Knowledge · Research · The blended 3-sleeve portfolio: what it really returns, and why equal capital is not equal risk

The blended 3-sleeve portfolio: what it really returns, and why equal capital is not equal risk

Strategy analysis 2026-07-05 6 sources
A real blended backtest of the three sleeves (Donchian overlay = de-risked BTC beta, funding carry = market-neutral income, XSECT = market-neutral momentum) as they actually run: isolated wallets, equal capital, geometric returns over the full period including flat days. The three are genuinely uncorrelated (avg pairwise +0.08), which turns three modest/uncertain sleeves into a smoother whole: honest forward expectation ~+10-13%/yr (Monte-Carlo median +11%, 74% positive years), maxDD -15 to -21% vs buy-and-hold BTC's -50%, and no liquidation. The key finding: equal CAPITAL is not equal RISK - the overlay carries 86% of portfolio risk, so a risk-aware weighting (down-weight the directional sleeve) lifts Sharpe from 0.72 to 1.23 and cuts drawdown to -9% for the same return.
  • The three sleeves are genuinely uncorrelated (avg pairwise correlation +0.08) - the diversification is real, not assumed. That is the whole reason the isolated-wallet design works.
  • Honest forward expectation ~+10-13%/yr (Monte-Carlo median +11%, 74% chance of a positive year), NOT the ~25% you get by summing the three sleeve headlines.
  • Idle time is a real drag: XSECT compounds at +12.5%/yr geometric over the full period, not its +26% 'active-day' headline - it is flat ~52% of the time.
  • Equal CAPITAL is not equal RISK: the overlay carries 86% of portfolio risk at 1/3 weight. Down-weighting it to ~0.2 lifts Sharpe from 0.72 to 1.23 and cuts maxDD from -21% to -9% for essentially the same return.
  • Portfolio maxDD -15 to -21% vs buy-and-hold BTC's -50% - the book roughly thirds the drawdown and never approaches liquidation (unlevered, worst single day -4.2%).
  • Over a BTC-bull window the diversified book does NOT beat holding BTC risk-adjusted (Sharpe 0.72 vs 0.91 at equal capital, drift). Its value is drawdown, stability and no-liquidation - it is insurance-shaped, not a bull-market outperformer.
  • Carry's daily-vol Sharpe (~11) is a measurement artefact; its real tail is basis/dislocation risk not present in the daily funding series. Treat carry's true risk as fatter than shown.
  • Caveats: only ~3 years of common history (carry data starts 2023-05, bull-heavy - no full bear in-sample); XSECT is DSR-borderline; the carry backtest uses a BTC/ETH/SOL proxy for the deployed HYPE carry.
Run the book at a RISK-aware weight, not equal capital: down-weight the directional overlay (it eats 86% of the risk budget at 1/3). ~0.2/0.4/0.4 gave Sharpe 1.23 and maxDD -9% for the same ~11% return.
Treat ~+10-13%/yr as the honest central expectation with a ~-15% fifth-percentile year. Never quote a sleeve's active-day return as its portfolio contribution - use geometric full-period CAGR.
Keep the book unlevered. The overlay's daily-close exit cannot escape an intraday gap; leverage reintroduces exactly the liquidation risk the whole design avoids.
Extend the carry funding history (python data/download_funding.py) so the common backtest window covers a full bear (2022), not just the 2023-2026 recovery - the current forward distribution is optimistic.
Score any future portfolio change (new sleeve, new weights, drift vs rebalanced) with backtesting/portfolio_ladder.py so variants are judged on the same 7-rung bar.

The question

Three isolated wallets, equal capital, left to run: what is the honest per-year return expectation - after accounting for the fact that each sleeve is out of the market part of the time? This is the portfolio-level answer, not three strategy headlines summed up.

The three sleeves and their roles:

  • Donchian overlay (single-BTC 20/20, LIVE) - de-risked directional crypto beta. Long above the 20-day high, flat below the 20-day low. Owns the crypto upside, steps aside in downtrends.
  • Funding carry (delta-neutral) - market-neutral income. Long spot + short perp, harvests the funding premium. No directional bet.
  • Cross-sectional momentum (XSECT) - market-neutral momentum alpha. Long the relative winners, short the losers over 20 coins, dollar-neutral (BTC beta ~0).

Method (and the one honesty move that matters)

Each sleeve is reduced to a daily net-return series (net of fees/turnover), where a day out of the market contributes exactly 0. Blended two ways: drift (three isolated wallets each seeded 1/3, compounding independently - how the bot actually runs) and rebal (daily-rebalanced equal weight - the theoretical 1/3-1/3-1/3). Common period 2023-05 to 2026-05 (~3 years, limited by how far back the funding data reaches).

The honesty move: every return here is geometric CAGR over the full period including flat days, never active-day annualisation. This matters enormously. XSECT's often-quoted +26%/yr is its return on the days it trades - but it is flat ~52% of the time (crash filter + warmup), so the number that actually lands in the wallet over the full period is +12.5%/yr geometric. Idle capital is a real drag, and it is already baked into these numbers - not modelled after the fact.

