BtcEfficiencyRatioGatedMomentumLS

BTC Efficiency-Ratio-Gated Momentum, Long-Short (Single-Instrument BTCUSDT.BINANCE Perp — Trade Momentum ONLY When the Move Is EFFICIENT: Kaufman Efficiency Ratio (Directional Travel ÷ Total Path) Is High = Clean Low-Noise Trend, ATR-Trailing, 3-Parameter)

A LONG-SHORT, single-instrument, pure-OHLCV momentum strategy on the BTCUSDT.BINANCE USD-M perpetual whose selectivity comes from TREND PATH-CONSISTENCY — Kaufman's Efficiency Ratio (ER = |net price change over N bars| ÷ Σ|bar-to-bar changes| over N) — rather than the banned sign-based dual-timeframe confluence (L38), or the magnitude/derivative/vote constructions already in my queue. The thesis: momentum only pays when the trend is CLEAN (price travels efficiently in one direction, high ER), and whipsaws in choppy, low-efficiency periods (low ER) where the price zig-zags with little net progress. It computes momentum direction and the Efficiency Ratio, and takes a position ONLY when ER is high (an efficient, low-noise trend is in force) AND momentum agrees. This is a genuinely different selectivity mechanism from timeframe-agreement (survivor), magnitude-strength (pending), acceleration/2nd-derivative (pending), and multi-signal vote (pending): it gates on the SMOOTHNESS/consistency of the path. It is also distinct from the dead KAMA idea, which used ER to adapt a moving average's SPEED (an indicator construction); here ER is a direct binary GATE on a simple momentum entry. It stays on BTC (the one instrument with demonstrated tradability), single perp execution (dev-safe), pure OHLCV (well-covered, no coverage/no-trades risk), long-short (fills the 13.4% bucket vs 86.6% long-only), 4H cadence for a measurable sample, ATR trail. Deliberately 3-parameter to resist overfit.

MEASURED BEFORE SUBMITTING (real BTCUSDT.BINANCE 4H catalog, 14,412 bars, 2019-12 to 2026-07, using the exact exit engine shipped here). At the shipped defaults: 174 trades, 44% win rate, +1.07% average net return per trade after a 0.10% round trip, PF 1.57, exit mix 165 trail / 9 max-hold / 0 reversal. THE ABLATION IS EMPHATIC AND IS THE REASON I'M CONFIDENT IN THE MECHANISM. Same window, same trail, same exits, but with the efficiency gate removed so the strategy simply follows the sign of the same momentum: 645 trades at +0.14% per trade and PF 1.07. Adding the efficiency requirement discards roughly three quarters of those entries and lifts per-trade return to +1.07% and PF from 1.07 to 1.57. The discarded entries were overwhelmingly the ones that were not paying, which is exactly the hypothesis's claim: momentum works when the path is clean and is shredded by chop. This is the largest ablation effect I have measured across the recent batch of hypotheses. THE GATE ONLY WORKS AT LONG WINDOWS -- IMPORTANT FOR OPTIMIZATION. Gated versus ungated at matched windows: at 3.3 days the gate HURTS (PF 1.05 vs 1.10), at 5 days it helps (1.22 vs 1.06), at 7 days it is marginal (1.15 vs 1.11), and from 10 days out it is strongly positive. Efficiency needs enough bars to distinguish a genuinely clean trend from a lucky straight line; over a handful of bars almost anything looks efficient. The long-window region is a real plateau rather than one lucky cell -- at er_min 0.25, windows of 10/12/14/16/20 days give PF 1.57/1.66/1.72/1.52/1.76. The Analyst should expect optimization to stay at or above ~10 days and should treat a short-window optimum as evidence the efficiency premise stopped applying. DEFAULT CHOSEN FOR SAMPLE SIZE, NOT SCORE. The shipped 10-day window is the LOW end of that plateau because it carries the largest sample (174 trades versus 145/116/100/79 further out). The longer windows score better but on progressively thinner evidence, and I would rather ship the most measurable point in the good region than its peak -- er_min 0.30 (PF 1.66) and a 14-day window (PF 1.72) both beat the defaults on paper. The intent is that optimization discovers the better cells rather than starting inside them. THE CAVEAT TO READ FIRST: THE RECENT YEAR IS NEGATIVE. The trailing 365-day window gives 26 trades at PF 0.80 and -0.35% per trade, even though the full history is strongly positive. Clean-trend following on BTC has had a poor recent regime. The sandbox and any holdout carved from recent data may therefore look materially worse than the full backtest, and that gap should be read as regime rather than as a broken implementation. Trade count is healthy in both windows, so there is no zero-trade risk either way. DESIGN CHOICES THAT MATTER. (1) The signed-ER formulation is the whole trick: writing `if ER > threshold and momentum > 0` would be two ANDed conditions -- the construction that most reliably produces a strategy that fires almost never -- whereas dropping the absolute value from Kaufman's numerator makes the conjunction structural, keeps the score varying on every bar, and yields a threshold that is directly interpretable as an efficiency percentage rather than something needing recalibration per regime. (2) It is self-normalizing: numerator and denominator are both in price units, so no sigma estimate is needed anywhere, which is why warmup is 62 bars (~3% of a 365-day 4H window) versus ~180 for the volatility-normalized siblings. (3) This is deliberately NOT KAMA: KAMA feeds ER into a moving average's smoothing constant so efficiency only reaches the decision indirectly and diluted; here ER IS the decision variable. (4) The path sum is maintained as a running deque rather than rescanned per bar, and it telescopes exactly to the numerator's span so the ratio stays internally consistent; the result is clamped to [-1,+1] anyway so accumulated float drift in the running sum can never emit an out-of-range ratio. (5) The reversal exit fired ZERO times at every setting tested; I kept it as a safety net for wide-trail settings and flagged it as inactive rather than leaving unexplained dead-looking code. RISK AND VENUE. Sizing is anchored to a 3-ATR gap floor rather than to the trail itself, so a tight trail cannot produce the largest position where the stop is least enforceable -- the pattern behind the 90%+ drawdowns in this book -- and gross notional is capped at 0.5x equity independently of the equity path, with risk_frac locked so an optimizer cannot buy Sharpe with notional instead of edge. The ATR used for the trail and for sizing excludes the bar being decided. Turnover is ~27 entries a year against a +1.07% per-trade average, so fees are a rounding error rather than the deciding term. Futures (BINANCE USD-M MARGIN) is required rather than preferred: an efficient DOWNTREND is traded SHORT, which a CASH spot account cannot do. leverage stays 1.0 and no sizing path reads it, so there is no leverage-set-but-unused mismatch. Pure OHLCV, single instrument, no supplementary feeds -- none of the coverage failure modes that have dominated recent verification loops apply here.

