NegFundingContrarianLongBasket

Dynamic Negative-Funding Contrarian Long Basket across Eight USD-M Majors (BINANCE, Hold Equal-Weight LONG Every Major Whose 8h Funding Is Deeply Negative — Crowded-Short = Oversold — Collect the Funding Tailwind AND the Mean-Reversion Bounce, Daily Rebalance, No Leverage, 3-Parameter)

A LONG-ONLY, single-venue, diversified CONTRARIAN basket that rides the ONE non-momentum long-only mechanism this factory has actually promoted: the negative-funding carry family (the deployed 5-asset basket at Sharpe 1.97 and the BTC cross-venue carry at 8.6). Its real edge is contrarian, not thin funding-collection: deeply NEGATIVE perp funding means shorts are crowded and paying longs — a mechanical oversold/capitulation signal that precedes mean-reversion bounces — and holding longs collects the funding tailwind WHILE capturing that bounce. It is deliberately distinct from the deployed FIXED-5 basket: the universe is DYNAMIC (hold only the majors whose funding is currently deeply negative, from 0 up to all 8), so capital concentrates in exactly the names the crowd is most short, and the book sits in cash when no name is crowded-short. It is engineered against this session's dominant death modes: DIVERSIFIED + LONG-ONLY + 1x leverage means it CANNOT be liquidated (unlike the three -100% single-name perp blowups) and has no single-name 81%-DD concentration (unlike the mid-cap equal-weight rotation); SINGLE-VENUE means no spot-perp basis convergence risk (which killed every alt cross-venue carry); it is not a grid (fat-tail), not a pairs bet (-82%), not options (fragile/blocked), not liquidations (data wall), not a price-only rotation (three already pending). Only 3 parameters.

Rides the one non-momentum long-only mechanism the factory has actually promoted — the negative-funding carry family (deployed 5-asset basket Sharpe 1.97, BTC cross-venue carry 8.6) — but reframes the edge as contrarian/oversold-bounce, which is what actually pays since the engine credits no funding cash flow (funding is the signal, price reversion is the return). Distinct from the deployed FIXED-5 basket via a DYNAMIC universe (hold only currently-crowded-short names, 0–8), and routed to a FUTURES venue (BINANCE) because the edge is funding-based and must hold the perp to read funding. Engineered against this session's death modes: LONG-ONLY + DIVERSIFIED (8 names) + 1x leverage means gross exposure never exceeds ~1x equity, so it cannot be liquidated (unlike the three −100% single-name perp blowups) and has no single-name 81%-DD concentration (unlike the mid-cap equal-weight rotation); SINGLE-VENUE removes the spot-perp basis-convergence risk that killed the alt cross-venue carries. Data plumbing is correct and proven: funding_rates_by_instrument is built unconditionally for every linear-perp leg (the same mechanism the deployed basket uses), read per-leg via a cached sorted-array + bisect nearest-<=-timestamp lookup with a lag tolerance clamped >= 8h; the primary leg also falls back to the single 'funding_rates' dict so the Layer-2 synthetic path still feeds a value. L5 (absolute-threshold no-trades) is mitigated by a modest cutoff (−0.003%/8h) plus an 8-name basket over years of daily data and a daily rebalance — a standalone sim produced hundreds of trades; only NET membership changes trade thanks to the 6% rebalance band, keeping fee drag low despite the daily clock. No silent edge-substitution: if per-leg funding is missing the name is skipped and the book goes to cash (only the diagnostic signal-return has a price-ROC variation fallback, never the trading decision). Sizing uses get_account_equity() directly because a MARGIN futures account already reflects full deployable equity (adding holdings value would double-count). Warmup is trivial (3 bars) so the sandbox produces trades immediately (no verification_loop, L21), and all eight legs are top, currently-listed, deeply-liquid USD-M majors so none triggers data_unavailable (L4). Three tuned parameters (funding_threshold, funding_avg_periods, rebalance_band) keep the optimization surface small. Avoids every confirmed dead class: not a grid, not a pairs bet, not options, not liquidations, not a price-only rotation.

Hard abandon on max_drawdown 86.68% (>50% floor, L19; DD CI low 97.7%). The +161x headline is entirely a 2021 bull artifact (annual 2021 +612.8x; every other year 2022/2024-partial/2025/2026 is negative or flat), the same single-regime majors-concentration that killed this family repeatedly. Implausible single-day returns (+102%, +82%, +43%) show the book is effectively heavily leveraged, falsifying the hypothesis's core 'long-only, 1x, liquidation-immune, drawdown-bounded' claim — so the very thing that was supposed to differentiate it from the -100% perp blowups is not actually present. Edge is also fee/impact-fragile (PF 1.14, impact 19% of gross, capacity only $2.78M, IR -0.17). Optimizing 3 parameters cannot manufacture drawdown control the sizing structurally lacks, nor create an edge outside 2021; it would only overfit the 2021 window. Note: this is a genuine strategy verdict — funding IS credited by the engine now, so the abandon is NOT about uncreditable funding.

