Strategy card
Black Manhattan 1.1
The night shift — a book held from one close to the next open, built on Centaurium 1.1.
Strategy card
- Name
- Black Manhattan 1.1
- Released
- September 2026
- Built on
- Centaurium 1.1 — the custom-trained AI forecaster
- Universe
- US mid-cap equities · long-only · 80 names
- Status
- Paper trading only, on a broker demo account. No real money, and no live track record yet.
- Risk shape
- Flat all day, exposed only overnight. Returns arrive in a small number of large moves, so the ride is lumpy even though the drawdown is moderate.
- Approach
- Shortly before the close, ranks the universe on how lopsided the model’s overnight forecast is, buys the eight most lopsided names, and sells them at the next morning’s open. It holds nothing during the trading day.
Who it’s for: Copiers who want a return stream that behaves differently from a daytime equity book — and who can accept that most of the profit comes from a few nights a year. The edge is risk-shaping and universe selection, not stock-picking — better-shaped equity exposure, clearly labelled.
How we name and version strategies → · About Centaurium 1.1 →
The 2022–2024 backtest
Three years that did the work of a cycle: a bear (2022), a choppy recovery (2023) and a bull (2024). Here is how Black Manhattan 1.1 behaved in each, next to the S&P 500.
In the 2022 bear market it gave back 4.9% against the S&P 500’s 18.1% — the shallowest bear year on the menu, because the book is flat all day and only carries risk overnight.
Year by year
| Setting | 2022 | 2023 | 2024 | 7yr | Max DD | Calmar |
|---|---|---|---|---|---|---|
| Standard · eight names, ~15% vol target | −4.9% | +25.4% | +18.1% | +17.5% | −13.5% | 0.95 |
| S&P 500 (SPY) | −18.1% | +26.3% | +25.0% | +18.2% | −25.0% | 0.44 |
Over a full cycle
Markets do not repeat 2024 forever. We assume a realistic cycle — of every 7 years, 1 looks like 2022 (bear), 3 like 2023 (neutral) and 3 like 2024 (bull) — and compound the three actual backtest years accordingly.
Compounds just under the index, at roughly half the index’s maximum drawdown. The return concentrates in a handful of nights a year, and a flat per-order fee needs position sizes large enough to stay small against it.
What a strategy with no model would have done
For this strategy a rule that needs no model at all is genuinely competitive, so the comparison belongs in the card rather than a footnote. Each control runs the same book — eight slots, same volatility target, same costs — and changes only how the names are picked.
| How the names are chosen | Risk-adjusted return | 7yr | Max DD |
|---|---|---|---|
| Black Manhattan 1.1 (the model) | +1.07 / +1.22 | +17.5% | −13.5% |
| No model: buy last week’s best overnight movers | +1.10 / +1.10 | +15.2% | −16.7% |
| No model: buy last month’s best overnight movers | +0.40 / +0.40 | +8.5% | −22.3% |
| No model: buy the most volatile names | +0.74 / +0.74 | +10.6% | −14.3% |
| No model: hold the whole universe overnight | +0.22 / +0.22 | +8.1% | −26.9% |
The free rule matches the model on risk-adjusted return. The model buys a shallower drawdown (−13.5% against −16.7%), more compounding (+17.5% against +15.2%), and it holds up better when retrained from a different random seed (+1.22 against +1.10). A real edge, a narrow one. Two Sharpe figures per row: one per training seed.
Lowest confidence on the menu. A costless rule is level with it — the table above. And the economics are size-dependent: the broker charges a flat fee per order and this book places two per name per night, so the cost is fixed in dollars however small the position. Below about $1,300 a position that fee alone breaches the cost line we halt the strategy at; even at $5,000 it is still 4 bps of pure fee. Not a strategy for small accounts. Published as a research line, not a recommendation.
Explore the full backtest
The complete reconstructed daily curve and every single trade, pulled live from the backtest file. Search it, sort it, zoom the chart — the receipts, not a summary.
Loading the backtest…
Every trade
Couldn’t load the backtest file. Please try again later.
Transparency
The in-sample backtest above sits next to real results once live — including the gap between them.
Published rules auto-halt this strategy if live performance breaks down against its backtest. No moving goalposts.
Better-shaped equity exposure, clearly labelled. No claim to beat the market by picking winners.
Version history
| Version | Released | Notes |
|---|---|---|
| Black Manhattan 1.1 | Sep 2026 | First published version. Numbered 1.1 because the version’s minor tracks the model it runs on, and it was born on Centaurium 1.1. |