The Thesis
Bollinger-reversion is a textbook mean-reversion agent: it buys when price slips below the lower Bollinger band and sells when it pushes above the upper band. The bet is that stretched prices snap back to their moving average. It trades a 24-name large-cap universe spanning tech (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare, staples, and industrials — a diversified, liquid field that suits a statistical rather than thematic approach. It is a built-in, hand-written strategy, approved by review at risk level 0.
Recent Activity
The scheduled run on 2026-10-29 executed 2 trades with none rejected: 3 shares of CAT at $647.16 and 8 shares of GOOGL at $283.35 — both buys, consistent with picking up names that had dipped below their lower band. After the run the book held $5,749.53 in cash against a total of $9,957.81. That total sits just under the $10,000 Season 2 starting line, so the agent is marginally underwater in live paper trading so far.
Backtest Performance
Over 451 days the strategy returned 17.55%, ending at $11,755.25 equity, a 9.46% CAGR. The win rate is an attractive 63.89% across 76 trades, with total fees of just $76 and no FX cost. But the quality metrics are more sober: a Sharpe of 0.66 and a maximum drawdown of 20.57% — you are risking roughly a fifth of capital to earn that return. Turnover of 1,720% also flags a busy, churn-heavy profile for a strategy with only 76 completed trades.
Validation: The Real Story
Here the picture turns cautionary. Validation failed. Walk-forward testing across 4 folds shows all four positive — a genuine strength — but the trend is a steady bleed:
| Fold | Return | Sharpe |
|---|---|---|
| 1 | 7.67% | 1.78 |
| 2 | 1.42% | 0.25 |
| 3 | 0.77% | 0.20 |
| 4 | 0.40% | 0.14 |
The out-of-sample return lands at just 0.4% with a Sharpe of 0.14. In other words, nearly all of the headline 17.55% was earned in the earliest, in-sample-adjacent window. The Deflated Sharpe Ratio of 0.342 — adjusting for the 6 trials run — confirms the edge does not clear the bar once you account for selection. The Probabilistic Sharpe Ratio of 0.814 is more flattering but cannot rescue a DSR that low.
Verdict
Bollinger-reversion is honest and cheap to run, with a credible win rate and four positive folds. Those are real positives. But the decay from a 1.78 Sharpe in fold one to 0.14 in fold four is the signature of an edge fading into noise — likely a regime-dependent effect that worked in 2024 and has thinned since. Treat the live deployment as an observation experiment, not a conviction allocation, and watch whether the snap-backs it is betting on still arrive.