Strategies

ETC Binance USD-M Dual-Timeframe Momentum Confluence — Long-Short 4H Trend Aligned with the Daily Trend, Flat on Disagreement, ATR Trailing Exit (Volatility-Scaled, Single-Bar, 3-Parameter)

Outcome: Abandoned

EtcDualTimeframeMomentumConfluenceLS4H

Outcome Summary

EtcDualTimeframeMomentumConfluenceLS honestly framed itself as an asset-lottery probe — acknowledging the 'clean large-cap' heuristic had just been falsified by DOT — and applied the factory's only surviving mechanism to ETC as a fresh Proof-of-Work major with idiosyncratic trend drivers. The draw came up a loser: Sharpe 0.111 with a CI deep into negative territory, profit factor 1.06, a 31% drawdown, and a catastrophic 2025-2026 regime, all carried by 2021/2022 and uninvestable at $510K capacity with 44% impact. The analyst abandoned it at backtest review, placing ETC with the DOT/ATOM choppy laggards rather than the ADA/BTC class. It never reached optimization, analysis, or risk review.

Strategy report

Hypothesis

A LONG-SHORT, single-instrument, pure-price MOMENTUM-CONFLUENCE trend follower on ETCUSDT.BINANCE (USD-M perpetual), 4H single-bar construction with a daily trend confirmation. HONEST framing: the confluence family is an asset-LOTTERY — the qualitative 'clean large-cap' heuristic was just FALSIFIED by DOT (Sharpe 0.208, below SOL/AVAX), so I do NOT claim ETC is definitely ADA-like. But it remains the ONLY family with survivors in 1125 hypotheses (BTC 1.99 paper-stage; ADA 1.75 near-miss that emphatically cleared the deflated-Sharpe gate), so probing fresh, untested moderate-volatility majors with the proven mechanism has the highest base rate available. ETC is chosen as a probe distinct from the queued L1/DeFi ports (AAVE, ATOM) and the survivor set: a Proof-of-Work major (ETH Classic) with ~5–6y Binance USD-M history, moderate ~4-5% daily vol (enough per-trade margin to clear fees, unlike lower-vol names, but not memecoin fat tails), and its own idiosyncratic trend drivers (mining-narrative flows, the 2021 run, 2024 rallies) that are somewhat decorrelated from the L1/DeFi complex. It applies the exact mechanism + vol-scaling that produced the BTC and ADA clears. NOT a mean-reversion fade (L53), NOT a squeeze breakout (L54), NOT a channel breakout (structurally-unmeasurable, abandoned), NOT a cross-sectional/always-on-ensemble basket (L52 — this is single-name), NOT a NON-PRICE-feed gate (L46), NOT options/COIN-M (L50/L51). Fills the long-short gap (13.5% vs 86% long-only). Single 4H bar type, 3 params, fixed fast-EMA to resist overfit (the #1 killer). Risk profile: ~1.5% equity risk/trade via ATR stop; vol-scaled notional = risk_budget / (atr_mult * ATR_4H) capped at 25% of equity*leverage; 2x leverage cap (reads self.config.leverage). Exactly 3 tunable parameters (slow/fast EMA ratio, daily-trend EMA length, ATR trailing multiple).

Ports the exact surviving BTC/ADA confluence mechanism - fast trend that must agree with a higher-timeframe trend, flat on disagreement, ATR trailing exit, vol-scaled risk-first sizing - to ETC as the hypothesis specifies, with no added filters and no per-asset special-casing. The daily confirmation is derived from the same 4H series (a long EMA) rather than a second bar feed, honouring the 'single 4H bar type' constraint and removing the multi-leg data risk that has burned iterations elsewhere. Offline dry-run on the real ETCUSDT 4H catalog (2020-01 to 2026-08, 14,362 bars) at the submitted defaults: 602 trades (~92/yr), avg +0.83% NET-of-fee per trade (5.5x the 0.15% viability floor), win rate 22% with trend-follower right-tail payoff, trade-close max drawdown ~26%, ~1.5x cumulative. The 3-parameter surface is positive at every grid point swept (ema_ratio 2.2-3.4, daily_ema_len 110-200, trail 3.0-8.0; Sharpe ~0.39-0.55) with no cliff, and defaults sit at the CENTRE of the flat region, not its maximum - _param_bounds is set to that plateau. Honest caveat matching the hypothesis's own framing: ETC is a probe, and the sweep shows the edge concentrated in 2021-2022 and 2024 with 2023, 2025 and 2026 negative across the WHOLE parameter surface, so the recent-regime OOS windows are the real risk here, not the parameter choice. Leverage is genuinely consumed: the risk-first quantity exceeds 1x-equity notional on ~8% of historical entries (max 0.43x equity), which the 2x cap permits and a 1x cap would clip.

