How Satis finds trading edges — and kills the rest. A closed loop with an honest gate: search every idea, correct for the search, prove it live, and feed reality back in. Most ideas die here. That's the point.
The loop that compounds
Every serious quant shop runs a version of steps 1–5. The differentiator is step 3 and step 6.
Step 3 — the honest gate. A search that keeps whatever "survives" is a false-edge factory. Ours corrects for how many things were tried, so it can't fool itself.
Step 6 — Compound (the one most never build). Each run compares the live record to the backtest that justified it. When reality diverges from the research, that becomes the next thing to investigate. The loop learns from the market, not just from itself.
The gate, in numbers
The most recent full search over the survivorship-bias-free universe — and why the survivors are read with humility, not hype.
Loading…
What survived
The strategies that cleared the honest gate. Read them in context: with dozens of tries, a couple pass by luck — the trustworthy ones also hold out-of-sample and rest on established factors.
Loading survivors…
Proving it live
Validated edges don't go straight to real money — they run forward as public paper records first. This is step 4 → 5 of the loop, live right now.
Loading live engines…
The graveyard
The ideas that failed the gate — including the ones every retail bot ships. Publishing these is the whole point: knowing what doesn't work, before risking money on it.
Naive "buy the dip"
No durable edge (Sharpe 0.42), dies out-of-sample. The default retail strategy — and a loser.
1-month mean-reversion
Worked in the first half, collapsed in the second (OOS Sharpe −1.3). A regime artifact.
Stock-picking momentum (alone)
Only ties a cheap index fund risk-adjusted. You can rent it as an ETF for 0.15%.
Crypto active-trading
Gap-through stops + fees erase the edge. Crypto is trend-capture, not day-trading.
The old "+128%" headline
A per-instrument envelope, not a tradeable portfolio. Retired on the spot when re-tested honestly.
Value / quality / size (alone)
Classic factor premia have decayed since ~2016 — flat-to-negative on their own here.
The point of all this. The engine's edge isn't a magic strategy — it's the discipline. Search widely, correct for the search, prove it forward in public, and let live reality tell you what to research next. In a field of confident promises that don't survive contact with reality, a research process whose defining trait is that it refuses to fool itself is the rarest thing there is.
Hypothetical, survivorship-bias-free research on a decade of US-equity data, net of modelled costs; walk-forward + out-of-sample + Monte-Carlo + a multiple-testing correction. Hypothetical results have inherent limitations and are not a live record or a promise of future performance. Satis Omnibus is in beta and pursuing NZ FMA licensing; nothing here is personalised financial advice.