PHOENIX → TITAN — BTC+ETH timing (current)

CRYPTO-TITAN — BTC + ETH Timing Strategy

Loading…

Strategy Overview

What this is: A BTC + ETH + cash timing strategy driven by a 21-model conviction ensemble. Every week the strategy decides: are we in BTC, are we in ETH, are we leveraged, or are we in cash. It never trades alts. Maximum 2 simultaneous positions. ~41% of historical days are full cash.

Tradable universe vs Signal universe:
Tradable: BTC, ETH only. Max 2 positions held simultaneously.
Signal: 50 coins (35 survivors + 15 explicitly-dead) used by the breadth signal, master gate, and survivorship-bias robustness check — but the strategy never opens a position in any alt.

Why no alts? Five alt-trading sleeves were tested empirically and dropped: ALT_XSMOM (OOS SR -0.21), ALT_DIVERS (0.16), BROAD_XSMOM top-5-of-100 (-0.26, -60% MDD), QUINTILE_SPREAD (-0.58, -67% MDD), DISPERSION_TIMED (0.31). Daily-bar alt momentum systematically picks parabolic alts right before they reverse — the dead coins in our universe (LUNA, FTT, etc.) were exactly the kind of "top momentum" names this would have bought.

Objective: ~50% CAGR with -32% MDD on the BTC+ETH+cash space, executable manually on OKX (leveraged tokens) or Hyperliquid (perpetuals). Long-only. Weekly rebalance Wednesdays.

How the timing decision is made: 21 weakly-correlated sleeves each output a per-coin target weight (mostly 0 or some BTC weight, occasionally ETH). The aggregated weights are sleeve-vol-targeted, risk-parity-blended, consensus-gated, and weekly-snapped. A daily vol-target overlay scales total exposure. A conviction-scaled leverage curve (1×→3.2×) amplifies high-agreement weeks. A vol-of-vol regime gate cuts leverage when vol is expanding. A self-stop halves exposure on drawdowns over -4% in 30 days.

21 timing models (every model outputs only BTC and/or ETH weights):

  • 17 BTC-only sleeves: BTC_VM, BTC_SLOW, TRIPLE_MOM, TURTLE20, TURTLE55, BOLLINGER, CONSOLIDATION_BREAK, MACRO_TREND (SPY+VIX gated), HALVING_BOOST, AUTOCORR_REGIME, CRASH_DIP, VOL_BREAKOUT, EFFICIENCY_TREND, RANGE_EXPANSION, INV_VOL_BTC (Moreira-Muir), VOL_PERCENTILE, SPY_ALPHA
  • 3 BTC+ETH sleeves: BREADTH_THRUST (Zweig kickoff), MTF_FRACTAL (NOVEL: long only when ALL 6 timeframes 5/21/63/126/252/365d aligned), CONVEX_BREAKOUT (NOVEL: NO vol-targeting → preserves right-tail convexity, trailing stops only)
  • 1 ETH-only sleeve: ETH_VM

The sleeves are deliberately heavy on BTC because empirically, 11 years of crypto data show BTC has the cleanest signal-to-noise ratio for daily-bar trend, breakout, and regime models. ETH adds a small uncorrelated layer; alts add noise.

Portfolio overlays (1-day lagged):

  1. Risk-parity sleeve weights: inverse 90d realised vol of each sleeve's returns.
  2. Consensus power-curve: scale = ((conv − 0.15)/0.55)1.1 × 1.35; below 0.15 sleeve agreement, full cash.
  3. Master gate: weighted vote over BTC>200MA, BTC DD>-30%, BTC 63d-mom>-8%, breadth>25%; full cash if score < 0.25.
  4. Weekly snap: positions held constant Wed-to-Wed; only re-evaluated weekly.
  5. Daily vol target: EWMA(60d) realized vol → 18% target, asymmetric clip [0.3, 1.5].
  6. Daily DD floor: linear scale-down from 0% to -12% NAV drawdown vs 365d HWM.
  7. Self-stop: NOVEL — when own 30d return < -4%, halve exposure for 21 days.
  8. Catastrophic-DD per coin: any coin in >28% DD from 60d high → forced to 0 weight.
  9. Conviction-scaled leverage: 1× base, ramps to 3.2× when sleeve agreement crosses 0.85.
  10. Vol-of-vol regime gate: NOVEL — when 14d-smoothed vol has risen >20% over 28d, cut leverage extra by 30%.

