MultiAssetExtremeFundingCashAndCarryBasketNeutral

Multi-Asset Extreme-Funding Cash-and-Carry Basket (Same-Venue Binance Spot + USD-M Perp, Per-Asset Extreme-Funding Gated, Market-Neutral)

A single strategy that runs the VALIDATED same-venue extreme-funding cash-and-carry across a BASKET of liquid Binance assets instead of one. For each asset in {BTC, ETH, SOL, BNB, XRP}, when that asset's OWN 8h perpetual funding rate is extreme (>= ~0.05%/8h), open a delta-neutral cash-and-carry: long the spot (BINANCE_SPOT) and short an equal-USD-notional USD-M perp (BINANCE), collecting the rich funding each 8h cycle while both legs are price-hedged. Positions are independent per asset and held only while that asset's funding stays elevated. This is the explicit reframe escalated by the analyst across five abandoned single-asset/cross-venue siblings: the same-venue extreme-funding gate is the only funding mechanism in the entire session shown to CLEAR fees (the BTC sibling: 184 trades, Sharpe 7.84, PF 1.15, genuinely neutral book). The single-asset version failed ONLY because BTC extreme-funding episodes cluster in 2020-2021, leaving recent holdout windows with zero trades. Spreading the SAME gate across 5 assets means some asset is almost always in an extreme-funding regime, giving continuous, validatable trading (non-zero OOS/holdout) while preserving the exact fee-clearing extreme-funding gate. Crucially this is NOT cross-exchange funding-differential carry (thin gap, repeatedly fee-dominated) and NOT thin-funding marginal carry (fee bleed) — it is single-venue, single-leg-per-asset funding COLLECTION gated to extreme levels that exceed the round-trip cost over the multi-cycle hold.

Iteration 3 fix for the Layer-3 (sandbox backtest) 300s wall-clock timeout. The funding lookup was already O(log n) via precomputed sorted arrays + np.searchsorted, so the timeout was driven by raw bar count across the 10 legs (5 spot + 5 perp) plus redundant per-bar funding recomputation. Two minimal, non-regressing changes: (1) coarsen every bar type from 4-HOUR to 8-HOUR — the natural funding-cycle resolution for a funding strategy — roughly halving processed bars and on-demand download volume per instrument; (2) compute the whole basket's funding once per bar into self._cur_funding and reuse it in _manage_asset, eliminating the second round of searchsorted calls. max_hold_bars adjusted 540->270 to preserve the ~90-day backstop on the coarser timeframe. Signal logic, imports, class structure, venue routing, and the fee-clearing extreme-funding gate are unchanged, so Layers 1 and 2 stay green.

Not worth optimizing: the basket loses money net-of-credited-funding in EVERY year (2020-2025), with profit_factor 0.99, negative expectancy (-$12.5/trade), Sharpe -2.81 (CI entirely negative), and capacity_usd of just $459 (impact_cost_pct 1475%) — there is no net edge to tune, at any scale. (This is NOT a funding-crediting issue: the current engine credits funding and the result is still a real loss, because the collected carry does not amortize the ~0.30% spot+perp round-trip given the tight 0.03%-enter/0.01%-exit band and ~3.6-day churn.) The reframe's central premise is falsified by the data: the claim was that spreading the extreme-funding gate across 5 assets yields continuous, validatable, fee-clearing trades, but activity clusters in 2020-2021 and is sparse AND negative in 2023-2025 — so the basket inherits the exact regime-concentration that abandoned the five prior single-asset/cross-venue siblings. This creates an un-tunable catch-22: raising funding_entry to the genuinely 'extreme' >=0.05% the hypothesis specifies (the code uses 0.03%) would re-concentrate trades in 2020-2021 and empty the recent holdout windows, while the current lower gate fee-bleeds. The only fee-clearing sibling (BTC extreme) already failed forward on zero-trade holdouts; a 6th reframe on the same mechanism does not escape it. Secondary flag for the record: a delta-neutral book should not show ±$48k single-trade swings — the hedge appears to leak directional exposure at times (worth a developer check), but fixing it would not rescue a PF-0.99, negative-every-year carry.

Market-neutral multi-asset extreme-funding cash-and-carry basket across {BTC, ETH, SOL, BNB, XRP}. When an asset's own 8h Binance perp funding is extreme (>= funding_entry), opens an independent delta-neutral book: long spot (BINANCE_SPOT) + short equal-USD-notional USD-M perp (BINANCE), collecting rich funding each cycle while price-hedged. Books close when funding cools to <= funding_exit (or on max-hold / per-book PnL backstops). Runs on 8-HOUR bars (funding-cycle cadence) with per-timestamp funding caching so per-bar cost stays O(log n).

Verification failed (Layer 3 — sandbox backtest): smoke test exceeded the 300s wall-clock limit. This almost always means per-bar work that scales with history — e.g. rescanning the full funding/supplementary series, or rebuilding a list and calling min()/sorted() inside calculate_signal()/on_bar() on every bar. Precompute sorted timestamp arrays ONCE in __init__ and use bisect, or cache lookups keyed by timestamp, so per-bar cost is O(log n) not O(n).

