Hyperliquid Dispersion-Release Episodes: Dollar-Neutral Relative Repricing

Hypothesis

Original research hypothesis: after unusually compressed cross-sectional returns, a sharp expansion in dispersion may mark token-specific information diffusing over several days. Trade a dollar-neutral relative-performance basket on synchronized 4-HOUR bars, not the common market direction. Baselines: formation_bars=6, history_bars=180, compression_quantile=0.30, expansion_multiple=1.8, holding_bars=18. Market orders execute at the completed signal-bar close, the registry equivalent of next-bar-open execution on a continuous venue. Intrabar protective stops are platform-matched. No queue advantage, exchange-hosted protection, selectable same-bar policy or inverse contracts are required. Use all available actual history; do not assume that Hyperliquid supplies more than approximately 5000 candles per interval. This is an unvalidated mechanism, with no preliminary performance study.

The supplied concentration map is empty, so instrument and timeframe shares cannot be verified from it. Snapshot prior-art searches found continuous relative-strength baskets, pair stress-dispersion trades and common-factor correlation-contagion continuation. This proposal instead detects a transition from low to high cross-sectional dispersion, remains dollar-neutral, and trades only discrete episodes. Its signal is relative divergence rather than synchronization or an isolated price shock. Hyperliquid is under target at 13.7% versus 20%, while long-only is above target at 75.3%; this multi-instrument long-short proposal addresses both. Options have a larger venue deficit, but no options dependency is added merely to satisfy a quota. Cross-venue trading is also under target, but an unnecessary second venue would change this mechanism and add overlap constraints.

Measured negative portfolio expectancy across 133 complete basket episodes in the visible development period. No specific implementation defect or evidence-supported repair justifies optimization. This rejects the tested dispersion-release continuation operationalization and draws no conclusion about the withheld holdout.

Implementation

Trades discrete cross-sectional dispersion releases in the fixed six-name Hyperliquid universe. Buys two relative winners and shorts two losers at equal initial dollar notionals, with fixed entry ATR protective stops, a synchronized-bar holding limit and six-bar cooldown. Signal-close market execution is the registry equivalent of next-open execution on this continuous venue.

Uses the preceding dispersion and a quantile strictly preceding that observation, then freezes four-leg membership and quantities. Shared notional obeys every specified equity, stop-risk, turnover and annualized volatility ceiling. All six named contracts must be listed with uninterrupted histories; no replacements or forward fills. Stops are platform-matched intrabar from actual fills. Research domains and fixed values are preserved; slippage is charged by the platform. Uses bounded actual-history buffers and no requested history extension. The unconditional fixed-clock comparator and basket-level statistical reporting belong to the evaluation pipeline, not an additional trading rule.

Backtest Review

Sharpe

-1.04

Total return

-0.99%

Max drawdown

1.16%

Trades

660

Win rate

46.7%

Profit factor

0.83

All 133 entry episodes contain four legs: two longs and two shorts. Sampled entry notionals are equal across legs.

Across 532 closed legs, mean net return is -0.322%, profit factor is 0.841 and Sharpe is -0.992.

Closed-trade PnL totals -$825.98, consistent with the -0.826% account return. Only 44 of 133 episodes are profitable.

Long legs earn $486.13 while short legs lose $1,312.12. Funding contributes only -$4.72.

Approximately 64 episodes per year exceeds the predicted 15–25; this forecast miss is not an eligibility floor.

Outcome Summary

Dollar neutrality did not establish predictive edge: reviewed short-leg losses outweighed long-leg gains.

The backtest reviewer voted to abandon because expectancy was negative and no specific implementation defect or evidence-supported repair justified optimization. Optimization and risk review were not reached; the withheld holdout remained unevaluated.

Trade a dollar-neutral basket of two longs and two shorts on Hyperliquid four-hour bars to capture relative-price continuation after cross-sectional return dispersion expands.

The backtest reported a -0.99219% total return, -1.037 Sharpe and 660 closed positions, averaging -0.333% net return per trade. Separately, the development-period review found negative expectancy across 133 complete basket episodes, with a 0.841 profit factor.