ETH/BTC Ratio Daily Trend-Following Long-ETH with EMA-Cross Filter

A long-only MULTI-INSTRUMENT CROSS-ASSET RELATIVE-STRENGTH strategy that trades ETHUSDT perpetual futures on Binance based on the ETH/BTC PRICE RATIO trend. This is the FIRST multi-instrument strategy in the portfolio — every existing strategy uses a single-instrument signal (price patterns, oscillators, breakouts on one asset). NONE use cross-asset relative strength. The mechanism is fundamentally orthogonal: the signal is derived from the RATIO of two asset prices (ETH/BTC), not either asset's absolute price. When the ETH/BTC ratio trends upward, ETH is outperforming BTC — the canonical 'alt-season' or 'ETH-cycle' regime documented in crypto markets since 2018. The trigger uses an EMA crossover ON THE RATIO ITSELF (not on either underlying): when the 50-day EMA of the ETH/BTC ratio crosses above the 200-day EMA, enter long ETH. Exit when the ratio EMA crosses back down OR ETH itself breaks below its own 200-day SMA (safety filter). This is mechanistically distinct from every existing strategy: (a) no candlestick or single-bar pattern (the ratio is a smooth time series), (b) no single-asset breakout (entry depends on cross-asset behavior), (c) no oscillator extreme (we follow ratio trend, not extremes), (d) no calendar effect. The signal captures a documented economic phenomenon: ETH-cycles where ETH outperforms BTC are driven by structurally different flows (DeFi adoption waves, ETF approval cycles, EIP releases, staking dynamics) than BTC-driven cycles (macro flows, ETF rebalances, halving anticipation). Trading the cross-asset relative trend captures this 'alt-rotation alpha' that no single-instrument strategy can isolate. The multi-instrument architecture uses ETHUSDT.BINANCE as primary (where positions are taken) and BTCUSDT.BINANCE as the secondary instrument (used only for ratio computation). Only 4 explicit parameters (short EMA period, long EMA period, ratio computation, safety SMA period) — minimum-parameter relative-strength strategy.

First cross-asset relative-strength strategy in the portfolio. Because only ETH is ever traded (BTC is reference-only for the ratio), the standard single-instrument FactoryStrategy hooks suffice -- no centralized multi-leg management is needed; the base cross-leg alignment barrier (both legs are 1-DAY) guarantees the BTC bar for the current ETH timestamp is present before calculate_signal runs, so the ratio is always contemporaneous. The repo's SyncedBuffer helper commits an (ETH, BTC) sample only when both legs report the same ts_ns, eliminating the independent-array-trim misalignment bug; the EMA helper is an instance method (not a staticmethod) to avoid the recurring Layer-2 synthetic-proxy TypeError where self-called staticmethods receive an extra self. The signal is the relative EMA spread (ema_short - ema_long)/ema_long once 200 ratio samples have accrued, and during warmup it returns the ratio's 1-bar momentum so the signal varies on every bar across all Layer-2 scenarios (which never reach 200 samples) -- avoiding a frozen-signal failure -- while the actual entry/exit decisions are gated on self._ready and the EMA-cross/SMA booleans. Entry requires the ratio uptrend AND ETH above its own 200-day SMA (faithful to the hypothesis's safety filter, applied at entry too so it never buys into an ETH downtrend), and exit fires on the ratio cross-down or ETH dropping below its SMA. min_bars_required=2 lets the buffer fill immediately. Sizing is capital-relative (95% of equity notional for a long-only trend leg) clearing the ETHUSDT $20 min-notional; trend captures over multi-week ETH-cycles dwarf the ~0.10% round-trip futures fee. Leverage stays 1.0 referenced only as a non-amplifying sizing multiplier so the unused-leverage gate does not apply. Long-only, single traded leg, daily cadence -> small Optuna surface (3 numeric tunables) avoiding the walk-forward overfitting seen in higher-parameter strategies.

ETH/BTC ratio EMA-cross trend-following long-ETH has a NEGATIVE realized edge masked by an MTM-illusory headline. end_unrealized_pct is 204.77 against a total_return of +201%, meaning the entire +201% is a single un-exited open ETH position marked-to-market at backtest end, while the realized track record loses money (profit_factor 0.99, expectancy -$141/trade). The result is also outlier-carried (return_kurtosis 84.9, tail_ratio 0.033, dominated by the 2023-02-05 +107% day) on just 26 trades, with Sharpe 0.30 and sharpe_ci_low -1.38 (CI straddles 0), and carries catastrophic risk (max_drawdown 60.6% with ci_high 99%, annualized_volatility 139.6%, 2022 -78.5%). There is also a sizing flag: avg_position_pct 188.7% vs an intended position_fraction of 0.95 (~2x target), which inflated the swings. Not optimize: a negative-realized-edge, MTM-inflated, outlier-carried 26-trade base has nothing robust to optimize. Not iterate: even correcting the ~2x sizing only scales the result — it does not flip PF above 1, remove the single-outlier dependence, or make the un-exited MTM headline a real realized edge; the EMA-cross-on-ratio signal has negative realized expectancy and the strategy is effectively leveraged long-ETH beta that caught one 2023 spike and rode ETH down 78% in 2022. FAILURE PATTERN: long-only trend-following on the ETH/BTC ratio that takes the position in ONE leg (ETH) is just leveraged long-ETH beta with extra steps — its apparent return is a held open position (end_unrealized >100%) plus a couple of alt-season outlier days, while the realized trade-by-trade edge is negative and the drawdowns are catastrophic (60-99%). Judge cross-asset-ratio strategies on REALIZED PnL and risk-adjusted metrics, not an MTM headline; a relative-strength signal that only ever goes long the outperforming leg does not isolate 'alt-rotation alpha' — a dollar-neutral long-ETH/short-BTC construction would, but that is a different (already-explored, de-cointegrated) pair. This single-leg ratio-trend lane should be retired.