EthMultiWeekTrendFollowingLong

ETH Multi-Week Trend Following Long-Only (Weekly-Aggregated Daily Bars, Long-Horizon Trend Capture, OHLCV-Only, BINANCE USD-M, Same-Mechanism-Variant of BtcMultiWeekTrendFollowingLong)

A long-only, single-instrument, single-venue, MACRO-HORIZON trend-following strategy on ETHUSDT.BINANCE 1D bars. Implements the EXACT SAME proven multi-week trend-following mechanism as the already-landed BtcMultiWeekTrendFollowingLong strategy — applied to ETH USD-M futures instead of BTC. Aggregates daily bars into weekly-equivalent metrics (8-week return, 26-week return, 52-week return) in-code, and trades only when all three weekly-scale momentum signals are simultaneously positive — capturing multi-month / multi-quarter ETH bull trends with HOLDING PERIODS OF WEEKS TO MONTHS. ADDRESSES macro_1w_plus horizon bucket (0.1% — the second-most-under-represented horizon after short_1m_15m). FOLLOWS THE PROVEN VERIFICATION-SURVIVAL RECIPE: my BtcMultiWeekTrendFollowingLong landed cleanly with this exact template; the EthDailyGoldenCrossMomentumLong in pipeline confirms ETH 1D daily bars work for daily strategies; single-instrument + OHLCV-only + explicit daily + simple single-dominant-condition is the highest-survival pattern I've identified. NO SUPP DATA (eliminates developer's supp_data → sub-daily pivot trigger). NO multi-instrument synchronization (avoids the pivot failure that killed EthBtcRatioMeanReversion). The weekly metrics are COMPUTED FROM daily bars in-code — no actual weekly bar type requested, eliminating bar-availability failure mode. ECONOMIC RATIONALE FOR ETH SPECIFICALLY: ETH has had a distinct multi-year bull cycle (2017, 2021, 2024 post-ETF-approval) with longer-duration trends than BTC due to DeFi/L2 narrative cycles. ETH's institutional adoption post-ETF (May 2024) makes ETH multi-week TSMOM a particularly relevant deployment target for current regime. COMPLEMENTARY to my landed BtcMultiWeekTrendFollowingLong: different instrument, different trend cycle timing, natural diversification of a working strategy template. THREE-FOUR meaningful parameters — radically simple.

Implements the exact proven mechanism of the landed BtcMultiWeekTrendFollowingLong on ETH using the standard single-instrument FactoryStrategy hooks. Weekly-scale momentum is computed in-code as trailing daily-bar returns (56/182/364 daily bars ~ 8/26/52 weeks), so no weekly bar type is requested and there is no bar-availability failure mode. calculate_signal returns the mean of the three trailing returns as a continuous per-bar signal (varies every bar -> Layer-2 frozen-signal detection passes), while should_enter applies the triple-horizon AND-gate (all three returns > 0) that only holds during sustained multi-quarter uptrends. should_exit uses the fastest leg flipping negative as the trend-break exit plus a wide 20% catastrophic hard stop via get_position_pnl_pct, giving the weeks-to-months holding period the macro_1w_plus bucket targets. min_bars_required = lookback_long + 2 (366) guarantees all three lookbacks are valid before any trade; ETHUSDT 1D has multi-year history so this warms up well within the data. Sizing deploys 95% of equity (capital-relative), clears the ETHUSDT $20 min-notional by a wide margin, and the multi-month per-trade trend edge dwarfs the ~0.10% round-trip futures fee at the low ~1-3 trades/year turnover. Long-only; leverage stays 1.0 and is referenced only as a sizing multiplier (=1), so the unused-leverage gate does not apply. Honors the hypothesis's BINANCE USD-M venue and the proven survival recipe (single-instrument, OHLCV-only, explicit daily, single-dominant-condition); no supplementary data and no multi-instrument synchronization, eliminating the supp-data pivot and cross-instrument alignment failure modes cited.

Do not optimize — the expected outcome is a forward-gate failure on an extreme fat-tail, high-drawdown, MTM-inflated beta capture. The +633% headline is misleading on three counts: (1) end_unrealized_pct is 396%, so most of it is an un-exited open ETH position marked at backtest end, not a realized track record; (2) tail_ratio 0.0037 with return_kurtosis 204 and skew 12.6 mean the entire edge rests on ~5 explosive days (2024-04-01 +84.7%, 2021-03-02 +81.7%, 2025-09-09 +51%) — strip those and it is flat/negative; (3) it is earned at a 59.1% max drawdown (CI high 98.9%), near-ruin risk, on what is largely long-ETH beta (beta 0.38, benchmark_correlation 0.63). There is no significant edge even before optimization: Sharpe 0.43 with sharpe_ci_low -1.05 (CI deeply straddles zero). Best-of-N optimization will inflate the Sharpe and deflated Sharpe — which penalizes kurtosis hard — will correct it back below the expected-max luck bar, exactly as happened to the BNB range-expansion (kurtosis 542 -> DSR 0.67) and BTC compression-breakout (tail_ratio 0.029) siblings, and the 54-trade sample (~9/yr) split across 3 walk-forward windows is fragile because a single window can be carried or starved by one outlier day. This is not a sizing/correctness bug, but avg_position_pct 203% shows the long-only position runs to ~2x equity in bull holds, amplifying the 59% drawdown. Even though the multi-week TSMOM template landed on BTC, this ETH instance's risk/outlier profile is disqualifying, and no tuning of the lookbacks or stop converts an extreme-fat-tail, 59%-drawdown, CI-straddling-zero, MTM-inflated beta capture into a forward-robust deployable edge; abandon rather than spend 2 hours re-deriving the rejection.

