SolHyperliquidFundingCarryLongSpotShortPerpNeutral

SOL Hyperliquid-Funding Harvest: Long Binance Spot + Short Hyperliquid Perp, Delta-Neutral Carry (Cross-Venue, Long-Short)

A delta-neutral, cross-venue carry strategy that harvests the structurally ELEVATED perpetual funding on Hyperliquid by holding a hedged spot-vs-perp book on SOL. Long leg = SOLUSDT.BINANCE_SPOT (CASH, reliable deep history); short leg = SOLUSD.HYPERLIQUID perp. When the Hyperliquid SOL perp funding is materially positive, the perp short RECEIVES funding each 8h cycle while the matched spot long neutralizes SOL price exposure. The edge is venue-specific, not generic: Hyperliquid's funding mechanism (capped open interest, DeFi-native long-biased flow, thinner arb capital than Binance) drives its perp funding to spike higher and stay elevated longer than Binance's on alt majors like SOL — so harvesting HL funding against cheap Binance spot is a structurally richer carry than a same-venue Binance basis trade. This is NOT a price-pattern strategy (the class abandoned repeatedly this session — sweeps, dip-buys, vol-expansion, cross-sectional reversal): it is a mechanical cash-flow harvest, the one mechanism family this factory has advanced through the full pipeline. It is also distinct from the two pending funding hypotheses: vs EthCrossVenueFundingBasisCarryNeutralLS (different asset SOL, and explicitly an HL-funding-harvest thesis rather than generic basis), and vs BtcInterExchangeFundingDiffCarryNeutralLS (spot-vs-perp single-asset structure, NOT perp-vs-perp differential). Fills four under-represented quota buckets at once (HYPERLIQUID 5.8%, cross-venue 6.8%, BINANCE_SPOT 12.1%, market-neutral vs 85.8% long-only). Deliberately 3 parameters to avoid the walk-forward overfitting that abandoned the RSI/Three-Soldiers strategies.

Iteration-3 fix for the Layer-3 'no trades' failure. Diagnosis: the signal read only the top-level `hl_funding_rates` supplementary key, but because the PRIMARY instrument is Binance spot, the backtest supp loader (`load_for_subprocess` → `_load_supp_with_per_leg_funding`) never populates that key — it only fills `hl_funding_rates` when the primary leg itself is on Hyperliquid. Instead, the HL perp leg's funding series is routed into `supplementary_data['funding_rates_by_instrument']['SOLUSD.HYPERLIQUID']` for per-leg accrual. So the signal was always 0.0 and `signal >= funding_entry` never fired → zero entries over 5276 evaluated bars. The minimal change adds `_extract_series()`, which reads the per-leg series keyed by `str(self._hl_iid)` first (verified to round-trip to the config string 'SOLUSD.HYPERLIQUID') and falls back to the flat `hl_funding_rates` dict (the synthetic-test / HL-primary path). No change to the smoothing, thresholds, entry/exit polarity, two-leg equal-notional order logic, or sizing — all of which already passed Layers 1–2 — so no earlier layer regresses. leverage stays 1.0 (spot venue cap) and `_leverage` is still referenced in `position_size`, so no leverage_set_but_unused concern.

Delta-neutral SOL cross-venue funding carry fails structurally at the base backtest, with funding correctly read (108 positive-funding entries) and credited by the engine. The book still bleeds every single day (Sharpe -4.4, profit_factor 0.765, omega 0.024, expectancy -$106/trade, zero positive days) because the cross-venue round-trip fee (~0.29%: Binance spot 0.20% + HL 0.09%) dwarfs SOL's tiny Hyperliquid funding carry (~0.01%/day), and funding_exit=0 with no hysteresis churns the position (avg hold 1d 7h, 108 round-trips, commissions 14% of gross + $7,280 impact). Three-parameter optimization cannot bridge a >20x gap between per-cycle cost and per-cycle carry. Two further blockers: (1) the tradeable HL window is only ~206 days (the HL perp leg has bars only from ~Oct 2025) — too thin to validate a carry in a walk-forward; (2) the 'delta-neutral' hedge is not clean — exposure 111%, a -3.70% single day, largest loss -$6,228, skew -12.3, kurtosis 166 — indicating the two legs are not reliably offsetting. This is not a code bug to iterate on (funding works) and not worth 2 hours of optimization. If the funding-harvest mechanism is to be pursued, it needs either a same-venue construction to avoid the cross-venue fee floor, or an asset with a far richer, more persistent funding spread than SOL — a hypothesis change, not a parameter tune.

Delta-neutral cross-venue carry harvesting elevated Hyperliquid SOL perp funding. Long leg = SOLUSDT Binance spot (primary, drives the funding signal); short leg = SOLUSD Hyperliquid perp (receives funding when positive). Enters the equal-notional hedged book when smoothed HL funding (per-hour units) is materially positive and flattens both legs when funding decays to/below the exit threshold, with a max-hold and combined-PnL kill-switch as infra backstops.

