The thesis
Mean-reversion is one of the oldest ideas in systematic trading, and this agent keeps it refreshingly literal: buy what looks oversold (RSI below 30), sell what looks overbought (RSI above 70). It trades a 24-name large-cap universe — the usual megacap tech (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare, staples and industrials. No exotic instruments, no leverage, just a bet that stretched prices snap back.
Recent activity
The strategy has been quiet. Its scheduled runs on 2026-11-02 and 2026-11-03 both executed zero trades — unsurprising in a tape where few names are hitting RSI extremes. The last real activity came at the end of October: on 2026-10-29 it bought 48 shares of BAC at $51.15, and on 2026-10-30 it added 4 shares of CAT at $622.54 and 8 of GOOGL at $280.28. A classic contrarian basket.
The live paper book tells a more sober story than the backtest. Total equity has slipped from $9,976.27 on 2026-10-29 to $9,847.17 on 2026-11-03 — a shade under the $10,000 starting line, with $2,742.49 sitting in cash. Early days, but not the launch it would have hoped for.
Backtest performance
On paper the record is attractive. Over 451 days the strategy returned 14.73% (final equity $11,473.17), a 7.98% CAGR, with a 70.59% win rate across just 38 trades. Fees were negligible ($38 total, no FX cost). The weak spots are visible even here: a Sharpe of only 0.58 and a max drawdown of 15.64% — the return is real but the ride is bumpy, and the reward-per-unit-risk is modest. Turnover of 879% also flags that this is an active book, not a buy-and-hold.
Validation: where it breaks
This is the part that matters, and it is why the strategy fails our validation gate. Across four walk-forward folds, three were positive — fold 2 (Jan–Jul 2025) was excellent at +11.10% and a 1.32 Sharpe. But the stitched out-of-sample result is −2.84% with a −0.33 Sharpe, dragged down by the most recent fold (Dec 2025–May 2026), which lost 2.84% and drew down nearly 15%. In other words, the strategy's edge decayed exactly in the newest, most relevant period.
The statistics echo this. The Probabilistic Sharpe Ratio of 0.785 is respectable, but the Deflated Sharpe Ratio of 0.304 — which penalises for the 6 trials run — says we cannot confidently reject the possibility that the headline Sharpe is a product of selection, not skill.
Verdict
Mean-reversion is a clean, cheap, interpretable strategy with a genuinely high hit rate — strengths worth keeping. But a high win rate with a soft Sharpe hints at the classic asymmetry of this style: many small wins, occasional painful losses. With out-of-sample returns turning negative in the latest regime and a DSR barely above 0.3, the honest read is promising thesis, unproven edge. Watch the live book, but size it with caution.