The thesis
Bollinger-reversion runs a textbook mean-reversion playbook across a 24-name large-cap universe: buy when price punches below the lower Bollinger band, sell when it pushes above the upper one. The bet is that short-term dislocations in liquid names like AAPL, MSFT, NVDA and XOM snap back toward their moving average. It's a clean, legible idea — the kind that either works because markets really do overreact, or quietly bleeds because bands get ridden rather than reverted.
Recent activity
The live book is running but quiet. Over the last two weeks most scheduled runs executed nothing: 2026-11-09 through 11-12 logged zero executed trades (with the odd rejection), and the account total has drifted in a $9,700–$9,995 range, sitting at $9,967.81 on the 12th with $1,055.32 in cash. The action came earlier: round-trips in AAPL (bought $329.82, sold $352.27) and MSFT (bought $514.53, sold $557.86) both captured the reversion they were designed to catch, and fresh longs went on in NVDA, XOM, GOOGL and CAT. So the logic is firing as intended — it's simply finding few qualifying setups right now.
Backtest
On paper the strategy looks solid. Over 451 days it returned 17.55% (final equity $11,755.25, a 9.46% CAGR) with a 63.89% win rate across 76 trades. The Sharpe of 0.66 is modest rather than impressive, and the 20.57% max drawdown is a real gut-check for a strategy that buys falling knives. Turnover of 1,720% is heavy, though at $76 total fees and zero FX cost, trading friction isn't what's hurting it.
Validation — the uncomfortable part
Here's where the story turns. The walk-forward validation did not pass, and the fold-by-fold numbers show why. All four folds were positive — a good sign of robustness — but the edge decays monotonically: fold 1 returned 7.67% at a 1.78 Sharpe, then 1.42%, then 0.77%, and finally just 0.40% at a 0.14 Sharpe in the most recent out-of-sample window. The headline 17.55% is overwhelmingly front-loaded into 2024–early 2025.
The statistics confirm the unease. The probabilistic Sharpe ratio (0.814) is encouraging, but the deflated Sharpe ratio — which penalises for the 6 trials run — sits at 0.342, below the level that would let us call the edge statistically real. An out-of-sample return of 0.40% is, functionally, flat.
Verdict
Bollinger-reversion is a well-behaved, understandable strategy that did what it promised early on and still executes cleanly. But the evidence points to an edge that was strongest in its discovery window and has faded toward noise since. Keep it live for observation, size it cautiously, and treat the 17.55% as a historical high-water mark rather than a forward expectation until the out-of-sample folds show signs of life again.