Hyperliquid Weekly MAX Distribution Spread Across Eight Altcoin Perpetuals

Hypothesis

Test whether the largest positive daily return in the preceding month predicts subsequent relative performance across a fixed eight-asset Hyperliquid basket. Long the two highest-MAX assets and short the two lowest, holding for one week. Build UTC daily closes from synchronized 4-HOUR bars. This is a prospective adaptation, not a faithful reproduction, of [Ozdamar, Akdeniz and Sensoy's MAX study](https://link.springer.com/article/10.1186/s40854-021-00291-9), which reports positive weekly high-minus-low MAX returns. Evidence is horizon-dependent: [Yadav's intraday study](https://pure.jgu.edu.in/id/eprint/9270/) reports the opposite relationship at an hourly horizon. Neither establishes profitability for this fixed, liquid perpetual universe.

Long-only experiments exceed their target by 26.8 percentage points; this dollar-neutral proposal addresses that largest imbalance, reduces single-symbol concentration, and uses underrepresented Hyperliquid. Corpus concentration was supplied empty, so instrument-level underuse cannot be verified. MAX measures the distribution of daily returns rather than cumulative momentum, funding willingness, turnover exhaustion or downside fragility. No funding confirmation or shock-reversal trigger is used. Options remain underrepresented, but a rolling options universe and its execution requirements have not been verified for this brief.

negative_expectancy: decline optimization of this frozen eight-asset Hyperliquid weekly high-minus-low MAX adaptation. Across 93 executed baskets, measured net economics are negative, with average leg return -0.676% versus the +0.15% review floor and PF 0.758. This verdict concerns the specified universe, sizing and protective exits; it does not refute the cited studies or MAX strategies generally. Rejection rests on measured losses, not drawdown alone or a predicted optimization outcome.

Implementation

Rank eight synchronized Hyperliquid perpetuals by their largest positive daily return over 30 completed UTC days. Each Monday at 00:00 UTC, long the two highest-MAX assets and short the two lowest for one week, allocating 15% equity per leg with intrabar 4% adverse stops and a 1.5% basket-equity loss exit.

Iteration 4 replaces only the unavailable StopMarketOrder.is_triggered attribute with order.status.name != 'TRIGGERED'. Signal logic, synchronization, weekly calendar scheduling, sizing and fixed research parameters are unchanged. Verification is left to the pipeline. Required local references could not be read because shell initialization failed with bwrap permission denied creating .agents. Writing the required task-specific outbox JSON also failed; this response contains the complete bound result, but the outbox file was not saved.

Verification Results

The sandbox reports -41.89% net return and -0.669% average leg return. Full-history analysis should assess whether weekly MAX ranking generates sufficient relative returns after commissions and impact. These smoke-test losses do not establish a correctness defect.

Backtest Review

Sharpe

-1.14

Total return

-42.36%

Max drawdown

56.95%

Trades

372

Win rate

26.3%

Profit factor

0.76

Trade records show 93 Monday 00:00 UTC baskets, each with two longs and two shorts; initial leg notionals are $15,000, consistent with the specified allocation. Code selects high-MAX longs and low-MAX shorts from completed daily observations.

QA's economic concern is borne out: 372 closed legs lose $42,356 net, with PF 0.758, Sharpe -1.140 and average leg return -0.676%, below the +0.15% Hyperliquid review floor.

The frozen specification has no tunable parameters, and no verified implementation defect supports developer iteration. These results reject the executed portfolio, without establishing that unstopped weekly MAX forecasts universally lack predictive information.

The long-only basket benchmark is not meaningful for this long-short strategy. No positions remain open at the evaluation boundary.

The required outbox file could not be written because this session's filesystem is read-only.

Iteration History

Verification failed (Layer 1.5 — hypothesis/config consistency) [class=hypothesis_mismatch]: - optimization_plan fixes leverage at 1.0

Track the protective order for each leg and synchronize its remaining quantity and trigger with position changes, including further entry fills and partial exits. Alternatively, use supported contingent protection that scales with parent fills. Verify a multi-fill entry followed by a stop closes the entire leg.

Protective stops are created only in on_position_opened(), using the position quantity and average entry price at that event. Entries are standalone GTC market orders, which can fill partially. Further opening fills change the existing position and generate PositionChanged events; this strategy has no handler to increase the stop quantity or update its trigger. Consequently, only the first filled portion is protected by the stated 4% adverse-price stop, and the remaining exposure can survive that stop.

Verification failed (Layer 3 — sandbox backtest): 'nautilus_trader.model.orders.stop_market.StopMarketOrder' object has no attribute 'is_triggered' Bar type used: LTCUSD.HYPERLIQUID-4-HOUR-LAST-EXTERNAL, Bars processed: 0 Ensure your strategy produces trades with the given data and parameters.

Outcome Summary

This fixed universe, sizing and protective-exit implementation failed economically, without disproving MAX strategies generally or the cited studies.

Backtest review issued an abandon verdict for negative expectancy; the frozen specification had no tunable parameters and no verified implementation defect supporting another iteration. Optimization, post-optimization analysis and risk review were not reached.

Across eight Hyperliquid altcoin perpetuals, the strategy bought the two assets with the highest preceding-month maximum positive daily return and shorted the two lowest for weekly holds.

Across 93 baskets and 372 closed legs, it returned -42.36%, with Sharpe -1.14, profit factor 0.758 and maximum drawdown 56.95%. Average leg return was -0.676%, below the +0.15% review floor.