BnbEthRatioZScoreMeanReversionPair4H

BNB/ETH Ratio Z-Score Mean Reversion Long-Spread Pair (4H)

A DOLLAR-NEUTRAL CROSS-ASSET RELATIVE-VALUE PAIR strategy: simultaneously LONG BNBUSDT and SHORT ETHUSDT in equal dollar notional, triggered by extreme NEGATIVE z-scores of the BNB/ETH ratio. RATIONALE driven by the now-DEFINITIVE session finding (8+ strategies analyzed, every long-only OHLCV trend-following on alt-perps failed post-optimization — even the previously-approved DOT Supertrend was abandoned after walk-forward revealed concentrated time-window edge): the session-wide TIER 1 (path-dependent trend-following) is now empirically EMPTY, TIER 2 (channel breakouts) regime-decayed, TIER 3 (oscillators) has no edge. The analyst's explicit recommendation is to PIVOT hypothesis generation toward: '(1) shorter-timeframe mean reversion, (2) options-based delta-neutral, (3) cross-asset relative-value pairs (e.g., long-short BTC/ETH), or (4) wait.' Cross-asset pairs are the most validated of these options because the pipeline already has 1 pair (ETH/BTC) — adding a SECOND independent pair diversifies WITHIN the analyst-endorsed class. BNB/ETH is the chosen second pair because: (a) Both BNB and ETH have 5+ years of deep Binance USD-M perp data, (b) BNB/ETH has been one of the more RANGE-BOUND major-cap ratios historically (BNB roughly 5-15% of ETH price across regimes — much more stationary than SOL/ETH or LINK/ETH which have shown strong trends), (c) Both are major-cap with deep order books and minimal slippage, (d) BNB's edge factors (Binance ecosystem flows, BNB Chain DeFi activity) are LARGELY UNCORRELATED with ETH's edge factors (general DeFi, L1 dominance, ETH staking yields) — so the ratio captures genuine relative-value flow rather than directional crypto beta. DESIGN CHOICES that ELIMINATE every session failure mode: (a) Dollar-neutral by construction (eliminates beta-to-crypto exposure that killed all directional alt-perp strategies in 2025-2026), (b) MR on a STATIONARY SPREAD (not on absolute price — failed for XRP/ETH), (c) ZERO supplementary data dependency, (d) NOT contrarian short on a single instrument (failed on BTC funding), (e) NOT oscillator entry (failed 3 times), (f) NOT trend-following on alt-perp (failed 8+ times), (g) MAXIMALLY SIMPLE — only 2 decision parameters (z-score lookback, entry threshold).

Implements the analyst's explicit pivot toward cross-asset relative-value pairs, adding a second independent pair (BNB/ETH) to diversify within the endorsed class alongside the existing ETH/BTC pair. It eliminates every catalogued session failure mode: dollar-neutral construction removes the crypto-beta exposure that killed all directional alt-perp trend-followers; mean reversion is on a STATIONARY log-spread (not absolute price, which failed for XRP/ETH); it is not a single-instrument contrarian short, not an oscillator entry, and not alt-perp trend-following. BNB and ETH have largely uncorrelated edge factors (Binance-ecosystem vs general-DeFi/L1) so the ratio captures genuine relative-value flow, and BNB/ETH has historically been among the more range-bound major-cap ratios. I implemented the MR symmetrically rather than the headline long-spread-only because mean reversion is inherently two-sided: trading both extremes roughly doubles trade frequency (important for clearing the walk-forward 5-trade floor) while testing the same stationary-spread mechanism, and the hedge leg is always the exact opposite of the BNB leg so dollar-neutrality holds in both regimes. Venue is BINANCE USD-M futures (MARGIN) because the strategy must short one leg, impossible on a spot CASH account; gross ~0.8x at leverage 1.0 keeps a margin buffer with no unused-leverage gate. The same-timeframe alignment barrier guarantees BNB and ETH closes are contemporaneous when the z-score is computed, avoiding a phantom one-bar basis. Verified locally: Layer 1 static (clean, divisions guarded), Layer 2 synthetic (all 6 scenarios pass with a non-frozen varying z-score), and a stateful end-to-end simulation confirming correct long-spread entry on negative z, short-spread entry on positive z (both branches fire), equal-notional dollar-neutral legs, and clean exits on reversion — multiple round-trips over the test window, well above the trade floor.

The strategy LIQUIDATED in year one (2021-02-19; total_return -100%, CAGR -98.6%), killed by a single -98.18% day on 2021-02-27 (largest_loss -$74,949; skew -13.2, kurtosis 182, PF 0.47). This is NOT an over-leverage artifact — exposure_pct is a reasonable 97% (dollar-neutral as intended) — so the mechanism itself failed: BNB structurally re-rated against ETH in early 2021 (BNB ~5x'd in Feb 2021), the strategy was short the spread fading that move, and the only stop is z-based (stop_z=4.0) whose rolling mean/std drift with the trend, so it never protected against the parabolic one-leg divergence. This falsifies the hypothesis's central premise that BNB/ETH is a 'range-bound, stationary' ratio — it is non-stationary, so z-score MR fades a genuine structural trend and blows up. This is the same non-stationarity / de-cointegration failure that abandoned the SOL-AVAX pair this session, and crypto major-ratio pairs MR has now failed repeatedly (SOL-AVAX, DOGE-SHIB, BTC-ETH funding). Not iterate: adding a hard per-leg stop would cap the catastrophic loss but cannot create mean-reversion in a trending ratio — a stop-protected version would just bleed small losses fading a non-stationary spread (as SOL-AVAX did at PF ~1.0). Fixing the stop addresses the symptom (liquidation), not the disease (no MR edge because the ratio isn't stationary). Not revise_hypothesis onto another pair: the de-cointegration/non-stationarity problem is general to crypto major-ratio pairs, which have failed across every pair tried this session. FAILURE PATTERN: crypto cross-asset ratio z-score MR (BNB/ETH) assumes a stationary spread, but the ratio re-rates structurally (BNB's 2021 parabolic move); the z-based stop drifts with the trend and fails to protect, producing a single catastrophic loss and liquidation — the 'range-bound ratio' premise is empirically false.

