Strategies

BTC Cross-Settlement Funding Arbitrage (BINANCE_CM Inverse Perp + BINANCE USD-M Perp, Same-Asset Cross-Margin-Type Hedged)

Outcome: Abandoned

BtcCrossSettlementFundingDiffCarryUsdmCmNeutral

Outcome Summary

BtcCrossSettlementFundingDiffCarryUsdmCmNeutral introduced a genuinely new mechanism — harvesting the funding differential between Binance's USD-M linear and COIN-M inverse BTC perps as a supposedly delta-neutral pair across four under-represented buckets. But the hedge was broken: dollar-matching the inverse leg ignored its nonlinear 1/price payoff, leaving a 247%-exposure directional, over-levered book that bled every month and was liquidated on 2024-04-19 to a -100% return, with 2,776 trades and $247k commission swamping a few-bps spread. The analyst ruled it a catastrophic, non-iterable failure — the inverse-hedge mis-sizing is technically fixable, but even corrected it hits the same fee-domination wall as every cross-venue funding-spread attempt this session. It ended after one iteration as abandoned, never advancing to optimization or risk review.

Strategy report

Hypothesis

Cross-settlement-type funding arbitrage on BTC perpetuals: capture the funding-rate differential between BINANCE_CM's BTC-margined inverse perp (BTCUSD_PERP.BINANCE_CM) and BINANCE USD-M's USDT-margined linear perp (BTCUSDT.BINANCE) by simultaneously taking opposite-direction positions on the SAME underlying asset across the two settlement-type venues. Both legs are BTC exposure — they CANCEL OUT directionally — leaving only the funding-rate differential between the two markets as the net PnL source. The two venues attract DIFFERENT participants: USD-M is dominated by USDT-speculator retail flow (higher leverage, more bullish bias, higher funding); CM is dominated by BTC-holding institutions and miners (lower leverage, hedging-driven, often lower funding). This structural participant-mix difference creates persistent funding-rate gaps between the two. The strategy SIMULTANEOUSLY FILLS FOUR severely under-represented portfolio buckets in a single hypothesis: (a) BINANCE_CM venue (0.2% → ≥5% target, the LARGEST gap I haven't filled), (b) cross-venue scope (4.1% → ≥15%), (c) market-neutral direction (8.7% → reducing 91.3% long-only), (d) pairs scope (13.8% → reducing 79.5% single). Mechanism class is COMPLETELY NEW to factory record — no failure pattern involves cross-settlement-type funding arbitrage. Distinct from BtcSpotPerpFundingCarryHedged (already abandoned due to BINANCE_SPOT data refill) because (1) both legs are perps (no spot dependency = no BINANCE_SPOT data refill risk), (2) funding source is the DIFFERENTIAL between two perp markets, not the absolute funding on one perp.

Implements the cross-settlement-type funding arbitrage hypothesis exactly: both legs are PERPS (no spot dependency, no BINANCE_SPOT refill risk), and the PnL source is the DIFFERENTIAL between two perp funding rates, not absolute funding on one. USD-M (USDT-margined, retail/leverage-heavy) structurally funds hotter than COIN-M (BTC-margined, miner/institutional hedging), creating a persistent gap captured by a market-neutral hedge. Fills the under-represented BINANCE_CM venue, cross-venue scope, market-neutral direction, and pairs scope buckets. Venue routing: futures on both legs is mandatory because the edge is funding-based and the strategy goes long+short. leverage=2.0 is consumed in position_size and the COIN-M leg notional (delta-neutral hedge keeps price risk offset, so moderate leverage amplifies the small carry safely). Fee-aware: round-trip ~0.20% is amortised by multi-day holds enforced via min_hold_bars + entry/exit hysteresis; sign-flip forces an immediate exit so the book never pays the differential it was built to collect.

Catastrophic, account-ending failure that is not an optimization candidate: the strategy was LIQUIDATED (2024-04-19) with total_return -100% and max_drawdown 100%. A delta-neutral pair cannot lose 100% if it is actually hedged, so the proximate cause is a mis-implemented inverse-contract hedge: the code dollar-matches the COIN-M leg as contracts = N/cm_price, but a BTC-margined INVERSE perp has nonlinear (1/price) payoff, so equal USD notional does NOT neutralize its delta against the linear USD-M leg. The result carried large residual directional exposure (avg_position_pct 247%, return_skew -24, kurtosis 721) that, levered, walked the account to liquidation, with relentlessly negative monthly returns 2020-2023 (e.g. 2023-12 -183%) — the opposite of a captured funding carry. This is not the 'engine doesn't credit funding' reason (funding is credited); it is a broken inverse-leg hedge plus fee/churn domination: 2,776 trades, commission 10.2% of gross ($247k), against a USD-M-vs-COIN-M BTC funding differential of only a few bps versus a ~0.20% round-trip. Not iterate: although the inverse-hedge mis-sizing is technically a fixable code defect, even a correctly delta-hedged version faces the same fee-domination wall that has killed every cross-venue/funding-spread strategy this session — a few-bps two-venue BTC funding gap cannot clear ~0.20% round-trip plus double-digit commission drag — so a hedge fix does not produce a viable edge. FAILURE PATTERN: cross-settlement (USD-M linear vs COIN-M inverse) BTC funding-differential carry blows up because inverse-perp delta cannot be neutralized by equal-USD-notional sizing, producing a directional, over-levered book that liquidates; and the targeted funding spread is too small to survive round-trip fees and churn. Inverse (COIN-M) legs require delta hedging in contract/BTC space (not USD-notional matching), and same-asset two-venue BTC funding spreads remain fee-bound.

