Strategies

Hyperliquid Scheduled Residual-Risk Allocation: Seven-Day Long–Short Cohorts

Outcome: Abandoned

HyperliquidScheduledResidualRisk

Outcome Summary

HyperliquidScheduledResidualRisk tested scheduled long–short residual-volatility allocations across a six-name candidate panel and underwent four iterations. The supplied backtest returned -3.68% with a -0.41 Sharpe across 264 closed positions, while the development-period review separately reported losses across 51 cohorts with both longs and shorts losing money. The strategy was abandoned at backtest review because negative expectancy lacked an evidence-supported repair, so optimization and later stages were not reached. The rejection applied to this tested operationalization, with no conclusion about the source author's unspecified strategy or the withheld holdout.

Strategy report

Hypothesis

Derived by the Research Lead from the prior hypothesis’s archived BigQuant summary of Huatai Securities, Smartbeta advantages in asset allocation: test a scheduled low-idiosyncratic-risk versus high-idiosyncratic-risk allocation rather than a rare dispersion event. The archived summary and linked PDF were not re-read in this revision; source provenance comes from the supplied original hypothesis. The source does not specify these trading rules. Use 4-HOUR bars on the original six-name candidate panel. Every fourteen UTC days, hold a dollar-neutral residual-volatility-ranked book for seven days, then remain flat until the next scheduled entry. The factor is constructed from the eligible traded panel, requiring no unlisted ETH signal leg. Market orders execute at the signal bar close, the registry equivalent of next-bar-open execution on continuous markets. Protective stops are platform-matched intrabar stops. No atomic basket execution or exchange-hosted protection is required.

The supplied evidence establishes a no-entry failure, not negative expectancy. Remove the absolute dispersion trigger and hard beta eligibility bounds rather than optimizing around an unobserved event. Replace the external ETH factor with an equal-weight panel factor so all required inputs are explicitly subscribed. Replace exact beta matching with dollar-neutral sides and conservative volatility-scaled exposure; this knowingly retains factor risk. Keep the original names and holding horizon to limit unrelated changes. Prior-art review found failed residual-return reversal and momentum books; this ranks residual variance, not past return. A fixed schedule is deliberately a new allocation experiment, despite advisory skepticism about continuously rebalanced Hyperliquid panels: it trades at most one cohort per fortnight and never rebalances a live cohort. Corpus counts were inspected; current portfolio percentages and timeframe concentration were not supplied, so no precise quota claim is made. Retain Hyperliquid and avoid the heavily tested Binance BTC series. All revised rules must be registered as another lineage look; the previously exposed history is not a fresh holdout.

Measured negative expectancy across 51 scheduled cohorts, with losses in both directional arms. No specific implementation defect or evidence-supported repair justifies optimization. This rejects the tested Research Lead operationalization, not the source author's unspecified strategy. No conclusion is drawn about the withheld holdout.

Implementation

Fortnightly UTC residual-risk allocation on the six-name Hyperliquid panel: buy the two lowest and short the two highest residual-volatility names, hold for 42 four-hour bars after fills, with 8% platform-matched intrabar stops. Market orders fill at the signal close, the registered continuous-market equivalent of next-bar-open execution. A stopped or failed leg causes cohort liquidation; no replacements or live reranking.

Implements the revised allocation experiment from scratch: an internal equal-weight panel factor, intercept OLS, deterministic residual-risk ranks, side-normalized inverse-risk weights and uniform downward scaling for every specified cap. Removes the former dispersion and beta filters and external ETH leg. Eligibility uses only point-in-time listing age and exact synchronized closes; unknown listing age is excluded rather than inferred. Only factor_lookback_bars is intended for optimization under the supplied research plan. The primary is a clock, never a forced holding. Per-leg holding counters and cohort state are scalar attributes; price histories use bounded framework buffers. Slippage is left to the platform.

Backtest Review

Sharpe: -0.41
Total return: -3.68%
Max drawdown: 9.24%
Trades: 264
Win rate: 45.1%
Profit factor: 0.91

All 51 cohorts entered on the fortnightly schedule with two longs and two shorts; position notionals varied.

Across 204 closed legs, average net trade return was -0.457%, profit factor 0.823 and portfolio return -5.475%.

Both arms lost: longs -$3,344.28 and shorts -$2,130.23. Aggregate price PnL was negative before funding (-$5,143.84).

Only 19 of 51 cohorts were profitable. The predicted positive residual-risk premium is unsupported in the visible development period.

Analysis

The Research Lead REVISED this hypothesis because the previous premise did not work. Implement the REVISED hypothesis (see the hypothesis description) from scratch — do not reuse the prior strategy. Revision rationale: The supplied evidence establishes a no-entry failure, not negative expectancy. Remove the absolute dispersion trigger and hard beta eligibility bounds rather than optimizing around an unobserved event. Replace the external ETH factor with an equal-weight panel factor so all required inputs are explicitly subscribed. Replace exact beta matching with dollar-neutral sides and conservative volatility-scaled exposure; this knowingly retains factor risk. Keep the original names and holding horizon to limit unrelated changes. Prior-art review found failed residual-return reversal and momentum books; this ranks residual variance, not past return. A fixed schedule is deliberately a new allocation experiment, despite advisory skepticism about continuously rebalanced Hyperliquid panels: it trades at most one cohort per fortnight and never rebalances a live cohort. Corpus counts were inspected; current portfolio percentages and timeframe concentration were not supplied, so no precise quota claim is made. Retain Hyperliquid and avoid the heavily tested Binance BTC series. All revised rules must be registered as another lineage look; the previously exposed history is not a fresh holdout.

Iteration History

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: COMPUSD.HYPERLIQUID-4-HOUR-LAST-EXTERNAL, Bars processed: 5497 Diagnostics: should_enter() returned a side 0 times over 5497 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: COMPUSD.HYPERLIQUID-4-HOUR-LAST-EXTERNAL, Bars processed: 5498 Diagnostics: should_enter() returned a side 0 times over 5498 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

hypothesis_revised: Verification exhausted 3 coding attempts without ever producing a trade — the entry condition never triggers on this instrument, so the HYPOTHESIS PREMISE (not the code) is the problem. Reframe it (di

Outcome Summary

Scheduled execution alone did not establish a positive residual-risk premium in the tested operationalization, so further work needs evidence for the edge before parameter optimization.

After four iterations, the backtest-review verdict was abandon because both directional arms lost money and measured expectancy was negative, with no identified implementation defect or evidence-supported repair to justify optimization. Optimization and later reviews were not reached; no conclusion was drawn about the withheld holdout.

Test a dollar-neutral Hyperliquid allocation long low-residual-volatility names and short high-residual-volatility names, entering every fourteen days for a planned seven-day hold.

The supplied backtest reported a -3.68% return, -0.41 Sharpe, 264 closed positions and 9.24% maximum drawdown. Separately, the backtest review cited 51 scheduled cohorts and 204 closed legs with a -5.475% portfolio return, -0.457% average net trade return and 0.823 profit factor.

Backtest and paper results are hypothetical. Trading involves risk of loss.