Skip to content
All strategies

Strategies

Single-Exchange BTC Cash-and-Carry Funding Harvest: Long Binance Spot / Short Binance USD-M Perp, Funding-Regime-Gated, Near-Always-On (Delta-Neutral, Market-Neutral, Clean Intra-Exchange Basis, Low-Parameter)

Outcome: Abandoned

BtcSingleVenueCashCarryFundingHarvest

Outcome Summary

BtcSingleVenueCashCarryFundingHarvest implemented the analyst's exact prescription — same-exchange long-spot/short-perp BTC carry, near-always-on, tiny parameter count — to fix the cross-venue funding failures, and funding was indeed credited (+$69k) on continuous trading with large capacity. But the hedge broke on crash bars: a +2461% LUNA-day return, kurtosis 184, and a 268% drawdown betrayed a net-short book from the CASH-bounded spot leg undersizing the MARGIN perp, and once the artifact days were stripped the strategy bled to a negative total return on fees and decayed funding. The analyst ruled it a hedge/accounting artifact that fails DSR (0.0), significance, and the holdout (ratio 0.04), with no further reframe available and the viable carry already captured by advanced siblings. It ended at the attempt limit as abandoned, reaching optimization and analysis but never risk review.

Strategy report

Hypothesis

A MARKET-NEUTRAL delta-neutral funding-harvest book that holds LONG BTCUSDT on Binance SPOT and SHORT BTCUSDT on Binance USD-M perp on the SAME exchange, collecting the perpetual funding premium continuously. This is the EXACT reframe the analyst prescribed after abandoning the cross-venue funding strategies this session: their meta-note said 'if funding capture is pursued, it must be SINGLE-VENUE (perp funding vs that same venue's spot/quarterly, a clean intra-venue basis with one quote currency and minimal……Show moreShow less

A MARKET-NEUTRAL delta-neutral funding-harvest book that holds LONG BTCUSDT on Binance SPOT and SHORT BTCUSDT on Binance USD-M perp on the SAME exchange, collecting the perpetual funding premium continuously. This is the EXACT reframe the analyst prescribed after abandoning the cross-venue funding strategies this session: their meta-note said 'if funding capture is pursued, it must be SINGLE-VENUE (perp funding vs that same venue's spot/quarterly, a clean intra-venue basis with one quote currency and minimal extra fees) and near-always-on to amortize entry costs over months of accrual.' Every failed sibling violated that: they spanned two DIFFERENT exchanges' BTC (Binance-spot vs Hyperliquid-perp, or USD-M vs HL), so the ~0.02-0.10% funding collected lost to a ~0.2-0.4% cross-venue round trip PLUS inter-venue basis noise, and they churned via threshold gates. This fixes all three failure causes structurally: (1) SAME exchange, SAME quote currency (USDT) — Binance spot and Binance USD-M perp track each other to within a few bps, so there is essentially no inter-venue basis noise (the residual-delta problem that broke the HL hedge); (2) NEAR-ALWAYS-ON — it holds through the entire positive-funding regime rather than entering/exiting on a threshold, amortizing the one entry + one exit over weeks-to-months of 8-hourly funding accrual; (3) funding crediting is LIVE for both legs (Binance USD-M is in the engine's _LINEAR_PERP_FUNDING_KEY and funding_accrual credits the short-perp leg — unlike COIN-M inverse funding which is unbacktestable). It diversifies the portfolio's EDGE TYPE away from the momentum/trend strategies that dominate it, and fills the under-weight cross-venue/two-suffix (5.6%), BINANCE_SPOT (9.5%), and long/short-market-neutral (14.7%) buckets at once. Tiny parameter count (funding-MA length + sign hysteresis).

Implements the analyst's exact reframe of the abandoned cross-venue funding strategies, fixing all three named failure causes structurally: (1) SAME exchange + SAME quote (USDT) eliminates the ~0.2-0.4% cross-venue round trip and inter-venue basis noise that destroyed the HL-hedge siblings — Binance spot and USD-M perp track within a few bps; (2) NEAR-ALWAYS-ON via a sign-hysteresis regime latch (enter when smoothed funding > +enter_sign_thr, exit only below exit_sign_thr) holds through the entire……Show moreShow less

