Strategy Tearsheet · Crypto Perpetuals
A market-neutral systematic book across 46 crypto perpetuals — a change-in-delivery trend signal, a beta-neutral funding-carry sleeve, vol-targeted and adversarially audited.
Read the headline honestly. The detailed backtest below runs on a survivor universe (46 coins that lived to today), which flatters it to Sharpe ≈ 1.7. On a survivorship-honest point-in-time universe — top-20 most-liquid, rebuilt each day and including delisted coins — the edge roughly halves to Sharpe ≈ 1.0 full, ≈ 2.0 out-of-sample. Plan around the point-in-time number; see §07 Survivorship, corrected.
Timing any single coin with a trend signal is a loser's game — you spend the whole backtest fighting buy-and-hold. The edge shows up only when you go cross-sectional: rank the whole market against itself, own the winners, short the losers, and collect the spread. This book blends two nearly-uncorrelated market-neutral sleeves — a change-in-state-of-delivery trend and a beta-neutral funding carry — across 46 perpetuals. It made money in 2022's bear and 2025's alt crash, at roughly zero correlation to Bitcoin. Here is the full, honestly-caveated tearsheet.
Buy & hold rode the alt supercycle to a bigger headline, but round-tripped through ~70–80% drawdowns to get there. The strategy compounds quietly and market-neutral: in 2025's alt crash it returned +156% while holding the same coins lost 55%, and in 2022's bear it fell only 16% against buy&hold's 77%. The curve below runs the vol-targeted (VT-40) setting; the shaded region is the held-out window never used to choose anything.
The edge is not one lucky year. It is positive in five of six calendar years; only 2022, a broad crypto bear, is negative — and even then it lost 16% against buy&hold's 77%. Because it is genuinely market-neutral, the good years and the bad market do not line up.
And it survives realistic trading costs. Turnover is low — the daily signal is sticky — so even at market-order taker fees (10 bps/side) the full-sample Sharpe still clears +1.0; at limit-order pricing it is +1.5.
A fair-fight comparison — matched, recent windows ending mid-2026 — makes the point plainly:
| Window (ends 2026) | Strategy | Buy & hold |
|---|---|---|
| Held-out (since Apr 2025) | 2.51× · −26% | 0.49× · −66% |
| Last 2 years | 3.16× · −26% | 0.43× · −80% |
| Last 12 months | 1.98× · −26% | 0.47× · −70% |
Leverage scales risk and reward together — the choice is a drawdown-appetite decision, not a source of alpha. The recommended policy, VT-40, targets 40% annualized volatility with a trailing-vol scalar, a 3× cap, and a drawdown kill-switch that flattens the book after a −25% trailing month. Funding-on-leverage and liquidation gaps are not modeled, so read the levered rows as upper bounds.
| Policy | Return/yr | Vol | Max DD | $1 → |
|---|---|---|---|---|
| 1× fixed | +26% | 21% | −18% | 3.6× |
| 2× fixed | +52% | 43% | −33% | 10.1× |
| 3× fixed | +76% | 64% | −46% | 22.3× |
| VT-40 (vol-target) | +71% | 43% | −32% | 19.2× |
No single coin drives this. The book holds ~45 names long and short every day; the largest single name is under 4% of gross exposure. Remove the best contributor (ZEC) entirely and it still compounds at +61% CAGR, Sharpe +1.3. The engine is two sleeves that are nearly uncorrelated — so averaging them keeps the return while cancelling much of the risk, and the carry sleeve earns net funding rather than paying it.
The core insight
Timing one coin fights buy-and-hold and loses. Ranking the whole market against itself pays you the spread — and that spread doesn't care which way Bitcoin goes.
A change-in-delivery trend sleeve and a beta-neutral funding carry sleeve correlate at just −0.03 to each other and ≈0 to Bitcoin. Blended 50/50 and vol-targeted, they produce an absolute-return stream that made money while the alt market fell 30%+ — the definition of a diversifier, not leveraged beta.
| Coin | Sharpe | Share of profit | Max DD | Avg gross wt |
|---|---|---|---|---|
| ZEC | +1.2 | 13% | −11% | 4.7% |
| AXS | +0.8 | 10% | −9% | 3.7% |
| ETC | +0.4 | 7% | −16% | 3.0% |
| GALA | +0.8 | 7% | −12% | 3.3% |
| FIL | +1.0 | 7% | −6% | 3.3% |
| SOL | +1.1 | 6% | −6% | 2.9% |
| CRV | +0.6 | 6% | −14% | 3.8% |
| THETA | +1.2 | 6% | −6% | 3.0% |
| LDO | +0.6 | 6% | −11% | 3.7% |
| INJ | +0.6 | 6% | −10% | 3.9% |
| DASH | +0.6 | 5% | −12% | 4.0% |
| TRX | +0.8 | 5% | −7% | 5.9% |
| MANA | +0.8 | 4% | −5% | 2.8% |
| MKR | +0.4 | 4% | −13% | 5.6% |
| GRT | +0.6 | 4% | −12% | 3.2% |
Two mechanically independent, market-neutral sleeves across 46 perpetuals, blended 50/50, one daily rebalance. Every signal uses only confirmed (closed) daily bars — no repainting, proven by a causality test.
One honest note on the signal: the trend leg reads only the color state of the CISD indicator (its change-in-state-of-delivery trend), not its drawn levels or liquidity-sweep markers — the parts that repaint. On a single coin that signal has no edge over holding; the value appears only when it is used to rank the cross-section.
A tearsheet that flags survivorship bias but never measures it isn't worth much, so we did the test. We pulled every USDT perpetual Binance ever listed — including the ones that died (LUNA, MATIC/POL, SRM, TOMO, ANC and others) — and rebuilt the book point-in-time: each day the universe is the top-N by trailing liquidity among the coins that were actually live and ≥90 days old that day. No hindsight, no excluded corpses.
The honest number
Survivor universe: Sharpe ≈ 1.6. Point-in-time (top-20 most-liquid): Sharpe ≈ 1.0 full, ≈ 2.0 out-of-sample. Roughly half the full-sample edge was the curated universe.
And the breadth story inverts. On the survivor set, more names helped; point-in-time, only the top ~20 most-liquid names hold up — going broader (top-30/40) drops the full-sample Sharpe toward ~0.4, because the honest universe churns through low-quality names that a survivor list quietly excludes. Diversification only pays when the things you diversify into are individually sound — survivorship guaranteed that; a live universe does not.
The takeaway isn't that the strategy is broken — a point-in-time top-20 book at Sharpe ~1.0 (OOS ~2.0), market-neutral, is still a legitimate edge. It's that the deployable version is narrower and smaller than the survivor headline, and the honest number to plan around is ~1.0, not ~1.6.