Hyperliquid Intrabar Recycling versus Persistent Repricing: Balanced Altcoin Basket

Hypothesis

Test whether large intrabar excursions without comparable close-to-close variance identify fragile speculative liquidity that subsequently underperforms assets with more persistent repricing. Use synchronized 4-HOUR bars across six Hyperliquid perpetuals. Buy the two lowest range-variance/close-variance scores and short the two highest for seven days. This is an original conjecture, not an established premium. Market orders execute at completed signal-bar closes, the registered equivalent of next-bar-open execution on continuous venues. Protective stops trigger intrabar through the platform. Evaluate actual available history; Hyperliquid's approximately 5000-candle limit bounds four-hour coverage. Base-tier taker fees are 0.045% each way according to [Hyperliquid fees](https://hyperliquid.gitbook.io/hyperliquid-docs/trading/fees).

The balanced book addresses the largest policy excess: long-only strategies constitute 76.9% versus a 55% ceiling. It also adds Hyperliquid exposure, currently 12.3% versus a 20% target, and reduces single-instrument concentration. The supplied recent concentration map is empty, so recent instrument shares cannot be asserted. Snapshot searches returned lottery-demand, failed-auction, semivariance and Amihud precedents but no matching cross-sectional range/close variance ratio design. This proposal changes the measured economic variable rather than substituting symbols into those failures. The supplied library excerpts do not establish this mechanism.

Negative observed expectancy after modeled costs: 12 closed legs across three basket episodes lose $160.47, with PF 0.956 and average net leg return -0.21761% against the +0.15% venue review reference. Funding is credited and accounting reconciles. Optimization is not warranted without an evidence-backed improvement path. This rejects advancing the measured baseline, not the intrabar-recycling conjecture universally; the sparse sample and untested parameter range do not establish premise falsification. No accounting, data or coding defect is asserted.

Implementation

Ranks six synchronized Hyperliquid perpetuals by Parkinson intrabar range variance divided by close-return sample variance. At Monday 00:00 UTC, buys the two lowest and shorts the two highest when their mean score gap qualifies, holding up to seven days with intrabar 8% protective stops. Market entries execute at completed signal-bar closes, the registered next-open equivalent on continuous venues.

Fixes the Layer-4 warmup defect by priming indicators once from the last lookback_bars+1 common buffered timestamps at or before the first decision. With 253 synchronized valid warmup bars, even the 252-return window is populated before trading resumes; timestamp deduplication prevents counting the current bar twice. Warmup still suppresses entries. All prior signal, schedule, sizing, exit and stop rules and the previous config remain unchanged. The -0.2075% smoke-test leg return across 12 positions is advisory and insufficient to establish no edge. Per-leg funding coverage and credits remain unconfirmed and must be checked by the full-evaluation pipeline; this implementation does not fabricate funding observations or alter the price-based mechanism.

Verification Results

The smoke test reports -0.20754% average net leg return across 12 closed positions, with $206.12 modeled impact. This does not establish a correctness defect or falsify the conjecture. Funding was not folded and no funding events were available; verify funding coverage and cash-flow accounting for all six perpetuals during full evaluation.

Backtest Review

Sharpe

-0.05

Total return

-0.16%

Max drawdown

1.83%

Trades

12

Win rate

41.7%

Profit factor

0.96

Trades match the weekly basket structure: two longs and two shorts, approximately 6.25% capital per leg, with seven-day or protective-stop exits.

Funding is credited, closed-trade PnL reconciles to the account loss, and no terminal inventory remains.

Net loss is $160.47 (-0.16048%), PF is 0.956, and average net leg return is -0.21761% versus the +0.15% Hyperliquid review reference. QA’s economic concern is borne out.

Only three basket episodes traded, all in August–September 2026; this does not establish that the broader conjecture is false.

Modeled impact of $206.13 overwhelms the baseline’s small pre-impact surplus. No demonstrated implementation defect or profitable parameter region supports specific iteration guidance.

Buy-and-hold comparisons are not meaningful for this balanced book.

Iteration History

Prime rolling indicators from the synchronized buffered bars before the first decision, processing lookback_bars+1 observations ending at the current timestamp exactly once, or arrange for synchronized indicator updates throughout warmup before the framework's trading-readiness gate. Keep entries suppressed during warmup. Verify that after 253 replayed bars all six histories contain the configured lookback (including 252) and the next eligible Monday can trade.

min_bars_required() returns 253, but indicators are updated only inside calculate_signal(). FactoryStrategy._maybe_process_primary() skips calculate_signal() entirely while len(_bars) < min_bars_required(). Consequently bar 253 only initializes previous_close, bar 254 supplies the first return, and the default 168-return window is ready only at bar 421 (bar 505 for lookback 252). Replaying the contract's 253 warmup bars leaves every rolling window empty. The strategy then suppresses weekly entries for an additional lookback window inside each evaluation segment, contrary to the frozen 253-bar warmup contract; those omitted weeks change the measured strategy and optimization comparison.

Outcome Summary

A future iteration needs evidence of an improvement that survives execution costs, while the sparse three-episode sample does not establish that the broader conjecture is false.

The backtest-review verdict was abandon because observed expectancy was negative after modeled costs and no evidence-backed improvement path justified optimization. Optimization, post-optimization analysis, and risk review were not reached; no accounting, data, or coding defect was asserted.

Rank six Hyperliquid perpetuals by intrabar range variance relative to close-to-close variance, buying the two lowest scores and shorting the two highest for seven days.

The backtest recorded 12 closed legs across three basket episodes, losing $160.47 (-0.16048%) with Sharpe -0.053 and profit factor 0.956. Average net leg return was -0.21761%, and modeled impact totaled $206.13.