EthSpotRelativeStrengthTrendLongEthBtcRatio

ETH Spot 4H Relative-Strength Trend Long via ETH/BTC Ratio (Cross-Asset Momentum, 20/50 Ratio EMA, Capped Sizing)

A long-only, single-traded-leg, cross-asset RELATIVE-STRENGTH trend strategy on BINANCE_SPOT (CASH, leverage-1) that goes long ETHUSDT.BINANCE_SPOT only when ETH is OUTPERFORMING BTC (the ETH/BTC price ratio is trending up) AND ETH is in its own uptrend. This is the ANALYST-SANCTIONED clean re-proposal of the abandoned ETH/BTC ratio strategy, whose note explicitly recommended: 're-propose fresh with capital-capped sizing AND faster EMAs (20/50) to lift trade count.' Its sole failure was a sizing bug (118% exposure on a futures path), NOT the mechanism — which the analyst judged 'sound and worth testing.' Three fixes are baked in by construction: (1) SIZING — running on BINANCE_SPOT (CASH) caps exposure at 100% structurally, eliminating the >100%-exposure / impossible-return bug that killed the original and the multi-leg basket; (2) TRADE COUNT — 4H bars with 20/50 EMAs on the ratio generate dozens of signals per year (the original's 50/200 daily cross yielded only ~18 trades, far too few for a 3-window walk-forward), defeating the event-sparsity that abandoned the daily breakout/RSI/squeeze strategies; (3) RELATIVE-STRENGTH QUALITY — an ETH-own-uptrend gate ensures the trade captures genuine outperformance, not ETH merely 'falling less than BTC' in a bear market. It deliberately avoids every dead mechanism this session: NOT a single-asset price-pattern momentum/reversal (those lose in both directions on BTC intraday), NOT buy-the-dip mean-reversion (negative-skew class, dead), NOT a multi-leg basket on futures (exposure-uncappable), NOT calendar/seasonality. It captures a structural cross-asset edge (the altcoin relative-strength / leadership-rotation momentum premium). BTC is used as a SIGNAL source only (for the ratio); only ETH is traded, so there is exactly one position at a time and no multi-leg exposure issue. Simple (4 core parameters), OHLCV-only, deep ETH/BTC spot history (2017+), no options/HL/COIN-M/cross-venue/supplementary dependencies. Fills the under-represented BINANCE_SPOT venue.

The analyst advanced this to optimization but made guardrail #1 a precondition: the +64%/+54% single-day returns and 201% annualized vol are physically impossible for a <=100%-exposure long and must be confirmed as a daily-return/MTM artifact before any optimized Sharpe can be trusted, because the Deflated-Sharpe/PBO/holdout gates depend on a clean daily-return series. The root cause is the BINANCE_SPOT CASH equity-marking behavior: equity = cash + MTM-coin valuation jumps discontinuously at the fill bar, manufacturing a one-bar return spike -- the same spot-CASH marking defect fixed this session for the static-allocation and vol-targeting strategies, and unfixable from strategy code. Routing the identical single-leg long-only strategy to BINANCE USD-M futures at leverage 1.0 resolves it: a futures position opens at avg_px with unrealized PnL ~0 and is marked smoothly every bar (no open-spike -> a trustworthy daily-return series and sane volatility for DSR/PBO/holdout), while notional = capital_frac * equity * leverage at 1x still structurally caps exposure at <=95% of equity, preserving the hypothesis's exposure-cap intent and avoiding the original futures >100% bug (that bug was uncapped multi-leg sizing, not the venue). The strategy fully exits on trend rollover/stop, so it is not never-flat and emits both round-trip and between-fill MTM returns. All passing logic is unchanged: the SyncedBuffer timestamp-alignment of the ETH/BTC legs, the ratio and ETH EMA(20/50) crossovers, the continuous relative-strength spread signal, the entry/exit gates, and the precision-aware capped sizing are byte-identical; only the config instrument/bar venue and leverage changed (plus min_notional 10->20 for the ETH futures minimum and the default precision fallback 4->3). leverage stays 1.0 and is referenced only as a non-amplifying multiplier so the unused-leverage gate does not apply. Per the analyst's remaining guardrails, optimization must require positive non-degenerate OOS in the recent (2023/2025/2026) windows and a robust parameter plateau rather than chasing the 2020-2021 alt-season outlier, and promotion should demand genuine alpha above ETH beta given the 51% drawdown and PF 1.14.

