Hyperliquid Funding-Willingness Confirmed Relative Momentum Basket

Hypothesis

Test whether agreement between relative price strength and sustained willingness to pay funding predicts multi-day continuation across LTC, BCH, ETC, UNI, AAVE and FIL Hyperliquid perpetuals. Use synchronized 4-HOUR bars and instrument-specific settled funding. Local catalog directories were confirmed for all six instruments. This is an original adaptation inspired by FMZ's funding-plus-momentum framework, not a source reproduction: https://www.fmz.com/strategy/548617. Use fixed factor weights, a fixed universe and seven-day holding periods; exclude its rolling regression and dynamic weight fitting. Estimate performance only over the intersection of actual bar and funding coverage.

Long-only experiments are the largest policy imbalance at 81.8% versus a 55% ceiling. This long/short, multi-instrument proposal also addresses Hyperliquid's 9.2% share versus its 20% target. Corpus concentration is empty, so individual instrument shares cannot be established; no BTC or ETH is used. Unlike the failed turnover-exhaustion and downside-variance baskets, the signal tests continuation confirmed by paid positioning demand rather than reversal. Unlike existing risk-adjusted momentum rotation, settled funding contributes an independent, pre-specified ranking input. The supplied success labels include negative Sharpes and are not treated as evidence of profitability.

negative_expectancy: This six-name, fixed-weight funding-confirmed momentum implementation loses money across 89 completed baskets. Average net trade return is -0.613%, below the +0.15% Hyperliquid review floor; PF is 0.915 and total return is -8.71%. Credited funding totals -$1,104.75, while the reported price-PnL component totals -$7,602.97. There is no measured positive parameter region or specific implementation defect supporting another developer iteration. Do not spend the optimization budget on this candidate. This verdict covers the tested configuration and universe; it does not claim every funding_days/momentum_bars variation or the broader mechanism is disproven.

Implementation

Long/short Hyperliquid basket over LTC/BCH/ETC/UNI/AAVE/FIL 4H perps: cross-sectional z of vol-normalised momentum (0.75) plus mean settled HL funding over funding_days (0.25); long top 2, short bottom 2, held 42 bars (7 days) with entry-anchored ATR stops per leg, a basket loss limit, and a 6-bar cooldown. Trades only where bar and funding coverage intersect.

Iteration 2 fixes Layer 1.5 only: the plan's fixed leverage (float 1.0) is now in parameters with the exact type and value, and the code reads it (mirroring config.leverage, a no-op at 1.0). All other fixed plan values are present with their exact types. Signal, sizing and schedule logic are unchanged from iteration 1.

Verification Results

Optionally place the stop from on_position_opened / on_order_filled so it rests as soon as the fill confirms.

Per-leg venue stops are placed only after the position exists (in _form_basket after in-event fills, otherwise on the next bar via _ensure_stops). In backtest the market entries fill in-event, so stops rest immediately. In live trading, fill latency can leave a leg without a stop until the next 4H bar.

No code change needed. Track how often baskets end on loss_hit versus hold expiry in the full backtest.

The basket loss limit (-2% of formation equity) is checked only at 4H bar closes on close prices, so intra-bar losses can go past it. The per-leg resting STOP_MARKETs bound this. The limit is also fairly tight against 60% gross (a 3.3% adverse basket move), so it may often end baskets well before the 7-day hold.

None. This is noted only so the static-analysis base_shadow warning is understood.

min_bars_required() returns 2, overriding the base warm-up. This is deliberate: the synchronized _sync buffer enforces volatility_bars+1 = 181 aligned snapshots before any factor is computed. That matches the plan's warmup_bars of 181.

In the 787-day sandbox (unoptimized): 356 closed legs (about 89 four-leg baskets), avg_trade_return_pct -0.61%, PF 0.915, Sharpe -0.50 (CI -1.81 to 0.80), max DD 16.3% against a pre-registered 15% constraint. Most of the loss is in the normal-vol tercile. Six mid-cap alts are a thin cross-section, so a top-2/bottom-2 split is dominated by idiosyncratic jumps (largest loss -$2969). The 0.25-weight funding z-score barely moves the ranking. The full backtest should check: (1) IC of the funding factor on its own; (2) how often the 2%-of-equity basket loss limit cuts baskets before 7 days, which would make the realized holding period much shorter than the hypothesis intends; (3) whether drawdown stays under 15% across the funding_days/momentum_bars range.

Backtest Review

Sharpe

-0.50

Total return

-8.71%

Max drawdown

16.34%

Trades

356

Win rate

40.7%

Profit factor

0.92

Trade records confirm 89 baskets, each with two long and two short legs. Average holding is 5.02 days, consistent with seven-day expiry plus earlier risk exits.

Accounting reconciles, funding is included, and no positions remain open.

Measured net expectancy is negative: average trade return -0.613% versus the 0.15% Hyperliquid review floor, profit factor 0.915, and total return -8.71%. QA's economic concern was borne out.

The 356 legs represent 89 basket decisions, not 356 independent observations. Sharpe -0.496 has a confidence interval spanning zero; this does not establish universal absence of the mechanism.

167 legs exited before seven days; trade records alone do not distinguish basket loss-limit exits from individual stops.

The supplied forecast diagnostic has IC -0.097 with t=-0.98; it does not independently measure the funding factor's contribution.

Drawdown reached 16.34% against the registered 15% risk constraint. Benchmark-relative metrics are inapplicable because benchmark_meaningful is false.

Iteration History

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

Outcome Summary

Any future iteration should establish evidence of positive net expectancy and isolate funding's contribution before spending optimization budget.

Backtest review abandoned the candidate for negative expectancy below the +0.15% review floor and drawdown above the registered 15% constraint, with no measured positive parameter region or specific implementation defect supporting another iteration. Optimization and subsequent review stages were not reached.

Combine relative momentum and settled funding across six Hyperliquid perpetuals to select a fixed-weight long/short basket with seven-day holding periods.

The backtest returned -8.71% with Sharpe -0.496 across 356 closed positions representing 89 baskets. Average net trade return was -0.613%, profit factor was 0.915, and maximum drawdown reached 16.34%.