The thesis
PanicFade v4 bets that negative headline shocks get overreacted to and revert toward fundamentals. The entry logic looks for names that are both sentiment-stricken (sentiment < -0.4) and already technically oversold — RSI14 below 35 and trading under their SMA20 — then sizes into the weakness. Exits are disciplined: it closes on reversion (RSI14 > 50 or a reclaim of SMA20) and hard-stops fresh lows on continued negativity, treating that as a genuine fundamental break rather than noise. The thesis is grounded in recent sentiment-feedback literature (arXiv:2509.11970 and corroborating MDPI/JRFM work on stock-level investor sentiment), which lends it more academic footing than most momentum-chasing bots.
A caveat baked into the backtest
There is an important nuance worth flagging up front. The improvement log shows that the original sentiment gate was structurally unreachable in historical simulation — per-symbol news coverage is too sparse, so the triple-AND condition never fired and every fold produced zero trades. The v2–v4 rewrites therefore redefined "panic" in price terms (deep RSI, distance below SMA20, crash-day drops), with sentiment demoted to an amplifier. So the validated performance below reflects a price-mean-reversion engine; the sentiment edge the thesis leans on is largely untested in-sample and will only prove out live.
Validation performance
The validation is genuinely encouraging. Across four walk-forward folds (2021–2026), all four were positive, with returns climbing from +1.6% to +21.7% and the final fold posting a standout 2.44 Sharpe at just 2.6% drawdown. Headline numbers: 37.93% total return, 0.96 Sharpe, 9.3% max drawdown, 62.88% win rate over 530 trades. The probabilistic Sharpe ratio is a strong 0.985, meaning the result is very unlikely to be pure luck.
The honest counterweight is the deflated Sharpe of 0.56. After adjusting for the 27 trials run during development, the edge shrinks considerably — respectable, not spectacular. The CAGR of 6.79% over ~1,233 days is modest, and turnover of 5,807% is extremely high, so this strategy lives and dies on frictions. Fees were a flat $530 (one per trade) with zero FX cost in the test, but live trading under the firm's ii fee model (£3.99/side plus 0.75% FX on non-GBP names) could meaningfully erode that 6.79%.
Recent live activity
PanicFade v4 went live with the Season 2 $10k reset. Its 2026-10-29 run executed nothing (cash $10,000, total $10,000); the following day it put money to work with three buys — BAC (20 @ $50.04), CAT (2 @ $622.54) and GOOGL (5 @ $280.28) — leaving cash of $6,309.33 and a total of $9,956.61. That total is fractionally below the $10k start, which is simply mark-to-market on day-old positions, not a realized loss.
Verdict
Strengths: a literature-backed thesis, four clean out-of-sample folds, a high win rate and controlled drawdown. Risks: a deflated Sharpe that signals a thin true edge, punishing turnover that makes fee discipline existential, and a sentiment premise the backtest could not actually exercise. The reviewer's 0.35 risk score (approved, with sizing caveats) feels about right — a credible strategy that still has to earn its keep in the open.