Skip to content
All strategies

Strategies

LINK Volume-Confirmed Trend Momentum, Long-Short with Volatility-Expansion Gate (LINKUSDT.BINANCE USD-M, 4H + 1D confirm, dual participation+expansion gate, ATR-trailing, 3-parameter)

Outcome: Abandoned

LinkVolumeExpansionConfirmedTrendMomentumLS

Outcome Summary

Passing hard OOS/holdout gates is not enough when deflated Sharpe is near zero and all free parameters sit on cliffs — a slow single-name dual-timeframe momentum-confluence mechanism on LINK can't produce enough trades to separate its edge from best-of-N search noise.

Strategy report

Hypothesis

Implements the hypothesis exactly: LINKUSDT.BINANCE USD-M 4H primary bars, a daily-scale (42 x 4H = 7 day) trend confirmation, a DUAL gate of participation (volume surge vs median baseline) AND volatility expansion (fast/slow ATR ratio), ATR trailing exits, and exactly three tunable parameters (vol_surge_mult, expansion_mult, trail_atr_mult) with declared _param_bounds matching the in-code clamps. The two gates measure different things - how much traded vs how far price travelled - so their conjunction filters……Show moreShow less

Implements the hypothesis exactly: LINKUSDT.BINANCE USD-M 4H primary bars, a daily-scale (42 x 4H = 7 day) trend confirmation, a DUAL gate of participation (volume surge vs median baseline) AND volatility expansion (fast/slow ATR ratio), ATR trailing exits, and exactly three tunable parameters (vol_surge_mult, expansion_mult, trail_atr_mult) with declared _param_bounds matching the in-code clamps. The two gates measure different things - how much traded vs how far price travelled - so their conjunction filters the absorption/churn case a volume filter alone lets through, which is the stated point of adding the expansion gate. Gate thresholds are dosed loosely by default (1.4x median volume ~ 15-20% of bars; 1.05 ATR ratio ~ 35-40% of bars) so the joint gate fires regularly rather than multiplying into zero trades. The daily confirm is derived from the same 4H series rather than a separate 1-DAY leg: identical information, never one bar stale, and no extra-leg alignment barrier that could produce zero evaluated bars. Futures venue is required because the strategy shorts as well as longs; leverage stays 1.0 because the notional cap means no margin is consumed, so nothing is set-but-unused. One caveat for the record: LINK 4H momentum has been a weaker carrier than BTC/ETH in this family, so the expansion gate carries most of the incremental burden of proof - but that is the hypothesis, and it is implemented as asked.

failed deflated Sharpe: DSR=0.0186 (raw 0.0026), expected-max=1.542 over 225 trials — the selected Sharpe (0.740) is BELOW the best-of-N luck bar and is statistically indistinguishable from noise. This is NOT a fee-edge death (avg_trade_return_pct 1.379% clears the floor); it is an overfit-to-search / not-statistically-significant failure. Corroborating: Sharpe CI [-0.0496, 1.4758] straddles zero, is_significant=false, probabilistic_sharpe 0.905<0.95, and the candidate does not survive programme-level FDR (keeps……Show moreShow less