The honest 'asset-lottery' probe drew a loser: ETC is a choppy laggard like DOT/ATOM, not in the ADA/BTC clean-trending class. Sharpe is only 0.111 with a bootstrap CI-low of -0.5509 (the CI straddles zero deeply), so there is no significant edge, and profit_factor is 1.06 (fee-edge zone, expectancy +$28/trade). It is also capacity-blocked: capacity_usd is only $510K with impact_cost_pct 44.2%, meaning thin ETC liquidity would consume nearly half of gross PnL — uninvestable at scale. The recent regime is catastrophic (2025 -13.1%, 2026 -21.7% YTD, rolling_sharpe to -22 in May 2026), with the record carried entirely by 2021 (+38%) and 2022 (+24.7%); max_drawdown 31.4% (CI to 63%), IR -0.35. The hypothesis was refreshingly honest that the 'clean large-cap' heuristic is a lottery, and this draw confirms it — ETC has no fee-clearing, significant, or deployable edge. Abandon at BACKTEST_REVIEW rather than spend the optimization budget.

Implementation

Long-short dual-timeframe momentum confluence on ETCUSDT.BINANCE (USD-M perpetual), built entirely from a SINGLE 4-HOUR bar series. Fast read: spread = (EMA12 - EMA_slow)/ATR14 with EMA_slow = round(12 * ema_ratio) - volatility-scaled and returned every bar as the continuous signal. Slow read: an EMA over daily_ema_len 4H bars (default 150 ~= 25 days) gives the daily-equivalent trend direction (close above/below). Entry only on CONFLUENCE (fast spread and slow direction agree): long if both positive, short if both negative, FLAT whenever they disagree. Exit on any of: ATR trailing stop of trail_atr_mult x entry-ATR from the best close since entry, slow-trend flip against the position, or fast-spread flip against the position. Risk-first vol-scaled sizing: qty = equity * 1.5% / (trail_atr_mult * ATR), capped at 25% of equity * self.config.leverage (2x), all off CURRENT equity so losses never compound into size. Exactly three tunables (ema_ratio, daily_ema_len, trail_atr_mult) with the fast EMA fixed at 12 bars.

Verification Results

Analyst/backtest_review: weight the recent regime decisively — it is negative at every parameter cell and that is exactly where the OOS windows and holdout sit; lean abandon rather than spend optimization budget, since best-of-N cannot manufacture a recent-regime edge the raw data does not carry.

RECENT REGIME STRONGLY NEGATIVE across the WHOLE parameter surface — recommend the analyst lean ABANDON at BACKTEST_REVIEW. This is a genuine no-recent-edge result, not a code defect (verified below). Sandbox (trailing year): total_return -23.8%, Sharpe -1.51, profit_factor 0.33, win_rate 0.099, avg_trade_return_pct -1.29%, probabilistic_sharpe 0.038, DD 24.3%. The developer honestly discloses the edge is concentrated in 2021-2022 and 2024, with 2023, 2025 and 2026 negative at EVERY grid point of the declared bounds — so the walk-forward OOS windows and the 15-day holdout all sit in a regime this construction loses in, and no in-bounds parameterization rescues it. The full-sample +0.83%/trade is carried by the earlier bull/volatile regimes. ETC is explicitly framed by the developer as a low-base-rate probe (the 'clean large-cap' heuristic was just falsified by DOT), and this probe lands with the SOL/AVAX/LTC no-recent-edge group, not the ADA/BTC winners.

