SolEthRatioZScoreMeanReversionPairs4H

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

A DOLLAR-NEUTRAL CROSS-ASSET RELATIVE-VALUE PAIR strategy: simultaneously LONG SOLUSDT and SHORT ETHUSDT in equal dollar notional, triggered by extreme NEGATIVE z-scores of the SOL/ETH ratio. RATIONALE driven by the now-DEFINITIVE session finding (8+ strategies analyzed, TIER 1 trend-following EMPTY post-walk-forward, every long-only OHLCV directional alt-perp strategy has failed): the only working strategy class identified by the analyst is cross-asset relative-value pairs. Pipeline already has ETH/BTC (cointegrated major pair) and BNB/ETH (cointegrated major pair). Adding SOL/ETH extends pair-trading coverage to a third RELATIONSHIP with distinct underlying drivers: SOL captures L1-throughput/Solana-DeFi/meme-coin-platform narratives, ETH captures DeFi-blue-chip/staking-yield/ETF narratives. These drivers are largely uncorrelated — SOL can outperform ETH during Solana ecosystem rallies (Jupiter, meme coins, mobile staking) while ETH can outperform during DeFi summer or staking-related events. The 10-day z-score captures local mean-reversion of the ratio around its trailing average regardless of the longer-term trend direction. 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), (b) MR on a CO-MOVING ratio (not absolute price — which failed), (c) ZERO supplementary data dependency, (d) NOT a contrarian short on a single instrument, (e) NOT oscillator entry, (f) NOT trend-following on alt-perp, (g) MAXIMALLY SIMPLE — only 2 decision parameters (z-score lookback=60, entry threshold=-1.5). DIFFERENTIATION FROM EXISTING PAIRS: ETH/BTC captures store-of-value vs smart-contract-platform divergence; BNB/ETH captures exchange-token-flows vs DeFi-blue-chip divergence; SOL/ETH captures throughput-L1 vs mature-L1 divergence — three independent edge sources for a robust pair-trading sleeve.

Implements the SOL/ETH relative-value pair, extending the pair-trading sleeve (ETH/BTC, BNB/ETH) to a third independent relationship (throughput-L1 vs mature-L1). ETH is configured as a SAME-4H-timeframe extra leg so the base class's cross-leg alignment barrier defers primary processing until the contemporaneous ETH bar arrives -- the z-score and fills use same-timestamp prices, avoiding the phantom-basis bug from a stale T-1 hedge leg. The ratio (SOL primary / ETH hedge) is stored in a single synced buffer (one append per processed primary bar) so the rolling mean/std stay index-aligned, per the multi-instrument alignment rule. Because the base template only submits the primary leg, the ETH hedge is submitted as a side effect in should_enter and closed in should_exit so both legs move together; equal 40%-equity notional per leg makes the position dollar-neutral with ~80% gross exposure, so leverage stays 1.0 (the leverage_set_but_unused gate does not trigger). The design eliminates the session's failure modes: dollar-neutral (no crypto beta), mean-reversion on a co-moving ratio (not absolute price), zero supplementary data, two decision parameters. calculate_signal returns the continuous z-score in natural units and thresholds compare in matching z-units (entry |z|>=1.5, exit through 0.5, regime stop 3.5), never clipped. Venue is BINANCE USD-M futures because the strategy shorts one leg; round-trip ~0.10%/leg is covered by 1.5-sigma ratio reversions.

Pre-optimization abandon: SOL/ETH ratio mean-reversion loses -92.2% (92.5% max DD) and is negative in every single year (2020 through 2026), with the Sharpe CI entirely below zero and probabilistic_sharpe 0.0096. SOL/ETH is non-stationary (it trends, not mean-reverts), so the z-score fade is wrong in every regime. The hypothesis's 'uncorrelated/independent drivers' framing is self-defeating — pairs MR needs cointegration (a reverting spread), and divergent legs guarantee a trending ratio. Optimization tunes lookback/z-thresholds/stops and cannot create stationarity that doesn't exist, and there is no positive year to tune toward. Premise failure, not a code bug — abandon rather than spend the compute; future pairs must be selected by a cointegration test.

Dollar-neutral SOL/ETH ratio z-score mean-reversion pair on Binance USD-M perpetuals, 4H bars. Primary leg SOLUSDT.BINANCE, hedge leg ETHUSDT.BINANCE. Each aligned 4H close computes ratio = SOL/ETH, a 60-bar (10-day) rolling mean/std, and z = (ratio-mu)/sigma. z <= -1.5 goes LONG SOL + SHORT ETH (long-spread); z >= +1.5 goes SHORT SOL + LONG ETH, equal 40%-equity USD notional per leg (dollar-neutral, ~80% gross). Exits when |z| reverts to <= 0.5, |z| >= 3.5 (regime-break stop), after a 42-bar (7-day) time stop, or a 15% per-leg drawdown. The signal is the continuous z-score (varies every bar); should_enter/exit threshold it in z-units. leverage 1.0.

