BtcCrossVenueBasisZScoreMeanReversionLSBinanceHl1H

BTC Cross-Venue Price Basis Mean Reversion (BINANCE Perp vs HYPERLIQUID Perp, OHLCV-Only, Long-Short Market-Neutral)

Cross-venue price basis arbitrage on BTC perps: when the price of BTCUSDT on BINANCE diverges from BTCUSD on HYPERLIQUID by more than a threshold (in basis points), enter a delta-neutral hedge that sells the over-priced venue and buys the under-priced venue. Both legs are the SAME underlying asset (BTC) on DIFFERENT venues — their prices should track within ~5-15 bps under normal conditions because arbitrageurs close gaps quickly. When the gap widens beyond ~20-30 bps (temporary liquidity dislocation, retail flow imbalance, latency arbitrage opportunity), the spread mechanically converges as smart money moves to capture it. This hypothesis fills FIVE under-represented portfolio buckets in ONE move: (a) cross-venue scope (4.3% → ≥15% target — STILL the biggest unfilled lever), (b) HYPERLIQUID venue (4.5% → ≥20% target), (c) market-neutral direction (8.9% → reducing 91.1% long-only), (d) pairs scope (14.1% → reducing 79.1% single), (e) different mechanism class than my own funding-carry / pairs-ratio / cross-settlement-funding hypotheses (which are all already in pipeline). CRITICAL DIFFERENTIATORS vs existing in-pipeline strategies: (1) BtcCrossSettlementFundingArbHedged captures FUNDING differential between BINANCE_CM and BINANCE USD-M (same exchange family, different settlement type); THIS captures PRICE basis between BINANCE and HYPERLIQUID (different exchange families). (2) ETH/BTC + SOL/BTC pairs MR capture intra-venue cross-asset ratios; THIS captures cross-venue same-asset basis. The mechanism is mechanically distinct. (3) NO funding data required — pure price-OHLCV — eliminates the supp:fundingRate data-gap failures that killed multiple recent funding-related hypotheses.

Implements the hypothesis exactly: cross-venue (BINANCE vs HYPERLIQUID), same-asset (BTC) PRICE basis mean reversion, market-neutral, OHLCV-only. Mechanically distinct from the in-pipeline funding-carry/cross-settlement-funding hypotheses (no funding data; captures price basis, not a funding differential) and from cross-ASSET ratio pairs (same underlying on both venues, eliminating ratio drift / asymmetric win-rate). Uses a rolling z-score of the bps basis rather than a static threshold so entries adapt to the prevailing basis regime (small structural offset + regime-dependent dispersion) and reliably trigger across the backtest window. Alignment is enforced by the base same-timeframe sync barrier plus an explicit timestamp tolerance, so the spread uses contemporaneous prices (no phantom one-bar basis). Equal-notional opposite legs give a true delta-neutral hedge; kill switch + time stop bound tail risk. leverage left at 1.0 (sizing references self.leverage so no unused-leverage gate), filling the cross-venue, HYPERLIQUID-venue, market-neutral, and pairs portfolio buckets simultaneously.

Structurally fee-dominated and timing-dead, shown directly at the trade level. The 'delta-neutral' legs leave a SYSTEMATICALLY adverse price residual every pair (net -$55 to -$72: 10-13 HL +210 / Binance -268 = -$58; 10-14 +357/-421 = -$64; 10-19 -355/+282 = -$72) — the leg bought as 'cheap' consistently underperforms the leg sold as 'rich', i.e. the basis does NOT mean-revert at 1H bar resolution. This is the stale-basis signature: a cross-venue BTC price gap is arbitraged away in seconds, so acting on a bar-CLOSE z-score enters AFTER the reversion and captures only adverse subsequent noise, then pays ~$47/pair in two-leg commission on top. Result: PF 0.667, omega 0.006, Sharpe -4.41, expectancy -$36.8/trade, negative on essentially every day. Not tunable: lowering entry_z captures rarer/larger gaps that are more likely genuine latency/real dislocations that don't revert, and even a perfectly neutral hedge captures only a few-bps convergence that cannot clear ~0.19% round-trip fees on two legs; there is no robust region to optimize toward. The hypothesis's own premise ('arbitrageurs close gaps quickly') concedes the edge requires sub-second execution that a 1H-bar strategy cannot achieve. Only ~8 months of single-regime HL data anyway. Same wall as this session's BTC cross-venue funding-differential strategies: BTC is the most-arbitraged perp, so cross-venue dislocations — funding OR price basis — are tiny, fast, and fee-dominated. FAILURE PATTERN: cross-venue same-asset price-basis z-score mean reversion on BTC (Binance vs Hyperliquid) at 1H resolution fails because the basis has already converged by the time a bar-close signal can act, leaving a systematically adverse leg residual plus two-leg fees; the cross-venue basis/MR family on a major needs sub-second execution and a thinner-history, higher-dislocation venue/asset, not 1H BTC. The Research Lead should stop proposing 1H-bar cross-venue basis/MR on majors.

