WeeklyCumulativeNegativeFundingLongCarryBasket1D

Weekly Cumulative-Negative-Funding Long Carry Basket, Single-Leg Long-Only (BINANCE USD-M, 8 Liquid Perps, Weekly: Hold Names Whose PAST-7-DAY CUMULATIVE Funding Was Deeply Negative — Sustained Net Short-Crowding Paid to Longs — Hold One Week for MAXIMAL Fee Amortization, Deep Funding Feed, Fee-Arithmetic Stated, 2-Parameter)

A LONG-ONLY, single-leg, funding-driven carry basket that harvests the SAME proven edge as the deployed negative-funding basket (Sharpe 1.97) but through a distinct, LOWER-FREQUENCY signal and horizon chosen to maximize fee viability and fill the rarest portfolio bucket (weekly/macro horizon, 0.6%). It respects the viable-space lessons: L30 (genuine NON-PRICE dependency required → uses the DEEP/backfilled FUNDING feed, not the shallow OI/taker/liquidation feeds), L29 (no options), L28 (two-leg carry fee-dies → SINGLE-LEG, one round trip PER WEEK, arithmetic stated). Distinct from the deployed instantaneous-level basket and from my pending 8h-stamp variants (consecutive-persistence, trough-reversal, positive-rollover-short): the signal here is the CUMULATIVE funding a name has PAID over the trailing 7 days (21 stamps summed), a smoother measure of SUSTAINED net short-crowding, evaluated and rebalanced only ONCE per week. Weekly cadence means each name incurs at most one entry + one exit fee per week while accruing seven days of funding — the maximal fee amortization available, which is precisely what separates a viable single-leg carry from a fee-dead one. FEE ARITHMETIC (L28): entry requires trailing-7d cumulative funding ≤ −0.15% (≈ sustained −0.02%/8h) of carry COLLECTED while long; held ~1–2 weeks that is ≥15–30bps of accrual versus a single-leg USD-M round trip of ~0.09% (~2–3× margin) BEFORE the contrarian bounce that follows sustained short-crowding. Universe: 8 deep-history liquid USD-M perps (BTC, ETH, SOL, XRP, DOGE, LINK, AVAX, LTC), each gated INDEPENDENTLY (no cross-sectional rank → not the dead top-K funding rotation), equal-weight, conservatively sized. NOT two-leg carry, NOT pure-OHLCV, NOT options. 2 core parameters (cumulative-funding threshold, evaluation cadence).

Implements the hypothesis inside the viable design space. (1) Non-price dependency (L30): the ONLY signal is the funding feed — the deep/backfilled non-price series — never OHLCV. (2) Single-leg (L28): one perp long per name pays one ~9bps round trip PER WEEK, and the weekly cadence maximizes fee amortization — the entry gate (trailing-7d cumulative funding <= -0.15%) guarantees >=15bps of accrued carry over a 1-2 week hold (~2-3x the fee) before the contrarian bounce, clearing the avg-trade floor. (3) No options (L29). (4) Independent per-name cumulative-funding gate, NOT a top-K funding rank rotation (no cross-sectional sort). The distinct signal vs the deployed instantaneous-level basket and my 8h-stamp siblings (consecutive-persistence, trough-reversal, positive-rollover-short) is the SMOOTH trailing-7d cumulative funding measure on a weekly macro horizon — filling the rarest portfolio bucket (weekly/macro, 0.6%). Verification robustness: I confirmed against the real funding parquet that SOL (the primary — the only per-leg series available in the Layer-3 sandbox) triggers the <= -0.15% cumulative gate in ~12% of weeks (37 of 301 weekly samples), so over a ~1-year sandbox window it reliably produces trades rather than looping on 'no trades'; the gate sits near the 12th percentile of the cumulative distribution — selective but not brittle. In the full backtest, funding_rates_by_instrument supplies each of the 8 legs its own real funding so all names trade. Long-only + leverage 1.0 + 10%/name sizing structurally avoids the 50%+ drawdown / -100% liquidation blowups that killed prior baskets (L15); the 15% price stop backstops the directional risk the single-leg (unhedged) structure carries. Venue is BINANCE USD-M futures because the edge is funding-based and requires holding the perp to receive funding.

