Strategies

DOGE Volume-Surge Momentum on 4H Bars

Outcome: Abandoned

DogeFourHourVolumeSurgeMomentumLongShort

Outcome Summary

DOGE Volume-Surge Momentum was a deliberately simple 4H long-short strategy that traded volume spikes confirmed by EMA slope, and its first backtest looked promising enough (+89% return, Sharpe 0.55, 488 trades, 25.9% drawdown) that the analyst approved it for full optimization. Optimization produced a flashier +190% headline and a marginally higher Sharpe of 0.586, but only by roughly doubling leverage and volatility while the risk-adjusted return barely moved. Under statistical scrutiny the edge dissolved: it failed the deflated Sharpe test (0.083 vs 0.95), landed below the 225-trial luck bar of 1.24, was flagged walk-forward-overfit with a negative average out-of-sample Sharpe (-0.748), and rested on a handful of parabolic outlier days plus 44% un-exited open MTM. With no robust region to tune toward, the analyst abandoned it on attempt 1 as a weak momentum edge statistically indistinguishable from best-of-N noise.

Strategy report

Hypothesis

Exploits the strong volume-price relationship in DOGE, a memecoin with outsized retail participation. When DOGE trading volume surges significantly above its recent average AND price is moving directionally (confirmed by EMA slope), enter in the direction of the move. The thesis is that DOGE volume spikes are driven by viral social media attention and retail FOMO, creating persistent momentum that lasts 1-5 days — long enough to capture 1-5% moves that far exceed trading costs. Uses 4-hour bars to balance signal frequency with per-trade return magnitude. This is a single-instrument, single-timeframe strategy with minimal parameters to avoid overfitting.

Iteration 2 fix for the Layer-2 failure 'DogeFourHourVolumeSurgeMomentumLongShort._ema() takes 2 positional arguments but 3 were given'. Under the verification proxy the @staticmethod _ema was invoked bound (self implicitly prepended), so a 2-parameter (values, period) signature received 3 args. The SMALLEST fix: convert _ema to a normal instance method 'def _ema(self, values, period)' so the call convention self._ema(closes, self.ema_period) matches exactly. No signal, entry, exit, or sizing logic changed and no imports touched, so Layer 1 stays green; this only repairs the crashing helper so the six synthetic scenarios can execute.

Failed deflated Sharpe on attempt 1: DSR=0.0832 (vs 0.95 bar), is_significant=FALSE, with the optimized Sharpe 0.586 BELOW the 225-trial expected-max luck bar of 1.2438 (sharpe_ci_low -0.4568 straddling 0, PBO 0.463) — after multiple-testing correction the selected config is statistically indistinguishable from best-of-N noise. The high probabilistic_sharpe (0.891) is the classic PSR-vs-DSR trap. The walk-forward is is_overfitted=TRUE with avg IS 1.21 collapsing to a NEGATIVE avg OOS of -0.748 (windows [-3.60, 0.61, 0.75], one catastrophic), and the holdout 'pass' is spurious (ratio 0 computed against the negative WF-OOS). Sensitivity FAILED with 2 cliffs on the core mechanism parameters (ema_period, vol_lookback) atop an otherwise uniform ~0.5-Sharpe plateau. The result is also outlier-carried and inflated: optimized end_unrealized_pct=43.7% (the +190% headline is mostly un-exited open MTM), return_skew 3.77 / kurtosis 36.7, an implausible +40.97% single-day return (2021-01-02), and the optimizer nearly doubled position sizing (avg_position_pct 31.6%->63.7%, ann. vol 22%->50%) to chase return while Sharpe barely moved (0.551->0.586). Not iterate (attempt 1 of 2): OOS is negative, the two core signal params are cliffs, and the selected Sharpe sits below the luck bar — there is no robust region to tune toward, so a second best-of-225 sweep re-finds the same sub-significant, outlier-carried config. Not revise_hypothesis: single-asset DOGE volume-surge momentum is not a proven mechanism stranded on a dead target — it is a weak intraday-momentum edge that fails deflation, matching the session's recurring single-asset momentum failure signature. FAILURE PATTERN: a clean, low-parameter single-asset 4H volume-confirmed momentum long-short on DOGE produces a modest ~0.55 base Sharpe that optimization cannot lift above the 1.24 best-of-225 luck bar (DSR 0.083); its walk-forward is overfit with a NEGATIVE average OOS (-0.748) and its edge rests on a handful of parabolic outlier days (kurtosis 37, +41% single bar, 44% open MTM). A stable sensitivity plateau and a flattering holdout ratio against a negative OOS baseline do not rescue an edge that fails deflated Sharpe.

Implementation

DOGEUSDT 4H volume-surge momentum, long+short, pure OHLCV on BINANCE USD-M futures. Continuous signal is the per-bar EMA(close) slope in percent (sign=direction, magnitude=conviction). Enters in the slope direction only when the current bar's volume surges >= vol_mult x the mean of the prior vol_lookback bars AND |slope| >= slope_thresh. Exits on EMA-slope reversal, ATR take-profit (tp_atr), ATR stop (sl_atr), or a max-hold time stop. ATR-risk-based sizing capped by max_notional_frac.

