BtcPerpSpotBasisCashCarryArb4H

BTC Perp-Spot Basis Cash-and-Carry Arbitrage (4H)

A DOLLAR-NEUTRAL CASH-AND-CARRY ARBITRAGE strategy: simultaneously SHORT BTCUSDT perpetual futures (Binance USD-M) and LONG BTCUSDT spot (Binance Spot) in equal dollar notional, triggered when the perp-spot premium exceeds a threshold. RATIONALE driven by the now-decisive session finding: 5 mechanism classes have ALL been falsified (supp-data, alt-perp trend-following, channel breakouts, oscillators, cross-asset pair MR). The analyst's remaining untested categories are: (a) sub-daily intraday (BTC 5-min Bollinger MR already in pipeline), (b) sector-basket (L1 cross-sectional momentum already in pipeline), (c) cross-VENUE arbitrage, (d) wait. This proposal tests a STRUCTURAL ARBITRAGE that is genuinely DIFFERENT from cross-asset pair MR (3 failed): instead of betting that the ratio between TWO DIFFERENT ASSETS will mean-revert (a non-stationary assumption that broke ETH/BTC, BNB/ETH, SOL/ETH), this strategy exploits the perp-spot basis on the SAME asset (BTC), which is STRUCTURALLY BOUNDED by arbitrage forces — specifically, the perpetual funding mechanism. When perp price > spot price by N bps, funding rate becomes positive, longs pay shorts, and this funding payment FORCES traders to close perp longs, mechanically driving perp back toward spot. This is the foundational cash-and-carry trade in crypto, run continuously by institutional arb desks. KEY DIFFERENTIATION from the falsified pair-MR class: (a) Pair MR assumes ratio reverts to mean — empirically false in crypto (ratios drift). (b) Basis arbitrage uses the SAME asset across venues with a STRUCTURAL convergence force (funding payments) — the basis CANNOT permanently drift because funding compensates for it. (c) The 'short' leg is on the SAME asset as the 'long' leg, eliminating the asymmetric win-rate trap that killed all 3 pair MRs (long_win_rate > short_win_rate due to one asset structurally outperforming). DESIGN CHOICES ELIMINATING ALL 5 PRIOR FAILURE MODES: (i) Not directional alt-perp (single-asset BTC), (ii) Not single-asset MR (uses cross-venue basis, not own-price reversion), (iii) Not pair MR on two different assets (uses same asset BTC on perp + spot — structurally hedged), (iv) ZERO supplementary data dependency (premium computed from OHLCV directly), (v) MAXIMALLY SIMPLE — only 2 parameters (entry premium threshold, exit premium threshold).

Fixes both iteration-1 failures. ECONOMICS (point 2): iteration 1 scalped ~5bps of OHLCV-basis convergence that could not beat the ~0.30% two-leg round-trip fee, giving negative return. This harvests the ACTUAL cash-and-carry edge the hypothesis names - FUNDING - by shorting the perp to receive positive funding and holding the hedged book until carry normalizes, so P&L is accumulated funding minus fees rather than a fee-eaten convergence scalp; entry_funding gates on a real positive carry regime, not a sub-fee threshold. NEUTRALITY (point 1, top priority): the two-leg hedge execution is unchanged from the proven, further-advanced delta-neutral funding-carry sibling (BtcDeltaNeutralFundingCarrySpotPerpBinance) - both legs are sized to the SAME dollar notional at fill time (perp_notional = size*perp_px; spot_qty = perp_notional/spot_px), contemporaneous 4H closes are guaranteed by the base template's same-timeframe alignment barrier so the hedge uses the current (not stale) spot price, and a net-dollar-delta log is emitted at every entry (should read ~$0) per the reviewer's explicit request. Cash-and-carry IS a funding-harvest trade - the hypothesis states the basis is structurally bounded BY the funding mechanism - so this is the truest implementation of the stated edge while eliminating the fee-drag that sank the convergence-scalp version. leverage=1.0 keeps the book dollar-neutral (leverage is used in sizing, so no unused-leverage gate).

Judged on the CREDITED funding result (5765 funding events available), the BTC perp-spot cash-and-carry loses money: total_return -9.74% over 6.5 years with profit_factor 1.08, because two-leg round-trip fees (commission 5.62% of gross) plus 14.34% market impact consume the entire thin funding carry. The apparent Sharpe 1.99 / PSR 0.9997 are artifacts of a 0.69%-annualized-vol drip with catastrophic fat tails (skew -5.11, kurtosis 108.8, rolling Sharpe to -114) — the pennies-before-a-steamroller signature, not a real edge. The carry has decayed to negative in every year 2022-2026; the only positive years (2020 +8.4%, 2021 +4.6%) came from a handful of crash-day basis-convergence spikes and residual directional exposure (a supposedly delta-neutral book showing +4-5% single days and end_unrealized -20.4% is not actually neutral), which do not repeat once funding normalizes. Capacity is a toy $4.86M — a standalone promotion blocker. With only 2 edge tunables (entry/exit funding thresholds), optimization can raise selectivity but cannot manufacture a carry that structurally clears ~0.30% two-leg fees + 14% impact in the normalized-funding recent regime. This is the second iteration (funding-harvest redesign after the iteration-1 basis-scalping failure) and it still loses; not worth 2 hours of optimization. NOTE: abandoning on the credited money-losing result and decayed/sub-fee carry — NOT on any 'funding is uncreditable' rationale.