Implementation

Delta-neutral, two-venue BTC funding-differential CARRY. Holds equal USD notional on opposite sides of BTCUSDT.BINANCE (USD-M linear perp) and BTCUSD_PERP.BINANCE_CM (COIN-M inverse perp). The two BTC exposures cancel directionally, leaving the funding-rate differential D = funding_USDM - funding_CM as the sole net PnL source. When D is rich on USD-M (the hotter, retail-driven venue) the strategy SHORTS USD-M and LONGS COIN-M to collect the gap each 8h settlement; when COIN-M is richer it does the reverse. Per-leg funding is read from supplementary_data['funding_rates_by_instrument'] (auto-built for both linear-perp legs and credited by the engine); the COIN-M leg falls back to a 0 baseline only when its series is unavailable (verification smoke test) so the carry logic still trades. Signal is the annualised differential in percent; entry/exit use hysteresis plus a min-hold floor and sign-flip stop to clear the ~0.20% round-trip fee. Sizing is equal USD notional per leg (qty=N/price each), with leverage wired into both legs' notional.

Backtest Review

Sharpe: -2.20
Total return: -100.00%
Max drawdown: 100.00%
Trades: 2776
Win rate: 46.3%
Profit factor: 0.90

Genuinely novel mechanism class (cross-settlement-type funding arb) targeting under-represented BINANCE_CM/cross-venue/market-neutral buckets

Funding wiring and atomic two-leg entry/exit are structurally reasonable for a linear pair

Account LIQUIDATED 2024-04-19, total_return -100%, max_drawdown 100% — a 'delta-neutral' pair cannot lose 100% if actually hedged, so the hedge is broken

Inverse-leg mis-hedged: equal-USD-notional (contracts = N/cm_price) does NOT neutralize a BTC-margined inverse perp's nonlinear 1/price delta against the linear USD-M leg; avg_position_pct 247%, skew -24, kurtosis 721 confirm large residual directional exposure

Relentless monthly losses 2020-2023 then liquidation — no funding differential captured

Fee/churn dominated: 2,776 trades, commission 10.2% of gross ($247k); USD-M vs COIN-M BTC funding gap is a few bps vs ~0.20% round-trip

positive

Outcome Summary

Inverse (COIN-M) legs must be delta-hedged in contract/BTC space, not by equal-USD-notional matching, or the residual nonlinear delta walks a levered book to liquidation — and even a correctly hedged version faces the same fee wall that killed every cross-venue/funding-spread attempt this session: a few-bps two-venue BTC funding gap cannot clear ~0.20% round-trip plus double-digit commission drag.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) as an account-ending failure, not an optimization candidate — a 'delta-neutral' pair cannot lose 100% if truly hedged, so the inverse COIN-M leg was mis-hedged (equal-USD-notional sizing does not neutralize a BTC-margined inverse perp's nonlinear 1/price delta), leaving a levered directional book that liquidated — so optimization and all later stages were never reached.

A novel delta-neutral cross-settlement-type funding arbitrage on BTC — holding equal-USD-notional opposite legs on the USD-M linear perp (BTCUSDT.BINANCE) and the COIN-M inverse perp (BTCUSD_PERP.BINANCE_CM) to cancel BTC price exposure and harvest the persistent funding-rate differential between the two participant mixes (retail USDT speculators vs BTC-holding institutions), filling four under-represented buckets (COIN-M, cross-venue, market-neutral, pairs) at once.

It was catastrophic: the account was LIQUIDATED on 2024-04-19 with total return -100% and 100% max drawdown over 2,776 trades (1,388 pairs), Sharpe -2.20, profit factor 0.90, expectancy -$99.7/trade, relentlessly negative monthly returns from 2020-2023, with avg_position_pct 247%, return skew -24, kurtosis 721, and commission at 10.2% of gross ($247k).

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