CrossSectionalFundingCarryTopKRotationNeutral

Cross-Sectional Funding Carry, Delta-Neutral Top-K Rotation (Data-Safe Binance Spot+Perp Basket, Relative Not Absolute Threshold)

A MARKET-NEUTRAL, CROSS-VENUE funding-carry book that harvests the perpetual funding premium by holding delta-neutral pairs on whichever data-safe perps CURRENTLY offer the richest funding, rather than gating on a fixed absolute threshold on one major. At each daily rebalance, rank a screened universe of confirmed-liquid Binance symbols by trailing 24h funding (mean of the last three 8h funding prints). Select the top-K (K=4) names whose trailing funding clears a small fee-breakeven floor. For each selected name, hold a delta-neutral pair: LONG <SYM>USDT.BINANCE_SPOT and SHORT equal-notional <SYM>USDT.BINANCE perpetual. The short-perp leg COLLECTS funding every 8h while the long-spot leg neutralizes price exposure, so PnL is dominated by accrued funding minus a small round-trip fee, decoupled from BTC's direction AND from BTC's absolute funding level. Because selection is RELATIVE (cross-sectional dispersion) rather than ABSOLUTE, the book stays deployed in every regime — it harvests whatever funding skew exists (often concentrated in narrative/meme/momentum alts) even when aggregate BTC/ETH funding has decayed to ~0. This directly fixes the dormancy that killed the single-major absolute-threshold version (zero trades in holdout, last trade March 2024). Universe is deliberately restricted to symbols with confirmed clean multi-year spot+perp history (BTC, ETH, SOL, BNB, XRP, ADA, DOGE, LTC) to screen out the ATOM/ARB/AAVE/Hyperliquid-alt data failures. Hysteresis on the ranking and a days-long minimum hold keep fee churn below the harvested carry. Simple by design: 4 core parameters (top_k, funding_floor, rebalance_hysteresis_band, max_basis_stop).

Fixes the reported Layer-2 failure: the timestamp is read INLINE inside calculate_signal (via bar.ts_event/ts_init) with no helper method that could be called with a mismatched positional argument — eliminating the '_bar_ts() takes 1 positional argument but 2 were given' TypeError across all six synthetic scenarios. All other logic (relative cross-sectional funding ranking, delta-neutral spot+perp pairing, hysteresis, min-hold, basis stop) is preserved unchanged. Selection is relative (cross-sectional dispersion) not absolute, so the book stays deployed in every regime, directly addressing the dormancy of the single-major absolute-threshold version. Venue routing: primary + secondary perps on BINANCE (MARGIN, short-capable) to collect funding; long legs on BINANCE_SPOT (CASH, long-only). leverage=1.0 since spot legs cannot use margin and the book is delta-neutral, avoiding any leverage_set_but_unused rejection.

The backtest is uninterpretable and the hypothesis's central premise-fix failed, so it is not worth a 2-hour optimization. (1) The dormancy this strategy was explicitly engineered to fix has re-appeared: returns are dense only 2020-09..2021-02, then a 5-year gap, then a single day 2026-07-07 (+10.9%); annual_returns has only 2020, 2021, 2026. Funding data is present across the span (35,075 events), so the 'relative ranking keeps the book deployed in every regime' claim demonstrably does not hold. (2) exposure_pct is 800% on a book that should run ~1.6x gross (per_pair_frac 0.20 x top_k 4 x 2 legs), with avg_holding_period 154 days against min_hold_days 3 and a daily rebalance — the cross-venue spot(CASH)+perp(MARGIN) pairs are accumulating and not closing, the recurring position-stacking artifact. (3) Consequently the metrics are artifact-driven and internally inconsistent: Sharpe 2.57 / probabilistic_sharpe 1.0 alongside total_return -2.33%, a +10.9% single day on a supposedly delta-neutral carry (funding is bps/day, so this is a mark-to-market artifact, not carry), return_kurtosis 360, tail_ratio 10.98, and end_unrealized_pct -25.84% (a large un-exited stacked book carried to the end). All 110 trades route through _submit_entry_instrument and bypass entry_diag/size guards. I state these as observed symptoms rather than asserting one unverified root cause. There is a concrete exposure/closing bug that alone might be iterable, but combined with the recurrence of the deployment/dormancy failure the hypothesis was designed to eliminate and the repeated failure of the funding-carry family, this should be abandoned. If the researcher wishes to salvage it, they must FIRST prove (a) the pair book closes properly and holds ~1.6x gross (not 800%), and (b) the book actually deploys and harvests carry across 2021-2025 — not just 2020-2021 — before any optimization is justified.

Market-neutral cross-sectional funding-carry book. Each daily bar, ranks a screened, data-safe Binance universe (BTC/ETH/SOL/BNB/XRP/ADA/DOGE/LTC) by trailing 24h funding (mean of last three 8h prints) and holds top-K delta-neutral pairs: LONG <SYM>USDT.BINANCE_SPOT + SHORT equal-notional <SYM>USDT.BINANCE perp. The short-perp leg collects funding while the spot long neutralizes price direction; PnL is accrued funding minus a small round-trip fee. Rank hysteresis and a multi-day min-hold suppress fee churn; a per-pair basis stop and funding-flip check force-close losers.

