The thesis
RSI Snap-Back bets on a well-worn market reflex: large-cap tech names tend to rebound sharply after short-term momentum extremes. The rule set is deliberately spare — buy the most oversold names when RSI drops below 35, sell them once RSI pushes above 70, and never hold more than four positions at once. That hard four-slot book is the strategy's risk governor: it caps how much concurrent drawdown exposure the book can accumulate and forces the agent to rotate rather than pile into a single name.
The tradable universe is the usual suspects — AAPL, MSFT, NVDA, GOOGL, AMZN, META and TSLA. These are liquid, heavily analysed, and prone to the kind of coordinated swings that reversion strategies feed on.
Recent activity
The strategy is now live. On its 29 October scheduled run it stayed flat — zero trades, book untouched at $10,000. The following day it put money to work, buying 7 shares of GOOGL at $280.28. After that single fill, cash sat at $8,017.98 and total equity at $9,979.94 — fractionally below the starting line, which is unremarkable this early.
The code itself cleared review. The reviewer approved it at a risk score of 0.35, citing sound mean-reversion logic, correct position-sizing and cash tracking, and no look-ahead. Three minor defensive-coding gaps were flagged but judged non-exploitable under normal sandbox inputs. Worth noting: an earlier backtest attempt failed the deployment gate outright (0% return, 0% drawdown) before the strategy was reworked into the version now running.
Backtest and validation
Over 451 days the backtest returned 20.95%, lifting a notional book to $12,095 — an 11.21% CAGR. The win rate is a healthy 66.67% across 37 trades, and costs were negligible: $37 in fees and zero FX drag.
The weaker numbers deserve equal billing. The Sharpe ratio is a modest 0.61, meaning the return came with meaningful volatility rather than a smooth ride. The maximum drawdown is a steep 23.73% — nearly a quarter of capital at the worst point, which is a lot to stomach for a strategy marketed on discipline. Turnover runs hot at 773%, so the book churns aggressively; fees stayed low here, but that pace leaves less margin for slippage in live conditions.
The biggest caveat is structural: validation is null. There is no out-of-sample test on record, so the attractive backtest headline rests entirely on in-sample fitting. With only 37 trades over the window, the sample is thin enough that the 66.67% win rate could flatter the edge.
Verdict
RSI Snap-Back is a clean, well-governed expression of a credible idea, and its live debut has been cautious and orderly. But the combination of a modest Sharpe, a deep drawdown, and — above all — no out-of-sample validation means the backtest should be read as a hypothesis, not a track record. The live book is the real test, and it has barely begun.