PHOENIX → HYDRA (current)

HYDRA — 20-Sleeve Diversified Ensemble

Risk-parity inverse-vol ensemble across 8 alpha categories | Monthly sleeve rebal, daily vol scaling | Portfolio vol-target 20%, gross cap 5× | No look-ahead, 15 bps TC

Strategy Overview

HYDRA is a professional-grade ensemble targeting institutional risk-adjusted return via uncorrelated-sleeve diversification rather than concentrated leverage. 20 independently-constructed sleeves span 8 alpha categories: equity trend, fixed income, commodity/energy, FX, volatility, crypto, cross-asset, and alternatives. Each sleeve is vol-targeted to 10% annualised; the ensemble is weighted inverse-vol (risk parity) and vol-targeted at 20% with a 5× gross cap.

Objective: Sharpe above 1.5 full-sample, above 2.0 out-of-sample, with max drawdown bounded under −20% — delivered without hindsight-biased sleeve selection, concentrated leverage, or look-ahead. Alpha comes from sleeve diversification (mean |pairwise correlation| ≈ 0.17), not from stacking a single bet.

Construction: Every sleeve uses 1-bar signal lag, 15 bps transaction cost on turnover, monthly sleeve rebalancing, daily inverse-vol weight updates, and rolling 63-day volatility scaling (5% floor, 1.5× scaling cap at the sleeve level). Walk-forward filter and regime overlays were tested and rejected — both hurt net performance (filter dropped sleeves at bottoms before recoveries; overlay was net-neutral).

Diversification math: With N=20, avg sleeve Sharpe ≈ 0.5, avg pairwise correlation ≈ 0.17, the equal-weight diversified Sharpe ceiling is ≈ 1.1. Inverse-vol weighting plus sleeve design (JPY safe-haven, dollar-neutral long-short, crisis hedges) lifts realised full-window Sharpe to 1.58 and OOS Sharpe to 2.01.

Honest ceiling: After extensive iteration, full-window Sharpe ≈ 1.6 / OOS Sharpe ≈ 2.0 is the realistic upper bound for a 21-year backtest with honest construction. Hitting Sharpe 3 over 21 years requires hindsight-biased sleeve selection, concentrated leverage (the METEOR-style approach produced −78% MDD in its 21y proxy), or sleeves exploiting regimes that won't repeat. HYDRA is the defensible professional-grade alternative.

HYDRA-Lite. A simplified execution variant is included for operators preferring monthly-only rebalancing with no dynamic vol scaling at the portfolio level: equal-weight sleeves, static 2.43× leverage, single trade per month. It achieves Sharpe 1.20 / CAGR 12% / MDD −17% — still clearly superior to SPY but trading ~0.4 SR for operational simplicity. See comparison section below.