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: SOLUSDT.BINANCE_SPOT-1-HOUR-LAST-EXTERNAL, Bars processed: 50615 Diagnostics: should_enter() returned a side 0 times over 5276 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

Funding data loads and is read correctly (108 entries fired on positive-funding threshold) — not a zero-trade or wrong-key bug; the mechanism executes

Fills wanted under-represented buckets (Hyperliquid, cross-venue, spot, market-neutral) and is a cash-flow mechanism rather than a dead price-pattern

Every single daily return is negative (zero positive days over ~206 trading days): Sharpe -4.4, profit_factor 0.765, omega 0.024, sortino -3.47, expectancy -$106/trade, PSR 0.0 — a systematic loser

Fee-floor failure: cross-venue round-trip ~0.29% (spot 0.20% + HL 0.09%) vs SOL HL funding ~0.01%/day; with 1d 7h avg holds and 108 round-trips, commission is 14% of gross plus $7,280 impact — churn buries the credited carry

Data wall: HL perp leg only has bars from ~Oct 2025, so tradeable history is ~206 days despite the 2020-2026 span — too thin to validate a carry via walk-forward

Hedge is not cleanly neutral: exposure 111%, a -3.70% single day, largest loss -$6,228, skew -12.3, kurtosis 166 — fat negative tails a truly delta-neutral book should not have

funding_exit=0 with smoothed funding oscillating around a small positive mean forces constant open/close, maximizing fee incidence

multi-year for walk-forward

The hedge code is correct (keep it) but the strategy is mis-parameterized so the backtest tests 'always hold a fee-bleeding hedge' rather than 'harvest materially elevated HL funding.' Fixes before optimization: (1) Raise funding_entry default to reflect genuinely elevated HL SOL funding (calibrate to the hl_funding_rates distribution; enter only top decile/quartile) — 3e-6/hr is near-zero. (2) Widen the optimizer search range for funding_entry so walk-forward can reach the material region (a ±20% sweep around 3e-6 never will), and allow funding_exit to go negative for hysteresis. (3) Add a minimum-hold / fee-amortization guard so accumulated funding exceeds the ~0.29% round-trip cost before exit. (4) Lower capital_frac to ~0.15-0.25/leg to cut the $7k impact drag. Full notes in workspace/discussions/2cac7524-b021-4081-8d78-500974604e67/iteration_1_feedback.md

This strategy sought a venue-specific edge, harvesting Hyperliquid's elevated SOL perp funding by shorting the HL perp against a matched Binance spot long to stay delta-neutral. The mechanism worked mechanically — funding data loaded and 108 positive-funding entries fired and were credited — but the book bled on every one of ~206 trading days, posting a -4.41 Sharpe, 0.765 profit factor, and -$106.55 expectancy per trade. The decisive problem was economic, not a code bug: the ~0.29% cross-venue round-trip fee was more than 20x SOL's tiny ~0.01%/day carry, funding_exit=0 churned the position, and the hedge showed fat negative tails inconsistent with true neutrality. The analyst abandoned it at the pre-optimization backtest-review gate, noting that no three-parameter tune could bridge the cost-vs-carry gap and that reviving the idea would require a same-venue build or a richer-funding asset — a hypothesis change, not a parameter tweak.

A funding-harvest carry only survives if the per-cycle carry exceeds the per-cycle transaction cost, so cross-venue fee floors demand either a same-venue construction or an asset with a far richer, more persistent funding spread than SOL — and closing with no hysteresis (funding_exit=0) guarantees fee-churning.

The backtest-review gate returned an 'abandon' verdict: funding was correctly read and credited (108 positive-funding entries), but the ~0.29% cross-venue round-trip fee dwarfed SOL's ~0.01%/day Hyperliquid carry, funding_exit=0 churned the position (avg hold 1d 7h), the hedge was not cleanly neutral (exposure 111%, skew -12.3, kurtosis 166), and only ~206 days of HL history existed — so it never reached optimization or the analyst.

It attempted a delta-neutral cross-venue carry that harvested structurally elevated Hyperliquid SOL perp funding by holding a long SOLUSDT Binance spot leg against a short SOLUSD Hyperliquid perp leg, opening the hedged book when smoothed funding exceeded a per-hour threshold.

Over ~206 tradeable days and 216 trades it lost money every single day: total return -11.95%, Sharpe -4.41, Sortino -3.47, profit factor 0.765, omega 0.024, expectancy -$106.55/trade, win rate 44.9%, with commissions at 14.12% of gross ($10,587) plus $7,280 impact and max drawdown 12.5%.