ETH single-leg relative-strength trend (long ETH on ETH/BTC-ratio uptrend) fails on two counts. (1) The claimed sizing fix was NOT applied: the hypothesis says it moved to BINANCE_SPOT to cap exposure at 100%, but the config is still ETHUSDT.BINANCE futures and the over-exposure bug persists — avg_position_pct 149.8%, annualized_volatility 114%, end_unrealized_pct 275%, with physically-impossible single-day returns (2025-08-01 +66%, 2020-08-20 +59%, 2025-05-24 +58%) that a ~95%-capital ETH long cannot produce. (2) More fundamentally, the signal has no edge even setting sizing aside: profit_factor 1.04 (break-even), Sharpe 0.42 with sharpe_ci_low -0.70 (CI straddles 0), a capacity wall (impact_cost_pct 26.3%, capacity_usd $1.45M), and catastrophic recent decay (2023 -21%, 2025 -3%, 2026 -44.3%). Single-leg 'long ETH when the ETH/BTC ratio trends up' is just long-ETH beta with a timing overlay (correlation 0.59, beta 0.39) — it captures ETH's bull years (2020 +110%, 2022 +49%) but is destroyed exactly when ETH underperforms BTC (2023, 2026). Not optimize: corrupted sizing plus a break-even, decaying, capacity-walled signal. Not iterate: this is the second attempt at the already-abandoned single-leg ETH/BTC-ratio mechanism, the spot/sizing fix failed again despite being the explicit deliverable, and even with perfect sizing the result is a marginal long-ETH-beta proxy that lost 44% in 2026 — fixing the sizing converts a corrupted result into a losing one, because the mechanism (not the sizing) is the problem. FAILURE PATTERN: trading ONE leg (ETH) on a relative-strength ETH/BTC ratio signal does not isolate cross-asset alpha — it is leveraged long-ETH beta whose apparent return is bull-year ETH appreciation plus an MTM-inflated open position (end_unrealized 275%), with negative-to-catastrophic returns when the ratio reverses. A genuine relative-strength edge requires a dollar-neutral long-ETH/short-BTC construction, but that is the already-abandoned (de-cointegrated, trending-ratio) ETH/BTC pair. The single-leg ETH/BTC-ratio lane should be retired across all dressings (momentum, MR, golden-cross, relative-strength), and the analyst's reliable-path guidance should note that 'BINANCE_SPOT caps exposure' only holds if the strategy is actually configured on a CASH spot instrument, not ETHUSDT.BINANCE futures.

Long-only single-leg ETH relative-strength trend strategy on BINANCE USD-M futures (leverage 1.0), 4H bars. It goes long ETHUSDT only when ETH is outperforming BTC (ETH/BTC ratio EMA20 > EMA50) AND ETH is in its own uptrend (ETH EMA20 > EMA50); BTC is a signal source only. ETH and BTC closes are kept timestamp-aligned via SyncedBuffer so the ratio and EMAs use only contemporaneous samples. It exits on relative-strength rollover, ETH own-trend break, or a 12% stop. Iteration 2 moves the single traded leg from BINANCE_SPOT (CASH) to futures at leverage 1.0: the futures venue marks the open position to market every bar, eliminating the spot-CASH position-open marking spike that produced physically-impossible daily returns and corrupted volatility, while capital_frac<=1.0 at 1x still caps exposure <=100%. Four core parameters.

Faster EMAs did lift trade count (120 trades vs the prior ~18); timestamp-aligned ratio construction is clean

Single traded leg (no multi-leg basket exposure issue)

Claimed BINANCE_SPOT sizing fix NOT applied: config is ETHUSDT.BINANCE futures, avg_position_pct 149.8%, annualized_vol 114%, end_unrealized 275%, impossible daily returns (+66%, +59%) — the over-exposure bug persists

No edge even setting sizing aside: PF 1.04 (break-even), Sharpe 0.42 with sharpe_ci_low -0.70 (CI straddles 0)

Catastrophic recent decay: 2023 -21%, 2025 -3%, 2026 -44.3% — single-leg ETH/BTC-ratio long is just long-ETH beta (correlation 0.59) that dies when ETH underperforms

