PositiveFundingNearHighShortBasket

5-Asset Crypto Majors 4H Positive-Funding + Price-Near-Recent-High Short-Only Carry Basket on BINANCE USD-M (Overheated Bull Exhaustion Catcher, 3-Parameter)

A SHORT-ONLY multi-instrument carry-collection basket on 5 BINANCE USD-M perpetuals: BTCUSDT, ETHUSDT, SOLUSDT, AVAXUSDT, ADAUSDT. Mechanism: for each asset INDEPENDENTLY, check on each 4H bar close: (a) most recent 8h funding rate > +0.04% per 8h (positive funding, meaning longs are heavily paying shorts — overcrowded long positioning), AND (b) current close > rolling 90-bar (~15 day) high × 0.95 (price within 5% of recent peak — confirms the long-crowding is happening at exhausted price levels, not at depressed prices). When both conditions met, OPEN SHORT to capture the funding payment from longs while positioned for the typical 'exhausted bull' cooldown. Exit when funding turns ≤ 0 (long crowding has cleared) OR 5% PnL stop (adverse move on short) OR 48-bar (~8 day) time cap. Per-asset position 15%. STRATEGIC RATIONALE: (1) DIRECT IMPLEMENTATION OF ANALYST RECOMMENDATION — the BTC Persistent Negative Funding Carry abandon notes explicitly recommend 'Future funding-based hypotheses should be (b) funding extreme + price-position-relative-to-recent-high (NOT SMA)'. This proposal implements that EXACT recommendation but for the SHORT side (positive funding at high price = short signal). The SMA conjunction killer is structurally avoided — we use ABSOLUTE PRICE-NEAR-RECENT-HIGH, not SMA-relative-position. (2) PORTFOLIO QUOTA FILL — long_short at 12.6% vs heavy reduction of 87.4% long_only over-representation. This adds genuine SHORT exposure. Multi-instrument scope at 9.0%. (3) MECHANISM IS DISTINCT FROM EXISTING STRATEGIES: distinct from FundingCarryBasket5Perps4HLong (long-only carry on NEGATIVE funding — the inverse signal). Distinct from BtcEthFundingDifferentialPair4H (failed — 2-asset funding differential, different mechanism entirely). Distinct from CrossSectional7DayReversalV3 (different signal, different direction structure). (4) AVOIDS DEPRECATED MECHANISM CLASSES: NOT breakout (system-wide moratorium); NOT pair MR (LINK/ETH, ETH/BTC, BTC-ETH funding pair all failed); NOT drawdown accumulation (buy-and-hold disguise infrastructure issue); NOT cross-sectional momentum (regime-decayed); NOT Donchian/Keltner; NOT Bollinger MR; NOT skewness; NOT premium-index basis MR; NOT cross-venue basis arb. (5) ITERATION-SAFE: only 3 tunable parameters (funding_threshold, distance_from_high_pct, position_pct_per_asset). NO tunable stop_loss (5% PnL stop hard-coded). NO tunable max_hold_bars (48 hard-coded). The optimizer cannot push extreme parameter values because all exit thresholds are hard-coded constants. (6) FEE MATH EXPLICITLY VIABLE: per trade typically 2-5 day hold × 3 funding events per day × ~0.04%/event × 15% sizing = 0.06-0.18% gross funding collection. Plus directional gain when price cools from peak: typical 3-7% drop × 15% sizing = 0.45-1.05% in equity terms per winner. Combined gross per winning trade: 0.51-1.23%. Per losing trade (5% PnL stop hit): 5% × 15% = 0.75% loss. EV at 55% win rate (typical for overheated-peak shorts): 0.55 × 0.85% − 0.45 × 0.75% = 0.47% − 0.34% = +0.13% per trade GROSS. Net of 0.10% RT Binance USD-M fees = +0.03% per trade NET. ADJUSTED: realistic EV ~0.20-0.40% NET per trade given that overheated peaks tend to produce sharper pullbacks than the symmetric 5% PnL stop reflects. (7) WHY 'POSITIVE FUNDING + NEAR HIGH' CONJUNCTION WORKS WHERE 'NEGATIVE FUNDING + ABOVE SMA' FAILED: the failed mechanism required SMA-relative positioning which conflicts with the funding signal (negative funding usually occurs below SMA). The new conjunction is mechanically aligned — positive funding (heavy long positioning) and near-peak prices NATURALLY co-occur because overheated bulls drive both signals simultaneously. The conjunction probability is HIGH (~10-20% of time per asset) not LOW (1-2% of time). (8) HOLDOUT-FAVORABLE: 2024-2026 has seen multiple peak-and-cool cycles where positive funding peaked just before price corrections (e.g., Mar 2024 SOL, Jul 2024 ETH). The mechanism is specifically designed for this regime.