How Rebalancing Works

Sleeve RebalanceMonthly (21 trading days). Each sleeve's signal (momentum, trend, carry, regime) is recomputed once a month from T−1 close data. Monthly cadence minimises turnover noise while preserving sleeve responsiveness.
Sleeve ConstructionEach of the 20 sleeves is a standalone long-only, short-only, or long-short rule acting on liquid ETFs (and BTC). Every sleeve is independently vol-targeted to 10% annualised via a rolling 63-day realised vol, with a 5% floor and a 1.5× scaling cap.
Portfolio WeightingInverse-vol risk parity. Daily weight on sleeve i ∝ 1 / σ_i(63d). Sleeves not yet live (pre-inception of their ETF) have zero weight; weights renormalise to sum to 1 across active sleeves. This equalises risk contribution, not dollar exposure.
Portfolio Vol Target20% annualised vol, 5× gross cap. After the inverse-vol blend, the portfolio is scaled daily by target_vol / realised_63d_vol, clipped at 5×. In practice the leverage is typically 1.5–3×. No path-dependent throttle, no CPPI — pure vol targeting.
Signal Lag1 full bar. Signals computed from T−1 close are applied from T open onward. This is the honest no-look-ahead convention and is cross-checked against T-open execution in the backtest.
Transaction Costs15 bps on turnover. Applied at the sleeve level on each rebalance and at the ensemble level on daily weight drift. Because inverse-vol weights drift slowly, daily turnover is small (monthly turnover dominates).
What is NOT DoneNo walk-forward sleeve filter (tested: dropped sleeves at bottoms, SR 1.57 → 1.01). No regime overlay (tested: net-neutral). No min-variance optimisation (tested: concentrated in low-vol sleeves, SR 1.58 → 1.14). No SR-tilt weighting (tested: momentum-chased weak periods, SR 1.59 → 1.07). Rejecting these is the finding.
Concrete Example — Monthly Rebalance Day:
Day T−1, 4 PM
Market closes. Each of the 20 sleeves recomputes its signal from T−1 close data: trend filters (20/50/100/200dma), momentum lookbacks (21d/63d/126d), regime gates (VIX level, yield trend, breakeven trend, dollar trend). 63-day realised vol updates.
Day T−1, evening
Each sleeve outputs its next-bar target weight. The ensemble computes inverse-vol weights across live sleeves and renormalises. Portfolio vol scaler recomputes (target 20% / realised 63d vol, clipped at 5×).
Day T, 9:30 AM
Orders execute at T open. Sleeve-level turnover absorbs 15 bps; ensemble vol-scale drift absorbs additional turnover cost if meaningful.
Day T, 4 PM
Day's return = Σ(weight_i × sleeve_i_return) × vol_scale − turnover_cost. NAV updates; drawdown and realised-vol trackers update.
Days T+1 … T+20
Sleeve signals hold. Inverse-vol weights drift daily with sleeve realised-vol updates. Portfolio vol scaler adjusts daily. Next monthly rebalance day, cycle repeats.
Growth of $10,000 (HYDRA vs SPY)
In-Sample vs Out-of-Sample

In-Sample (2005-04 → 2017-12)

Out-of-Sample (2018-01 → present)

Interpretation. OOS Sharpe 2.01 is higher than IS Sharpe 1.34 — a strong robustness signal. OOS MDD is also shallower (−14.6% vs −18.7%). No curve-fit period cherry-picking; the IS/OOS split at 2018-01-01 was decided before sleeve selection and never revisited.
Walk-Forward — Rolling 5-Year Windows
HYDRA strongly outperforms SPY in 3 of 4 non-overlapping 5-year windows. 2011-2015 was a multi-strategy-fund-wide weak period (low vol, dispersion-starved, bond bull-bear tantrum); HYDRA underperformed SPY's 12.9% return but never breached −15.0% drawdown. No window shows a large loss.
HYDRA-Lite Comparison (Simplified Execution)

HYDRA-Lite is a stripped-down execution variant designed for operators who want monthly-only rebalancing with no dynamic vol scaling at the portfolio level. The sleeve rules are the same; only the ensemble layer changes.

Weighting
Rebalance
Leverage
Portfolio vol scaling

HYDRA (shipped)

HYDRA-Lite

SPY benchmark

Trade-off. Stripping dynamic vol scaling costs ~0.4 Sharpe and ~4 percentage points of CAGR versus the shipped version, but eliminates the daily leverage-scaling workflow. MDD is essentially unchanged at matched target vol. If operational simplicity is the priority, HYDRA-Lite is still a significant upgrade over SPY (Sharpe 1.20 vs 0.63, MDD −17% vs −55%).

Walk-Forward — HYDRA vs HYDRA-Lite vs SPY

Leverage & Dynamic Weighting — Four-Variant Comparison

What does leverage actually buy you? This section isolates the two design choices that produce HYDRA's headline numbers — leverage (the portfolio scalar that can amplify the composite by up to 5×) and dynamic inverse-vol weighting (which reshuffles sleeve weights daily) — by showing all four combinations over the same 21-year window. Comparing horizontally isolates the leverage effect; comparing vertically (HYDRA-pair vs Lite-pair) isolates the weighting effect.