Dollar-neutral BTC cash-and-carry funding harvest on 4H bars: SHORT BTCUSDT perpetual (Binance USD-M, primary) and LONG equal-dollar-notional BTCUSDT spot (Binance Spot, extra/CASH hedge). Enters when the 8h funding rate >= entry_funding (positive funding => longs pay shorts => the short perp RECEIVES carry) and HOLDS the delta-neutral book to accumulate funding, exiting only when funding normalizes (<= exit_funding) or a combined-PnL kill switch trips. Both legs are sized to identical dollar notional at fill time and the net dollar delta is logged each entry to prove neutrality. Two edge tunables: entry_funding, exit_funding.

Genuinely structural, same-asset delta-neutral construction — avoids the pair-MR non-stationarity trap and the supplementary-data walls it was designed around

Funding is credited (5765 events), both legs trade in balance (212 short perp / 212 long spot), no data or execution-drop failures

Dense enough firing (424 legs over 6.5yr) to populate walk-forward windows

Loses money on the credited result: total_return -9.74%, profit_factor 1.08 — two-leg fees (commission 5.62% of gross) plus impact (14.34% of gross) consume the entire carry

Edge decayed: positive only in 2020/2021, NEGATIVE every year 2022-2026; rolling Sharpe collapses to -72/-114 in stress periods

Sharpe 1.99 / PSR 0.9997 are misleading artifacts of a 0.69%-vol funding drip with fat tails (skew -5.11, kurtosis 108.8) — the classic pennies-before-a-steamroller pattern, not a real risk-adjusted edge

The 'delta-neutral' book shows +4-5% single days on crash dates and end_unrealized_pct -20.4% — neutrality is imperfect; 2020/2021 profits are incidental basis-spike/directional gains, not repeatable carry

Toy capacity ($4.86M) — a promotion blocker even if the edge were real

Only 2 tunables; no funding-threshold setting overcomes a carry that structurally sits below two-leg fees + 14% impact in the normalized-funding recent regime

Two issues must be fixed before optimization is meaningful: 1) HEDGE/NEUTRALITY IS BROKEN (top priority). A dollar-neutral BTC perp-short / spot-long book is producing +100% to +200% SINGLE-DAY returns and +624% in 2020 — impossible if the legs were offsetting. Investigate why: (a) verify the spot leg notional actually matches the perp leg in dollars at fill time (you size the perp from equity*per_leg_pct but compute spot_qty from notional=size*perp_px / spot_px — confirm both legs end up with equal dollar exposure and that the spot leg is actually filling, not silently dropped). (b) Confirm perp and spot 4H closes are contemporaneous (alignment barrier) so the computed basis isn't a stale-leg phantom. (c) Add an assertion/log of net dollar delta (perp_notional - spot_notional) after each entry — it should be ~0; the giant daily swings prove it isn't. Until daily P&L looks like a delta-neutral book (sub-1% daily moves), the metrics are uninterpretable. 2) ECONOMICS: entry_premium=5 bps cannot beat ~0.30% two-leg round-trip fees, which is why total_return is -4.59% with PF 1.09. Either (a) raise the entry threshold well above total round-trip cost, or (b) redesign to harvest the actual cash-and-carry edge — FUNDING — by holding the hedged position to collect positive funding (now credited by the engine) rather than exiting the moment basis converges. As written, the strategy captures a few bps of convergence and pays it all back in fees. Re-run the backtest after fixing the neutrality bug; if daily returns then look delta-neutral and funding capture shows a real positive carry, resubmit for optimization.

Designed as a structural escape from five falsified mechanism classes, this second-iteration strategy shorted the BTC perpetual against an equal-notional spot long to collect funding carry on a delta-neutral, same-asset book — deliberately avoiding the pair-MR non-stationarity trap and supplementary-data walls. It executed cleanly (212 balanced leg pairs, 5,765 funding events credited, dense enough for walk-forward) but lost 9.74% over 6.5 years at a profit factor of 1.08, because two-leg fees and 14.34% impact consumed the thin carry, and its headline Sharpe 1.99 was a low-vol fat-tailed illusion. The reviewer found the carry had decayed to negative in every year from 2022 onward, the only profitable years (2020, 2021) stemmed from non-repeatable crash-day basis spikes and imperfect neutrality, and capacity was a toy $4.86M — so it was abandoned at the backtest-review gate as a sub-fee money-loser that two funding thresholds could not rescue.

A structurally sound, same-asset delta-neutral construction is not enough: the BTC funding carry has normalized to sit below the ~0.30% two-leg round-trip fees plus ~14% impact, so a thin-carry harvest becomes a pennies-before-a-steamroller money-loser, and its 2020/2021 gains came from non-repeatable crash-day basis spikes and residual directional exposure rather than repeatable carry.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) on the credited, money-losing result — a decayed sub-fee carry with imperfect neutrality (end_unrealized -20.4%, +4–5% single days) and toy capacity that its two funding-threshold tunables cannot fix; optimization, analyst, and risk-review stages were never reached.

A dollar-neutral cash-and-carry strategy on BTC that shorts the Binance USD-M perpetual and holds an equal-notional Binance spot long, entering when funding is positive to harvest the perpetual funding carry on a same-asset, delta-neutral book.

On the funding-credited backtest (5,765 funding events) it fired densely and in balance (212 short-perp / 212 long-spot legs over ~6.5 years) but lost money: total return -9.74%, profit factor 1.08, with a superficially high Sharpe 1.99 / PSR 0.9997 that was a low-vol (0.69% annualized) fat-tailed artifact (skew -5.11, kurtosis 108.8). Two-leg fees (5.62% of gross) plus 14.34% market impact consumed the entire carry, the edge was positive only in 2020/2021 and negative every year 2022–2026, and capacity was a toy $4.86M.