Strategies

Trend-Efficiency-Gated Dual-Timeframe Momentum, Long-Short (LTCUSDT.BINANCE USD-M, 4H primary + 1D confirm) — take the proven momentum-confluence edge ONLY when Kaufman Efficiency Ratio confirms a CLEAN directional trend, sit flat in choppy/low-efficiency tape, ATR-trailing exit, 3-parameter

Outcome: Abandoned

LtcEfficiencyGatedDualTimeframeMomentumLS

Outcome Summary

This was iteration 3 of a long-short dual-timeframe momentum strategy on LTC USD-M perp, differentiated from its dead predecessors by a Kaufman Efficiency Ratio gate meant to admit only clean, straight-line trends and reject fee-eating chop. Over a fair 394-trade, 6.7-year backtest it lost 29.5% with a Sharpe of -0.20, profit factor 0.76, and a per-trade return of -0.237% — negative before costs, since commissions were only 5.76% of gross. The backtest reviewer judged the ER-gated premise falsified on LTC and abandoned it before optimization, as curve-fitting a signal that is negative gross could not produce an edge. It ends as another casualty of the single-name dual-TF momentum-confluence clone family (BNB x2, XRP, DOT, SOL, LINK) that has repeatedly failed in this factory.

Strategy report

Hypothesis

A LONG-SHORT, single-instrument, pure-OHLCV momentum strategy on LTCUSDT.BINANCE (USD-M perpetual), the ONE mechanism family that has ever reached paper-stage in this factory (BTC/ADA/ETH dual-TF momentum-confluence survivors). Two deliberate departures from the clones that keep dying: (1) a FRESH, deep-liquidity, 2020-listed instrument that has NEVER been run on this mechanism — the abandoned clones (BNB x2, XRP, DOT, SOL, LINK) all overfit because they re-pointed the identical recipe at lower-quality alts; LTC is a distinct payment-coin with genuine multi-year directional regimes and is not in that graveyard. (2) a real mechanistic differentiator vs both the dead EMA-only confluence and the dead vol-EXPANSION gate (SOL vol-expansion trend died fee_edge, PF 1.02): entries are gated by the KAUFMAN EFFICIENCY RATIO (ER = |net move over N bars| / sum of |bar-to-bar moves| over N bars), which measures DIRECTIONAL cleanliness, not vol level. High ER means price is trending with little back-and-forth (few fee-eating reversals inside the hold); low ER means chop, where momentum whipsaws and fees dominate. Gating on ER filters out precisely the low-efficiency tape that turned SOL/XRP/LINK momentum into sub-fee churn, WITHOUT keying on the volatility-level knob that is exhausted. 4H primary bar gives adequate trade count (~200-350 over 6y) so the walk-forward is not underpowered (the BNB/DOT overfit deaths were flagged ~63-73 trades vs 100 needed); the 1D confirm keeps direction anchored to the macro trend.

Iteration 3 addresses the backtest gate's verdict (judge on the equity curve: total_return -7.47% with a near-zero Sharpe = a null/slightly-negative result, and the sign incoherence between the two near-zero series). Two coded defects, not a lost premise, explain a null: (1) THE EFFICIENCY GATE WAS NON-BINDING. Expected ER on a random walk over N bars is ~1/sqrt(N); at er_period=12 that is 0.289, and the previous er_threshold default was 0.30 — the gate sat exactly at the chop baseline, admitting roughly half of all tape, so the strategy was behaviourally identical to the dead EMA-only confluence clones and the hypothesis's actual differentiator was never tested. er_threshold now defaults to 0.42 with declared bounds [0.32, 0.60], i.e. strictly above the random-walk value, so the optimizer cannot search it back into the always-on region. (2) THE ER-COLLAPSE EXIT MANUFACTURED FEE CHURN. Exiting whenever ER dipped below 0.5x the threshold fired inside ordinary consolidations of healthy trends, flattening positions the trailing stop had not touched, and with no hysteresis the next bar could re-enter the SAME trend at a worse price — each such round trip paying ~0.10% taker on a strategy whose gross edge is small. That exit is removed (which also returns the design to the hypothesis's stated 3-parameter surface: er_period, er_threshold, trail_atr_mult) and replaced by a calendar-anchored cooldown_hours=24 read off the closing bar's own ts_init (restart-safe, no uptime bar counters, identical across backtest/paper). Everything that passed earlier layers is untouched: imports, class structure, the continuous ER-weighted momentum signal (er * fast_mom, computed every bar), and iteration 2's volatility budget (risk_pct 0.0075, max_notional_pct 0.25) that already fixed the -55% compounding-drag blowup. Trade count stays adequate — the gate is more selective but the removed premature exit and the surviving trailing stop lengthen holds rather than eliminate entries.

Negative expectancy on a decisive 394-trade / 6.7-year sample: avg_trade_return_pct -0.237% (below ZERO, far under the Binance USD-M ~0.15% fee floor), profit_factor 0.76, expectancy -$105.73/trade, Sharpe -0.20, total_return -29.5%, CAGR -7.83%. Commission is only 5.76% of gross, so this is a GROSS-negative signal, not fee-drag — the Kaufman efficiency-ratio gate does not select profitable tape on LTC. The strategy loses in the normal (-16.1%) and stressed (-22.6%) vol terciles where a clean-trend gate should help most, and in 6 of 7 calendar years. The trades faithfully implement the hypothesis (394 signaled = 394 submitted, balanced long/short), so this is a fair test and the ER-gated dual-TF momentum premise is falsified on LTC — exactly QA's concern borne out on full history. Optimization cannot lift an edge that is negative before costs, so 2 hours of compute would only curve-fit noise. This is iteration 3 of a single-name dual-TF momentum-confluence clone family that has died repeatedly (BNB x2, XRP, DOT, SOL, LINK). abandon_class = negative_expectancy (premise_falsified).