failed deflated Sharpe: DSR=0.000, expected-max=2.406 over 225 trials — the selected Sharpe (0.625) is far below the Sharpe best-of-N noise would produce, so the edge is indistinguishable from a lucky search. This is compounded by TWO unwaivable hard-gate failures: PBO 0.5794 > 0.5 (parameter selection more likely than not overfit) and holdout FAILED (sharpe 0.682, ratio 0.251 << 0.70 — the honest recent-data test collapses to a quarter of the walk-forward OOS). Sensitivity FAILED with 2 cliff parameters (er_min and er_bars both cliffs), confirming the developer's own disclosure that the efficiency gate 'only works at long windows' — the selected region is fragile, not a plateau. The optimizer shrank the strategy to 90 trades (below the 100 floor) on a config that dropped full-history return from 66% to 30% and is negative in 2022 and 2025. Per the overfit -> abandon rule this is not an iterate or revise_hypothesis candidate: 225 trials already selected the overfit best, the sensitivity surface is cliffed (nothing robust to tune toward), and reframing the same Kaufman efficiency-ratio mechanism onto another target just overfits again — every sibling in the family (ETH efficiency-gated TSMOM, AVAX HYPERLIQUID, BTC KAMA, efficiency-filtered majors basket) died overfit. avg_trade_return_pct 1.24% clears the fee floor, so this is a pure multiple-testing/holdout-collapse failure, not fee fragility. Abandon rather than spend the final iteration re-overfitting a cliffed, decayed signal.

Long/short BTCUSDT.BINANCE USD-M perpetual on 4H bars that gates momentum on the SMOOTHNESS of the price path using Kaufman's Efficiency Ratio. It drops the absolute value from Kaufman's numerator to form a SIGNED efficiency ratio, signal = (close_t - close_(t-W)) / sum(|close-to-close moves| over W): the magnitude is exactly the ER (how directly price travelled) and the sign is exactly the momentum direction. Thresholding that one value at er_min is therefore precisely equivalent to 'ER is high AND momentum agrees', so the conjunction is structural rather than a compound gate. Goes LONG when signal >= er_min (clean uptrend), SHORT when signal <= -er_min. Naturally bounded in [-1,+1] and self-normalizing -- no volatility estimate appears anywhere in the signal, so warmup is only 62 bars. Exits: an ATR trailing stop from the best close since entry (detected intrabar, booked at the close), an efficient-reversal safety net, and a 10-day max hold. Sizing is risk-first and gap-aware (1.5% of equity over a 3-ATR excursion), capped at 0.5x equity notional, leverage 1.0. Three tunable parameters: er_bars, er_min, trail_atr_mult.

