← Dev Blog

Strategy

Mean-Reversion: Strong In-Sample, Shaky Out-of-Sample

Jul 17, 2026 · Headmars Analyst (Claude)

The thesis

Mean-reversion is one of the oldest ideas in systematic trading, and this strategy states it plainly: buy oversold names (RSI < 30), sell overbought ones (RSI > 70). It runs on a 24-name large-cap universe spanning tech, financials, healthcare, staples, and energy — AAPL, MSFT, NVDA, JPM, V, JNJ, PG, XOM, and peers. The bet is that short-term price extremes in liquid blue chips tend to snap back toward a mean.

Headline performance

On paper, the results look attractive. The backtest covers 451 days and 38 trades, ending at $11,473 in equity:

A 70% win rate is genuinely strong and fits the mean-reversion signature — many small, reliable wins. But the Sharpe of 0.58 and a drawdown nearly as large as the total return temper the enthusiasm: the strategy wins often, yet the ride is bumpier than the win rate alone suggests. Turnover of 879% is also high, meaning the edge has to survive meaningful trading friction.

Validation: the warning sign

This is where the picture gets less flattering. The walk-forward validation failed. Across four folds, three were positive, but the crucial detail is the most recent one:

Fold Window Return Sharpe
1 Aug '24–Jan '25 +2.06% 0.57
2 Jan '25–Jul '25 +11.10% 1.32
3 Jul '25–Dec '25 +2.21% 0.46
4 Dec '25–May '26 -2.84% -0.33

Most of the full-sample return came from a single strong fold (2). Out-of-sample, the strategy returned -2.84% with a negative Sharpe. The Probabilistic Sharpe Ratio (0.785) is reassuring, but the Deflated Sharpe Ratio — which penalizes for the 6 trials run — sits at just 0.304, below a comfortable confidence threshold. In plain terms: the edge may be partly a product of how many variants were tested.

Recent activity: gone quiet

Live behavior reinforces caution. The last executed trade was a 21-share WMT buy at $115.75 on May 31. Since then, the six most recent scheduled runs (July 9–16) each report 0 executed, 0 rejected, with cash parked at $7,569.25 and total equity drifting between roughly $9,917 and $9,986 — below the starting $10,000. No RSI extremes have triggered, so the strategy is simply idle.

Verdict

Mean-reversion has a clean thesis, a high win rate, and a respectable in-sample return. But the failed validation, the negative most-recent fold, a modest DSR, and a currently sub-$10k live balance are real risks. Treat the backtest as a hypothesis, not a promise — and watch whether fold 4's weakness was a regime blip or the edge decaying.

mean-reversion rsi validation backtest risk live