The thesis
Mean-reversion is one of the oldest ideas in markets, and this agent implements the textbook version: buy oversold names (RSI below 30), sell overbought ones (RSI above 70). It runs across a 24-name universe of large-cap U.S. equities spanning tech (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare (JNJ, UNH, PFE, ABBV), and consumer and industrial staples (PG, KO, WMT, COST, XOM, CVX, CAT). The bet is simple: extreme moves overshoot and snap back.
Headline backtest numbers
On paper, the strategy looks appealing. Over 451 trading days it returned 14.73% (final equity $11,473 on a $10,000 base), a 7.98% CAGR, with a 70.59% win rate across 38 trades. Fees were negligible at $38 total and FX cost was zero.
The risk profile is more sobering. The Sharpe ratio of 0.58 is modest, and max drawdown of 15.64% is meaningful for a strategy earning single-digit annualized returns. Turnover of 879% is also high — this is an active book, not a buy-and-hold. A high win rate paired with a middling Sharpe usually means the losers, when they come, are larger than the frequent small winners.
Validation: the warning light is on
Here is where enthusiasm should cool. The walk-forward validation failed. Across four folds, three were positive, but the most recent out-of-sample window (2025-12-16 to 2026-05-29) returned -2.84% with a negative Sharpe of -0.33 and a 14.96% drawdown. The earlier folds carried the record: fold 2 alone delivered 11.1% at a 1.32 Sharpe, doing much of the heavy lifting.
The statistical flags reinforce the concern. With 6 trials run, the Deflated Sharpe Ratio sits at just 0.304 — below the threshold you would want to trust an edge that survived multiple attempts. The Probabilistic Sharpe Ratio of 0.785 is healthier, but the deflated figure is the honest one once you account for selection.
Recent activity: quietly holding cash
Live behavior matches a strategy waiting for its setup. The last executed trade was a buy of 21 WMT shares at $115.75 on 2026-05-31. Since then, every scheduled run from 2026-07-08 through 2026-07-15 reported 0 executed, 0 rejected — no RSI extremes triggered. The book holds roughly $7,569 in cash against a total near $9,957, drifting modestly below its $10,000 base with the market.
Verdict
Mean-reversion has a coherent thesis, a clean win rate, and disciplined, low-cost execution. But the failed walk-forward and depressed Deflated Sharpe suggest the backtest edge may be partly fitted to the past. The recent negative out-of-sample fold is the tell. Treat this one as live-for-observation: worth watching for whether the next real oversold signal reverts as advertised, but not yet worth conviction capital.