Long-only, single-venue, diversified CONTRARIAN basket on 8 Binance USD-M perp majors (BTC/ETH/SOL/BNB/AVAX/LINK/ADA/XRP), 1-DAY bars, leverage 1.0. Each day it reads every leg's recent 8h funding (averaged over the last funding_avg_periods updates, from supplementary_data['funding_rates_by_instrument'], per-leg, keyed by instrument-id string) and holds an EQUAL-WEIGHT LONG in every major whose funding is deeply negative (<= funding_threshold). Deeply negative perp funding = crowded shorts paying longs = a mechanical oversold/capitulation signal; going long collects the funding tailwind (as a signal) while capturing the mean-reversion bounce (the realized PnL, since the engine does not credit funding cash flow). The universe is DYNAMIC — 0 up to all 8 names — so capital concentrates in exactly the names the crowd is most short, and the book sits fully in cash when nothing is crowded-short. Allocation runs inside calculate_signal on each timestamp-aligned daily bar: it computes each name's equal-weight target dollar (of full account equity, capped below 1x via a cash buffer), then issues DELTA orders toward those targets — sells first (freeing margin before buys) with a rebalance band that skips trades smaller than max(min_notional, 6% of equity) to keep turnover and fees low. Sells only ever reduce existing longs, so the book stays strictly long-only. calculate_signal returns a continuous cross-sectional funding gap (peer-average minus primary funding, scaled, plus a tiny price term) for signal-variation; should_enter/should_exit/position_size are inert.

At BACKTEST_REVIEW confirm full-history trade count clears ~100 and expectancy stays positive across sub-periods.

Trade count marginal: 15 closed round-trips over 363 days → ~75 over a ~5y history, borderline vs the ~100-trade floor. Sharpe 1.18 CI [-0.46, +2.58] crosses zero and the result leans on high kurtosis (41.9)/skew (3.69). Edge looks real (avg_trade +1.29%, PF 2.95, win_rate 0.67, max_dd 6.4%) but not yet robustly sampled.

Let optimization sweep funding_threshold across clearly deep values (-0.02% to -0.10%/8h) to test whether the contrarian edge strengthens with depth.

funding_threshold=-3e-05 (-0.003%/8h) is labeled 'deeply negative' but is only a mild buffer below zero. Structure matches the hypothesis and magnitude is tunable, so not a mismatch — just closer to 'any negative funding' than deep capitulation.

Trades across the full window (1030 trades, funding data present: 15079 events).

avg_trade_return_pct 0.416% clears the futures fee floor.

Funding-driven signal is now backtestable (engine credits funding) — the mechanism is not structurally blocked.

max_drawdown 86.68% (CI low 97.7%) — far beyond the 50% hard-abandon threshold; drawdown CI high of 61238% indicates leverage/blow-up dynamics.

Return is entirely 2021 (+612.8x that year); 2022 -17.7%, 2025 -9.5%, 2026 -16.3% — negative/flat in every non-2021 regime. Single-regime concentration, not a durable edge.

Implausible single-day returns (+102%, +82%, +43%) prove effective leverage far above the claimed 1x — the hypothesis's central 'diversified, 1x, liquidation-immune, drawdown-bounded' safety claim is falsified by the result.

skew 17.5 / kurtosis 638: the headline is carried by a few extreme days, not a stable edge.

profit_factor 1.14 is thin and fee-fragile; impact_cost_pct 19% with capacity_usd only $2.78M — the edge disappears at any real scale (information_ratio -0.17).

deployable scale

2021-only; 2022/2025/2026 negative

multi-regime

NegFundingContrarianLongBasket bet that deeply negative perp funding marks crowded-short, oversold majors, and that a dynamic long-only basket could collect the funding tailwind plus the mean-reversion bounce while being structurally safe from the -100% single-name blowups that plagued the session. The backtest ran cleanly across the full window (1,030 trades, funding credited, avg trade return clearing fees) and posted a +161x headline at Sharpe 1.2, but that came with an 86.7% max drawdown and implausible +80–100% single-day gains that exposed effective leverage far above the promised 1x. With returns concentrated entirely in 2021, negative in every other regime, and a fee/impact-fragile 1.14 profit factor at just $2.78M capacity, the analyst hard-abandoned it at the review gate — the drawdown control and liquidation-immunity the hypothesis sold as its differentiator simply were not present, and three parameters could not manufacture them or an edge outside 2021.

A 'long-only, 1x, liquidation-immune' framing does not by itself bound drawdown — sizing must actually enforce the 1x claim, and a headline return dominated by a single bull regime with a thin, fee/impact-fragile profit factor is a 2021 artifact rather than a durable, deployable edge.

The analyst issued a hard 'abandon' at the pre-optimization backtest-review gate on the 86.7% max drawdown (well past the 50% hard-abandon line), compounded by single-regime concentration (a 2021-only edge negative in every other year) and implausible single-day returns (+102%, +82%, +43%) that revealed effective leverage far above the claimed 1x — falsifying the strategy's central 'diversified, 1x, liquidation-immune, drawdown-bounded' safety thesis.

A long-only, single-venue contrarian basket on BINANCE USD-M perps that daily-rebalanced an equal-weight long into every major (BTC/ETH/SOL/BNB/AVAX/LINK/ADA/XRP) whose smoothed 8h funding was deeply negative — reading crowded-short funding as an oversold/capitulation signal — to collect the funding tailwind and capture the mean-reversion bounce, sitting in cash when no name qualified, claimed to be liquidation-immune via diversification and 1x leverage.

Over ~2019-12 to 2026-07 it took 1,030 trades with funding data present (15,079 events), a 52.3% win rate, avg_trade_return_pct of 0.416% (clearing the futures fee floor), total_return +161x, and Sharpe 1.204 — but with an 86.7% max drawdown (CI low 97.7%, CI high 61,238%), profit_factor of just 1.14, skew 17.5 / kurtosis 638, impact_cost 19% of gross, and capacity of only $2.78M. Returns were concentrated almost entirely in 2021 (+612.8x that year) while 2022, 2025, and 2026 were negative or flat.