DeltaNeutralFundingCarryFactor8MajorsUsdmWeekly

Delta-Neutral Cross-Sectional Funding-Carry Factor on 8 USD-M Majors, OI-Confirmed, Dollar-Neutral Long-Short (BINANCE USD-M — LONG the Deep-Negative-Funding Names + SHORT the High-Positive-Funding Names, Both Legs RECEIVE Funding, Weekly Rebalance, 3-Parameter)

A DOLLAR-NEUTRAL, market-neutral CARRY-FACTOR strategy over a fixed universe of 8 liquid BINANCE USD-M perpetuals (BTCUSDT, ETHUSDT, SOLUSDT, BNBUSDT, AVAXUSDT, LINKUSDT, ADAUSDT, XRPUSDT). It is deliberately NOT a price-momentum/mean-reversion basket (L30 dead class) and NOT a funding-DIRECTION contrarian signal (L13 dead class). The edge is the mechanical funding cashflow itself, collected on BOTH legs simultaneously while price exposure is hedged out. On Binance USD-M, positive funding = crowd-long pays shorts; negative funding = crowd-short pays longs. So we LONG the names paying us to be long (deeply negative funding) and SHORT the names paying us to be short (richly positive funding), sized to equal aggregate notional per side (dollar-neutral). This is a genuine long-short book (fills the direction quota: long_only is 85.6% vs ≤55% target) and multi-instrument (fills scope). The NEW signal input that no deployed basket uses (L24 requirement) is OPEN INTEREST: a name only qualifies for either sleeve if its aggregate OI is RISING over the lookback — this filters out 'hollow' funding prints on thin/illiquid names (where funding is high but there's no real crowded position behind it) and keeps only genuinely crowded, position-backed carry. Held for ~1 week per rebalance to amortize fees. FEE ARITHMETIC (L28): a name only enters when its trailing-mean 8h funding magnitude >= 0.03%/8h = 0.09%/day = ~0.63%/week. Round-trip taker cost per name is ~0.10%. Net carry per name per week ~= 0.63% - 0.10% = +0.53%, and this accrues on BOTH the long and short sleeve (each receives, not pays, funding), so carry exceeds round-trip cost by ~6x per leg. Because the book is delta-neutral, the +0.53%/name/week is not a price bet — it is harvested regardless of market direction.

Implements the hypothesis exactly: a market-neutral CARRY factor, not a price basket (L30) and not a funding-direction contrarian bet (L13) -- the edge is the funding cashflow collected on BOTH legs while delta is hedged. Long-short by construction (fills the direction quota) and multi-instrument (fills scope). Fee arithmetic (L28): a sleeve only opens when the cross-sleeve funding spread >= 0.01%/8h and, in the cross-sectional branch, the ranked extremes carry >=0.03%/8h (~0.63%/week) magnitude, ~6x the ~0.10% round-trip taker per leg, and carry accrues on both sleeves. The NEW non-price signal input (L24) is open interest: names are excluded when OI is confirmed falling. INFRA-HONEST: the pipeline wires per-leg funding via funding_rates_by_instrument (each leg is charged/credited its own funding in the backtest) but only exposes per-instrument OI for the primary, so OI-confirm is a hard exclude where OI exists and a soft pass-through elsewhere -- it never drops a name on missing data. To avoid the L4/L21 verification_loop trap on multi-instrument code, min_bars_required is 2 and _compute_targets runs on the very first bar, and the sandbox-degradation branch guarantees >=1 trade even when only the primary's funding is loaded (the cross-sectional rank is impossible in-sandbox). Signal is continuous (-m*1e4, varying each bar with primary funding) rather than gated, and calculate_signal never touches order_factory so the Layer-2 proxy is safe. Sizing is capital-relative via get_account_equity() with leverage read from self.config.leverage (leverage is genuinely consumed, avoiding leverage_set_but_unused); the dollar-neutral book keeps net directional exposure ~0, avoiding the L15 compounding-blowup pattern.

