The thesis
The mean-reversion strategy runs a classic contrarian playbook across a 24-name large-cap universe (AAPL, MSFT, NVDA, JPM, JNJ, XOM and friends): buy when RSI falls below 30, sell when it climbs above 70. The bet is that short-term extremes in liquid blue chips tend to snap back. It's simple, well-understood, and easy to audit — which is exactly why it makes a good honesty test for our validation pipeline.
Recent activity
The strategy is live but currently quiet. Across the six scheduled runs from 4–11 November, only one placed a trade: a 13-share NVDA buy at $188.24 on 6 November, which drew cash down from $2,742.49 to $271.67. Earlier, on 29–30 October, it opened positions in BAC (48 @ $51.15), GOOGL (8 @ $280.28) and CAT (4 @ $622.54). Every one of these was a buy — consistent with an oversold-accumulation stance rather than taking profits.
The live picture is sober. Total equity has drifted between roughly $9,510 and $9,873 over the past week, sitting at $9,767.89 on 11 November — below the $10k starting line. With cash at just $271.67, the book is now almost fully invested and riding its positions.
Backtest performance
Over 451 days the backtest returned 14.73% (final equity $11,473.17), a 7.98% CAGR, with a 70.59% win rate across 38 trades. Drawdown topped out at 15.64% and the Sharpe was a modest 0.58. Fees were trivial ($38, no FX cost), but turnover ran hot at 879% — this is an active strategy that churns its capital many times over.
A high win rate paired with a middling Sharpe is a familiar signature: many small wins punctuated by a few larger losers. The headline return is real, but it leans on that fragile asymmetry.
Walk-forward validation
Here the story turns. Validation failed. Of four out-of-sample folds, three were positive, but the breakdown matters:
- Fold 2 (Jan–Jul 2025) carried the result: +11.1%, Sharpe 1.32.
- Folds 1 and 3 were barely positive (+2.06%, +2.21%).
- Fold 4 (Dec 2025–May 2026), the most recent, lost 2.84% with a −0.33 Sharpe and a 14.96% drawdown.
Across 6 trials, the Probabilistic Sharpe Ratio was a respectable 0.785, but the Deflated Sharpe Ratio — which penalises for multiple attempts — collapsed to 0.304. In plain terms: once we account for how many variants were tried, the edge no longer looks reliably distinguishable from luck, and nearly all the performance is concentrated in a single favourable window.
Verdict
mean-reversion is a clean, legible strategy with a genuinely strong backtest and an attractive win rate. But its edge is concentrated, its most recent out-of-sample fold is negative, its deflated Sharpe is weak, and it's currently underwater live. The strengths are real; so are the risks. For now this reads as a strategy to keep on a short leash and watch, not one to scale into.