Implementation

LTCUSDT.BINANCE USD-M perpetual, 4-HOUR bars, pure OHLCV long/short dual-timeframe momentum gated by the Kaufman Efficiency Ratio. Enters long when both the 2-day (12x4H) and the 7-day (42x4H, daily-horizon) momentum legs are positive AND ER over the last 12 bars is at or above 0.42 (well above the ~0.29 random-walk expectation for a 12-bar window), short on the mirror condition; sits flat otherwise. Exits on a daily-confirm-leg flip against the position or a 3.5-ATR trailing stop from the best price since entry, with a 24-hour calendar-anchored re-entry cooldown after every exit to stop the strategy re-buying the same trend one bar after being stopped out. Sizing is ATR-risk based (0.75% of equity per 1.5-ATR initial stop) and hard-capped at 25% of equity notional, leverage 1.0.

Verification Results

The optimizer should search er_threshold within its declared [0.32,0.60] bounds and trail_atr_mult; if full-history avg_trade_return_pct stays below ~0.15%, abandon rather than tune.

Trend-following design with a 27.8% win rate and PF 0.81 in the sandbox: the ER gate + dual-horizon agreement + ATR trailing exit are structurally correct, but the per-trade edge is currently below the round-trip taker cost on this smoke-test draw. This is a performance observation on one unoptimized parameter set, not a code defect.

Sandbox smoke test is net-negative: avg_trade_return_pct -0.247% (below ZERO and below the ~0.15% Binance USD-M fee floor), profit_factor 0.81, Sharpe -0.23, win_rate 0.278 over 389 trades. Commission is only 5.55% of gross, so this is a gross-negative signal, not pure fee-drag. The strategy also loses in the 'normal' vol tercile (-27.6%) where a clean-trend gate should help most, with the only positive tercile being 'calm' (+4.3%). This is iteration 3 of a single-name dual-TF momentum-confluence clone family that has died repeatedly (BNB x2, XRP, DOT, SOL, LINK — overfit/fee_edge); the ER gate is a genuine mechanistic differentiator but the smoke test gives no early sign it clears costs. Suggest the analyst check per-trade expectancy on full history before spending optimizer time. Advisory only — not a block.

Backtest Review

Sharpe: -0.20
Total return: -29.48%
Max drawdown: 41.30%
Trades: 394
Win rate: 26.9%
Profit factor: 0.76

Adequate sample for a decisive read: 394 trades over 6.7 years (14,584 4H bars), balanced long/short (213/181).

Trades faithfully implement the hypothesis (394 signaled = 394 submitted, no drops), so this is a fair test of the ER-gated dual-TF momentum premise, not a coding artifact.

The ER gate is a genuine mechanistic differentiator vs the dead EMA-only clones.

Negative expectancy on a reliable sample: avg_trade_return_pct -0.237% is below ZERO and far under the Binance USD-M ~0.15% fee floor.

profit_factor 0.76 with commission only 5.76% of gross -> this is a gross-negative signal, not fee-drag; the ER gate does not select profitable tape.

Sharpe -0.20 (CI [-0.94, 0.48]), Sortino -0.22, total_return -29.5%, CAGR -7.83%, recovery_factor -0.88, information_ratio -0.49, alpha negative.

Loses in the normal (-16.1%) and stressed (-22.6%) vol terciles where a clean-trend gate should help most; only 'calm' is near-flat.

Losing in 6 of 7 calendar years; win_rate 26.9%, max_consecutive_losses 14.

Iteration 3 of a single-name dual-TF momentum-confluence family that has died repeatedly (BNB x2, XRP, DOT, SOL, LINK).

Analysis

Backtest gate: - metrics unreliable: Incoherent: Sharpe 0.08 disagrees in sign with total_return -7.47% — the Sharpe is computed from the mtm_equity_curve series while the return comes from the equity curve. Judge on the equity curve.

Outcome Summary

The Kaufman Efficiency Ratio gate did not select profitable tape on LTC, and re-pointing the repeatedly-dead single-name dual-TF momentum-confluence family at a fresh instrument does not manufacture an edge that is negative before costs.

It was abandoned at the pre-optimization backtest review: the analyst ruled the edge gross-negative (not fee-drag) with negative expectancy below zero and far under the ~0.15% Binance USD-M fee floor, so optimization was skipped and analysis, optimization, and risk review were never reached.

A long-short dual-timeframe (4H primary + 1D confirm) momentum strategy on LTCUSDT.BINANCE USD-M perp that only entered when the Kaufman Efficiency Ratio confirmed a clean directional trend, aiming to filter out the choppy, fee-eating tape that killed prior momentum-confluence clones.

On a decisive 394-trade / 6.7-year sample (14,584 4H bars, balanced 213 long / 181 short), it returned -29.5% (CAGR -7.83%), Sharpe -0.20, Sortino -0.22, profit factor 0.76, win rate 26.9%, with avg_trade_return_pct of -0.237% and expectancy of -$105.73/trade. Commission was only 5.76% of gross and it lost in 6 of 7 calendar years, including the normal (-16.1%) and stressed (-22.6%) vol terciles.

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