The Thesis
RSI Snap-Back rests on a straightforward premise: the largest-cap tech names — Apple, Microsoft, Nvidia, Alphabet, Amazon, Meta, and Tesla — tend to recover sharply after short-term momentum exhaustion. The strategy enters when a name's RSI drops below 35 (oversold) and exits when it climbs above 70 (overbought). A hard four-slot book caps concurrent exposure, enforcing capital discipline and limiting the damage any single drawdown can inflict on the overall portfolio.
The logic is clean and well-understood. Mean-reversion in liquid large-caps is one of the more academically supported edges in equities, and concentrating on Mag-7 names means the strategy operates in some of the deepest, most-scrutinised markets in the world — reducing the risk that a signal is merely a liquidity artefact.
Deployment and Early Activity
The agent passed code review on May 31 with a risk score of 0.35. The reviewer flagged correct position-sizing and cash tracking, no unsafe code, and no look-ahead bias — three minor defensive-coding gaps were noted but deemed non-exploitable under normal sandbox conditions. A first backtest attempt failed an automated gate (returning 0% with 0% drawdown, indicating a data or initialisation issue), but a corrected run produced the metrics used for approval.
Since deployment, RSI Snap-Back has completed two scheduled runs — May 31 and June 1 — and executed zero trades in both. With $10,000 in cash and the full book empty, the strategy is simply waiting: no Mag-7 name has crossed the RSI < 35 threshold required to open a position. That patience is a feature, not a bug. A strategy that fires only when conditions are met is preferable to one that forces trades.
Backtest Performance
Over 451 days of simulated history, RSI Snap-Back returned 20.95% on a $10,000 stake (final equity $12,095), implying an annualised rate of roughly 11.21%. The agent executed 37 trades and won on 66.67% of them — a solid hit rate for a mean-reversion system.
The Sharpe ratio of 0.61 is modest but positive, suggesting risk-adjusted returns above cash without exceptional smoothness. Turnover was high at 773% annualised, which matters less here because the $1-per-trade fee structure kept total friction to $37 — negligible relative to the gain.
The most prominent concern in the backtest is a 23.73% maximum drawdown. For a strategy trading the world's most liquid stocks with a four-name ceiling, that figure deserves scrutiny. A drawdown of that magnitude can trigger forced exits in real portfolios, particularly if it coincides with broader market stress — precisely when Mag-7 names are most likely to show extreme RSI readings in the first place.
Strengths and Risks
What works: The entry/exit logic is explicit and verifiable. Two-thirds win rate over 37 trades provides a reasonable sample. The book-size constraint genuinely limits runaway concentration.
What warrants caution: Validation is currently null — no walk-forward or out-of-sample test has been run. A 20.95% gain across 451 backtest days is encouraging, but without an unseen holdout period, overfitting to the specific regime captured in training data remains a live concern. The 23.73% max drawdown also deserves stress-testing against a sharp-correction scenario where all four book slots are underwater simultaneously.
RSI Snap-Back is a disciplined, legible agent that earned its deployment. The real test begins when the first RSI threshold triggers and the paper capital goes to work.