Why a 50-coin signal universe if we only trade BTC/ETH?

  1. Survivorship-bias robustness — Including 15 explicitly-dead coins (LUNA1, FTT, USTC, MATIC, UNI, XEM, BTG, STEEM, SC, OMG, WAVES, CELR, ANT, GRT, KNC) is the test. The strategy SR is higher on the 50-coin universe than on survivors-only, ΔSR -0.10 — the eligibility + catastrophic-DD overlay correctly excludes the dead coins as they crash, and their presence doesn't generate spurious alpha.
  2. Breadth signal — BREADTH_THRUST and the master gate use "% of 50 coins trending above 100MA". This produces a market-wide measure that's more meaningful than just BTC's own state. The resulting position is still BTC+ETH only.
  3. Validation of the catastrophic-DD overlay — the dead-coin time series confirm the per-coin -28% DD filter triggers at the right moments (LUNA May 2022, FTT Nov 2022, etc.).

35 survivor coins (signal-only): BTC, ETH, SOL, ADA, DOGE, LTC, BCH, XRP, LINK, DOT, AVAX, ATOM, XLM, TRX, ALGO, BNB, XMR, DASH, EOS, NEO, ETC, ZEC, HBAR, VET, FIL, THETA, XTZ, QTUM, MKR, BAT, IOTA, DCR, ZRX, LRC, ICX.
15 dead coins (signal-only): LUNA1 (Terra collapse), USTC (depeg), FTT (FTX collapse), MATIC (migrated to POL), UNI (delisted), XEM (NEM -99%), BTG (51%-attacked), STEEM, SC, OMG, WAVES (USDN collapse), CELR, ANT, GRT, KNC.
Of these 50 coins, only BTC and ETH ever appear in the actual position book.

Honest assessment — what this strategy actually is:

This is a BTC + ETH + cash timing strategy, not a "broad crypto" strategy. The 21 models all output BTC and/or ETH weights — they never trade alts. Across the 4,238-day backtest the strategy held: 0 positions (cash) 41% of days, 1 position (usually BTC) 30% of days, 2 positions (BTC+ETH) 29% of days, 3+ positions 0% of days. Across all alts in the 50-coin universe, total days held: zero. The wide universe is for signal/bias-test only.

After v1 → v10 iterations, CRYPTO-TITAN reaches Full SR 1.47, OOS SR 1.13, CAGR 51.4%, MDD -32%, NAV 124× over 11.5 years. CAGR target (50%+) is met via conviction-scaled leverage (1×→3.2×) on Hyperliquid perps / OKX leveraged tokens.

OOS Sharpe plateau is real: across all configurations tested (universe sizes 20→111, sleeve counts 4→23, daily vs weekly rebal, leverage 1× to 3.5×, several short variants, multiple regime gates and conviction curves), OOS Sharpe stays in the 1.10-1.50 range. The published crypto-quant ceiling for long-only daily spot bars without funding-rate or options data is ~SR 2.

Empirical findings worth flagging:
  • Shorts don't work on daily bars in this sample. Tested SHORT_BTC_BEAR, SHORT_TREND_BREAK, SHORT_PERMABEAR, LONG_SHORT_XSMOM. All -SR. Bounce-squeeze + funding drag + structural BTC uptrend over 11 years punishes systematic short bias. Daily-bar shorting needs intraday data + funding signals.
  • Naive cross-sectional XSMOM across 100+ coins blows up (-60% MDD). Picks "trending" alts right before they crash. Survivorship-bias test passes (no false alpha) but signal is too noisy.
  • Asset concentration kills — going top-K per day with leverage amplified losses too. Time-concentration (cash when uncertain, lever when sure) was the better interpretation.
  • Calendar anomalies inverted post-2022. Turn-of-month had OOS SR -0.63. Day-of-week effects are now noise.
  • Survivorship bias is NEGATIVE in this strategy — the full universe SR is 0.10 HIGHER than survivors-only. Including dead coins makes the strategy better (the eligibility + catastrophic-DD overlay correctly excludes them before damage; their historical dispersion just adds noise to weak baselines, not to ours).