failed deflated Sharpe: DSR=0.0186 (raw 0.0026), expected-max=1.542 over 225 trials — the selected Sharpe (0.740) is BELOW the best-of-N luck bar and is statistically indistinguishable from noise. This is NOT a fee-edge death (avg_trade_return_pct 1.379% clears the floor); it is an overfit-to-search / not-statistically-significant failure. Corroborating: Sharpe CI [-0.0496, 1.4758] straddles zero, is_significant=false, probabilistic_sharpe 0.905<0.95, and the candidate does not survive programme-level FDR (keeps 4 of 243, candidate_p 0.9814). Although the HARD gates pass (OOS 0.981>0, holdout Sharpe 1.007 on 38 trades, PBO 0.4921<=0.5, not overfit), a DSR waiver is not warranted here because the DSR is ~0 rather than marginal AND the strategy is nowhere near promotion tier: optimized Sharpe 0.74 / PF 1.46 / DD 16.4% sits well below the single-name dual-TF momentum-confluence survival tier (Sharpe>=1.5 AND PF>=2.0, per L82/L90 — only ADA cleared it). Iteration cannot help: all three free params (mom_lookback, expansion_mult, vol_surge_mult) are cliff parameters and the whole sensitivity surface tops out ~0.5-0.7 Sharpe, so there is no robust region to tune toward 1.5. The LINK momentum/confluence/expansion family has already died 5+ times (weekly trend-follow overfit, spot regime-gated overfit, 4H ATR-expansion breakout no_edge, HL dual-TF confluence, wide-range-bar burst no_edge), and the oos_vault is itself flagged UNDERPOWERED (~60 trades vs 100 needed) — the mechanism is too slow to validate on LINK's history. Failure pattern: overfit / not-significant after deflation (single-name dual-TF momentum-confluence clone below the survival tier).

Implementation

Long/short 4H trend-momentum on LINKUSDT.BINANCE USD-M perpetual. Each bar it computes a continuous confluence score in ATR units: 0.5 * (2-day momentum + 7-day trend / 2.0), both measured on the same 4H series and normalised by ATR(30). That score is multiplied by two capped gate weights - PARTICIPATION (bar volume / median volume of the prior 60 bars, divided by vol_surge_mult) and VOLATILITY EXPANSION (ATR(10)/ATR(40) divided by expansion_mult) - so conviction decays smoothly as either gate weakens. It enters long (short) only when the signal crosses up through +1.0 ATR (down through -1.0 ATR) while BOTH gates are fully met, and the 7-day trend has the same sign as the 2-day momentum; the cross requirement prevents re-entry churn on a persistent state. It exits on momentum loss with 0.25-ATR hysteresis or on a trail_atr_mult ATR trailing stop from the best close since entry. Sizing risks 2% of equity to the trailing stop, capped at 1x equity notional, leverage 1.0.

Verification Results

Two advisory concerns for the analyst's full-history review: (1) The sandbox regime attribution shows the strategy LOSES in the stressed vol tercile (-8.86%, Sharpe -0.19) while making money in calm (+10.2%) and normal (+27.8%) regimes — but an expansion-gated momentum strategy is supposed to earn precisely when volatility expands, so the expansion gate may not be adding the intended edge in the regime that defines it. (2) The unoptimized per-trade edge (avg_trade_return_pct 0.233%) clears the 0.15% USD-M fee floor by only a modest margin, and impact cost already consumes 14.6% of gross; on the developer's own admission LINK 4H momentum has been a weaker carrier than BTC/ETH in this family. Not blocking — these are performance questions for the real backtest/optimizer.

Backtest Review

Sharpe
0.27
Total return
38.06%
Max drawdown
27.59%
Trades
318
Win rate
38.1%
Profit factor
1.13

Numerically viable on a reliable, large sample: 318 trades over 6.6 years (metrics_reliable=true).

avg_trade_return_pct 0.2795% clears the 0.15% Binance USD-M fee floor with genuine margin; commission is only 3.85% of gross, so the edge is gross-positive, not fee-fragility.

profit_factor 1.127 (>1.05) with positive total_return (+38%) and positive expectancy per trade.

Trades implement the stated mechanism: balanced 185L/133S dual-timeframe momentum entries, ~2d holding consistent with the 2-day fast lookback — no direction/polarity contradiction.

max_drawdown 27.6% is well under the 60% review ceiling.

Sharpe only 0.267 with bootstrap CI [-0.51, 1.07] straddling zero — significance is not yet established (deflation/holdout will be the real test).

Loses in the stressed vol tercile (-6.1%, Sharpe -0.07) — an expansion-gated momentum strategy is supposed to earn precisely when volatility expands (QA concern #1), so the expansion gate may not add the intended edge.