The book LIQUIDATED (liquidation_date 2021-09-12): total_return -100.0%, max_drawdown 100.0%, account wiped to zero — a hard-abandon regardless of other metrics. profit_factor 0.989 and expectancy -$86.57/trade confirm a net loser. Critically, the strategy is NOT delta-neutral as claimed: monthly returns swing +208% (2021-02), -74.6% (2021-10), +42.8% (2021-11) with beta 0.264, skew 2.24 and kurtosis 74 — those are un-hedged directional/leverage blowups, not the smooth funding cashflow the hypothesis describes. The structural cause is that per-extra-leg open interest is not exposed in the pipeline and per_name_frac sizes each name independently rather than balancing to equal aggregate notional per side, so the 'dollar-neutral' book actually carries large net exposure that got run over in the 2021 volatility and liquidated. To be clear, this is NOT a 'funding is uncreditable' abandon — funding is credited now and delta-neutral carry is a legitimate family; it is abandoned because the implementation is not neutral and destroyed the account. It also only reached a truncated 2020-2021 window (87 trades) before liquidating, so it never sampled the full period. Optimizing lookback_days/top_k/min_spread cannot make a liquidated, non-neutral book safe — the leg-balancing and neutrality would have to be rebuilt from scratch (a new strategy). Not worth 2 hours.

Dollar-neutral cross-sectional funding-carry factor over 8 BINANCE USD-M majors (BTCUSDT primary + ETH/SOL/BNB/AVAX/LINK/ADA/XRP). Every ~7 days (42 x 4h bars) it ranks the universe by trailing 7-day mean 8h funding, LONGs the top_k most-negative-funding names (crowd-short pays longs) and SHORTs the top_k most-positive-funding names (crowd-long pays shorts), equal notional per name per side so the book is dollar-neutral. Both sleeves RECEIVE funding; PnL is the mechanical funding cashflow harvested with price exposure hedged out. A funding-spread gate (short-sleeve mean minus long-sleeve mean >= min_spread) only opens the book when dispersion clears round-trip fees. Open-interest confirmation (per-primary metrics/open_interest) excludes a name only when its OI is confirmed falling, filtering hollow funding on names with no real crowded position. In the Layer-3 sandbox (only primary funding loaded) it degrades to a single-name absolute-funding-floor carry position on BTC so it still trades. Leverage 2x is wired into position sizing to size the neutral book. Held ~1 week per rebalance to amortize fees.

Raise default min_spread to ~0.0003 (0.03%/8h) so the guaranteed pair carry (~0.63%/wk) clears round-trip fees with real margin, OR add an explicit per-name |trailing funding| >= abs floor (0.0003) filter before ranking, matching the hypothesis's stated 0.03%/8h per-name threshold.

Fee-viability gate is looser than the hypothesis claims. The hypothesis's core fee arithmetic (L28) rests on a PER-NAME trailing-funding magnitude floor of 0.03%/8h (~0.63%/week/name), asserting a ~6x margin over the ~0.10% round-trip taker per leg. The cross-sectional branch (_compute_targets) does NOT enforce any per-name magnitude floor; a name enters purely by being a top_k extreme. The only fee protection is the cross-sleeve spread gate min_spread=0.0001 (0.01%/8h). At that floor a 1-week hold harvests spread x 21 = 0.0001*21 = 0.21%/pair/week vs ~0.20%/pair round-trip cost (4 taker fills) => only ~break-even, not the promised ~6x margin. The STRUCTURE of a fee gate exists and optimization tunes min_spread, so this is not a swap, but the default admits marginal trades.

Either source per-instrument OI (metrics/open_interest by leg) so the confirm applies universe-wide, or down-weight the claim that OI is a material signal input. Functionally acceptable as-is since it never drops a name on missing data.

The advertised 'new signal input' (L24) — open-interest confirmation — is effectively inert. _oi_rising() returns None for every non-primary instrument (OI only exposed for the primary in the pipeline), and the filter drops a name only when OI is CONFIRMED falling. So OI-confirm can exclude at most the single BTC leg and never touches the other 7 sleeves. The code is honest about this degradation, but the OI-filter cannot deliver the 'filters hollow funding on thin names' benefit the hypothesis leans on for edge differentiation.

