The thesis
RSI Snap-Back bets that large-cap tech names overshoot in the short term and then revert. The rule is deliberately mechanical: buy a name when its RSI drops below 35, sell when it climbs back above 70, and never hold more than four positions at once. That hard 4-slot book is the interesting part — it forces the strategy to rank its convictions and caps how much concurrent drawdown the portfolio can absorb. The universe is the obvious one for this style: AAPL, MSFT, NVDA, GOOGL, AMZN, META, and TSLA — seven liquid, heavily-traded names where short-term momentum extremes are common and reversions are usually fast.
Recent activity
The live record is young and quiet. The strategy went live on 22 October with $10,000 and, across six scheduled runs, has pulled the trigger exactly once: a 6-share buy of GOOGL at $321.67 on 23 October. Every other run reported 0 executed — unsurprising for a system that only acts on an RSI threshold being breached. As of the 28 October run the book held $8,064.65 in cash against a total value of $10,015.85, so after nearly a week live the strategy is essentially flat. There is nothing alarming here, but there is also nothing yet to judge: one fill is not a track record.
Backtest and validation
The backtest is the headline number, and it is genuinely decent. Over 451 days the strategy returned 20.95% (final equity $12,095.14), an 11.21% CAGR, with a 66.67% win rate across 37 trades. Trading costs were trivial — $37 in fees, zero FX — despite a hefty 773% turnover, which tells you the names are cheap to trade and the slot discipline keeps churn productive rather than wasteful.
The risks sit right next to those strengths. The Sharpe ratio is only 0.61, meaning the returns came with a lot of volatility for the reward. More pointedly, the maximum drawdown was 23.73% — a near-quarter peak-to-trough fall that any live investor would have felt acutely. A high win rate paired with a deep drawdown is the classic mean-reversion signature: you win often and small, then occasionally wear a painful move when a name keeps falling instead of snapping back.
The most important caveat is what is missing. The validation field is null — there is no out-of-sample or hold-out result reported. A single-period backtest on the Mag-7 over a strong tech run is exactly the setup where overfitting hides, and without validation we cannot distinguish a real edge from a well-curated in-sample fit.
Verdict
RSI Snap-Back has a clean, disciplined thesis and an encouraging backtest, but it is unproven where it matters. Until it shows out-of-sample validation and accumulates a real live record, treat the 20.95% as a hypothesis — not evidence.