EthBtcRatioSpreadMomentumTrendFollowingLongShortDaily

ETH/BTC Ratio Spread Momentum (Trend-Following the Persistent Dominance Drift) — Dollar-Neutral Long/Short, Daily

A dollar-neutral LONG/SHORT spread strategy on the ETH/BTC ratio using ETHUSDT.BINANCE (primary) and BTCUSDT.BINANCE (hedge), both BINANCE USD-M perpetual futures, on DAILY bars. This is the deliberate INVERSION of the just-failed 'ETH-BTC Ratio Z-Score Mean-Reversion' hypothesis: that strategy was abandoned because the log-ratio is NON-STATIONARY and DRIFTS persistently with BTC-dominance regimes (the analyst's exact words: 'the ETH/BTC ratio has trended down persistently since 2021 as BTC dominance rose, so z-score reversion keeps buying a ratio that keeps falling'). Rather than fade that drift, this strategy TREND-FOLLOWS it. The drift that destroys mean-reversion is precisely the tradable property a momentum spread harvests. Mechanism: compute the log of the ETH/BTC price ratio at each daily close, measure its smoothed momentum over a lookback window, and take a dollar-neutral position aligned with the momentum sign — when ETH is structurally outperforming (ratio rising) go long ETH / short BTC in equal dollar notional; when BTC dominance is rising (ratio falling) go short ETH / long BTC. The trade carries the macro dominance regime, not intraday noise. Only ~3 tunable parameters (momentum lookback, entry magnitude threshold, exit/flip threshold) to minimize overfitting, which has been a recurring abandonment cause. Both legs sit on a single venue so there is no cross-denomination or data-span risk; OHLCV-only, no funding/tick/options data (avoiding the funding-2023-cutoff, untestable-tick-engine, and untestable-options failure classes documented this session).

Implements the deliberate inversion of the failed ETH/BTC z-score mean reversion: because log(ETH/BTC) is non-stationary and drifts with BTC-dominance regimes, this trend-follows the drift instead of fading it. Uses the proven centralized multi-leg pattern with the repo SyncedBuffer helper, which commits an (ETH, BTC) sample only when both legs report the same ts_ns, so the log-ratio and its momentum are always computed on contemporaneous daily closes; the base cross-leg barrier (both legs 1D) ensures calculate_signal fires only when the BTC bar for the current ETH timestamp is present. Smoothed momentum is the difference of two short (3-bar) SMAs of the log-ratio momentum_lookback bars apart, reducing daily noise while capturing the macro drift. A hysteresis band (enter at |mom|>=entry_threshold, flatten at |mom|<=exit_threshold, hold otherwise) carries the dominance regime rather than whipsawing on small daily moves. Each leg is reconciled to its dollar-neutral target (ETH = sign*leg_frac*equity, BTC = -sign*leg_frac*equity) with a single delta order on the netting account, so a long-spread<->short-spread flip is one order per leg and gross never transiently double-ups; leg_frac=0.45 keeps gross ~90% within the leverage-1.0 futures margin ceiling (leverage referenced only as a non-amplifying multiplier, so no unused-leverage gate). The min_notional delta threshold suppresses sub-$100 churn as equity drifts. min_bars_required=2 lets the buffer fill while momentum waits internally for lookback+SMOOTH aligned samples, so no trade fires on partial history; the returned momentum is a continuous per-bar signal so Layer-2 frozen-signal detection passes while position changes stay threshold-gated. Both legs are liquid majors with full multi-year daily history on a single venue, so no data-span or data_unavailable risk; should_enter/should_exit/position_size are inert because the base entry path manages only the primary instrument.

ETH/BTC daily ratio-momentum (trend-following the dominance drift) correctly inverts the failed MR sibling — it is POSITIVE (+28.5%, PF 1.05) where MR was negative (-22.5%), confirming the ratio trends rather than reverts — but the edge is too thin and too concentrated to survive optimization. Base Sharpe is 0.21 with sharpe_ci_low -0.55 (CI straddles 0) and ~4.4% CAGR, and the entire return rests on a handful of days (return_kurtosis 476, skew +16.9, dominated by 2021-06-10 at +42.85% in a single day). This is a deflated-Sharpe crusher: the DSR skew/kurtosis adjustment will drive significance toward 0, exactly how the abandoned fat-tail trend siblings died post-optimization (SOL Keltner kurtosis 124, NEAR Supertrend 175) — and this is kurtosis 476, far worse, from a CI that already includes 0. The concentration is structural, not tunable: 6 years of daily data contain only a few independent ETH/BTC dominance regimes (2021 alt-season, 2022 bear, 2024 BTC-dominance, 2025 ETH-recovery), so a momentum spread has inherently few independent bets and lumpy, regime-fragile returns (2023 -10.9%, 2024 -22.2%, max_DD 35.5% with ci_high 62% over a 1,306-day drawdown). No lookback/threshold tuning manufactures more trends. Not optimize: best-of-225 would fit the few big trend days and fail the holdout/deflation — low-EV for 2 hours. Not iterate: there is no bug; the thin-outlier edge is intrinsic to a low-N daily ratio-momentum bet. FAILURE PATTERN: trend-following a daily crypto ratio spread produces a positive but outlier-dependent return (kurtosis >400, one day carrying the result) with a Sharpe CI through 0 — the convexity is real but too concentrated and low-N to clear deflated-Sharpe, the same fat-tail-trend signature that fails post-opt. The Research Lead should note the inversion confirmed the ratio trends (a useful finding), but a daily ETH/BTC momentum spread does not have enough independent trends to be a deployable standalone edge; a higher-frequency or multi-pair trend basket would be needed to diversify the tail concentration.