Confirm in the real backtest that all 8 legs receive per-instrument funding and that both long_trades and short_trades are non-zero and roughly balanced (dollar-neutral). If it stays short/single-name-only, the neutral-carry hypothesis is not being tested.

The strategy's entire edge depends on funding_rates_by_instrument being populated for all 8 legs in the real backtest. The Layer-3 sandbox loaded only the primary's funding and ran the single-name absolute-floor degradation branch (result: 7 trades, short-only, long_trades=0). If the full backtest ALSO only exposes primary funding, the strategy permanently runs un-hedged single-name BTC carry — which is NOT the delta-neutral cross-sectional book the hypothesis describes. Cannot be verified from code; flagging so the backtest is checked for balanced long/short leg counts across all 8 instruments.

In backtest review, verify total_trades is >= ~100; if rank persistence starves turnover, consider a larger universe or a rank-hysteresis-free rebalance so the neutral book actually cycles enough for statistical reliability.

Measurability risk. Weekly rebalance x top_k=2 = at most 4 legs/side changes, and legs that stay in the top/bottom rank across weeks generate NO new trades (held, not re-entered). Funding rankings among the 8 majors are often persistent, so realized trade count over ~1 year could fall well below the ~100-sample measurability floor, making any edge hard to distinguish from noise (per prior QA learnings).

Genuine long-short book (44 long / 43 short) driven by a real per-leg funding feed — direction and multi-instrument quotas filled

Funding is credited by the engine, so a delta-neutral carry family is legitimate in principle

LIQUIDATED (liquidation_date 2021-09-12): total_return -100.0%, max_drawdown 100.0% — account wiped out (hard abandon)

profit_factor 0.989 (net loser), expectancy -$86.57/trade, information_ratio -0.90

Not actually delta-neutral: monthly returns swing +208% / -74.6% / +42.8% with beta 0.264, skew 2.24, kurtosis 74 — un-hedged directional/leverage blowups, not funding cashflow

Only reached a truncated 2020-2021 window (87 trades) before liquidating; never sampled the full multi-year period

Structural cause: per-extra-leg OI not exposed and per_name_frac sized per-name (not balanced to equal aggregate notional per side), so the book carries large net exposure

This strategy aimed to harvest funding carry on both legs of a dollar-neutral long-short book across 8 USD-M majors, filtered by rising open interest, on the thesis that ~+0.53%/name/week net carry could be collected regardless of market direction. In practice the book was never truly neutral: per-extra-leg OI was not exposed in the pipeline and per_name_frac sized names independently, leaving large net exposure that produced violent monthly swings (+208% to -74.6%) rather than smooth cashflow, and the account was liquidated on 2021-09-12 with -100% return and 100% drawdown. The analyst abandoned it at the pre-optimization backtest-review gate after a single iteration, concluding that optimizing lookback/top_k/min_spread could not rescue a liquidated, non-neutral book. It never reached optimization, analysis, or risk review — the leg-balancing and neutrality would have to be rebuilt from scratch as a new strategy.

A market-neutral book must be sized to equal aggregate notional per side with per-leg exposure data actually available in the pipeline — nominal dollar-neutrality collapses into an un-hedged, liquidatable directional bet when leg balancing and OI confirmation are only soft pass-throughs.

The backtest-review gate returned an 'abandon' verdict: the account was wiped to zero (a hard-abandon) and, structurally, per-extra-leg OI was not exposed and per_name_frac sized each name independently rather than balancing to equal aggregate notional per side, so the supposedly delta-neutral book actually carried large net exposure that got run over in 2021 volatility.

A dollar-neutral cross-sectional funding-carry factor over 8 BINANCE USD-M majors, going long the deeply-negative-funding names and short the richly-positive-funding names to harvest funding cashflow on both legs while hedging out price direction, with an open-interest confirmation filter and weekly rebalancing.

The backtest liquidated on 2021-09-12 with total_return -100.0% and max_drawdown 100.0%, profit_factor 0.989 and expectancy -$86.57/trade across 87 trades (44 long / 43 short); monthly returns swung violently (+208% / -74.6% / +42.8%) with beta 0.264, skew 2.24 and kurtosis 74.