failed deflated Sharpe: DSR=0.0, expected-max=1.82 over 225 trials — the selected Sharpe is statistically indistinguishable from no-skill. This is a decisive overfit: walk-forward flags is_overfitted=true (IS avg 2.72 vs OOS avg 0.497) with OOS windows [4.08, -5.36, 2.76] — one deeply negative window shows the edge does not generalize — and PBO 0.627 > 0.5. Both are unwaivable hard gates, so promote is impossible. Optimization also thinned the sample to just 31 trades (below the 100 floor) by tightening cum_threshold, so the attractive headline (Sharpe 0.78, PF 6.58, DD 5.0%, skew 12.9, kurtosis 318) is a concentration artifact on a handful of episodes, not a robust edge. The decay I flagged at review is confirmed in the optimized run: 2021 (+33%) carries the entire result while 2024/2025/2026 are flat/negative (+0.9%, -1.9%, -1.2%) — the funding-carry edge existed in the 2021 high-funding regime and has faded since, which is exactly why the OOS window collapses to -5.36 and DSR is 0.0. This was the strongest of the funding-timing family, yet every sibling has already died the same way: the near-identical persistent-funding basket is flagged is_duplicate=true (similarity 0.851) and abandoned overfit, the reversal-long variant abandoned overfit, and the short variant abandoned fee_edge. A second optimization pass would only re-select the same 2021-concentrated overfit from the same decayed edge — there is no robust parameter region to tune toward. Per the overfit-not-revise policy (high IS, collapsed/negative OOS, DSR 0.0), abandon.

Long-only, single-leg weekly funding carry basket over 8 liquid BINANCE USD-M perps (SOL primary + BTC, ETH, XRP, DOGE, LINK, AVAX, LTC) on 1-DAY bars, each gated INDEPENDENTLY (no cross-sectional rank). Once per week it holds any name whose trailing-7-day CUMULATIVE 8h funding (21 stamps summed) is <= -0.15% — a smooth measure of sustained net short-crowding where crowded shorts pay longs — collecting the funding plus the contrarian bounce that follows sustained short-crowding. Weekly cadence means at most one entry + one exit fee per name per week while seven days of funding accrue (maximal fee amortization). Exits at the weekly evaluation when the trailing-7d cumulative funding normalizes (>= exit_threshold, after a >=1-week min-hold), on a 6-week max-hold timeout, or a 15% protective price stop. No leverage, conservative 10%-equity-per-name sizing (<=0.80 max gross) so the book cannot liquidate and drawdown is bounded by spot moves. Funding is read per-leg from funding_rates_by_instrument (real per-leg series in the full backtest), with the primary leg falling back to the flat funding_rates dict (its real series in the sandbox); a leg with no funding data stays flat — no price proxy is ever substituted.

At backtest-review, confirm the full 8-leg total_trades and metrics_reliable before trusting Sharpe/PF; if trades stay well below ~100, judge on carry+bounce economics and the deployed sibling rather than on small-sample statistics.

Measurability is the main risk and this is the LOWEST-frequency member of the funding-carry family. Sandbox exposes only the primary (SOL) funding series and produced 3 trades over 363 days (metrics_reliable=false); the other 7 legs trade only in the full backtest via funding_rates_by_instrument. Weekly evaluation + long holds (avg 23 days, min-hold 1 week, max-hold 6 weeks) make this intrinsically sparse: the rationale itself estimates the SOL gate fires in ~12% of weeks, and holds suppress re-entry. Full 8-leg multi-year trades are plausibly ~50-100 -- borderline for statistical reliability. Not a code fix (tightening the gate would violate the hypothesis's weekly-cumulative design).

Optional: evaluate the price stop every bar (daily) while keeping ENTRY/funding-exit on the weekly cadence, so the protective stop reacts within a day without adding entry churn or extra entry fees.

The 15% protective price stop is evaluated ONLY at the weekly rebalance (exit logic runs inside _manage, which is gated by bar_count % eval_cadence == 0). Between weekly evaluations a name can fall well past -15% and will not be cut until the next weekly check, so realized stop losses can materially exceed 15% on a fast intra-week crash. This matches the hypothesis's stated 'exits at the weekly evaluation ... or a 15% protective price stop' design and is bounded because the book is long-only + unlevered (no liquidation), but the stop is looser in practice than its nominal 15%.

Ignore the PF=0.0 field for this sandbox; assess PF on the full 8-leg run where losses will exist.

Metrics artifact, not a defect: sandbox reports profit_factor=0.0 while win_rate=1.0, avg_loss=0.0, and all 3 trades are wins (total_return +2.83%). This is the metrics engine's 0/0 fallback when gross_loss is zero (no losing trades), NOT the all-losses PF=0.0 code-defect signature. Flagged only so the 0.0 is not misread as a polarity/exit bug. With only 3 trades (metrics_reliable=false) the positive result is encouraging but not conclusive.

