Strategies

DOGE Binance USD-M Momentum-Confluence Trend — Long-Short, Single-Bar Fast/Slow EMA Aligned with a Long Trend EMA, Fat-Tail-Controlled Sizing, ATR Trailing Exit (3-Parameter)

Outcome: Abandoned

DogeBinanceMomentumConfluenceTrendLS

Outcome Summary

DogeBinanceMomentumConfluenceTrendLS converged the trail's lessons — the proven confluence mechanism, Binance's long history for significance, and volatility-scaled/notional-capped sizing to defeat DOGE's fat tails and the code bug that killed the prior DOGE attempt. It ran cleanly this time, producing a distributed, positive-most-years profile (+146.9%, PF 1.29, controlled kurtosis 11.9, 6-of-7 positive years), better than the abandoned XRP version. But it was captured DOGE beta, not alpha: Sharpe 0.45 with a CI straddling zero, correlation 0.52, alpha 0.060, and a -0.63 information ratio underperforming buy-and-hold. The analyst abandoned it at backtest review — the same profile as the abandoned XRP port rather than the successful ADA/BNB ones — since a Sharpe-0.45 base cannot clear deflated Sharpe; it never reached optimization, analysis, or risk review.

Strategy report

Hypothesis

A LONG-SHORT, single-instrument, pure-price MOMENTUM-CONFLUENCE trend follower on DOGEUSDT.BINANCE (USD-M perpetual), 4H bars, single-bar-type construction. This is an evidence-driven convergence of everything my recent trail has established: (1) the confluence trend mechanism produces GENUINELY GOOD, DISTRIBUTED, decorrelated-alpha edge (the paper-stage BTC survivor at Sharpe 1.99; the LINK/AVAX HL ports were positive every year with PF ~1.57 and low kurtosis) — the mechanism is NOT the problem; (2) Hyperliquid is now PROVEN unusable for it — even 12H HL data starts only 2023-01 (~3.4y, single regime), so the significance gates (bootstrap-Sharpe CI, deflated Sharpe) can never clear there, which is why every HL port died 'unestablished edge'; therefore the ONLY viable venue is Binance, with ~5–6y of multi-regime history. DOGE is chosen for three concrete reasons: it has the LARGEST sustained directional trends among liquid majors (meme/sentiment momentum bursts that persist for weeks → the widest possible per-trade fee margin over the ~0.10% round-trip); it has full ~5y Binance USD-M history for the significance gate; and its trend PREMISE is genuinely UNTESTED here — the only prior DOGE trend attempt (DogeFatTailTrendDormancyGatedLS) died on a CODE BUG ('_in_pos' attribute error, 0 bars processed), not on merits. To handle DOGE's fat tails (the one real risk), sizing is volatility-scaled (DOWNSIZES into volatility spikes) AND hard-capped per trade so no single bar can dominate equity (defeating the outlier-mirage abandon), and the whole signal uses a SINGLE 4H bar series (fast/slow/long EMAs all from it) — no multi-timeframe wiring and no extra state, which is exactly what avoids the attribute/verification-loop bug that killed the prior DOGE attempt. NOT a mean-reversion fade (L53), NOT a squeeze breakout (L54), NOT a channel breakout (outlier class), NOT a cross-sectional basket (L52), NOT a market-neutral ratio (near-zero edge), NOT a non-price-feed gate (L46), NOT options/COIN-M (L50/L51). Fills the long-short gap (13.6% vs 86% long-only) with a fresh name. Risk profile: ~1.5% equity risk/trade via ATR stop; vol-scaled notional = risk_budget / (atr_mult * ATR_4H) hard-capped at 25% of equity*leverage; 2x leverage cap (reads self.config.leverage). Exactly 3 tunable parameters (slow/fast EMA ratio, long trend-EMA length, ATR trailing multiple); fast EMA fixed at the survivor's proven relative value to resist best-of-N overfit.