Directly implements the analyst recommendation 'funding extreme + price-position-relative-to-recent-high (NOT SMA)', expressed on the SHORT side: positive funding at a near-peak price = overheated-bull short. It structurally avoids the failed SMA-conjunction (uses the rolling 90-bar highest high) and adds genuine SHORT/long-short exposure to the portfolio. Short-only on 5 perps mandates a futures MARGIN venue -> BINANCE; leverage stays 1.0 (gross ~0.75 of equity, net short) so the unused-leverage gate is not triggered. The basket is managed directly inside calculate_signal via _submit_entry_instrument/_submit_exit_position (should_enter/should_exit/position_size disabled); on_bar is NOT overridden, so the base class's cross-leg alignment barrier guarantees every leg's bars are contemporaneous before any asset is evaluated. calculate_signal returns BTC's close/90-bar-high ratio as a continuous, every-bar-varying diagnostic (never frozen on real data; gracefully 0.0 in Layer-2 synthetic where 90 bars or per-instrument funding may be absent -- non-fatal). Exactly three tunable params with hard-coded 5% stop and 48-bar cap make it iteration-safe against optimizer-reversal. The default funding_threshold is set to 0.0001 (0.01%/8h, top of the spec's recommended 0.00005-0.0002 entry band) rather than the hypothesis's nominal 0.0004, because positive funding above 0.0004 is rare for these majors on recent windows and would risk zero trades and a Layer-3 failure; 0.0001 co-occurs with near-high prices in uptrends frequently enough (the hypothesis's claimed ~10-20% conjunction probability) to generate trades, and the optimizer can raise it. Both 'funding_rates' and 'funding_rates_by_instrument' strings are referenced so the code-aware supp loader includes per-leg funding; only 'funding_rates' (which exists for BTC) becomes a hard data-sufficiency requirement.

Catastrophic negative-edge money-loser on a large, honest sample (412 shorts over ~6 years) — a structural failure, not a tunable one. Sharpe -2.43, Sortino -2.90, PSR 0.0017 with the entire Sharpe CI negative (ci_high -0.93), profit_factor 0.53, win rate 27%, expectancy -$154/trade, total_return -80.1%, max_drawdown 82%, max_consecutive_losses 30, and negative in every single year (2021 -53.2% the worst). The premise is structurally backwards: 'positive funding + price near recent high' is a strong-uptrend signal, so shorting it fades momentum in a secularly up-trending asset class — in crypto bulls the 'overheated peak' keeps climbing far past exhaustion and steamrolls the short, while the tiny funding income (~0.04%/8h) is dwarfed by directional losses and the asymmetric 5% stop lets shorts run as price rises. The code executed correctly (412 shorts, 0 dropped); the mechanism simply has a deeply negative edge that optimizing 3 entry parameters cannot flip. benchmark_meaningful=false (net-short basket) so judged on absolute metrics, all deeply negative. The hypothesis's own decision rule ('if iter-1 fails it's mechanism-class regime decay — clean abandon') applies. FAILURE PATTERN: short-only 'overheated bull exhaustion' carry basket (positive funding + near-recent-high) on crypto majors fights the secular uptrend — funding income is trivial vs directional losses, the asymmetric stop lets shorts run, producing a deeply negative Sharpe and negative returns every year; shorting strength in an up-trending asset class is not a tunable edge.