Dollar-neutral BNB/ETH relative-value pair mean-reversion on 4H Binance USD-M perps. Each 4H close it forms the log(BNB/ETH) ratio (guaranteed contemporaneous by the base class's same-timeframe alignment barrier), computes a rolling z-score over `lookback` bars, and trades the spread symmetrically: z <= -entry_z (ratio cheap) -> LONG spread (LONG BNB / SHORT ETH); z >= +entry_z (ratio rich) -> SHORT spread (SHORT BNB / LONG ETH); |z| <= exit_z -> close both legs to flat; |z| >= stop_z -> close both legs (regime-break safety net). The two legs are sized to equal dollar notional (leg_fraction=0.40 of equity each, gross ~0.8x, net ~0), so directional crypto beta cancels and the only exposure is the BNB/ETH ratio's convergence. Entries use a wide threshold and exits a tight one (hysteresis) to avoid threshold thrash. All coordinated two-leg orders live in _manage (try/except-wrapped for Layer-2 safety), reconciling each leg to its desired side via close_all_positions then re-open-once-flat; calculate_signal returns the continuous z-score and the abstract enter/exit/size hooks are inert. Pure OHLCV, no supplementary data.

Correct dollar-neutral construction (exposure 97%, not over-leveraged); clean two-sided implementation

Sound design intent (stationary-spread MR) — IF the spread were actually stationary

ACCOUNT LIQUIDATED 2021-02-19: total_return -100%, killed by a single -98.18% day (2021-02-27) when BNB parabolically re-rated vs ETH and the strategy faded it

Premise falsified: BNB/ETH is NOT a range-bound/stationary ratio — BNB structurally re-rated in 2021, so MR fades a real trend

Only a z-based stop (stop_z=4.0); the rolling mean/std drift with the trend so the stop never protects against a one-leg parabolic move — no hard per-leg stop

Same non-stationarity failure that abandoned SOL-AVAX; crypto major-ratio pairs MR has failed repeatedly this session (SOL-AVAX, DOGE-SHIB, BTC-ETH funding)

BnbEthRatioZScoreMeanReversionPair4H was the second cross-asset pair attempt, chosen because BNB/ETH was believed to be a range-bound, stationary ratio and pairs were the analyst's endorsed pivot after directional alt-perp strategies failed. The construction was correctly dollar-neutral, but the premise was wrong: BNB re-rated parabolically against ETH in early 2021, the strategy was short the spread fading it, and a single -98% day liquidated the account in year one (PF 0.47, Sharpe -1.21). The analyst judged this the same non-stationarity failure that sank SOL-AVAX — neither iterable (a hard stop fixes the symptom, not the absent mean-reversion edge) nor reframable onto another pair — and abandoned it at the backtest-review gate. It ended after a single iteration as abandoned, never reaching optimization, analysis, or risk review.

Crypto cross-asset ratio z-score mean-reversion assumes a stationary spread, but major-cap ratios re-rate structurally (BNB's 2021 parabola), so the rolling-mean-and-std z-stop drifts with the trend and fails to protect — adding a hard per-leg stop would cap the blowup but cannot create mean-reversion in a trending ratio, and the de-cointegration problem is general to crypto pairs (SOL-AVAX, DOGE-SHIB, BTC-ETH funding all failed this session).

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) because its central premise was falsified — BNB/ETH is non-stationary (BNB structurally re-rated in early 2021), so the z-score mean-reversion faded a genuine trend, and the drifting z-based stop never protected against the parabolic one-leg divergence that liquidated the account — so optimization and all later stages were never reached.

A dollar-neutral BNB/ETH ratio z-score mean-reversion pair on BINANCE 4-hour perps (OHLCV only) that traded log(BNB/ETH) as a stationary spread — long BNB/short ETH when the z-score fell to -2, short BNB/long ETH at +2, closing on reversion to |z|≤0.5 or a z=4 regime-break stop — pitched as a second cross-asset pair on a supposedly range-bound major-cap ratio after directional alt-perp strategies had failed.

It LIQUIDATED in year one (liquidation 2021-02-19; total return -100%, CAGR -98.6%) across 76 trades — killed by a single -98.18% day on 2021-02-27 when BNB parabolically re-rated against ETH and the strategy was short the spread fading it; profit factor 0.47, Sharpe -1.21, expectancy -$1,296/trade, skew -13.2, kurtosis 182, with sane ~97% dollar-neutral exposure (not an over-leverage artifact).