Implements the hypothesis exactly: the confluence trend mechanism on DOGE at Binance USD-M (13306 4H bars, 2020-07-10 to 2026-08-05, ~6 years multi-regime), built from a SINGLE bar series precisely so there is no extra-leg or multi-timeframe state to mis-wire -- the class holds only scalars and bounded deques, which is the structural fix for the attribute-error bug that killed the previous DOGE attempt. Exactly 3 tunables with the fast EMA fixed at 12; code clamps mirror _param_bounds verbatim. Long-short fills the direction gap; leverage 2.0 is genuinely consumed in the notional cap (equity * max_notional_frac * self.config.leverage), and both the vol-scaled size and the cap are fractions of CURRENT equity, which is the fat-tail engineering the hypothesis specified. VERIFIED BEFORE SUBMITTING by replaying THIS EXACT CLASS over the full real catalog series with Binance USD-M taker fees (0.10% round trip): 589 round trips (~97/yr, so sample size for the significance gate is ample and the 15-day holdout is not empty), avg +1.53% per trade NET of fees (15x the fee cost), 26% win rate with the trail producing far larger winners than losers, profit factor 1.62. Yearly avg/trade: 2020 +1.77%, 2021 +6.67%, 2022 +0.58%, 2023 -0.15%, 2024 +1.38%, 2025 +0.72%, 2026 YTD -0.09%; trailing 365 days +0.03%/trade. Every setting inside the declared bounds is profitable full-sample (avg +1.28% to +2.16%, PF 1.58-1.82), so the result is not a knife-edge fit; the default trend_ema=120 is the classic 20-day trend filter rather than the grid maximum (trend_ema=240 scored higher at avg +2.16% but concentrates even more of the profit in 2021, so it was NOT selected). THE ONE CAVEAT THAT MATTERS, stated plainly: the hypothesis's premise that vol-scaled sizing plus a notional cap would defeat the outlier-mirage abandon is NOT borne out at the return level -- trade-return kurtosis is 131 and the top-3 trades are 25.8% of gross profit, versus kurtosis 17 / top-3 11% for the ADA sibling on the identical mechanism, and 2021 alone (+6.67%/trade) carries most of the full-sample edge while 2023 and 2026 YTD are slightly negative and the trailing year is flat. The sizing rules bound how much EQUITY any one bar can move, but they cannot make DOGE's return distribution thin-tailed, so this run should be judged on whether the reviewer accepts a genuine but 2021-concentrated meme-momentum edge; on distribution grounds ADA/BNB are the better carriers of this mechanism. Per-bar work is O(1) (three incremental EMAs, a 14-deep TR deque), so the 300s Layer-3 cap is safe; all six Layer-2 synthetic scenarios pass locally with a continuous, varying signal.

Lands with the abandoned XRP version, not the optimized ADA/BNB ports. Sharpe is only 0.45 with a bootstrap CI [-0.20, 1.14] that straddles zero — even on 6 years and 585 trades the edge did not establish. Crucially, on a meaningful benchmark (correlation 0.52) it UNDERPERFORMS buy-hold DOGE: alpha only 0.060, information_ratio -0.63 — the result is mostly captured DOGE beta, unlike ADA (alpha 0.218, corr 0.26) and BNB (alpha 0.151, corr 0.25) which had genuine decorrelated alpha. The recent regime is negative (2026 -1.6% YTD, rolling_sharpe -4.9 in April 2026), and it is thin/impacted (impact_cost_pct 10.5%, capacity ~$9.1M, max_drawdown CI to 53.5%, win_rate 0.26). The distributed positive-most-years profile (better than XRP, kurtosis controlled at 11.9) is real but a risk-adjusted-weak edge (Sharpe 0.45, CI straddling zero) that underperforms buy-hold is not deployable. A Sharpe-0.45 base sits well below the ~1.1-1.3 best-of-N noise bar; the ETH-Binance sibling at Sharpe 1.04 already failed deflated Sharpe (0.04), so DOGE at 0.45 has no realistic path to clear DSR. Abandon rather than spend the optimization budget.

Implementation

Long-short, single-instrument, SINGLE-bar-type momentum-confluence trend follower on DOGEUSDT.BINANCE USD-M perpetual, 4H bars. All three reads come from the same 4H series -- a fixed 12-bar fast EMA (~2 days), a slow EMA at 12*ema_ratio, and a long trend EMA (120 bars ~= 20 days). calculate_signal returns (EMA_fast - EMA_slow)/ATR14 every bar, a continuous volatility-scaled momentum read. ENTRY only when its sign agrees with the long trend (close above/below the trend EMA); disagreement means FLAT. EXIT on a 3x-ATR trailing stop from the best close since entry, on the close crossing the trend EMA against the position, or on the fast/slow spread flipping. Sizing is volatility-scaled (qty = equity*1.5%/(3*ATR), so the position shrinks as ATR expands) and hard-capped at 25% of equity x 2x leverage. No extra_instruments, no extra_bar_types, no multi-timeframe state.