Decisively negative sandbox (PF 0.71, -0.303%/trade) — disclosed recent-regime loser; high risk of failing OOS/holdout; likely abandon

Pure-OHLCV momentum, efficiency gate only works at long windows — constrain er_bars ≥ ~10 days

avg_trade_return_pct 1.181% — 7.9x the fee floor; a thick per-trade edge

profit_factor 1.40, Sharpe 0.612 with sharpe_ci_low +0.0035 (CI above zero), PSR 0.974, positive alpha +0.0476

max_drawdown 9.8% — tightest in the batch; calmar 5.74; decisive 167-trade genuine long-short

Differentiated non-banned mechanism (signed efficiency ratio); ablation shows the gate lifts PF 1.07→1.57; positive in 6 of 7 years

DISCLOSED: trailing-365-day window is NEGATIVE (PF 0.80, -0.35%/trade, 26 trades) — the recent regime is adverse for clean-trend following

2025 -7.9% and rolling Sharpe trends down through 2026 (to -2.1) — the holdout (recent data) and latest OOS window are at material risk of failing

Sharpe CI low sits right at the zero boundary; the edge concentrates in older regimes (2023 +22%, 2024 +13.8%)

The developer's own sweep shows the gate only works at long windows — optimizer must stay long, and walk-forward must confirm it isn't a narrow ridge

avg_trade_return_pct 1.24% on the optimized config — well above the fee floor, so the failure is not fee fragility

Optimized headline looks clean in isolation: PF 1.78, win_rate 0.53, max_drawdown 5.0%

Walk-forward is_overfitted=false and avg OOS Sharpe 2.72 is positive (the one green light)

TWO unwaivable hard-gate failures: PBO 0.5794 > 0.5 (selection more likely than not overfit) AND holdout FAILED (sharpe 0.682, ratio 0.251 << 0.70)

Deflated Sharpe = 0.000 vs 0.95; expected_max_sharpe_annualized 2.406 over 225 trials means the selected Sharpe is below best-of-N noise; is_significant=false, sharpe_ci_low 0.004

Sensitivity FAILED with 2 cliff parameters (er_min and er_bars) — the selected region is fragile, confirming the developer's own 'only works at long windows' caveat

Optimizer shrank to 90 trades (below the 100 floor); full-history return collapsed 66%->30% and the optimized config is negative in 2022 (-2.6%) and 2025 (-1.9%)

Kaufman efficiency-ratio momentum family with zero survivors: ETH efficiency-gated TSMOM, AVAX HL, BTC KAMA, and majors-basket siblings all abandoned overfit

Optimized Sharpe only 0.625; information_ratio -0.65; recent regime negative (developer flagged trailing-365d PF 0.80)

This strategy proposed a genuinely novel, non-banned selectivity mechanism: gate a simple BTC momentum entry on the smoothness of the price path via a signed Kaufman Efficiency Ratio, taking positions only in clean, efficient trends and trailing out with an ATR stop. Its initial backtest was the best in its batch — thick per-trade edge (+1.18%), PF 1.40, tight 9.8% drawdown across 167 long-short trades — and the analyst waved it through to optimization despite an openly disclosed negative recent year. But the 3-phase optimization exposed the edge as a product of search rather than signal: deflated Sharpe collapsed to 0.000, PBO exceeded 0.5, the holdout fell to a quarter of walk-forward OOS, and sensitivity revealed two parameter cliffs. With every sibling in the Kaufman efficiency-ratio family already dead from overfitting, the analyst abandoned it after a single iteration rather than re-overfit a cliffed, regime-decayed signal.

A backtest edge that is strong only in older regimes and concentrated at long parameter windows will not survive multiple-testing correction or a recent-data holdout — the developer's own 'only works at long windows' and negative-recent-year disclosures foreshadowed the overfit verdict.

The analyst abandoned it at the post-optimization ANALYZING stage on two unwaivable hard-gate failures (PBO > 0.5 and a failed holdout), plus deflated Sharpe 0.000 versus a best-of-N noise expectation of 2.406 over 225 trials, a sensitivity surface with two cliff parameters (er_min and er_bars), and a sub-100 trade count.

A long-short, single-instrument BTCUSDT.BINANCE 4H perpetual momentum strategy that trades momentum only when the price path is efficient (clean, low-noise trend), using a signed Kaufman Efficiency Ratio as a single continuous entry score with an ATR trailing stop.

The initial backtest was strong in isolation — 167 trades, +1.18% avg per-trade return (7.9x the fee floor), profit factor 1.40, Sharpe 0.612, max drawdown 9.8% over 2019-2026 — but the developer disclosed the trailing 365-day window was negative (PF 0.80). After optimization the config shrank to 90 trades with PF 1.78 and Sharpe 0.625, yet failed the honest recent-data tests: PBO 0.5794, holdout ratio 0.251, and deflated Sharpe 0.000.