EthBtcRatioMeanReversionPairsLS4H

ETH BTC Ratio Mean Reversion Pairs Trade

Market-neutral pairs trade exploiting the mean-reverting nature of the ETH/BTC price ratio. ETH and BTC are ~85% correlated but their ratio oscillates within regimes — when ETH outperforms BTC sharply (ratio spikes), it tends to revert, and vice versa. The strategy goes long the underperformer and short the outperformer when the ratio deviates significantly from its rolling mean. This provides true diversification from the existing pipeline — it is market-neutral (profits regardless of crypto direction), uses a different signal source (cross-asset ratio), and is uncorrelated with both trend-following and funding-rate strategies. In the current extreme fear environment, correlations spike then break down, creating frequent ratio dislocations as ETH and BTC respond differently to catalysts (e.g., ETH staking ETF inflows vs BTC ETF outflows).

Implements the ETH/BTC ratio mean-reversion thesis on the proven SyncedBuffer pairs pattern already validated in the portfolio. The two legs are kept timestamp-aligned (a sample commits only when both ETH and BTC report the same ts_event), so the log-ratio and z-score are computed on contemporaneous prices — avoiding the independent-trim desync bug that corrupts naive two-array pair strategies. The z-score (returned in natural units, not clipped, so it can exceed the >1 entry threshold) drives a symmetric long/short with equal-dollar legs for true dollar-neutrality, making the P&L pure relative value and uncorrelated with the book's trend and funding strategies. Risk is bounded three ways — z-reversion exit (hysteresis: z_exit < z_entry), a regime-break stop that refuses to open and force-closes when |z|>=z_stop (the cointegration has broken), an emergency raw-ratio stop that catches blow-ups before z updates, and a time stop. Both legs settle on one BINANCE USD-M MARGIN account (matched fees, deepest liquidity, both in the known-precision list at 0.001 step). Leverage is left at 1.0 — a hedged market-neutral book does not need margin amplification and the MARGIN account already permits the short leg at 1x — so the sizing's self.config.leverage reference is present but the inert-leverage gate (which only fires for leverage>1) does not apply.

Consistently money-losing with no edge to optimize: Sharpe -1.51 with the ENTIRE confidence interval below zero (-2.25 to -0.78), probabilistic_sharpe 0.0001, profit_factor 0.816, total_return -60.5%, expectancy -95, and negative in 6 of 7 years (only 2026 +1.67%) across an ample 616-trade / 6.5-year sample. The book is genuinely market-neutral (beta -0.013) but reliably negative -- the ETH/BTC log-ratio z-score fade has no edge on 4H because the ratio undergoes long secular regimes (it trends for months), so the |z|>2 fade is repeatedly run over (return_skew -2.37, tail_ratio 0.60, omega 0.58 -- small wins, fat left tail), and the 4H churn adds fee drag (commission 4.22% of gross). This is a dead mechanism on this pair, not a tuning/sizing problem -- and the identical ETH/BTC ratio-MR construction was already abandoned on its HL 4H sibling. No parameter optimization converts a 0.816-PF strategy with a wholly-negative Sharpe CI into an edge; tuning z-thresholds only shifts where the trending ratio runs you over. Do not spend optimization compute. FAILURE PATTERN: ETH/BTC ratio mean-reversion (z-score fade) on intraday timeframes loses consistently because the major-pair ratio trends rather than oscillates around a stable mean -- the fade gets run over every regime; market-neutral does not rescue a no-edge spread.

Dollar-neutral ETH/BTC ratio mean-reversion pairs trade on Binance USD-M futures, 4H bars, pure OHLCV. It maintains a timestamp-aligned (ETH, BTC) history via SyncedBuffer, forms log(ETH/BTC), and computes its z-score over a rolling 60-bar window. When the ratio is rich (z>=+2.0) it shorts ETH and longs BTC; when cheap (z<=-2.0) it longs ETH and shorts BTC — equal notional per leg (market-neutral). Exits on reversion inside |z|<0.3, a regime-break stop (|z|>=3.5), an emergency 10% ratio move, or a 42-bar time stop. Per-leg notional = equity * per_leg_pct * leverage; leverage 1.0.

Genuinely market-neutral and clean: beta -0.013, benchmark_correlation -0.10, benchmark_meaningful=FALSE, balanced 308L/308S, 0 dropped, metrics_reliable=true

Ample sample (616 trades / 6.5 years) -> the negative result is statistically reliable, not thin-sample noise

Decisively money-losing: Sharpe -1.51 with ENTIRE CI negative (-2.25 to -0.78), probabilistic_sharpe 0.0001, profit_factor 0.816, total_return -60.5%, expectancy -95

Negative in 6 of 7 years (only 2026 +1.67%); rolling Sharpe negative across nearly the whole history

Wrong payoff shape for a fade: return_skew -2.37, tail_ratio 0.60, omega 0.58 -> the ETH/BTC ratio trends through the z-stop rather than mean-reverting

Same ETH/BTC ratio-MR mechanism was already abandoned on the HL 4H sibling; 4H churn adds fee drag (commission 4.22% of gross)

1 of 7

majority

EthBtcRatioMeanReversionPairsLS4H pitched a clean, market-neutral relative-value trade — fading z-score dislocations in the ETH/BTC ratio — as true diversification from the book's directional strategies. The construction delivered on neutrality (beta -0.013, balanced legs, no artifacts), but the edge was decisively absent: -60.5% return, a -1.51 Sharpe with a wholly-negative CI, profit factor 0.816, and losses in 6 of 7 years over 616 trades. The cause was a falsified premise — the ETH/BTC ratio trends through months-long regimes rather than mean-reverting, so the fade is run over with a fat left tail while 4H churn adds fees. The reviewer abandoned it pre-optimization as a dead mechanism on this pair, noting the identical construction had already failed on its Hyperliquid 4H sibling and that no z-threshold tuning can convert a 0.816-PF trending-ratio fade into an edge.

ETH/BTC ratio mean-reversion (z-score fade) on intraday timeframes loses consistently because the major-pair ratio trends rather than oscillates around a stable mean — the fade is run over every regime — and a genuinely market-neutral construction cannot rescue a no-edge spread; tuning z-thresholds only shifts where the trending ratio runs you over.

The analyst abandoned it at the backtest-review gate as a reliably negative, no-edge result: the ETH/BTC log-ratio trends through long secular regimes rather than oscillating around a stable mean, so the |z|>2 fade gets repeatedly run over (skew -2.37, small wins / fat left tail) with 4H churn adding fee drag — and the identical ETH/BTC ratio-MR construction had already been abandoned on its Hyperliquid 4H sibling.

A market-neutral, dollar-neutral pairs strategy on the ETH/BTC ratio (ETHUSDT primary, BTCUSDT hedge, Binance USD-M, 4H) that z-scores log(ETH/BTC) over a rolling window and fades dislocations — long the underperformer / short the outperformer at |z|>=2, exiting on reversion inside ±0.3, a |z|>3.5 regime-break stop, an emergency ratio move, or a time stop — aiming for true diversification uncorrelated with the book's trend and funding strategies.

It was decisively money-losing over an ample 616-trade / 6.5-year sample: Sharpe -1.51 with the entire CI negative (-2.25 to -0.78), probabilistic_sharpe 0.0001, profit factor 0.816, -60.5% total return, expectancy -$95/trade, 46.6% win rate, 62.3% max drawdown, return_skew -2.37, negative in 6 of 7 years — while genuinely market-neutral (beta -0.013, benchmark_correlation -0.10, balanced 308L/308S, metrics_reliable=true).