The thesis
PanicFade v4 bets that markets overreact to bad news. When a large cap gets hit by a negative headline shock (sentiment < −0.4) and is already technically oversold — RSI14 below 35 and trading under its SMA20 — the strategy buys into the fear and sizes up as the selling deepens. It exits on reversion: RSI14 climbing back above 50 or price reclaiming the SMA20. A hard stop on fresh lows under continued negativity treats persistent weakness as a genuine fundamental break rather than a buyable panic.
The logic is grounded in the equity-sentiment overreaction literature (arXiv:2509.11970, 2025, and corroborating JRFM work), and the universe is deliberately conservative — 24 mega- and large-cap names (AAPL, MSFT, JPM, UNH, XOM, and peers). Fading panic is far safer in COST than in a microcap.
Backtest and validation
The headline backtest is attractive: +37.93% total return over 1,233 trading days (a 6.79% CAGR), a 0.96 Sharpe, a contained 9.3% max drawdown, and a 62.88% win rate across 530 trades. Fees were a trivial $530 with zero FX cost.
More importantly, the validation holds up where many backtests fall apart. Across 4 walk-forward folds, all 4 were positive, and performance improved over time — fold 1 scraped +1.6% (Sharpe 0.19), while the most recent fold (May 2025–Aug 2026) returned +21.7% at a 2.44 Sharpe with just a 2.6% drawdown. The probabilistic Sharpe ratio is a near-certain 0.985.
The risks
Two numbers temper the enthusiasm. First, the deflated Sharpe ratio is only 0.56 — after adjusting for 27 trials of selection, confidence that the true edge is real drops to a coin-flip-plus. That gap between a 0.985 PSR and a 0.56 DSR is exactly what multiple-testing deflation is designed to expose; treat the strong recent fold as promising, not proven.
Second, turnover is 5,807% — this is a high-churn strategy whose edge depends on cheap, reliable execution. The flat per-trade fee flatters it; under realistic slippage the margin narrows.
Live activity
Reality has been quieter than the backtest. The live paper book sits around $9,812 total with $8,642 in cash — essentially flat-to-slightly-underwater since inception. Recent sessions (Oct 14–21) show almost no action: mostly 0 executed, 0 rejected, with a single fill on Oct 15. The last meaningful rotation came in late September and early October — buys in UNH, NKE, HD, MCD, and BAC, then exits in MCD, NKE, HD, and UNH as names reverted or hit stops.
That inactivity is the strategy behaving as designed: no panic in a calm, grinding-higher tape means no trades. The discipline is encouraging. The open question is whether the recent out-of-sample strength survives contact with live markets once volatility actually returns — and whether a 0.56 DSR is enough conviction to keep sizing into the next real shock.