Path to genuinely higher OOS SR (3+) requires NEW DATA:
  • Perp funding rates from Hyperliquid/Bybit/OKX — long spot + short perp captures basis (cash-and-carry). Historical Sharpe 3-5. Adds one extra trade leg per coin per week.
  • Higher-frequency intraday data — orderbook microstructure, funding-rate changes, basis spreads. Sharpe 2-4 typical for tactical entries.
  • Options vol-risk-premium — sell premium on Deribit BTC options. Sharpe 2-3 typical. (User has indicated no options.)

v1 → v8 progression: v1 OOS 0.54 → v2 1.21 → v3 1.49 → v4 1.42 → v5 1.40 → v6 1.01 (shorts hurt) → v7 1.15 (leverage added) → v8 1.13 OOS / 1.47 Full / CAGR 51%.
Headline metrics
Growth of $10,000 (2014-2026)
Log scale. Net of 20 bps round-trip TC, 30 bps short TC (where applicable), and ~15%/yr leverage-funding cost on the levered portion. Includes 15 dead coins (LUNA1, USTC, FTT, etc.) and the daily catastrophic-DD overlay that pulls long positions to zero on -28% drops from 60d high.
Drawdown
Drawdown from prior peak NAV. Strategy-level DD floor at -12% NAV begins linearly de-risking; full cash at -12%. The leverage in v7/v8 raised peak MDD to -32% from a base of -13% in unleveraged v3-v6.
Rolling 12-month Sharpe
Rolling 12-month Sharpe shows the strategy is positive-SR most of the time, with regime-driven volatility around the mean. The 2018-2019 transition and 2022-2023 transition show the gates kicking in (SR collapses to near-zero rather than going deeply negative).
Current week allocation
Important: despite the universe of 50 coins, the strategy in practice ONLY ever allocates to BTC and/or ETH. Other coins (SOL, ADA, XRP, etc.) and the 15 dead coins (LUNA1, FTT, USTC, etc.) are in the universe to stress-test survivorship-bias robustness — they are never bought. The 21 sleeves are designed around BTC/ETH; alt sleeves (ALT_XSMOM, ALT_DIVERS) were tested and dropped because alt selection on daily bars introduces noise without alpha (and blew up on dead-coin crashes).
Recent 12 weekly snapshots
Historical exposure (BTC + ETH stack, 2014-2026)
Stacked weekly exposure to BTC and ETH. The whitespace gaps are CASH periods where the master gate or self-stop fired (notably most of 2018, all of 2022 winter, and large parts of 2025). Above 1.0 = leveraged exposure (1× to 3.2× via Hyperliquid perps / OKX leveraged tokens).
Coin holding frequency (entire backtest)

Coins ever allocated to

Position count distribution

The strategy holds at most 2 positions simultaneously (BTC + ETH). Median day in-market: 1 position. The high cash percentage reflects the strict master gate that pulls everything to zero in confirmed bear regimes (2018, 2022) and chop (much of 2025).
In-sample vs Out-of-sample

In-Sample 2014-09 → 2022-06

Out-of-Sample 2022-07 → present

No model parameters are fit on OOS data. Each sleeve's signal logic (lookbacks, thresholds, stops) was settled before the OOS period began. The OOS period covers 2022 crypto winter, 2023-24 recovery, and 2025 chop — three distinct regimes.
Yearly performance
vs Benchmarks
BTC buy-and-hold and equal-weight survivors are both DAILY-RETURN-CAPPED at ±25% (same realism floor used for the strategy backtest — without it, dead coins like LUNA bouncing from $0.00006 → $0.0002 would show fictional 200% daily prints). EW survivors only includes the 35 alive coins; this is the survivorship-biased benchmark. The strategy beats both benchmarks on Sharpe (1.47 vs 1.02 BTC and 1.06 EW) and dramatically on MDD (-32% vs -84%/-90%).
Per-sleeve standalone metrics
Each sleeve's metrics are computed independently — daily-rebalance backtest with 20 bps TC, no portfolio-level overlays. Higher OOS than full-sample SR for several sleeves (BREADTH_THRUST, CRASH_DIP) reflects that they specifically target conditions that have appeared more frequently OOS. Risk-parity sleeve weighting in the ensemble equalises their contribution; no sleeve has >10% of total risk budget.
Survivorship bias robustness (3-way test)
Test design: Run the same strategy code path against three universes — (a) SURVIVORS-only 35 coins (the classical biased view that drops failed coins); (b) STRATEGY universe 50 coins (35 survivors + 15 dead); (c) EXTENDED universe 111 coins (everything dead we have data for). The strategy's SR should be roughly equal or HIGHER on the wider universes — that is the negative-bias-delta test.

