The thesis
Mean-reversion is one of the oldest ideas in systematic trading: prices that overshoot tend to snap back. This strategy encodes that mechanically with RSI — buy when a name is oversold (RSI below 30), sell when it is overbought (RSI above 70). It runs live across a 24-name large-cap universe spanning tech, financials, healthcare, staples, and energy — AAPL, MSFT, NVDA, JPM, JNJ, WMT, XOM and peers. It is a clean, interpretable rule, which is exactly why it makes a good honesty test for our validation pipeline.
Backtest: the good news
On paper, the numbers flatter the strategy. Over 451 days it returned 14.73% (final equity $11,473 on a $10k base), a 7.98% CAGR, with a 70.59% win rate across 38 trades. Costs were negligible — $38 in fees, zero FX. That win rate is the headline a casual reader latches onto, and mean-reversion strategies characteristically win often.
But the risk-adjusted picture is more sober. Sharpe is just 0.58, and max drawdown reached 15.64% — larger than the total return would suggest is comfortable. Turnover of 879% also signals a strategy that trades a lot to earn that Sharpe; frequent winners with occasional deep losers is the classic mean-reversion payoff shape, and it shows here.
Validation: the strategy failed
This is where discipline matters. Our walk-forward validation did not pass. Across four sequential folds, three were positive — folds 1 through 3 returned 2.06%, 11.10%, and 2.21% — but the most recent fold (Dec 2025 to May 2026) returned -2.84% with a -0.33 Sharpe and a near-15% drawdown. The out-of-sample Sharpe is negative.
The deflated metrics tell the same story. With 6 trials searched, the Probabilistic Sharpe Ratio sits at 0.785 but the Deflated Sharpe Ratio is only 0.304 — below the threshold we treat as evidence the edge survives multiple-testing correction. In plain terms: the full-sample 14.73% is partly a story the model tells about the past, and the newest unseen period didn't cooperate.
Recent activity: mostly silence
Live behaviour reinforces the caution. The last six scheduled runs (July 6–13) each executed zero trades — no oversold or overbought signals crossed the thresholds. Cash has held flat at $7,569.25 while total equity drifted between roughly $9,898 and $9,987, still below the starting mark. The last actual fill was a 21-share WMT buy at $115.75 on May 31. A quiet book is not itself a failure — mean-reversion waits for extremes — but combined with a failed validation, it means we are holding, not adding conviction.
Verdict
Mean-reversion is a legitimate, readable strategy with a genuinely high hit rate, but the evidence says treat it as unproven, not deployable. The recent negative fold and low Deflated Sharpe are exactly the signals our pipeline exists to catch. We keep it live in paper form, watch whether new signals confirm the thesis out-of-sample, and resist the temptation to trust that 70% win rate on its own.