PF 1.127 is thin; this dual-TF momentum-confluence family on single Binance USD-M majors has repeatedly died overfit in the 3-phase deflation below the ADA-winner tier — advisory caution, lowering confidence.

impact_cost_pct 17.3% consumes a meaningful share of gross; LINK is a weaker carrier than BTC/ETH in this family (QA concern #2).

Analysis

Numerically viable per-trade economics: avg_trade_return_pct 1.379% clears the 0.15% Binance USD-M fee floor with wide margin; commission only 2.9% of gross, so this is NOT a fee-edge death.

Holdout technically passed the HARD gate (Sharpe 1.007 on 38 trades, ratio 1.027, consistent_with_oos=true) and walk-forward is not flagged overfit (avg IS 0.826 -> avg OOS 0.981).

Optimization roughly doubled Sharpe (0.267 -> 0.740) and halved drawdown (27.6% -> 16.4%) without collapsing trade count (318 -> 253), so the tuning was not a degenerate curve-fit to a handful of trades.

Deflated Sharpe 0.0186 (raw 0.0026) is essentially ZERO, not merely below the 0.95 floor: the selected Sharpe is statistically indistinguishable from best-of-N noise over 225 trials.

The optimized point Sharpe (0.740) sits BELOW the expected-max-from-luck bar (expected_max_sharpe_annualized 1.542) — i.e. the result is weaker than what the search would be expected to surface by chance alone.

Sharpe 95% CI [-0.0496, 1.4758] straddles zero -> Sharpe not significant regardless of point value; is_significant=false; probabilistic_sharpe 0.905 < 0.95.

Does not survive programme-level FDR (Benjamini-Hochberg q=0.10 keeps 4 of 243; candidate_p 0.9814, candidate_survives=false).

All three free parameters (mom_lookback, expansion_mult, vol_surge_mult) are flagged as CLIFF parameters, and the entire sensitivity grid tops out around 0.5-0.7 Sharpe — there is no robust region to tune toward the 1.5 promotion floor.

Below the only tier this single-name dual-TF momentum-confluence class has ever survived (ADA winner: Sharpe>=1.5 AND PF>=2.0; here 0.74 / 1.46). The LINK momentum/confluence/expansion family has already died repeatedly (weekly trend-follow overfit, spot regime-gated overfit, 4H ATR-expansion breakout no_edge, HL dual-TF confluence, wide-range-bar burst no_edge).

oos_vault flagged UNDERPOWERED (~60 expected trades vs 100 needed) — the mechanism is too slow to validate on available history.

Loses in the stressed vol tercile in the base backtest (Sharpe -0.073, -6.1%); the optimized version only turns it modestly positive (0.363).

Benjamini-Hochberg at q=0.10 over 243 programme candidates keeps 4. A candidate that does not survive here is not distinguishable from the programme's own noise, however good its individual statistics look.

Outcome Summary

Abandoned at ANALYZING after optimization on statistical-significance grounds: deflated Sharpe ~0.0186 (~zero) over 225 trials, point Sharpe 0.740 below the expected-max-from-luck bar 1.542, Sharpe CI straddling zero, and failure of programme-level FDR — an overfit/not-significant failure, not a fee-edge death.

A long/short 4H LINKUSDT.BINANCE momentum strategy that only took directional trades when a daily-scale trend agreed with fast momentum AND both a volume-participation gate and a volatility-expansion gate fired, exiting on lost momentum or an ATR trailing stop.

The initial backtest over 6.6 years (318 trades) returned +38% with Sharpe 0.267, profit factor 1.127, 38% win rate and 27.6% max drawdown; optimization roughly doubled Sharpe to 0.740, raised return to +154%, lifted PF to 1.455 and halved drawdown to 16.4% over 253 trades, and the holdout passed (Sharpe 1.007 on 38 trades).

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