BtcCrossVenuePriceConvergenceArbHlBinanceNeutral1H

BTC Binance-Hyperliquid Cross-Venue Price Convergence Arbitrage (1H)

A DOLLAR-NEUTRAL CROSS-VENUE ARBITRAGE strategy: simultaneously LONG one venue's BTC perpetual and SHORT the other venue's BTC perpetual when their prices diverge meaningfully, capturing the convergence as cross-venue arbitrageurs close the dislocation. RATIONALE driven by the now-definitive session finding: 22+ strategy attempts across 6 mechanism categories have all failed. The analyst's honest final conclusion is 'every mechanism we've tested this session has failed; the only valid pivots are categories that have not been tested at all OR waiting.' The four genuinely-untested categories remaining are: (1) Sub-daily single-instrument intraday on BTC/ETH — partially covered, (2) Cross-VENUE arbitrage (Binance vs Hyperliquid), (3) Expanded cross-sectional momentum (10-20 alts), (4) Volatility-targeting overlays. This proposal directly addresses untested category #2 — the only genuinely-novel mechanism class not yet attempted in this session. KEY STRUCTURAL DIFFERENCES from all 22 failed strategies: (a) Not cross-ASSET pair MR — this is cross-VENUE on the SAME asset (BTC), eliminating the ratio-drift problem that killed 5 pair MR attempts. (b) Not single-instrument directional — fully market-neutral by construction. (c) Not trend-following or cross-sectional momentum. (d) The convergence force is REAL and MECHANICAL: when BTC perp diverges meaningfully between Binance and Hyperliquid, professional arb desks (Jump, GSR, Wintermute, Cumberland) and retail arbitrageurs actively close it within minutes-to-hours. This is a STRUCTURALLY ENFORCED arbitrage, not a statistical pattern. (e) The same asset (BTC) on both legs eliminates the asymmetric long/short win-rate signature that killed all 4 pair MR attempts. MAXIMALLY SIMPLE — 2 parameters (entry spread threshold, exit spread threshold). ZERO supplementary data dependency — both legs use only OHLCV from their respective venues. ADDRESSES ANALYST'S #2 RECOMMENDED UNTESTED CATEGORY EXACTLY.

Iteration-2 fix for the Layer-3 'no trades' failure. The previous version thresholded an ABSOLUTE de-meaned basis percent (entry 0.35%), but the raw cross-venue basis between two BTC perps is only a few basis points, so the transient de-meaned component never reached 0.35% and should_enter never fired. The minimal, targeted change standardizes the de-meaned spread into a z-score (divide by rolling std) and thresholds in sigma (entry_z=2.0, exit_z=0.5), which reliably triggers regardless of the basis's tiny absolute magnitude. Imports, class structure, neutral-pair submission/unwind logic, naked-leg guard, and capital-relative sizing are unchanged, so the previously-passing Layer-1/Layer-2 checks are preserved while the entry condition now activates.

Fundamental fee-vs-spread flaw, not a tunable problem. The Binance-vs-Hyperliquid BTC perp basis is only a few bps, but the strategy pays ~0.19% round-trip across BOTH legs (Binance ~0.10% + HL ~0.09%), so it bleeds on every trade: every day and every month is negative, Sharpe -19.85, PF 0.75, avg_trade_return -$51.58, and commissions are 27.58% of gross. With only 2 parameters (entry_z/exit_z), no optimization can manufacture an edge below trading costs — higher entry_z merely trades less while the captured convergence stays smaller than the double-leg fee. Cross-venue convergence arbitrage at 1H bar granularity with retail taker fees is structurally dead: professional desks close the dislocation at sub-second latency and zero/negative fee tiers before a 1H bar even prints. Not worth 2 hours of optimization compute.

Dollar-neutral cross-VENUE BTC perpetual convergence arbitrage on 1H bars: long the cheap venue and short the rich venue when the Binance-vs-Hyperliquid basis dislocates, capturing the mechanical convergence. The cross-venue basis is de-meaned with a rolling mean and STANDARDIZED by its rolling std into a z-score, so entry/exit thresholds are expressed in sigma units (scale-invariant). Equal USD notional per leg keeps net BTC delta ~0. Only two user knobs (entry_z, exit_z); OHLCV-only, no supplementary data.

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: BTCUSDT.BINANCE-1-HOUR-LAST-EXTERNAL, Bars processed: 56136 Diagnostics: should_enter() returned a side 0 times over 5316 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

Correctly delta-neutral by construction (beta 0.0012, equal-notional both legs); not an inflated-Sharpe artifact

Contemporaneous leg alignment handled (no phantom >0.5% basis; the traded basis is realistically tiny)

Every day and every month is negative; total_return -9.6%, Sharpe -19.85, profit_factor 0.75, expectancy -$51.58/trade

Fee bleed dominates: commission_pct_of_gross 27.58%, $15,439 commissions over 728 legs (2.3 round-trips/day)

Cross-venue BTC-perp basis is a few bps — structurally smaller than the ~0.19% round-trip cost of trading both legs; convergence captured cannot beat costs

Only 2 tunable params (entry_z/exit_z); raising entry_z reduces frequency but never makes a sub-bp spread exceed double-leg fees

Proposed as the analyst's recommended untested 'cross-venue arbitrage' category after 22+ prior session attempts failed, this strategy went dollar-neutral long the cheap and short the rich BTC perp across Binance and Hyperliquid whenever their standardized basis diverged. Although it was genuinely delta-neutral by construction (beta 0.0012), it bled on every trade: across 364 trades it returned -9.6% with a Sharpe of -19.85, profit factor 0.75, every day and month negative, and commissions consuming 27.58% of gross. The analyst abandoned it at the backtest-review gate, identifying a fundamental fee-vs-spread flaw — the few-bps cross-venue basis is structurally smaller than the ~0.19% double-leg round-trip cost, and with only two tunable parameters no optimization could manufacture an edge below trading costs. Cross-venue convergence arbitrage at 1H bars with retail taker fees was judged structurally dead, since professional desks close such dislocations at sub-second latency and near-zero fees before a 1H bar even prints.

The Binance-vs-Hyperliquid BTC perp basis is only a few bps — structurally smaller than the ~0.19% round-trip cost of trading both legs — so cross-venue convergence arbitrage at 1H granularity with retail taker fees is fee-dead and cannot be rescued by tuning entry/exit thresholds.

It was abandoned at the pre-optimization backtest-review gate with an 'abandon' verdict, failing every criterion (Sharpe -19.85<0.5, profit factor 0.75<1.2, commission 27.58%>15%, negative avg trade return); the analyst, risk, and optimization stages were never reached.

A dollar-neutral cross-venue arbitrage on BTC perps that simultaneously longed the cheaper and shorted the richer of the Binance and Hyperliquid contracts whenever their z-scored basis hit |z| >= 2.0 on 1H bars, betting that arbitrageurs would close the dislocation and capture the convergence.

Over the backtest it fired 364 trades (728 legs) with a 45% win rate but lost money continuously: total return -9.6%, Sharpe -19.85 (CI -23.35 to -17.33), profit factor 0.75, negative expectancy (-$51.58/trade), every day and month negative, with commissions at 27.58% of gross PnL ($15,439).