Verification Results

Verification failed (Layer 2 — synthetic scenarios): Parameters used: ['sl_atr', 'tp_atr', 'risk_pct', 'vol_mult', 'atr_period', 'ema_period', 'min_notional', 'slope_thresh', 'vol_lookback', 'max_hold_bars', 'size_precision', 'max_notional_frac'] Check that __init__ sets all attributes from self.parameters.get(). - steady_uptrend: TypeError: DogeFourHourVolumeSurgeMomentumLongShort._ema() takes 2 positional arguments but 3 were given (bar timestamp: 1735690980000) - steady_downtrend: TypeError: DogeFourHourVolumeSurgeMomentumLongShort._ema() takes 2 positional arguments but 3 were given (bar timestamp: 1735690980000) - flat_ranging: TypeError: DogeFourHourVolumeSurgeMomentumLongShort._ema() takes 2 positional arguments but 3 were given (bar timestamp: 1735690980000) - volatility_spike: TypeError: DogeFourHourVolumeSurgeMomentumLongShort._ema() takes 2 positional arguments but 3 were given (bar timestamp: 1735690980000) - zero_volume: TypeError: DogeFourHourVolumeSurgeMomentumLongShort._ema() takes 2 positional arguments but 3 were given (bar timestamp: 1735690980000) - price_gap: TypeError: DogeFourHourVolumeSurgeMomentumLongShort._ema() takes 2 positional arguments but 3 were given (bar timestamp: 1735690980000)

Backtest Review

Sharpe: 0.55
Total return: 89.10%
Max drawdown: 25.93%
Trades: 488
Win rate: 36.9%
Profit factor: 1.16

Ample, well-distributed sample: 488 trades over 6 years (280 long / 208 short), returns spread across multiple years (2022 +29%, 2024 +53%, 2025 +42%) rather than one outlier

Genuine positive edge: total_return +89%, Sharpe 0.55, Sortino 0.89, profit_factor 1.155, expectancy +$142/trade, positive alpha 0.062, PSR 0.877, healthy asymmetry (avg_win $2,864 vs avg_loss $1,449, tail_ratio 1.73)

Well-behaved risk and costs: max_drawdown 25.9%, avg_position_pct 31.6% (correctly sized, not over-levered), commission 3.48% of gross, impact 13%, capacity $5.8M — fee-viable at deployable scale

Moderate kurtosis (10.8) — not an outlier-carried artifact

Modest Sharpe (0.55) with sharpe_ci_low -0.34 straddling 0 — may not clear the deflated-Sharpe luck bar after best-of-225 selection; the optimizer/robustness gates must scrutinize this

Recent-regime decay: 2026 -9.7% with rolling Sharpe running negative through 2026 — the holdout must confirm the edge persists into the untouched recent window

Some years negative (2020 -15.7%, 2023 -23%), so the edge is regime-dependent

Analysis

Clean, low-parameter single-instrument design with a coherent volume-confirmed momentum thesis

Non-degenerate trade count (488 base / 416 optimized) over a long 1088-day sample

Sensitivity plateau is broadly stable (most params non-cliff) and commission drag is modest (3.5% of gross)

Failed deflated Sharpe: DSR=0.0832 (vs 0.95), is_significant=false, optimized Sharpe 0.586 BELOW the 225-trial expected-max luck bar of 1.2438 — indistinguishable from best-of-N noise

Sharpe CI straddles 0 (sharpe_ci_low -0.4568)

Walk-forward is_overfitted=TRUE: avg IS 1.21 collapses to avg OOS -0.748, with OOS windows [-3.60, 0.61, 0.75] (one catastrophically negative)

Holdout 'pass' is spurious — ratio 0 measured against a negative WF-OOS baseline, not a real forward validation

Sensitivity FAILED with 2 cliffs on core mechanism params (ema_period, vol_lookback)

Optimized headline is 43.7% un-exited open MTM (end_unrealized_pct), outlier-carried (skew 3.77, kurtosis 36.7, +40.97% single day 2021-01-02)

Optimizer levered up (avg_position_pct 31.6%->63.7%, ann. vol 22%->50%) to chase return; Sharpe barely improved (0.551->0.586)

Outcome Summary

A clean single-asset momentum edge that clears the initial backtest gate on a ~0.55 Sharpe still fails once multiple-testing deflation is applied — a high probabilistic Sharpe means nothing if the deflated Sharpe, walk-forward OOS, and sensitivity all reject it, so future single-asset momentum ideas should be screened against the best-of-N luck bar before investing optimization time.

The analyst abandoned it after optimization on attempt 1 of 2: it failed the deflated Sharpe test (DSR 0.083 vs 0.95 bar, not significant), with the optimized Sharpe 0.586 sitting below the 225-trial expected-max luck bar of 1.244 and a Sharpe CI straddling zero (low -0.457). Walk-forward was flagged overfit (avg in-sample 1.21 collapsing to avg out-of-sample -0.748), sensitivity failed with 2 cliffs on core params (ema_period, vol_lookback), and the gains were outlier-carried (kurtosis 36.7, a +41% single-day bar, 43.7% un-exited open MTM).

A single-instrument, low-parameter long-short strategy that entered DOGEUSDT 4H moves in the direction of an EMA-slope-confirmed trend only when volume surged well above its recent average, betting that retail-FOMO volume spikes create persistent 1-5 day momentum.

The initial backtest showed a positive but modest edge: +89.1% total return, Sharpe 0.55, profit factor 1.155, and 25.9% max drawdown over 488 trades (36.9% win rate) across a 1088-day sample. Optimization lifted the headline to +190% return and Sharpe 0.586, but nearly doubled position sizing (avg 31.6%→63.7%) and volatility (22%→50%) while barely improving risk-adjusted return.

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