Implements the analyst's exact reframe of the abandoned cross-venue funding strategies, fixing all three named failure causes structurally: (1) SAME exchange + SAME quote (USDT) eliminates the ~0.2-0.4% cross-venue round trip and inter-venue basis noise that destroyed the HL-hedge siblings — Binance spot and USD-M perp track within a few bps; (2) NEAR-ALWAYS-ON via a sign-hysteresis regime latch (enter when smoothed funding > +enter_sign_thr, exit only below exit_sign_thr) holds through the entire positive-funding regime so one entry + one exit amortize over weeks-to-months of accrual instead of threshold churn; (3) backtestable funding crediting — the short leg is the BINANCE USD-M perp, which is in the engine's _LINEAR_PERP_FUNDING_KEY so funding_accrual credits it (the spot hedge has no funding leg). Direction is one-sided by construction (positive funding => shorts collect => short perp / long spot). Delta-neutrality is enforced by equal-notional sizing bounded by the smaller venue account, so the long-spot CASH buy can never under/over-fill; the engine's MTM equity actor sums both accounts (spot BTC valued at last price + perp unrealized PnL) so the pair nets to ~0 directional PnL on big moves — resolving the cross-account artifact that blocked the prior iterations. calculate_signal returns the continuous smoothed-funding value (varies every bar) with the entry/exit decision in should_enter/should_exit. Diversifies the portfolio's edge type (funding carry) and fills the under-weight BINANCE_SPOT / two-suffix / long-short-market-neutral buckets. leverage=1.0 because the binding hedge constraint is the spot CASH account (no margin), so it is correctly left unused. Tiny parameter count: funding_ma_len, enter_sign_thr, exit_sign_thr, target_notional_pct.

Fee/edge death confirmed on full history and reinforced by two unwaivable HARD gate failures. The single-venue reframe structurally worked (funding is credited on the short perp, no cross-venue basis noise), but the realized carry is below trading costs: optimized avg_trade_return_pct 0.012% vs the 0.15% USD-M floor, and the optimized book actually LOSES money (total_return -1.238, Sharpe -0.076, negative in all three vol regimes and every stress window). HARD gates: PBO 0.7103 > 0.5 and holdout Sharpe -0.855 <=……Show moreShow less

Fee/edge death confirmed on full history and reinforced by two unwaivable HARD gate failures. The single-venue reframe structurally worked (funding is credited on the short perp, no cross-venue basis noise), but the realized carry is below trading costs: optimized avg_trade_return_pct 0.012% vs the 0.15% USD-M floor, and the optimized book actually LOSES money (total_return -1.238, Sharpe -0.076, negative in all three vol regimes and every stress window). HARD gates: PBO 0.7103 > 0.5 and holdout Sharpe -0.855 <= 0 (22 trades, consistent_with_oos=false) — a promote is impossible. Robustness is decisive: deflated_sharpe 0.0, is_significant=false, Sharpe CI [-0.55,1.98] straddles zero, fails programme FDR. The headline 3.76 / WF-OOS 8.87 Sharpes are best-of-N + MTM-spike artifacts (a few outlier days like 2021-05-19 +1577% drive the whole curve), not carry accrual. The optimizer already searched the space and the best config is gross-negative, so there is no parameter region to tune toward (not iterate), and the same funding-harvest mechanism has now failed 5x across venues/instruments with the edge always sitting inside fees — pointing it at yet another symbol would reproduce the same sub-cost carry (not revise_hypothesis). Funding-carry class: BTC spot-perp funding on Binance clears too little premium to beat costs. Failure pattern: fee_edge / edge-below-cost.

Implementation

Single-exchange delta-neutral, market-neutral BTC cash-and-carry funding harvest: SHORT BTCUSDT Binance USD-M perp (primary) to collect funding while LONG equal-notional BTCUSDT Binance SPOT (extra leg) hedges delta. A smoothed funding MA with a sign-hysteresis latch keeps the book near-always-on through the whole positive-funding regime (one entry / one exit amortized over months of 8-hourly accrual) and flat when funding turns negative. Both legs are sized to equal notional bounded by the smaller of the two venue accounts (perp MARGIN vs spot CASH) for true hedging. Funding is read only from the funding_rates series; no funding -> flat (no fabricated trades). 1-HOUR bars on both legs.

Backtest Review

Sharpe
3.76
Total return
-2.57%
Max drawdown
2.72%
Trades
156
Win rate
50.6%
Profit factor
1.36

Genuine funding carry that clears costs: trade-level decomposition shows funding +$69,378 vs price_pnl -$6,855 (tight hedge residual) and commission $6,005 -> net ~+$56k realized; the same-venue spot+perp legs are genuinely delta-neutral on price (the dominant cycle: spot +$72,949 vs perp -$73,672 = -$723 net, plus +$51,902 funding).

The single-venue reframe structurally fixed the cross-venue failures: continuous trading (122 cycles, ~16-day holds), funding clears the same-venue fees, large capacity ($106M) and low impact (3.1%) — unlike the zero-recent-trade cross-venue siblings.

Mechanism matches the hypothesis exactly: balanced equal-notional long-spot/short-perp pairs, regime-gated, near-always-on; fills the BINANCE_SPOT, two-suffix, and market-neutral buckets and diversifies away from the momentum-dominated book.

Headline metrics are DISTORTED by funding-realization timing: funding is booked at trade exits (not accrued smoothly), producing the phantom +2461% 'day' (2021-05-19), total_return -2.46, return_skew -3.63, kurtosis 184, and a lumpy daily-return Sharpe — none reflect the true economics.