Code is correct and genuinely dollar-neutral (cross-leg barrier, synced ratio buffer, beta -0.03); large non-sparse sample (828 trades).

Catastrophic across the board: Sharpe -0.98, total_return -92.2%, max_drawdown 92.5%, profit_factor 0.83, negative expectancy (-$93/trade), sharpe_ci_high -0.16 (entire CI below 0), probabilistic_sharpe 0.0096.

EVERY year is negative (2020 -8.4%, 2021 -47.6%, 2022 -19.4%, 2023 -15.1%, 2024 -18.0%, 2025 -15.5%, 2026 -0.7%) — persistent loss across all regimes, not a single blowup.

Structural premise failure: SOL/ETH is non-stationary (it trends, does not mean-revert). The hypothesis markets 'largely uncorrelated drivers' as a feature, but that is exactly what makes the ratio trend — pairs MR requires cointegration, not divergence.

High exposure (112%) and turnover (40x) means it is continuously short the trend and bleeding fees (commission 2.98% of gross).

Not worth 2 hours of optimization. The strategy loses -92.2% with a 92.5% max drawdown and is negative in EVERY year (2021 -47.6% down to 2026 -0.7%), with the Sharpe confidence interval entirely below zero (ci_high -0.16) and probabilistic_sharpe 0.0096. The root cause is structural, not a tuning issue: the SOL/ETH ratio is non-stationary — SOL repeatedly out- and under-performs ETH in sustained trends — so fading every z-score deviation stands in front of a trend that keeps running. The hypothesis's own selling point ('largely uncorrelated drivers / independent edge sources') is the disqualifier: a pairs mean-reversion trade requires the spread to be cointegrated/reverting, whereas divergent drivers guarantee a trending ratio. Optimization only adjusts lookback/z-thresholds/stops and cannot make SOL/ETH stationary, and there is no positive year anywhere for it to tune toward. This is a premise failure, not a code bug, so it is not a developer iterate. If the pairs sleeve is to grow, pairs must be admitted by an actual cointegration/stationarity test, not assembled from deliberately uncorrelated legs.

SolEthRatioZScoreMeanReversionPairs4H was pitched as the next leg of a relative-value pairs sleeve after directional alt-perp strategies failed session-wide, fading extreme z-scores of the SOL/ETH ratio in a dollar-neutral book. The code was correct and genuinely market-neutral and traded densely (828 trades), but it lost 92.2% with a 92.5% drawdown and was negative in every single year across all regimes. The root cause was that SOL/ETH is non-stationary — SOL repeatedly out- and under-performs ETH in sustained trends — so the very 'uncorrelated drivers' the hypothesis marketed as independent edge sources were exactly what made the ratio trend instead of revert. The analyst abandoned it before optimization as a premise failure rather than a code bug, echoing the ADA/DOT pairs outcome and recommending that future pairs be selected by a cointegration test instead of assembled from deliberately divergent legs.

Pairs mean-reversion requires a cointegrated, reverting spread, so 'largely uncorrelated / independent drivers' is a disqualifier rather than an edge — divergent legs guarantee a trending ratio, and candidate pairs must be admitted by an actual cointegration/stationarity test, not assumed.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) for a structural premise failure: the SOL/ETH ratio is non-stationary and trends rather than mean-reverts, so fading every z-score deviation continuously stood in front of a running trend — a problem optimization cannot fix since it only tunes thresholds and there was no positive year to optimize toward.

A dollar-neutral SOL/ETH relative-value pair on 4H bars that went long SOL + short ETH (and vice versa) in equal notional when the 60-bar (10-day) rolling z-score of the SOL/ETH ratio breached ±1.5, betting on mean reversion of the ratio with a 0.5 z-exit target, 3.5 z regime-stop, 7-day time stop, and 15% per-leg drawdown stop.

Over 1,407 days it took 828 trades (genuinely dollar-neutral, beta -0.03) with a 48% win rate but deeply negative results: Sharpe -0.98 (entire CI below zero, ci_high -0.16), profit factor 0.83, expectancy -$93/trade, a -92.2% total return, and a 92.5% max drawdown — and it was negative in every single year from 2020 (-8.4%) through 2026 (-0.7%), including 2021 at -47.6%.