Clean same-venue construction (spot long + perp short, no cross-exchange fill artifact), genuinely low beta (-0.002), and a healthy raw trade count (638) — the plumbing works and funding is credited

Differentiated intent (market-neutral funding collection) that would fill under-represented buckets if it had a net edge

Negative total return (-20.4%) and NEGATIVE in every year of the sample; profit_factor 0.99, expectancy -$12.5/trade, Sharpe -2.81 with the entire CI negative (-4.14 to -1.41) — net-of-credited-funding, the carry does not clear costs

No deployable edge: capacity_usd $459 and impact_cost_pct 1475% — net edge is ~zero, so impact obliterates it at any real size; total_commission $33.8k reflects churn from the tight 0.03%/0.01% entry-exit band and ~3.6-day holds

Gate mismatch: funding_entry 0.03%/8h is below the hypothesis's 'extreme' >=0.05% spec, admitting marginal episodes that fee-bleed on the 0.20% spot + 0.10% perp round-trip

The core premise is falsified: trades cluster in 2020-2021 and are sparse AND negative in 2023-2025 — the 5-asset basket did NOT deliver continuous fee-clearing extreme-funding trades, so it inherits the regime-concentration that killed the single-asset siblings

Hedge-neutrality concern: ±$48k single-trade swings and ±$2500 avg win/loss on ~$18k legs suggest occasional directional leakage in a book that should have tiny per-trade PnL

Do NOT optimize at the current extreme 0.0005 gate — the reframe's goal (continuous, recent-regime trading) is NOT met: 83% of trades are in 2020/2021 and there are ZERO in 2022/2025/2026, so the holdout will be empty and DSR will fail exactly as it did on the single-asset BTC sibling. The mechanism itself is sound (funding +$71,124 credited, dominates the -$20k basis residual, net ~+$35k realized, clean neutral distribution). Two specific changes before any optimization: 1. VERIFY FUNDING DATA COVERAGE TO 2026 (do this FIRST — it determines everything). Trading stops at 2024-12. Confirm whether funding_rates_by_instrument (and the venue-wide fallback) for all 5 perps actually extends through 2025/2026. If the funding supplementary data is truncated at ~end-2024, the recent emptiness is a DATA-INJECTION bug, not regime decay — fix the data coverage and re-backtest; recent trades may appear at the existing gate. If the data IS complete to 2026 and still no trades fire, recent extreme funding genuinely did not occur -> proceed to (2). 2. LOWER funding_entry ON THE ALT-HEAVY BASKET (distinct from the single-asset BTC iterate). SOL/XRP/BNB carry richer and more frequent funding than BTC, so constrain the gate lower (~0.0002-0.0003) to catch moderate alt-funding episodes in 2022-2026 while the 5-asset diversification keeps the distribution clean. CRITICAL fee-viability guard: report mean funding-per-book vs total round-trip fees-per-book on the denser sample — the spot+perp round trip is ~0.20%, so the captured carry over the multi-cycle hold must exceed it. If the lower gate bleeds fees per trade, do not keep lowering. DECISION RULE: data complete/fixed + lower-gate alt basket produces 2025/2026 trades that clear fees + non-empty holdout -> proceed to optimization. If, after verifying data coverage, a fee-clearing threshold still leaves recent windows empty, the extreme-funding cash-and-carry family is structurally confined to bull-mania regimes (correlated major-coin funding that hasn't recurred) and should be ABANDONED — the multi-asset basket was the designated fix for sparsity and would have empirically exhausted that path.

This was the sixth reframe of the same-venue extreme-funding cash-and-carry, escalated after five abandoned single-asset and cross-venue siblings, betting that spreading the validated gate across five Binance assets would keep some asset always in an extreme-funding regime and deliver continuous, fee-clearing trades. The plumbing worked — 638 genuinely delta-neutral trades with funding credited — but the economics did not: it lost -20.4% net, was negative in every single year (PF 0.99, Sharpe -2.81), and had a capacity of just $459 with impact cost at 1475% of gross. The premise was falsified because trades still clustered in 2020-2021 and went sparse-and-negative through 2023-2025, inheriting the exact regime concentration that had sunk the earlier siblings, with a gate (0.03%) below the hypothesis's own 'extreme' spec compounding the fee bleed. The analyst abandoned it at the pre-optimization gate after 3 iterations, noting an un-tunable catch-22 and flagging occasional ±$48k swings as possible hedge leakage worth a developer check — though fixing that would not rescue a carry that loses every year.

Spreading a regime-concentrated funding gate across more assets does not manufacture a continuous edge — it creates an un-tunable catch-22: a genuinely 'extreme' gate (≥0.05%) re-concentrates trades in 2020-2021 and empties recent holdouts, while the lower 0.03% gate used here admits marginal episodes that fee-bleed on the ~0.30% round-trip, and credited funding cannot rescue a carry that doesn't amortize its own trading costs.

Also flagged for the record: a delta-neutral book showed ±$48k single-trade swings, suggesting occasional directional hedge leakage, but this was secondary to the absent net edge.

The analyst issued an 'abandon' verdict at the pre-optimization backtest-review gate: there was no net edge to tune (loses every year, PF 0.99, negative Sharpe) and no deployable capacity, and the reframe's central premise was falsified — the 5-asset basket did not produce continuous fee-clearing trades but inherited the same 2020-2021 regime concentration that killed five prior siblings. Optimization and all later stages were never reached.

A market-neutral, multi-asset extreme-funding cash-and-carry basket that runs the validated same-venue gate across {BTC, ETH, SOL, BNB, XRP} — for each asset, when its own 8h Binance perp funding is extreme, go long spot (BINANCE_SPOT) and short an equal-notional USD-M perp to collect the rich funding while price-hedged, spreading the gate across five assets so some asset is almost always in an extreme-funding regime.

Over the 2018-2026 sample it took 638 delta-neutral trades (319 long / 319 short) and was genuinely low-beta (-0.002) with funding credited, but lost money net of that funding: total return -20.4%, negative in every year, profit factor 0.99, expectancy -$12.5/trade, and Sharpe -2.81 with the entire CI negative (-4.14 to -1.41). It was also structurally undeployable — capacity only ~$459 with impact cost 1475% of gross — and trades clustered in 2020-2021, staying sparse and negative in 2023-2025.