Verification failed (Layer 2 — synthetic scenarios): Parameters used: ['top_k', 'assets', 'min_notional', 'funding_floor', 'min_hold_days', 'per_pair_frac', 'qty_precision', 'funding_prints', 'max_basis_stop', 'default_qty_precision', 'rebalance_hysteresis_band'] Check that __init__ sets all attributes from self.parameters.get(). - steady_uptrend: TypeError: CrossSectionalFundingCarryTopKRotationNeutral._bar_ts() takes 1 positional argument but 2 were given (bar timestamp: 1735689720000) - steady_downtrend: TypeError: CrossSectionalFundingCarryTopKRotationNeutral._bar_ts() takes 1 positional argument but 2 were given (bar timestamp: 1735689720000) - flat_ranging: TypeError: CrossSectionalFundingCarryTopKRotationNeutral._bar_ts() takes 1 positional argument but 2 were given (bar timestamp: 1735689720000) - volatility_spike: TypeError: CrossSectionalFundingCarryTopKRotationNeutral._bar_ts() takes 1 positional argument but 2 were given (bar timestamp: 1735689720000) - zero_volume: TypeError: CrossSectionalFundingCarryTopKRotationNeutral._bar_ts() takes 1 positional argument but 2 were given (bar timestamp: 1735689720000) - price_gap: TypeError: CrossSectionalFundingCarryTopKRotationNeutral._bar_ts() takes 1 positional argument but 2 were given (bar timestamp: 1735689720000)

Genuinely intended market-neutral construction (beta -0.0003, benchmark_correlation -0.09) and a data-safe curated universe; fee drag is low (commission 1.21% of gross).

The delta-neutral spot+perp pair idea is sound in principle and funding data is broadly available (35,075 events).

DORMANCY RE-APPEARED: trades/returns are dense only 2020-09..2021-02, then a 5-year gap, then a single artifact day 2026-07-07 (annual_returns has only 2020/2021/2026) — the exact 'stays deployed in every regime' promise failing again despite funding data being present.

exposure_pct 800% on a book intended at ~1.6x gross, with avg_holding_period 154 days vs min_hold 3 days — pairs are accumulating/not closing (cross-venue spot-CASH + perp-MARGIN stacking artifact).

Internally inconsistent, artifact-driven metrics: Sharpe 2.57 / PSR 1.0 vs total_return -2.33%; +10.9% single day on a delta-neutral book; return_kurtosis 360; tail_ratio 10.98; end_unrealized_pct -25.84%.

Negative alpha (-0.0023) and information_ratio (-0.87); all 110 trades bypass the standard entry path (entry_diag 0), so size/exposure guards are not exercised.

Do not optimize yet — the mechanism is not functioning as designed. (1) DORMANCY: selection still uses an ABSOLUTE funding_floor gate (eligible = tf[a] > 0.00005), so in low-funding regimes the eligible set empties and the book goes dormant — the exact failure the 'relative not absolute' premise was meant to fix (returns only exist 2020-09->2021-02 and one 2026-06-26 bar). Make selection genuinely cross-sectional: rank the whole universe and take top-K by dispersion (e.g. above universe median or top-K unconditional with only a per-name fee-breakeven check), not an absolute floor. (2) HEDGE NOT NEUTRAL: end_unrealized_pct = -25.76% is impossible for an equal-notional same-asset spot+perp pair — check that both legs are populated in self._prices and time-aligned before opening (avoid opening one leg unhedged), and that spot/perp notionals match; report per-pair price-PnL residual (should be ~0, only funding should accrue). (3) ROTATION: avg_holding_period 153 days vs a daily/3-day-min-hold design — confirm pairs actually rotate as the ranking changes and _close_pair fully flattens both legs. (4) Confirm funding is actually accruing to the short-perp leg across the whole period (34,943 funding events available); a working carry book should show steady positive drift, not flat 2022-2025 with all P&L in one early window plus a final-bar spike. Re-run the initial backtest after fixing. Detailed feedback: workspace/discussions/dcd539f5-61e0-4a8f-8619-d5d58c4e70c2/iteration_1_feedback.md

This strategy was the redesigned successor to a dormant single-major funding-carry version, switching to relative cross-sectional selection over a data-safe spot+perp basket precisely to stay invested in every regime. It achieved genuine neutrality but reproduced two fatal problems: the pairs never closed — accumulating to 800% gross exposure with 154-day average holds against a 3-day minimum — and the promised all-regime deployment failed, with activity concentrated in 2020-2021 followed by a five-year gap and a single 2026 artifact day. The resulting metrics were artifact-driven and self-contradictory (Sharpe 2.57 yet total return -2.33%, kurtosis 360, a +10.9% day on a delta-neutral carry). The analyst abandoned it at the backtest-review gate after 4 iterations, requiring proof that the pair book closes properly and truly harvests carry across 2021-2025 before any re-entry; it never reached optimization.

A cross-venue delta-neutral pair book must be proven to actually close its legs and hold target gross exposure before its metrics mean anything — position stacking at 800% gross with 154-day holds invalidated an apparently strong Sharpe, and the relative-ranking fix still failed to keep the book deployed across regimes.

The backtest review returned 'abandon': the dormancy the strategy was explicitly built to fix reappeared (trades dense only 2020-09 to 2021-02, then a 5-year gap, then one artifact day in 2026 despite funding data being present throughout), and a position-stacking bug left the spot(CASH)+perp(MARGIN) pairs un-closed at 800% gross, making optimization unjustified.

A market-neutral, cross-venue funding-carry book that each daily rebalance ranked a curated 8-name Binance universe by trailing 24h funding, selected the top-K richest above a fee floor, and held each as a delta-neutral pair (long spot, short equal-notional perp) to collect funding on the short leg — using relative rather than absolute selection to stay deployed in every regime.

The construction was genuinely neutral (beta -0.0003, correlation -0.09) with low fees (1.21% of gross), but the backtest was uninterpretable: exposure hit 800% against a ~1.6x target with an average 154-day hold versus a 3-day minimum (pairs accumulated and never closed), and metrics were internally inconsistent — Sharpe 2.57 and PSR 1.0 alongside total return -2.33%, return kurtosis 360, a +10.9% single day on a supposedly delta-neutral book, and end-unrealized -25.84%.