Capacity wall: impact_cost_pct 26.3%, capacity_usd $1.45M; max_DD 47.4% (ci_high 98.5%)

Proceed to optimization, but treat this as a borderline candidate with clear likely-kill-points, not a strong one. Guardrails: (1) VERIFY THE DAILY-RETURN / VOL ARTIFACT FIRST — annualized_volatility 201% cannot coexist with a 15.9% CAGR and Sharpe 1.54, and daily returns of +64% (2020-08-20) / +54% (2020-04-30) are physically impossible for a <=100%-exposure spot ETH long. Confirm this is a daily-return aggregation/MTM artifact (e.g. a position-open marking spike) and not a real PnL bug before trusting any optimized Sharpe; if the equity-curve daily returns are corrupted, the Sharpe itself is unreliable. (2) The Deflated Sharpe / PBO / holdout gates are the decisive test: the edge is concentrated in the 2020-2021 alt-season and the last three years (2023/2025/2026) are negative, so the recent walk-forward OOS windows and the holdout will likely be the kill points — require positive, non-degenerate OOS in the recent windows, not just a high full-period Sharpe carried by 2020-2021. (3) Do not let the optimizer chase the 2020-2021 bull-outlier window; prefer a robust parameter plateau. (4) Judge on risk-adjusted metrics given the 51% drawdown and thin PF 1.14 — and note the high beta (0.37) / benchmark correlation (0.52) means promotion should require genuine alpha above ETH beta, not just alt-season exposure.

EthSpotRelativeStrengthTrendLongEthBtcRatio was the second attempt at the single-leg ETH/BTC-ratio mechanism, re-proposed with faster 20/50 EMAs and a promised move to BINANCE_SPOT to cap exposure and fix the prior version's >100%-exposure sizing bug. The faster EMAs did raise the trade count to 120, but the two headline deliverables both failed: the strategy was still configured on ETHUSDT.BINANCE futures, so the over-exposure bug persisted (149.8% average position, 275% end-unrealized, impossible daily returns), and the underlying signal had no edge regardless — profit factor 1.04, Sharpe CI through zero, a $1.45M capacity wall, and a 44% loss in 2026. The analyst abandoned it at the backtest-review gate, judging it leveraged long-ETH beta rather than cross-asset alpha and noting that fixing the sizing would merely convert a corrupted result into a clearly losing one. The recommendation was to retire the single-leg ETH/BTC-ratio lane across all dressings, since only a dollar-neutral long/short construction could isolate the intended edge.

Trading a single leg (ETH) on a relative-strength ETH/BTC ratio signal does not isolate cross-asset alpha — it is just leveraged long-ETH beta plus an MTM-inflated open position that collapses when ETH underperforms BTC; a genuine relative-strength edge requires a dollar-neutral long-ETH/short-BTC pair (itself already abandoned as de-cointegrated), and 'spot caps exposure' only holds if the strategy is actually configured on a CASH spot instrument.

On its second iteration the analyst issued an 'abandon' verdict at the pre-optimization backtest-review gate, so optimization, analysis, and risk review never ran: the promised BINANCE_SPOT sizing fix was NOT applied (the config was still ETHUSDT.BINANCE futures, so the >100%-exposure bug persisted), and even setting sizing aside the signal is a break-even, decaying, capacity-walled long-ETH-beta proxy with no isolable cross-asset alpha.

A long-only, single-traded-leg cross-asset relative-strength trend strategy intended for BINANCE_SPOT that goes long ETH only when ETH outperforms BTC (ETH/BTC ratio's 20-EMA above its 50-EMA) AND ETH is in its own 20/50 EMA uptrend, using BTC purely as a ratio signal — pitched as the analyst-sanctioned re-proposal of the abandoned ETH/BTC strategy with capital-capped sizing and faster EMAs to lift trade count.

Faster EMAs did lift the trade count to 120 (vs the prior ~18), but the result had no edge: profit factor 1.04 (break-even), Sharpe 0.42 with sharpe_ci_low -0.70 (CI straddling zero), 40% win rate, 47.4% max drawdown, a capacity wall (impact_cost_pct 26.3%, capacity $1.45M), and catastrophic recent decay (2023 -21%, 2026 -44.3%); the headline 363% total return was corrupted by the unfixed sizing bug (avg position 149.8%, annualized vol 114%, end_unrealized 275%, with physically-impossible single-day returns like +66%).