The sleeves, solo (daily net, to 2026)

Sleeve CAGR (geom) Sharpe maxDD in-market
Overlay (BTC 20/20) +39% 0.87 -48% 65%
Carry (BTC/ETH/SOL, 2x cap) +6.3% 11.2* -0.7% 100%
XSECT (deployed) +12.5% 1.02 -19% 48%

*The carry Sharpe of 11 is a daily-volatility artefact: the daily funding series is almost monotonic, so measured daily vol is near zero. Carry's REAL risk is not in the daily series - it is basis blow-outs, exchange/liquidity dislocations and single-coin (HYPE) idiosyncrasy. Treat the carry Sharpe as meaningless-high and its true tail as fatter than shown.

The result that carries the whole thesis: they are genuinely uncorrelated

Average pairwise correlation of the three daily-return series: +0.08 (overlay-carry 0.12, overlay-xsect 0.11, carry-xsect 0.00). These are three real bets, not one dressed up as three. That low correlation is exactly what lets three modest/uncertain sleeves combine into a much smoother whole - it is the entire reason the isolated-wallet design exists.

The blended portfolio (equal capital, common period)

Convention CAGR Vol Sharpe maxDD worst day
drift (realistic) +11.7% 16% 0.72 -21.3% -4.2%
rebalanced +13.1% 13% 1.00 -15.3% -2.8%

Monte-Carlo (block bootstrap, preserves autocorrelation, 1-year forward, drift): return p5/p50/p95 = -15% / +11% / +51%, prob(positive year) 74%, prob(down >10%) 10%, 1-year maxDD median -12% and 95%-worst -23%. So the honest central expectation is ~+11%/yr, with a survivable worst case and no path to liquidation (the book is unlevered).

The insight the ladder surfaced: equal CAPITAL is not equal RISK

At 1/3-1/3-1/3, the overlay carries 86% of the portfolio's risk (carry ~0%, xsect ~14%). The directional sleeve dominates, which is why the equal-capital drift book has a middling Sharpe (0.72) - lower even than XSECT alone (1.02). Re-weighting to respect risk changes everything:

Weighting CAGR Sharpe maxDD Calmar
equal capital, drift +11.7% 0.72 -21% 0.55
equal capital, rebalanced +13.1% 1.00 -15% 0.85
risk-aware 0.2/0.4/0.4, rebal +11.3% 1.23 -9% 1.28
neutral-only (carry+xsect) +8.2% 1.25 -8% 0.98

Down-weighting the overlay to 0.2 keeps essentially the same return (+11.3%) while lifting Sharpe to 1.23 and cutting maxDD to -9%. The pure market-neutral book (drop the directional overlay entirely) has the cleanest risk profile but gives up the crypto-bull upside and ~3 points of return.

The honest verdict (portfolio evaluation ladder)

Seven rungs, geometric throughout. The equal-capital drift book (how it actually runs) fails rung 4: over this 2023-2026 window - which was mostly a BTC recovery bull - simply holding BTC beat the diversified book risk-adjusted (Sharpe 0.72 vs 0.91). That is the uncomfortable truth: in a strong bull, diversification and drawdown control cost you. The book earns its keep on rungs 5-7 (drawdown -21% vs BTC -50%, worst calendar year -3%, no liquidation) - it is insurance-shaped, not a bull-market outperformer. The risk-aware and rebalanced variants pass all rungs (SOLID).

Caveats (do not over-trust the numbers)

  • Only ~3 years of common history, and it was bull-heavy (carry funding data starts 2023-05). The bootstrap has never seen a full brutal bear in this window (2025 was only -3%), so the forward distribution is optimistic. Extend the funding history to cover 2022 before trusting the tails.
  • Carry's Sharpe is flattered (daily-vol artefact); its real risk is basis/dislocation, not shown.
  • XSECT is DSR-borderline (~0.93, just under the 0.95 bar) - real but not proven; its +12.5% could be less forward.
  • The carry backtest uses BTC/ETH/SOL funding as a proxy for the deployed HYPE carry (no HYPE history on disk). A single-coin live wallet is noisier than this cross-asset average.

Bottom line for "good return over years without getting liquidated"

Honest central expectation ~+10-13%/yr, a ~-15% fifth-percentile year, a worst-case drawdown around -20 to -23%, and no liquidation (nothing is levered). The book is a smoother ride with a lower ceiling than holding BTC - exactly insurance-shaped, which matches the goal. The single most useful lever is not more return but risk-aware weighting: at equal capital the overlay eats the risk budget; down-weighting it to ~0.2 buys a materially cleaner ride for the same return.

Reproduce: python backtesting/portfolio_backtest.py (full report) and python backtesting/portfolio_ladder.py (the 7-rung score + variants).