HyperliquidMultiSymbolDonchianBreakoutLong4H

Hyperliquid Multi-Symbol 4H Donchian Range Breakout Long-Only Portfolio

A long-only multi-instrument breakout strategy on Hyperliquid perpetual futures (HL venue, ~0.09% round-trip fee) — a venue ENTIRELY ABSENT from the current pipeline and the entire failure log. Every strategy tested this session has been on Binance USD-M, Binance Spot, or BINANCE_CM. Hyperliquid is a fundamentally distinct venue: on-chain decentralized order book, dominated by retail/degen flow rather than institutional market-makers, hourly funding cycles (vs Binance's 8-hour), and a completely different participant base whose microstructure behavior has NOT been arbitraged away to the same degree as Binance. The strategy uses 4-HOUR bars on a 3-symbol basket: BTC-USD.HYPERLIQUID, ETH-USD.HYPERLIQUID, SOL-USD.HYPERLIQUID — the three highest-volume HL perps with the longest available history. Mechanism is a textbook 20-bar Donchian channel breakout: on each 4H bar close, for each symbol, if close > prior 20-bar high (~3.3 day window) AND the symbol is not currently long AND fewer than 2 portfolio positions are open, enter long sized at 5% of equity per leg (15% max gross exposure). Exits use a trailing stop at the prior 10-bar low (~1.7 day window), a hard stop at -7% from entry, and a time stop at 24 bars (~4 days) if PnL is below +2%. NO volume filter, NO trend regime gate, NO ADX, NO EMA confirmation — pure price-based Donchian, the simplest possible breakout mechanism. Two free parameters total (entry lookback=20, exit lookback=10) — the lowest count among all hypotheses this session. This hypothesis is structurally orthogonal to every confirmed failure pattern: (1) NOT a BTC-on-Binance strategy (avoiding the 'BTC-2024+-microstructure-regime-decay' pattern that killed 4 independent BTC strategies on Binance — seasonality, momentum, mean-reversion, vol-spike); (2) NOT a single-instrument alt-daily breakout (3-symbol portfolio aggregates trade frequency 3x AND uses 4H timeframe, multiplicatively avoiding the 9 single-alt-daily-breakout failures); (3) NOT a single-bar 4H pattern (Donchian aggregates 20 bars of high/low context, not just the current bar's properties); (4) NOT a pullback-in-uptrend (pure breakout entry, opposite mechanism); (5) NOT mean-reversion-with-technical-gates (no Bollinger/RSI/SMA); (6) NOT a seasonality strategy (no time-of-day or day-of-week logic); (7) Does NOT depend on OI/L-S-ratio supplementary data (avoiding the verification-framework no-trades failure). The cross-venue cross-section test is the experimental crux: if Binance-validated mechanisms decay due to 2024+ Binance-specific microstructure shifts (institutional adoption, ETF flows, market-maker crowding), then Hyperliquid — a venue with a fundamentally different and younger ecosystem — should still exhibit the original momentum-continuation premium. If it doesn't, that's itself a meaningful negative result that informs future hypothesis selection. Expected trade frequency: 3 symbols × ~30-40 Donchian breakouts/year/symbol on 4H bars × 50% no-overlap entry rate ≈ 90-120 trades/year, well above the walk-forward viability floor.

Implements the cross-venue cross-section test exactly: a textbook Donchian breakout on Hyperliquid — a venue entirely absent from the pipeline and failure log — to check whether the momentum-continuation premium that decayed on 2024+ Binance microstructure survives on a younger, retail-dominated on-chain venue. Correctness details: the HL instrument-id format is the catalog-confirmed BTCUSD.HYPERLIQUID (no dash; the hypothesis's 'BTC-USD' form would not resolve), and all three 4H bars exist in the catalog. Each symbol keeps its own high/low deques and the prior-N-bar Donchian levels exclude the current bar, so the breakout/trailing tests are computed against history, not the live bar. The portfolio cap (max 2 concurrent legs) is enforced across all symbols by counting open positions, and the secondary legs are driven on the primary 4H clock where the base's same-timeframe alignment barrier guarantees ETH/SOL bars are buffered at the BTC timestamp (no stale-leg artifact). Per the prompt's guidance and prior lessons, calculate_signal returns a continuous BTC-Donchian-distance value (clamped ±50) that varies every bar — verified through the actual Layer-2 proxy with both extras fed: non-frozen (460 unique) and zero exceptions — while the entry decision is the pure Donchian test in should_enter. Long-only with 5% legs and a 7% hard stop keeps risk controlled; leverage is 1.0 so sizing references no margin leverage (no inert-leverage gate). Two free parameters is the lowest-complexity breakout possible, minimizing overfit.

Weak edge that has broken down in the most recent regime — not worth optimizing. The Donchian breakout on HL majors shows only a Sharpe 0.64 (sharpe_ci_low -1.09, CI straddles 0), PF 1.22, thin per-trade expectancy after fees+impact, with annual returns DECLINING every year (2024 +6.4% -> 2025 +1.6% -> 2026 -3.1%) and the rolling Sharpe deeply, persistently negative through all of 2026 (-3 to -12 for ~4 straight months). On only ~2.3 years of HL history concentrated in the 2024 launch bull, a 3-window walk-forward puts the recent OOS in the negative 2025-2026 regime, so it will fail walk-forward OOS, and a 0.64 base Sharpe selected over 225 trials will fail deflated-Sharpe — optimizing would curve-fit to the 2024 bull and waste the compute. CREDIT: the developer correctly solved the HL sample-size wall (4H + 3-symbol basket -> 210 trades, populated walk-forward, unlike the zero-trade daily HL versions), so this is NOT the structural can't-validate failure; it is a genuine weak-and-decayed-edge result. The hypothesis's own experimental crux is thereby answered NEGATIVELY: the momentum-continuation premium does not survive on Hyperliquid any better than on Binance — the breakout worked in HL's 2024 launch and decayed to negative by 2026. FAILURE PATTERN: long-only Donchian/trend breakout on Hyperliquid majors (single-asset daily OR multi-symbol 4H) produces an edge concentrated in HL's 2024 launch bull that decays to negative by 2026; the multi-symbol 4H basket fixes the trade-count/walk-forward-population problem but cannot fix the underlying decay — Sharpe ~0.6 with a 0-straddling CI and a 4-month deep-negative recent rolling Sharpe. The cross-venue momentum premium is not durably present on HL's short, regime-concentrated history.