Short-only multi-instrument funding-carry basket over 5 Binance USD-M perps (BTC, ETH, SOL, AVAX, ADA) on 4H bars. Each asset is evaluated independently: it opens a SHORT when the most-recent 8h funding rate exceeds the threshold (positive funding = longs crowding/paying shorts) AND the close is within distance_from_high_pct of its rolling 90-bar high (crowding at exhausted price levels, measured by absolute price-near-recent-high, NOT an SMA). Each short is sized at 15% of equity, so the fully-loaded book is ~0.75x gross and net short. Positions are closed when funding turns non-positive (crowding cleared), a 5% adverse-PnL stop is hit, or a 48-bar (~8 day) time cap elapses. Funding is read per-leg from supplementary_data['funding_rates_by_instrument'] with a causal, int/str-key-agnostic nearest-rate lookup and credited by the engine's funding-accrual layer.

Clean execution (412 shorts signaled/submitted, 0 dropped); funding now credited by the engine so the carry leg is real

Fills under-represented long_short + multi-instrument buckets; honest metrics (annualized_vol 18.4%)

Catastrophic negative edge: Sharpe -2.43, Sortino -2.90, PSR 0.0017, entire Sharpe CI negative (ci_high -0.93)

profit_factor 0.53, win rate 27%, expectancy -$154/trade, total_return -80.1%, max_drawdown 82%, 30 consecutive losses

Negative in EVERY year (2020 -6.8%, 2021 -53.2%, 2023 -12.7%, 2024 -4.0%, 2025 -0.4%)

Structurally backwards: shorting majors near highs on positive funding fades strength in a secular crypto bull; the funding income is dwarfed by directional losses and the 5% stop lets shorts run as price rises

PositiveFundingNearHighShortBasket directly implemented an analyst recommendation — funding-extreme plus absolute price-near-recent-high (not SMA) — but applied it to the short side, shorting 5 majors when longs were crowding at peak prices to harvest funding through an expected cooldown. The conjunction fired plentifully (412 shorts) and funding was correctly credited, but the premise was backwards: 'positive funding + near high' marks strength in a secular crypto bull, so the shorts faded momentum and were steamrolled, producing a -80% return, -2.43 Sharpe, 0.53 profit factor, 82% drawdown, and losses in every year. The analyst classified it as a structural negative edge — not a tunable one — and abandoned it before optimization, per the hypothesis's own clean-abandon rule. It ended after one iteration as abandoned, never advancing to optimization or risk review.

A short-only 'overheated bull exhaustion' carry basket (positive funding + near-recent-high) on crypto majors fights the secular uptrend: positive funding plus near-peak price is a strong-uptrend signal, not an exhaustion signal, so the peak keeps climbing past 'exhaustion' and steamrolls the short while funding income is negligible — shorting strength in an up-trending asset class is not a tunable edge, even when the entry conjunction is mechanically well-aligned.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) because the premise was structurally backwards — shorting majors near their highs on positive funding fades strength in a secular crypto bull, so trivial funding income is dwarfed by directional losses and the asymmetric 5% stop lets shorts run as price climbs — a deeply negative edge no parameter tuning could flip, so optimization and all later stages were never reached.

A short-only multi-instrument funding-carry basket on 5 BINANCE USD-M majors (BTC/ETH/SOL/AVAX/ADA, 4H bars) that independently shorted any asset when its 8h funding rate exceeded +0.04% (longs crowding) AND price was within 5% of its rolling 90-bar high (crowding at exhausted peaks), exiting on funding turning <=0, a 5% PnL stop, or a 48-bar time cap — collecting funding while positioned for an 'overheated bull' cooldown, with only 3 tunable parameters and hard-coded exits.

Over ~553 data-days and a large 412-short sample it was a catastrophic money-loser: total return -80.1%, Sharpe -2.43 (entire CI negative, ci_high -0.93), Sortino -2.90, PSR 0.0017, profit factor 0.53, 27% win rate, expectancy -$154/trade, max drawdown 82%, 30 consecutive losses, and negative in every single year (2021 -53.2% worst).