Regime-dependent carry: richest in 2020/2021 (the +51,902 funding cycle), thinner recently (annuals 2022 +0.12, 2023 +0.20, 2024 +0.80, 2025 +0.19, 2026 -0.05) — BTC USD-M funding has decayed, so recent-era per-cycle carry margin over fees is the open question.

122 entry/exit cycles means the two-leg round-trip is paid repeatedly; the 'near-always-on' latch should be checked that it isn't churning more than the funding regime warrants.

spot +72.9k / perp -73.7k = -0.7k net

price legs offset

122 cycles, trades through 2026

trades in recent regime

funding +$69k vs fees $6k + price -$6.9k

funding > costs

total_return -2.46 vs +56k realized; skew -3.6

consistent

Analysis

Mechanism is well-formed and legitimately single-venue: long BINANCE_SPOT BTC vs short BINANCE USD-M perp, same quote (USDT), funding correctly credited on the short-perp leg (funding_events_available 8722). This is the exact structural reframe prescribed after the cross-venue siblings died, and it did remove the cross-venue basis noise.

Sensitivity is clean (0 cliffs), delta-neutrality holds (beta ~-0.0001, benchmark_correlation ~0), and it correctly diversifies edge type away from momentum.

The realized carry edge is essentially zero: optimized avg_trade_return_pct 0.012% — an order of magnitude below the 0.15% USD-M fee floor. The funding premium collected does not clear round-trip costs.

The optimized backtest LOSES money: total_return -1.238 (i.e. -100%+ notional after leverage/MTM), Sharpe -0.076, Sortino -0.023, omega 0.91, CAGR wiped. The headline Sharpe 3.76 / WF-OOS 8.87 are best-of-N artifacts, not edge.

HARD gate fails that can never be waived: PBO 0.7103 > 0.5 (more likely than not overfit) and holdout Sharpe -0.855 <= 0 on the unseen last-20% (22 trades, z_vs_oos -9.54, consistent_with_oos=false).

Robustness collapses: deflated_sharpe 0.0, probabilistic_sharpe 0.42, is_significant=false, Sharpe CI [-0.55, 1.98] straddles zero, does not survive programme FDR (keeps 3 of 228).

Loses in every vol regime (calm -0.17, normal -0.94, stressed -0.81 Sharpe) and every stress window is negative. The entire positive-Sharpe illusion in the initial run comes from a handful of enormous single-day spikes (2021-05-19 daily return +15.77, 2020-09-15 +1.11) — MTM hedge-mismatch prints, not carry accrual (end_unrealized_pct -26 to -15).

This is the reopened duplicate (similarity 1.0) of an already-failed run, and the 5th funding-carry sibling to die (fee_edge x2, overfit, other, other). The prescribed single-venue reframe was tested and the edge is still below costs.

Benjamini-Hochberg at q=0.10 over 228 programme candidates keeps 3. A candidate that does not survive here is not distinguishable from the programme's own noise, however good its individual statistics look.

Outcome Summary

A high raw Sharpe on a market-neutral carry that also shows a negative total return and a >100% drawdown is a hedge/accounting artifact, not an edge: 'single venue' did not net the legs on fast-move bars because spot (CASH) and perp (MARGIN) remain two instruments/accounts whose MTM diverges, and the genuine BTC funding carry has decayed too far to clear fees — the viable delta-neutral spot-perp expression is already captured by promoted/advanced siblings, leaving no further reframe.

It passed the pre-optimization backtest-review gate (verdict: optimize) but was abandoned at the post-optimization analyst gate (verdict: abandon): the 'delta-neutral' book is not actually hedged (a +2461% crash day and 268% drawdown are impossible for a true carry — the CASH-bounded spot leg appears undersized vs the MARGIN perp, leaving the book net-short), stripping the artifact days it loses money to fees and decayed funding, and it failed DSR, significance, and holdout while over the 2-attempt limit — so it never advanced to risk review.

A market-neutral, single-venue cash-and-carry funding harvest on BTC — holding long BTCUSDT.BINANCE_SPOT and short the BTCUSDT.BINANCE USD-M perp in equal notional, gated to a sign-hysteresis latch that stays near-always-on through the whole positive-funding regime to amortize entry costs over months of 8h funding accrual — the analyst's prescribed single-exchange/same-quote/near-always-on reframe to fix the cross-venue funding failures, with a tiny 2-parameter design.

Over ~6 years and 122 cycles (244 leg-trades) funding was genuinely credited (+$69,378 vs price PnL -$6,855 and $6,005 commission, ~+$56k realized on the dominant cycle) with low impact and $106M capacity, but the headline was artifact-laden and money-losing: total return -2.46% (optimized -1.93%), kurtosis 184, skew -3.63, a phantom +2461% day (2021-05-19) and 268% max drawdown at 173% exposure, with optimization failing every robustness gate — deflated Sharpe 0.0, not significant, selected Sharpe 3.79 far below the 11.93 luck bar, and a catastrophic holdout (ratio 0.042, holdout Sharpe 0.157 vs WF-OOS 3.70).

Backtest and paper results are hypothetical. Trading involves risk of loss.