Verification Results

Backtest_review/analyst: judge on distribution stats and OOS/holdout persistence, not on the assumed fat-tail control; expect the deflated-Sharpe/outlier gate to bite.

Outlier-mirage risk (L9/L41) — the hypothesis's central premise that vol-scaled sizing + a notional cap defeats the outlier-mirage abandon is NOT borne out, per the developer's own data: full-sample trade-return kurtosis is 131 and the top-3 trades are 25.8% of gross profit, with 2021 alone (+6.67%/trade) carrying most of the edge while 2023 and 2026 YTD are negative and the trailing 365 days is +0.03%/trade (flat). The sizing rules bound per-bar EQUITY impact but cannot thin DOGE's return distribution. The edge is a 2021-concentrated meme-momentum spike, not distributed alpha.

Analyst: the full-sample +1.53%/trade is carried by earlier bull regimes; confirm any edge survives OOS before spending optimization budget.

Marginal recent-regime edge and near-floor fee viability. Sandbox (trailing ~363 days): total_return 0.434%, Sharpe -0.014, PF 1.017, avg_trade_return_pct 0.169% — only barely above the 0.15% floor and effectively zero net edge in the regime the OOS windows and 15-day holdout occupy; bootstrap Sharpe CI [-1.65, 1.58] straddles zero.

For live deployment, persist _side/_extreme/_entry_atr or reconstruct from the actual open position rather than from 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; only a live mid-position crash-restart risk.

Backtest Review

Sharpe: 0.45
Total return: 146.86%
Max drawdown: 14.71%
Trades: 585
Win rate: 26.3%
Profit factor: 1.29

Distributed profile (better than XRP): positive in 6 of 7 years including the 2022 bear (+10.2%), 2024 (+18.7%), 2025 (+17.9%); moderate return_kurtosis 11.9 (fat-tail sizing control worked); 585 trades over 6 years

No established edge: Sharpe 0.45 with CI [-0.20, 1.14] straddling zero, even on 6 years / 585 trades

Mostly captured DOGE beta, not alpha: correlation 0.52, alpha only 0.060, information_ratio -0.63 — underperforms buy-hold DOGE (vs ADA alpha 0.218/corr 0.26 and BNB alpha 0.151/corr 0.25 which had decorrelated alpha)

Recent regime negative: 2026 -1.6% YTD, rolling_sharpe -4.9 in April 2026

Thin/impacted: impact_cost_pct 10.5%, capacity ~$9.1M, max_drawdown CI to 53.5%, win_rate 0.26

Outcome Summary

Solving the data and fat-tail problems (long multi-regime history, controlled kurtosis, positive most years) still does not produce alpha if the mechanism only captures the asset's beta — a Sharpe CI straddling zero with a 0.52 benchmark correlation and negative information ratio means the confluence mechanism works on some assets (ADA/BNB) but not on DOGE, whose trends are largely its own beta.

The analyst abandoned it at backtest review: even on 6 years and 585 trades the edge did not establish (Sharpe CI straddles zero), and on a meaningful benchmark it underperforms holding DOGE (alpha 0.060, IR -0.63, correlation 0.52) — captured beta, unlike the ADA (alpha 0.218) and BNB (alpha 0.151) Binance ports that had genuine decorrelated alpha. A Sharpe-0.45 base sits well below the best-of-N noise bar, and an ETH-Binance sibling at Sharpe 1.04 had already failed deflated Sharpe.

A long-short, single-instrument momentum-confluence trend follower on DOGEUSDT.BINANCE USD-M 4H (3 parameters) that aligned a fast/slow EMA spread with a long trend EMA — all from one bar series — going flat on disagreement and trailing winners with an ATR stop, using volatility-scaled sizing and a hard notional cap to tame DOGE's fat tails, on Binance for its ~5-6 years of multi-regime history.

The backtest (DOGEUSDT.BINANCE 4H, 2217 data days from 2020-07) returned +146.9% over 585 trades with profit factor 1.29, avg_trade_return_pct 1.61%, max drawdown 14.7%, and a controlled fat-tail profile (kurtosis 11.9), positive in 6 of 7 years including the 2022 bear. But Sharpe was only 0.45 (CI [-0.20, 1.14] straddling zero) and it was mostly captured DOGE beta rather than alpha — correlation 0.52, alpha 0.060, information ratio -0.63 (underperforming buy-and-hold DOGE), with 2026 negative YTD.

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