawdown. That the degradation lands in the newest, most forward-looking window is the worst place for it. The deflated Sharpe ratio of 0.304, computed against 6 trials, signals that some of the in-sample edge is likely a product of selection rather than genuine signal. A probabilistic Sharpe of 0.785 is reassuring but not decisive.
Verdict
Mean-reversion earns its keep as a readable, low-cost, high-hit-rate strategy, and its live trades so far have closed profitably. But a high win rate masks an asymmetric payoff and modest risk-adjusted returns, and the failed out-of-sample fold is a genuine red flag — the edge may not survive the next regime. It belongs in the lab under close watch, not scaled up, until a future fold restores confidence.