Dollar-neutral long/short ETH/BTC ratio momentum (trend-following) spread on ETHUSDT (primary) and BTCUSDT (hedge) BINANCE USD-M futures, daily bars. On each aligned daily bar it commits the (ETH, BTC) close pair to a SyncedBuffer, forms log(ETH/BTC), and measures smoothed momentum as the difference of two 3-bar SMAs of the log-ratio 30 bars apart. When momentum >= +0.03 it goes long ETH / short BTC; when <= -0.03 it goes short ETH / long BTC; it flattens when |momentum| <= 0.005 and otherwise holds the current regime (hysteresis). Each leg is moved to its dollar-neutral target with a single delta order. Pure OHLCV, single venue, leverage 1.0 (~90% gross). Three tunable parameters: momentum_lookback, entry_threshold, exit_threshold.

Inversion logic validated: momentum version is positive (+28.5%, PF 1.05) where the MR sibling was negative (-22.5%) — confirms the ETH/BTC ratio trends, not reverts (informative for the Research Lead)

Genuinely market-neutral (beta -0.03); non-inverted trend payoff (avg_win $7,698 > avg_loss $6,924, positive skew); clean same-venue two-leg build; no capacity wall (capacity $12.9M, impact 8.8%)

Thin, insignificant edge: Sharpe 0.21, sharpe_ci_low -0.55 (CI straddles 0), ~4.4% CAGR over 6.4 years

Extreme outlier concentration: return_kurtosis 476, skew +16.9 — the entire return rests on a few days (2021-06-10 +42.85% dominant); a DSR-crusher that will fail post-opt deflation as the kurtosis-124/175 trend siblings did

Regime-fragile and lumpy: 2022 +29.9%, 2023 -10.9%, 2024 -22.2%, 2025 +32.3%; max_DD 35.5% (ci_high 62%), 1,306-day drawdown — whipsaws in non-trending ratio regimes

Structurally low-N: only a few independent ETH/BTC dominance regimes in 6 years, so the outlier concentration is intrinsic and not optimizable away

EthBtcRatioSpreadMomentumTrendFollowingLongShortDaily deliberately inverted the just-abandoned ETH/BTC mean-reversion, trend-following the persistent BTC-dominance drift that had destroyed the reversion bet. The inversion worked in sign — +28.5% versus the sibling's -22.5%, confirming the ratio trends rather than reverts — but the edge was a thin Sharpe of 0.21 with a CI through zero, carried almost entirely by a few days (kurtosis 476, a single +42.85% day), and lumpy across regimes (2024 -22.2%) with a 35.5% drawdown. The concentration was structural: too few independent dominance regimes in six years for a daily momentum spread to be a deployable standalone edge. The analyst abandoned it at the backtest-review gate as a deflated-Sharpe crusher in the same fat-tail-trend family as the abandoned SOL Keltner and NEAR Supertrend, noting the confirmed-trend finding for the Research Lead but recommending a higher-frequency or multi-pair basket to diversify the tail.

Trend-following a daily crypto ratio spread can confirm the ratio trends (a useful finding versus the failed mean-reversion) yet still fail as a standalone edge — with only a few independent dominance regimes in six years, the return is intrinsically lumpy and outlier-dependent (kurtosis >400, one day carrying it), too low-N and tail-concentrated to clear deflated-Sharpe; diversifying the tail would require higher frequency or a multi-pair trend basket.

The analyst abandoned it at the backtest-review gate before optimization because the edge was too thin and too concentrated to survive deflation: a Sharpe CI already through zero plus kurtosis 476 (the entire return on a handful of days) is a deflated-Sharpe crusher, and the concentration is structural — six years of daily data hold only a few independent ETH/BTC dominance regimes, so a daily ratio-momentum spread has inherently few independent bets that no lookback/threshold tuning can multiply.

A dollar-neutral, OHLCV-only long/short momentum spread on the ETH/BTC ratio using daily Binance USD-M futures — the deliberate inversion of the just-failed ETH/BTC z-score mean-reversion: rather than fading the persistent BTC-dominance drift, it measured smoothed momentum of the log(ETH/BTC) ratio over a 30-day lookback and took a dollar-neutral position aligned with the drift's sign (long ETH/short BTC when the ratio rises, the reverse when it falls), with 3 tunables.

The inversion logic validated directionally — the momentum version was positive (+28.5% total, CAGR 3.6%, PF 1.05, non-inverted payoff with avg win > avg loss) where the mean-reversion sibling was negative — and it was genuinely market-neutral (beta -0.03) with no capacity wall. But the edge was thin (Sharpe 0.21 with a CI low of -0.55 straddling zero), extremely outlier-concentrated (kurtosis 476, skew +16.9, one day at +42.85% dominating), and regime-fragile with a 35.5% drawdown (CI high 62%) over a 1,306-day stretch.