No code change needed — this is an edge/regime finding for the analyst, not a bug.

Verified the strongly negative sandbox is genuine no-recent-edge, NOT a code/polarity bug. Signal polarity is correct (spread=(EMA12-EMA_slow)/ATR: rising -> spread>0 and px>daily-EMA -> daily_dir +1 -> LONG), both long and short legs lose symmetrically (long_win 0.081, short_win 0.109 — an inversion would make one side win), and none of the L17 code-defect signatures are present (win_rate 0.099 not 0.0, |Sharpe| 1.51 not >5, PF 0.33 not 0.0). The 9.9% win rate nets negative because the trailing-stop trend-follower's wins (avg 1187 vs avg loss 392, ~3x) did not run far enough in ETC's choppy recent regime — correct behaviour of the coded rule, not a defect. The mechanism is identical to the DOT/ATOM single-bar confluence siblings I verified.

For live deployment, reconstruct _side/_extreme/_entry_atr from cache.positions_open() rather than the current signal.

should_exit() infers _side from the live signal sign on restart (_side==0) and re-seeds _entry_atr/_extreme from current values. Unreachable in backtest (state persists); only a live mid-position crash-restart risk.

Backtest Review

Sharpe: 0.11
Total return: 27.75%
Max drawdown: 31.39%
Trades: 589
Win rate: 21.2%
Profit factor: 1.06

589 trades over 6.6y is a measurable sample; honest hypothesis framing (acknowledged asset-lottery, not overclaimed)

Sharpe 0.111 with CI-low -0.5509 — straddles zero deeply, not significant; far below the family DSR bar

profit_factor 1.06 (fee-edge zone), expectancy +$28/trade

Capacity blocker: capacity_usd $510K with impact_cost_pct 44.2% — thin ETC liquidity eats ~half of gross PnL

Recent regime catastrophic: 2025 -13.1%, 2026 -21.7% YTD, rolling_sharpe to -22 in May 2026; record carried by 2021/2022 only

max_drawdown 31.4% (CI to 63%), IR -0.35 — ETC is a choppy laggard, not ADA/BTC class

Outcome Summary

Confirming the confluence family is an asset lottery: probing fresh moderate-vol majors with the proven mechanism keeps drawing choppy laggards (DOT, ATOM, ETC) with Sharpe CIs straddling zero and thin capacity, so a favorable base rate does not make any individual untested name a survivor — most are not.

The analyst abandoned it at backtest review: the honest asset-lottery probe drew a loser — ETC is a choppy laggard like DOT and ATOM, not in the ADA/BTC clean-trending class. There is no significant edge (Sharpe CI straddling zero, PF in the fee-edge zone), the recent regime is deeply negative, and it is capacity-blocked ($510K, 44.2% impact) so it is uninvestable at scale.

A long-short, single-instrument dual-timeframe momentum-confluence trend follower on ETCUSDT.BINANCE USD-M 4H (3 parameters) that aligned a fast/slow EMA spread with a ~25-day trend EMA — all from one bar series — going flat on disagreement and trailing winners with an ATR stop, an honestly-framed asset-lottery probe of the only surviving mechanism on a fresh moderate-vol Proof-of-Work major.

The backtest (ETCUSDT.BINANCE 4H, 2393 data days from 2020-01) showed no significant edge: Sharpe 0.111 (CI-low -0.551, straddling zero deeply), profit factor 1.06 (fee-edge zone), expectancy +$28/trade, max drawdown 31.4% (CI to 63%) over 589 trades. Capacity was $510K with impact cost 44.2% (nearly half of gross PnL), and the record was carried by 2021/2022 with a catastrophic recent regime (2025 -13.1%, 2026 -21.7% YTD, rolling Sharpe to -22).

Backtest and paper results are hypothetical. Trading involves risk of loss.