Long-only 20/10 Donchian channel breakout across a 3-symbol Hyperliquid 4H perp basket (BTCUSD/ETHUSD/SOLUSD .HYPERLIQUID, pure OHLCV). Per symbol: enter long when the 4H close exceeds the prior 20-bar high, the symbol is flat, and fewer than 2 portfolio legs are open; size 5% of equity per leg (≤15% gross). Exit on a trailing stop at the prior 10-bar low, a hard -7% stop from entry, or a 24-bar time stop if PnL < +2%. No volume/trend/oscillator filters — two free parameters. BTC trades via the standard interface; ETH and SOL are orchestrated on the primary 4H clock with the base's alignment barrier ensuring contemporaneous bars. Leverage 1.0.

Developer solved the HL frequency problem: 4H + 3-symbol basket gives 210 trades (vs 15-35 for daily HL), so walk-forward windows are populated (not the zero-trade wall that killed daily HL versions)

Genuinely novel venue (experimental/diversification value); clean multi-symbol code, controlled exposure (max_dd 5.6%), positive overall (PF 1.22, alpha +0.0085)

Weak, uncertain edge: Sharpe 0.64, sharpe_ci_low -1.09 (CI straddles 0), thin per-trade expectancy after HL fees + 6.5% impact

Declining and recently broken: annual 2024 +6.4% -> 2025 +1.6% -> 2026 -3.1%, with rolling Sharpe deeply negative through all of 2026 (-3 to -12 for ~4 months)

Short, regime-concentrated history (~2.3 years, edge concentrated in HL's 2024 launch bull) -> recent walk-forward OOS windows fall in the negative 2025-2026 regime

A 0.64 base Sharpe selected over 225 trials will almost certainly fail deflated-Sharpe; same HL-trend-decay signature as 4 prior HL daily abandons

~2.3 years (HL)

multi-regime

~0.5+ but CI must exclude 0

marginal

2026 rolling Sharpe -8 to -12 (4 months)

stable/positive OOS

HyperliquidMultiSymbolDonchianBreakoutLong4H was a deliberately designed cross-venue experiment — the first Hyperliquid strategy in the book — testing whether the breakout/momentum premium that decayed on Binance would still be present on HL's distinct, younger ecosystem, using the simplest possible 2-parameter Donchian on a 3-symbol 4H basket. It earned real credit for engineering: the 4H basket generated 210 trades and populated the walk-forward windows that zero-trade daily HL versions could never fill. But the edge was weak (Sharpe 0.64, CI straddling 0, PF 1.22) and decaying — positive in HL's 2024 launch bull, flat in 2025, negative in 2026 with a four-month deeply negative rolling Sharpe. The reviewer abandoned it pre-optimization, noting that on ~2.3 years of regime-concentrated history it would fail both walk-forward OOS and deflated-Sharpe, and that the hypothesis's own experimental crux was thereby answered: the momentum-continuation premium is not durably present on Hyperliquid, decaying just as it did on Binance.

Long-only Donchian/trend breakout on Hyperliquid majors produces an edge concentrated in HL's 2024 launch bull that decays to negative by 2026 — a multi-symbol 4H basket fixes the trade-count/walk-forward-population problem that killed daily HL versions but cannot fix the underlying decay — so the cross-venue experiment answers negatively: the momentum-continuation premium does not survive on Hyperliquid any better than on Binance.

The analyst abandoned it at the backtest-review gate as a weak edge that has broken down in the most recent regime: on only ~2.3 years concentrated in HL's 2024 launch bull, a walk-forward puts recent OOS in the negative 2025-2026 regime (would fail) and a 0.64 base Sharpe over 225 trials would fail deflated-Sharpe — so optimization would merely curve-fit the 2024 bull; the developer did, however, correctly solve the HL sample-size wall (210 trades, populated windows).

A long-only, OHLCV-only multi-instrument 20/10 Donchian breakout portfolio on three Hyperliquid 4H perps (BTC/ETH/SOL) — enter long on a close above the prior 20-bar high (max 2 concurrent legs, 5%/leg, ≤15% gross), exit on a 10-bar-low trailing stop, -7% hard stop, or a 24-bar time stop — explicitly a cross-venue experiment to test whether the momentum-continuation premium that decayed on Binance survives on Hyperliquid's younger, retail-driven microstructure, with just 2 free parameters.

It showed a weak, decaying edge over ~2.3 years of HL history: +5.4% total return (CAGR 2.06%), Sharpe 0.64 with sharpe_ci_low -1.09 (CI straddles 0), profit factor 1.22, +$22.66/trade expectancy across 210 trades, a low 5.6% max drawdown and 6.5% impact cost — but with annual returns declining every year (2024 +6.4% → 2025 +1.6% → 2026 -3.1%) and the rolling Sharpe deeply negative through all of 2026 (-3 to -12 for ~4 months).