Strongest funding-carry variant by far: profit_factor 2.24, avg_trade_return_pct 7.74%, Sharpe 0.68 (vs siblings' 0.19/0.02) — the weekly + cumulative-funding construction validated its fee-amortization thesis

Excellent risk control: max_drawdown 9.2%, no liquidation, calmar 9.25; economically-grounded carry edge (shorts pay longs), proven by the deployed sibling (Sharpe 1.97)

Funding feed available (4900 events, L30-compliant, credited); huge capacity (~$4B); fills the rarest weekly/macro bucket

sharpe_ci_low positive (0.003), PSR 0.987

Edge concentrated in 2021 (+49%) and 2023 (+16%); recent years flat/negative (2024 +2.4%, 2025 -3.7%, 2026 +3.6%) — funding-carry decay risk that weakened the siblings

Lumpy: return_skew 6.5, kurtosis 126 — a few big wins carry the result; only 86 trades over 6 years

Sharpe 0.68 below the 1.5 floor; the deflated-Sharpe/OOS gates in the 2024-2026 window are the binding risk given the concentration

Best funding-carry variant of the family; economically-grounded carry edge (shorts pay longs), funding feed available (no data wall)

Optimized risk profile is excellent on paper (max_drawdown 5.0%, profit_factor 6.58) with positive sharpe_ci_low (0.195)

HARD FAIL: walk-forward is_overfitted=true — IS avg Sharpe 2.72 vs OOS avg 0.497, with OOS windows [4.08, -5.36, 2.76] (one deeply negative — no generalization)

HARD FAIL: PBO 0.627 > 0.5 — parameter selection is overfit

deflated_sharpe 0.0, is_significant=false — selected Sharpe indistinguishable from no-skill over 225 trials (expected-max 1.82)

Optimization thinned the sample to 31 trades (< 100 floor); PF 6.58 / skew 12.9 / kurtosis 318 is a concentration artifact

Edge decayed post-2021: optimized annual returns 2021 +33% carries all; 2024 +0.9%, 2025 -1.9%, 2026 -1.2%

Whole funding-timing family is dead: persistent-funding sibling is_duplicate=true (sim 0.851) abandoned overfit; reversal-long abandoned overfit; short variant abandoned fee_edge

This was the strongest of the session's funding-timing basket family: a weekly, cumulative-negative-funding long carry whose lower-frequency construction genuinely delivered the fee amortization its siblings lacked, earning an 'optimize' verdict on a 2.24 profit factor, 7.74% per-trade return, and controlled 9.2% drawdown. But the 3-phase optimization confirmed the decay risk flagged at review: in-sample Sharpes of 2.72 collapsed to a 0.497 out-of-sample average with one window at -5.36, PBO hit 0.627, deflated Sharpe was 0.0, and tightening the threshold thinned the book to 31 trades whose 6.58 profit factor was a 2021-concentrated artifact (every later year flat or negative). The analyst abandoned it — the negative-funding edge that made the deployed sibling a Sharpe-1.97 winner has faded since funding normalized, so no robust region remained to tune toward, and the near-identical persistent-funding, reversal-long, and short variants had all already died overfit or fee-dead.

Even the best-constructed funding-carry variant — weekly cadence and cumulative funding genuinely validated the fee-amortization thesis at review (Sharpe 0.68, PF 2.24) — cannot survive because the underlying negative-funding edge has decayed to a 2021-concentrated signal since funding normalized, so optimization only re-selects the same overfit, and the whole funding-timing family is exhausted.

The analyst issued an abandon verdict on unwaivable overfit hard gates (is_overfitted=true, PBO 0.627 > 0.5, deflated Sharpe 0.0 against expected-max 1.82 over 225 trials, one OOS window deeply negative at -5.36): the funding-carry edge existed in the 2021 high-funding regime and has decayed since, the optimizer's attractive headline was a concentration artifact on ~31 trades, and every sibling of the funding-timing family had already died the same way with no robust parameter region to tune toward.

A long-only, single-leg funding-carry basket over 8 liquid BINANCE USD-M perps that rebalances weekly on 1-day bars, holding names whose trailing-7-day cumulative 8h funding was deeply negative (sustained short-crowding paid to longs) for one week to maximize fee amortization — the same proven edge as the deployed negative-funding basket (Sharpe 1.97) via a smoother, lower-frequency signal filling the rare weekly/macro bucket.

It was the strongest funding-carry variant of the family at review — profit factor 2.24, avg_trade_return_pct 7.74%, Sharpe 0.68 (CI low positive at 0.003), 9.2% max drawdown, no liquidation over 86 trades — but optimization exposed overfit: is_overfitted=true (IS avg Sharpe 2.72 → OOS 0.497 with windows [4.08, -5.36, 2.76]), PBO 0.627, deflated Sharpe 0.0, and the tuned config thinned the sample to 31 trades with returns concentrated in 2021 (+33%) and flat/negative 2024-2026.