In one line: a small, genuine, repeatable
edge on a thin slice of capital — it beat 100% of 40 random copies of itself, in and out of sample,
and stepped around a 2008 that took the index down 55%.
Portfolio CAGR
4.0%/yr
$100k book, 0% on cash
Exposure-adj. CAGR
~11%/yr
in-market only, 0% on cash
Concurrent symbols
≤20
~1–2 typical (rarely more)
Max drawdown
−13%
| Kind of system | Mean-reversion stock-picking — buys a strong name only after it sells off hard, holds briefly, then returns to cash. Selective and episodic, not always-on. |
|---|---|
| Universe | S&P 500. |
| Average period | ~4 trading days per trade — in a position on roughly a third of days, in cash the rest. Sell into the bounce or after about a week. |
| Concurrent symbols | Up to 20 slots, but it rarely fills more than one or two at a time — the selectivity is why average deployment stays near ~8%. |
| Return | Portfolio CAGR 4.0% / yr — $100k book, 0% earned on idle cash, no exposure adjustment; low because the book sits in cash most of the time. Exposure-adjusted CAGR ~11% / yr — credited only while capital is deployed; this is what the in-market slice earns once cash drag is stripped out. The gap between the two figures is the strategy: a real edge on a small deployed slice. Risk-adjusted numbers are on the Details and Deep Dive. |
| Exposure | Seldom in the market — ~8% of capital deployed on average, with long flatlines in cash through 2008 and 2020 and brief bursts when many stocks sell off at once. |
| Distinguishing feature | A genuine selective edge — it cleared every one of 40 random-pick copies of itself on a risk-adjusted basis, in-sample and out-of-sample, at a ~−13% worst drawdown against the index’s ~−55%. The edge lives in being selective, so it cannot simply be scaled up. |
| Who it’s for | A portfolio manager looking for an uncorrelated, capital-light sleeve — a crash-resistant complement that runs on a thin slice and leaves the rest in cash, not a standalone engine meant to compound aggressively. |
| Out-of-sample | Tested 2005–2026 (21.4 yrs), but the author disclosed the rules in January 2025, so the strictly out-of-sample window is short — ~1.4 yrs (2025–2026). A separate 2005–17 / 2018–26 temporal-stability split shows the edge held on later data it was not tuned on (profit factor 1.55 → 1.61), but that is a robustness check, not the disclosure pivot. |
→ Details for why “it trails the index” is the wrong scorecard, the skill-vs-luck proof, and how it was tested — or the full Deep Dive with methodology, the random-portfolio test, the in/out-of-sample work, and risk-adjusted tables.
See how it pairs with your book. Its low correlation and large cash
cushion make it a natural sleeve — we can re-run it on your universe, or send you the Python, daily
trade logs, and data behind every number.
· start a conversation →