HyperliquidEthBtcRatioMomentumPairsLS

Hyperliquid ETH/BTC Intraday Ratio-Momentum Pairs, Market-Neutral Long-Short (Two Clean-Sizing HL Instruments, 1H Ratio EMA-Crossover with Hysteresis Band, Beta-Canceling Relative-Value Trend, Low-Parameter, Fills HYPERLIQUID + Long-Short + Relative-Value Quota Gaps Simultaneously)

A MARKET-NEUTRAL, LONG-SHORT, TWO-INSTRUMENT relative-value momentum strategy that trades the ETH/BTC ratio on HYPERLIQUID — the ONLY venue/account path the factory has confirmed sizes cleanly (HL BTC 90-97%, HL ETH 64.5%), deliberately AVOIDING every confirmed-broken path: BINANCE USD-M over-exposure (106-1062%), BINANCE COIN-M (399% + phantom returns), HL SOL (early-2024 data poisoning), and the HL daily-cadence data-sparsity wall. Both legs (ETHUSD.HYPERLIQUID, BTCUSD.HYPERLIQUID) are individually clean-sizing, and the position is dollar-neutral (equal-notional long one / short the other), so it is structurally immune to single-instrument over-deployment and cancels overall crypto market beta. The signal is the relative-strength trend of the ETH/BTC price ratio captured by a fast-vs-slow EMA crossover on 1H bars with a hysteresis band — simple and low-parameter (fast EMA, slow EMA, band width, stop), directly addressing the recurring overfitting failures. Intraday (1H) cadence accumulates many more independent trades over the ~3-year HL window than a daily strategy, escaping the sparsity wall that abandoned the HL daily Donchian runs, while a hysteresis band keeps flips infrequent (multi-day holds) so each round-trip clears the ~0.36% two-leg HL fee. This is distinct from the existing BINANCE BTCETHBollingerTrendLS (different venue, ratio-momentum not Bollinger, intraday not daily) and from all the now-saturated cross-venue funding carries.

Implements the hypothesis precisely on the only HL path confirmed to size cleanly: primary ETHUSD.HYPERLIQUID + extra BTCUSD.HYPERLIQUID, both on 1-HOUR bars. Because both legs share the same timeframe, the base class's cross-leg alignment barrier defers the signal until BTC has a contemporaneous bar, so the ratio is always computed on same-timestamp closes (no stale-leg phantom move). The edge is the relative-strength trend of the ETH/BTC ratio via a fast/slow EMA crossover (fast_period=24, slow_period=120 hours) with a hysteresis band (0.004) that holds the prior direction inside the band, producing infrequent multi-day flips so each two-leg round-trip clears the ~0.36% HL fee — exactly the sparsity/fee reasoning in the brief, and escaping the HL daily-cadence wall by running intraday. Dollar-neutral equal-notional legs (leg_fraction 0.45, gross ~90% of equity, net ~0) cancel market beta and structurally prevent single-instrument over-deployment, deliberately avoiding the broken BINANCE USD-M / COIN-M / HL-SOL paths. Only four edge parameters (fast, slow, band, stop) keep it low-parameter to resist overfitting. leverage=1.0 (90% gross needs no margin headroom) and is referenced in sizing so there is no leverage_set_but_unused mismatch. calculate_signal returns the continuous EMA spread (with a tanh price-proxy fallback when BTC data is absent, e.g. the Layer-2 synthetic proxy) so it varies bar-to-bar and never trades on incomplete data, while every cache/order call is exception-guarded. Fills the HYPERLIQUID, long-short and relative-value quota gaps at once, and is distinct from the BINANCE daily Bollinger pair.

Abandon on two independent disqualifiers, either sufficient. (1) DATA WALL: the hypothesis's load-bearing premise — that 1H accumulates many trades 'over the ~3-year HL window' — is FALSE; 1H HL data spans only ~7 months (2025-10-11 to 2026-05-20, 5195 bars, data_days 212), so the strategy generated just 37 pairs over a single 7-month regime and cannot be validated with a 3-window walk-forward + 15-day holdout. This is the THIRD confirmation this session of the unbackfilled 1H HL catalog (after the BTC-1H and ETH-1H probes); the standing ESCALATION to the data engineer (backfill BTCUSD/ETHUSD.HYPERLIQUID 1H to ~2023, which the Hyperliquid API supports) covers this too. (2) ANTI-EDGE: independent of the data wall, the strategy is a clear loser — total_return -18.4%, Sharpe -2.91, profit_factor 0.70, avg_trade_return -$282, avg_win<avg_loss, tail_ratio 0.011, losing 7 of 8 months. Ratio-MOMENTUM (trend-following the ETH/BTC EMA crossover) is the wrong sign: the ETH/BTC ratio is mean-reverting, not trending, at intraday cadence, so EMA-crossover trend-following bleeds. Not worth 2 hours: there is no validatable data and no edge to tune toward. Not iterate: no code-level bug (sizing clean, neutral, custom routing works — entry_signaled:0 is an artifact, 74 real trades executed); no parameter change adds ~3 years of history or flips an anti-edge. PRESERVED FINDINGS for the research lead: (a) the ETH/BTC ratio MEAN-REVERSION direction (sibling 417c8d41, deep 4H data, PF 0.91, near-breakeven) is the supported direction and is already in optimization — the MOMENTUM/trend direction on the same ratio is a confirmed anti-edge, so do not re-queue ratio-trend books; (b) 1H HL remains data-walled at ~7 months — keep all HL relative-value/pairs work on 4H (deep, ~2.3yr) or daily until the 1H backfill lands; (c) the market-neutral ETH/BTC pair construction itself is sound and clean-sizing on HL, so the venue/leg choice is fine — only the signal direction and the 1H cadence are the problems.