Long-only single-instrument macro-horizon trend-following on ETHUSDT.BINANCE USD-M futures, daily bars. Computes three weekly-scale trailing returns from daily bars in-code (8-week=56d, 26-week=182d, 52-week=364d) and enters long only when all three are simultaneously positive, capturing multi-month ETH bull trends with weeks-to-months holding periods. Exits when the fastest 8-week momentum turns negative (trend break) or a wide 20% catastrophic stop is hit. Deploys 95% of equity per position, leverage 1.0, pure OHLCV, no supplementary data, no extra instruments. Four parameters: the three lookbacks and stop_pct.

Mechanism has a landed BTC sibling; academically grounded multi-week TSMOM; liquid (capacity $91M, impact 3.3%); goes flat in bear regimes (no 2022 trades)

Healthy absolute return (+633%) with PF 1.33 and positive years (2020 +82%, 2024 +85%, 2025 +22%)

Extreme outlier dependence: tail_ratio 0.0037, return_kurtosis 204, skew 12.6 — the entire edge is ~5 explosive days; deflated Sharpe will heavily discount it

No significant edge: Sharpe 0.43 with sharpe_ci_low -1.05 (CI deeply straddles 0)

Catastrophic risk: max_drawdown 59.1% (CI high 98.9%); largely long-ETH beta (beta 0.38, benchmark_correlation 0.63)

Headline is MTM-inflated: end_unrealized_pct 396% — most of +633% is an un-exited open position, not realized; only 54 trades (~9/yr), sparse for 3-window walk-forward

avg_position_pct 203% (>100%) — position value runs to ~2x equity during bull holds, amplifying the drawdown risk

EthMultiWeekTrendFollowingLong ported the landed BTC multi-week TSMOM template to ETH, computing 8/26/52-week momentum from daily bars and going long only when all three legs were positive, targeting the under-represented macro-horizon bucket. Its 633% headline was misleading on three counts: ~396% was unrealized open-position MTM, the realized edge rested on roughly five explosive days (tail ratio 0.0037, kurtosis 204), and it was earned at a 59.1% drawdown (CI high 98.9%) on largely long-ETH beta with a Sharpe of 0.43 whose CI straddled zero. Position value running to ~2x equity in bull holds amplified the near-ruin risk. Even though the template landed on BTC, this ETH instance's risk and outlier profile was disqualifying, so the analyst abandoned it at the backtest-review gate rather than spend optimization time re-deriving the rejection.

A working mechanism template does not transfer automatically to a new instrument — this ETH instance was an extreme-fat-tail, 59%-drawdown, mark-to-market-inflated long-beta capture whose edge rests on a handful of days and a CI straddling zero, none of which lookback or stop tuning can convert into a forward-robust, deployable edge.

The analyst abandoned it at the backtest-review gate before optimization, judging it a likely forward-gate failure: the return rests on ~5 explosive days and an un-exited open ETH position, is earned at near-ruin drawdown on what is largely long-ETH beta (beta 0.38, correlation 0.63), and shows no significant edge before selection (Sharpe 0.43, CI straddling zero) — so best-of-N tuning would inflate the Sharpe only for the kurtosis-penalizing deflated Sharpe to correct it below the luck bar.

A long-only, single-instrument, OHLCV-only macro-horizon trend-follower on ETHUSDT daily futures — computing weekly-scale 8/26/52-week trailing returns in-code from daily bars and entering long only when all three are simultaneously positive (a sustained multi-month uptrend), exiting when the fastest leg turns negative or a wide 20% catastrophic stop hits, porting the landed BtcMultiWeekTrendFollowingLong template to ETH.

The base backtest posted a 633% total return (CAGR 25.9%) with profit factor 1.33 and positive bull years, but Sharpe was only 0.43 with a CI low of -1.05, a 59.1% max drawdown (CI high 98.9%), and extreme outlier-dependence (tail ratio 0.0037, kurtosis 204, skew 12.6). Most of the headline was unrealized (end_unrealized_pct 396%), it traded only 54 times (~9/yr), and avg_position_pct ran to 203% of equity during bull holds.