Result: Strategy SR is HIGHEST on the 50-coin set with dead coins included. Survivors-only is the LOWEST. This means the strategy's signal logic + catastrophic-DD overlay correctly exclude dead coins before they damage performance, AND the dead coins' historical correlation structure helps the ensemble. ΔSR (survivors − strategy) = -0.10 — classical survivorship bias would produce a positive delta; our strategy passes.

15 dead coins explicitly in universe: LUNA1 (Terra collapse May 2022), USTC (depeg), FTT (FTX collapse Nov 2022), MATIC (migrated to POL Mar 2025), UNI (delisted Apr 2025), XEM (NEM -99% from ATH), BTG (Bitcoin Gold 51%-attacked repeatedly), STEEM, SC (Siacoin abandoned), OMG Network (delisted), WAVES (USDN stablecoin collapse), CELR, ANT (Aragon DAO dissolved), GRT, KNC.
v12 deep-dive: thorough alt-trading tests (and why they all failed)

After observing that the strategy only ever holds BTC and/or ETH, we ran two systematic tests of "scan a wide universe and allocate to alts when conditions are right." Both failed empirically. Test results below — committed to the feature branch for full audit.

Test 1: MULTI_SCAN — 5-signal confluence scanner

Per-coin scoring on 5 independent signals: long-trend (price>200MA AND 50MA>200MA), short-mom (21d>5%), persistence (63d AND 252d positive), not-overbought (price < 0.93×21d-high — filters parabolic tops), stable-vol (21d vol < 1.5× 252d median). Strict eligibility: 252d minimum history, not in >25% DD, positive 365d return, BTC>200MA AND BTC 63d-mom > 5%. Top-K by 63d momentum among coins with confluence ≥ min_signals.

config (top_k, min_signals)Standalone OOS SRStandalone Full SREnsemble OOS SREnsemble CAGR
k=3, min=3+0.17 (50-univ) / -0.35 (111-univ)0.66 / 0.501.1252.5%
k=5, min=4-0.11 / -0.090.64 / 0.531.0950.2%
k=7, min=5-0.98 / -0.69-0.04 / 0.03+1.16 (best)42.7%
k=10, min=3+0.09 / +0.040.94 / 0.891.1052.0%
BASELINE (no scan)——1.1351.4%

Verdict: All 12 configurations have negative or near-zero standalone OOS SR. The best ensemble config (k=7, min=5) gives +0.03 OOS SR but loses 8.7% CAGR (NAV halves from 124× to 62×). Net-negative trade-off.

Test 2: Per-coin alt trend sleeves (parallel ETH_VM-style)

Eight new sleeves: SOL_VM, BNB_VM, AVAX_VM, LINK_VM, ADA_VM, XRP_VM, LTC_VM, DOGE_VM. Each is a dedicated vol-managed trend follower on one coin (same logic as BTC_VM and ETH_VM). All BTC-regime gated (only fire when BTC>200MA AND BTC 63d mom > 0).

Coin sleeveStandalone Full SRStandalone OOS SR
SOL_VM0.86+0.36
BNB_VM0.78+0.78
AVAX_VM0.51+0.12
LINK_VM0.55+0.08
ADA_VM0.60+0.49
XRP_VM0.39+0.54
LTC_VM0.35-0.22
DOGE_VM0.25-0.03

Ensemble impact (with all 8 alt sleeves added):

Full SROOS SRCAGRMDDNAV2025+
Baseline (BTC+ETH only)1.471.1351.4%-32.1%124×-3%
+ 8 alt sleeves1.170.9230.7%-35.8%22×-14%
Δ-0.30-0.21-20.7%-3.7%-102×-11%

Verdict: Adding 8 alt sleeves to the ensemble DROPS OOS SR by 0.21 and 80% of NAV. The 2025+ year goes from -3% to -14% because alts crashed harder than BTC in 2025 chop. Cherry-picking just the OOS-strong sleeve (BNB_VM at 0.78) would be look-ahead bias.