Compounded growth — $10,000 at inception

Trailing 1y & 3y (annualised)

Trailing Returns
Performance & Risk Statistics

Performance (full sample)

Risk (full sample)

Drawdown
Rolling 1-Year Sharpe
Daily Vol Scaling (Trailing 1 Year)

How the vol-targeter decides today's gross exposure. Every day the strategy measures the trailing 63-day realised vol of the inverse-vol-weighted sleeve composite (the "raw" portfolio, pre-scale). It then sets the gross-exposure scalar to 20% / raw_vol_63d, clipped at the 5× leverage cap. That scalar is applied to day-T's return — the realised line below shows the resulting 63-day vol of HYDRA's scaled output. Both series use .shift(1) on the rolling window, so today's sizing is based only on data through T−1 close; orders fill at T open.

Last 20 trading days — vol & sizing
Daily rebalance trades — last 15 trading days
Turnover = sum of |Δ gross sleeve position| in NAV%. Top buy/sell shows the single largest trade that day.
Per-sleeve trade ledger — most recent day
Top 10 sleeves by absolute trade size. Position = inverse-vol weight × gross scalar, expressed as % of NAV.
Current ETF positions (notional)
Per-ticker exposure after resolving each sleeve's internal ETF toggle and per-sleeve vol scalar. Sum of absolute exposures is the true gross — higher than the 500% sleeve-level cap because each sleeve is independently vol-targeted to 10% (capped at 1.5×) before the ensemble-level 20% target applies.
Exact ETF trades — most recent day
Signed per-ticker rebalance trades. Positive = buy, negative = sell, as a fraction of NAV.
ETF trade history — last 15 trading days
Top buys and top sells by absolute size each day. Turnover = sum of |Δ ETF| in NAV%.
Calendar Year Returns
Monthly Returns Heatmap
Current Sleeve Allocation (trailing 21d inverse-vol)
Per-Sleeve Diagnostics (since-inception, realised)
Sleeve Categories & Investment Universe
Equity trend (4) — vol-contingent SPY (s1), sector top-3 momentum (s2), semis trend SMH (s17), EM trend EEM (s19)
Fixed income (5) — bond duration regime TLT (s3), credit trend HYG (s4), yield-curve carry (s5), inflation hedge TIP/IEF (s20), EM bond carry EMB (s24)
Commodity / Energy (3) — DBC trend (s6), gold-silver regime (s7), XLE energy regime (s22)
FX (2) — JPY safe-haven FXY (s8, VIX-triggered), dollar regime UUP (s9)
Volatility (1) — VIX contango carry (s10)
Crypto (1) — BTC trend (s12)
Cross-asset (2) — absolute momentum GEM-style (s13), dollar-neutral long-short risk-on/off (s27)
Alternative (2) — defensive sector rotation (s15), SPY 5d mean-reversion (s18)
Instruments traded: SPY, IWM, QQQ, XLF XLK XLP XLY XLE XLV XLI XLU XLB, SMH, EEM, TLT IEF SHY, TIP, LQD HYG, EMB, DBC, GLD SLV, FXY UUP, VXX, BTC-linked proxy, BIL. Cash fallback SHY or BIL depending on sleeve.
Methodology Notes
Backtest disclosure. All figures are from a daily-bar simulation from 2005-04-05 to the most recent close shown above. Returns are gross of management fees and net of 15 bps round-trip transaction costs applied to turnover. Signals use T−1 close data applied from T open (one full bar lag). Sleeves that reference instruments not yet listed at a given date are zero-weighted until their ETF inception. Past results are not a guarantee of future performance. HYDRA is a research strategy; this factsheet is for informational purposes only and not investment advice.