Market-neutral cross-venue price-basis mean reversion on BTC perps. PRIMARY leg BTCUSDT.BINANCE (USD-M perp, 1H) drives signal/entry/exit; EXTRA leg BTCUSD.HYPERLIQUID (perp, 1H) is the hedge. Computes basis_bps = (binance_close - hl_close)/hl_close * 10000 each aligned bar and trades its ROLLING Z-SCORE over zscore_window bars. When |z| >= entry_z the basis is mechanically stretched: z>=+entry_z (Binance rich) -> SHORT Binance + LONG HL (primary SELL); z<=-entry_z (Binance cheap) -> LONG Binance + SHORT HL (primary BUY). Closes BOTH legs when |z| <= exit_z (reverted) after min_hold_bars, or on a time stop (max_hold_bars) or combined-PnL kill switch. Equal USD notional per leg => delta-neutral on BTC price, so PnL comes only from convergence of the cross-venue dislocation. Pure OHLCV, no supplementary/funding data. Both legs futures/MARGIN => genuinely long+short.

Cleanly implemented: atomic two-leg equal-notional hedge, z-score signal, sync barrier, no naked legs, fills under-quota cross-venue/HYPERLIQUID/market-neutral/pairs buckets. Pure OHLCV (no funding-data dependency).

Sensible holds (1-13h) so the trade-level economics are a fair, direct test of the edge.

Every 'delta-neutral' pair leaves a large, systematically adverse price residual (net -$55 to -$72 per pair: the leg bought as 'cheap' consistently underperforms the leg sold as 'rich') — the basis does NOT mean-revert at 1H resolution.

Classic stale-basis signature: a cross-venue BTC gap is arbitraged away in seconds, so acting on a bar-close z-score enters after reversion and captures only adverse noise. The hypothesis itself concedes arbitrageurs 'close gaps quickly' — too fast for a 1H-bar strategy.

Two-leg fees (~$47/pair, $9,231 total, 30.5% of gross) compound the bleed; captured convergence (a few bps) is far smaller than ~0.19% round-trip on two legs.

Negative on essentially every day: PF 0.667, omega 0.006, Sharpe -4.41, expectancy -$36.8/trade, return_skew -13.4 / kurtosis 188.

Only ~8 months of single-regime HL data (data_days 203, all returns Oct 2025-Jun 2026) — not optimizable regardless; lowering entry_z captures rarer/larger gaps that are more likely genuine dislocations that don't revert.

~8 months single regime

multi-regime

systematic -$55 to -$72 net price residual/pair

~0 residual

positive

BtcCrossVenueBasisZScoreMeanReversionLSBinanceHl1H traded the Binance-vs-Hyperliquid BTC price basis as a delta-neutral z-score mean reversion, pitched as a pure-OHLCV mechanism distinct from the session's funding-carry failures and filling five quota buckets. The code was clean — atomic equal-notional legs, sync barrier, sensible holds — but the basis does not mean-revert at 1H resolution: every pair left a systematic -$55 to -$72 adverse residual because a bar-close signal acts after the seconds-fast arbitrage, and two-leg fees (30.5% of gross) compounded the bleed to PF 0.67 and Sharpe -4.41. The analyst ruled it the same wall as the BTC cross-venue funding strategies — tiny, fast, fee-dominated dislocations on the most-arbitraged perp — un-tunable and on only ~8 months of single-regime data. It ended after one iteration as abandoned, never advancing to optimization or risk review.

Cross-venue same-asset price-basis mean reversion on BTC at 1H resolution is timing-dead: the gap converges in seconds (the hypothesis's own 'arbitrageurs close gaps quickly' premise concedes it needs sub-second execution), so a bar-close signal catches only adverse residual plus two-leg fees — like the session's cross-venue funding attempts, BTC is too arbitraged; this family needs sub-second execution and a thinner-history, higher-dislocation venue/asset, and 1H-bar cross-venue basis/MR on majors should stop being proposed.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) as structurally fee-dominated and timing-dead — the cross-venue BTC basis is arbitraged away in seconds, so a bar-close z-score enters after reversion and captures only adverse noise plus ~$47/pair two-leg fees — with no tunable region and only a single thin regime of data, so optimization and all later stages were never reached.

A market-neutral cross-venue price-basis mean reversion on BTC perps — trading the rolling z-score of the basis between BTCUSDT.BINANCE (USD-M) and BTCUSD.HYPERLIQUID, shorting the richer venue and longing the cheaper in equal USD notional when |z| crossed an entry threshold and flattening on reversion to fair — a pure-OHLCV (no funding data) two-leg hedge meant to harvest cross-venue dislocation convergence and fill the cross-venue, Hyperliquid, market-neutral, and pairs quota buckets.

Over ~8 months of single-regime overlapping HL data (203 days) and 205 pairs (410 leg-trades) it was decisively negative: profit factor 0.67, omega 0.006, Sharpe -4.41 (CI [-19.7, -3.7]), expectancy -$36.8/trade, total return -5.87%, return skew -13.4 / kurtosis 188, with commission at 30.5% of gross and each 'delta-neutral' pair leaving a systematically adverse -$55 to -$72 price residual.