Conclusion from v12 thorough testing

The empirical answer is: no, this framework cannot beat BTC+ETH+cash by adding alt allocation, regardless of selection method. Three approaches tested across v6, v12:
  1. Naive momentum rotation (BROAD_XSMOM, QUINTILE_SPREAD): catastrophic, MDD -60% to -90%
  2. Multi-signal confluence scanner (MULTI_SCAN): neutral to slightly negative, all 12 configs
  3. Dedicated per-coin trend sleeves: -0.21 OOS SR, -80% NAV impact
The structural reason is that crypto alts are very highly correlated with BTC (~0.85), so adding them doesn't diversify — it amplifies. In bear regimes (2018, 2022, 2025) alts crash 30-50% harder than BTC. The win-rate in alt seasons (2017, 2021) doesn't compensate for the loss-rate in winter and chop. Empirically, BTC+ETH+cash with conviction-scaled leverage is the upper envelope of what this signal stack produces. Code for all dropped sleeves is preserved in sleeves.py for audit.
v1 → v8 iteration progression
Every commit is preserved on the feature branch. Key learnings: (1) more sleeves > fewer (Phoenix philosophy holds in crypto too); (2) shorts on daily bars are -SR for crypto in this sample; (3) leverage scales CAGR linearly but not Sharpe; (4) the OOS plateau ~1.5 is structural, not parameter-tunable.
Live execution — manual rebalance flow
Recommended setup: Start manually on OKX with leveraged tokens for 4-8 weeks. Once the signal is proven live, port to a Hyperliquid bot for automated weekly rebalance.

Weekly Wednesday workflow (15 min)

  1. Fetch latest closing prices into data/crypto/ (yfinance via scripts/expand_universe.py).
  2. Run python live_signal.py --venue okx --capital <USD> after market close.
  3. Output is a target-weights table with a "trades vs last week" diff (OPEN / CLOSE / FLIP / ADD / TRIM).
  4. Execute the diff manually:
        OKX: use spot for <1.5× weights, BTC3L/ETH3L for long-leveraged, BTC3S/ETH3S for short-leveraged (currently disabled but framework ready).
        Hyperliquid: open/close/resize perp positions at the indicated leverage.
  5. Save snapshot of executed weights for next-week diff.

Hyperliquid bot path (after live validation)

  • Hyperliquid has a public Python SDK (hyperliquid-python-sdk).
  • Bot would: cron weekly Wed 00:00 UTC → fetch fresh prices → run strategy → diff vs current Hyperliquid positions → submit signed order batch.
  • Defensive overlays: per-coin position cap, daily loss-limit kill-switch, manual approval mode for first month.

Key execution costs & haircuts (already in backtest)

  • Spot taker fee: ~10 bps. Modelled at 20 bps round-trip.
  • Perp taker fee: ~5-7 bps. Modelled at 30 bps round-trip (conservative).
  • Funding rate on perp leverage: ~3 bps/day in long-heavy regimes (~15%/yr) on the levered notional above 1×. Modelled.
  • Slippage on illiquid alts: NOT modelled. Limit positions to 1% of daily volume (manual check on coins like KNC, CELR).
Disclaimer: This is research output, not investment advice. Past performance does not guarantee future results. Crypto markets are highly volatile. Leverage amplifies losses. Backtests include transaction costs (20-30 bps round-trip) and modelled perp funding (~15%/yr on levered notional) but no slippage on illiquid coins. Survivorship-bias-aware backtest design — universe includes 15 dead/collapsed coins (LUNA1, USTC, FTT, MATIC, UNI, XEM, BTG, STEEM, SC, OMG, WAVES, CELR, ANT, GRT, KNC) explicitly. The OOS Sharpe ceiling on daily-bar spot crypto data appears to be ~1.5 across many configurations tested.