Market-neutral, long-short relative-value pairs strategy trading the ETH/BTC ratio on Hyperliquid with 1H bars. Each aligned hourly bar it computes ratio = ETH_close / BTC_close and a fast-vs-slow EMA crossover on that ratio; a hysteresis band turns the crossover into a sticky direction (long ETH/short BTC when the ratio trends up, short ETH/long BTC when it trends down, hold inside the band). Both legs carry equal notional (45% of equity each) so the book is dollar-neutral and crypto-beta-canceling. A per-trade ratio stop flattens both legs on a stop_pct adverse move and blocks re-entry in that direction until the EMA spread flips. Order management is centralized in calculate_signal via netting deltas with a rebalance_band that suppresses micro-churn; the single-instrument hooks are inert. Pure OHLCV, no leverage.

Genuinely market-neutral construction (beta 0.001, benchmark_correlation 0.004, benchmark_meaningful=false) on the clean-sizing HL ETH+BTC legs; sizing is clean (avg_position_pct 46.8% ≈ 45%/leg, ~90% gross, metrics_reliable=true). entry_signaled:0 is a harmless custom-routing artifact (74 real trades = 37 balanced pairs).

Low-parameter and dodges the broken Binance/COIN-M/HL-SOL paths.

DATA WALL: 1H HL data spans only ~7 months (2025-10-11 to 2026-05-20, 5195 bars), NOT the ~3 years the hypothesis assumed — the load-bearing premise is false. 37 pairs over a single 7-month regime cannot power a 3-window walk-forward + holdout. This is the THIRD confirmation of the unbackfilled 1H HL wall (after BTC-1H and ETH-1H).

Clear loser: total_return -18.4%, Sharpe -2.91, Sortino -3.11, profit_factor 0.70, avg_trade_return -$282 (negative), losing 7 of 8 months. avg_win $1492 < avg_loss $1710, tail_ratio 0.011, omega 0.32.

Wrong/weak sign: ratio-MOMENTUM (EMA-crossover trend-follow of ETH/BTC) bleeds, consistent with the ratio being mean-reverting rather than trending at intraday cadence — the mean-reversion sibling (4H, deep data) is the better-supported direction.

HyperliquidEthBtcRatioMomentumPairsLS aimed to fill the Hyperliquid, long-short, and relative-value gaps at once with a beta-canceling ETH/BTC ratio-momentum pairs book on two clean-sizing HL legs, betting that 1H cadence would generate enough trades to escape the daily-sparsity wall. Both pillars failed: 1H HL data turned out to span only ~7 months (the third confirmation of the unbackfilled 1H catalog), and the signal was the wrong sign — trend-following the ratio lost 18.4% with a -2.91 Sharpe because the ETH/BTC ratio mean-reverts intraday rather than trending. The reviewer abandoned it pre-optimization, since no parameter change adds years of history or flips an anti-edge. The construction itself — market-neutral, clean dollar-neutral sizing — was judged sound, leaving three preserved findings: the ratio mean-reversion direction (its 4H sibling) is the supported edge already in optimization, 1H HL remains data-walled, and HL relative-value work should stay on 4H or daily until backfill.

On the ETH/BTC ratio the mean-reversion direction (sibling on deep 4H data, near-breakeven, already in optimization) is the supported edge while ratio-momentum/trend is a confirmed anti-edge — so don't re-queue ratio-trend books; keep HL relative-value work on 4H or daily until the 1H backfill lands, since the market-neutral pair construction itself is sound and clean-sizing.

The analyst abandoned it at the backtest-review gate, before optimization, on two independent disqualifiers: the load-bearing data premise was false (1H HL data spans only ~7 months / 5,195 bars, not ~3 years — the third confirmation of the unbackfilled 1H HL wall), and the signal is the wrong sign (ratio-momentum trend-following bleeds because the ETH/BTC ratio mean-reverts rather than trends at intraday cadence).

A market-neutral, dollar-neutral long-short pairs strategy trading the ETH/BTC ratio on two clean-sizing Hyperliquid legs (ETHUSD/BTCUSD, 1H bars) via a fast-vs-slow EMA crossover on the ratio with a hysteresis band — a low-parameter relative-value trend book intended to cancel market beta and use intraday cadence to escape the HL daily-sparsity wall.

It was a clear loser over the available ~7-month window: -18.4% total return, Sharpe -2.91, Sortino -3.11, profit factor 0.70, avg_trade_return -$282, avg_win $1,492 < avg_loss $1,710, losing 7 of 8 months across 74 trades (37 balanced pairs) — though the construction was genuinely market-neutral (beta 0.001) with clean sizing (avg_position_pct ~46.8%/leg, ~90% gross, metrics_reliable=true).