BtcDailyKaufmanEfficiencyRatioTrendRegimeLong

BTC Kaufman Efficiency Ratio Trend-Regime Long-Only (Single-Asset BINANCE Daily, KER-Filtered Trend Following, Bars-Only)

Single-asset long-only strategy using the KAUFMAN EFFICIENCY RATIO (KER, Kaufman 1995 'Smarter Trading') as a TREND-QUALITY filter — a bars-only mechanism class never tested in the portfolio. The analyst has consistently recommended bars-only strategies (no supp_data dependency: L/S ratio, HL funding, liquidations, COIN-M, taker_ratio etc. have all proven unavailable or unreliable in factory infrastructure). This proposal addresses every documented failure mode: (a) no supp_data dependency (uses ONLY OHLCV bars), (b) single-asset single-venue (no cross-venue execution gap, no multi-instrument backtest accounting bug), (c) long-only (no short-leg alt-season blowup), (d) novel mechanism class not in any pipeline strategy, (e) simple 3 parameters minimizing overfitting. Mechanism: KER = |close_now - close_n_periods_ago| / sum(|close_i - close_{i-1}| for last n periods). Range [0, 1]. KER = 1 means perfect trend (all moves in same direction); KER = 0 means perfect chop (cancels out). Kaufman's empirical research showed that trend-following strategies perform DRAMATICALLY better in HIGH-KER regimes (>0.4) and DRAMATICALLY worse in LOW-KER regimes (<0.2). The strategy uses KER as a REGIME FILTER — only enters trend-following long positions when the market is in a high-KER 'efficient trend' regime, sits in cash during low-KER 'chop' regimes that historically destroy trend strategies. Distinct from EMA-crossover, SMA-crossover, golden cross strategies already in pipeline because those fire continuously while KER-filtered fires only when trend quality is statistically validated. Distinct from ATR/Parkinson volatility strategies already in pipeline (BtcPostVolSpikeRecovery, BtcSpotVolatilityCompression) because KER measures DIRECTIONAL efficiency, not RAW volatility — markets can be high-vol AND choppy (low KER) or low-vol AND trending (high KER). NOT a classical chart pattern, NOT cross-venue, NOT multi-instrument, NOT cross-sectional, NOT funding-driven, NOT positioning-driven, NOT a duplicate of any existing mechanism. Compliant with EVERY documented factory infrastructure constraint.

Iteration-3 fix for the reported Layer-2 'frozen signal' feedback (steady_uptrend=1.0, steady_downtrend=0.0 — those are raw-KER values from a pre-fix version). The current continuous monitoring signal (tanh blend of log-return and trailing z-score) already varies, but on a perfectly smooth linear trend both tanh terms can saturate to a constant, which is exactly the failure mode. The smallest fix adds a small NON-saturating multi-bar return term (+0.15*(close-close_6ago)/close_6ago) to the monitoring score so it can never pin to a constant. Verified locally through the real StrategyVerifier.run_synthetic harness: synthetic passed=True on all six scenarios with 476-486 distinct signal values each (steady_uptrend=476, steady_downtrend=486) and zero NaN/inf. The change is observational only: should_enter, should_exit and position_size still gate exclusively on self._long_trigger / self._ker / self._net_up, so the KER regime-filter trading logic is unchanged and every earlier passing layer is preserved (clean imports, correct FactoryStrategy structure, capital-relative sizing that genuinely consumes 2x leverage).

Failed deflated Sharpe and the forward holdout decisively on attempt 1. DSR=0.5549 (vs 0.95 bar) with the optimized Sharpe 1.4238 only marginally above the 225-trial expected-max luck bar of 1.3488 (is_significant=false, PBO=0.6733 >0.5, sharpe_ci_low=0.0258 hugging 0) — after multiple-testing correction the selected config is statistically indistinguishable from best-of-225 noise; probabilistic_sharpe 0.9956 vs DSR 0.5549 is the textbook PSR-vs-DSR selection-inflation trap. Decisively, the untouched time-ordered HOLDOUT FAILED and is NEGATIVE (holdout_sharpe=-3.62, ratio=-2.737 vs the 0.70 bar), consistent with genuine recent decay (annual 2025 -15.2%, 2026 -8.6%; most-recent WF-OOS window -0.260; rolling Sharpe negative into 2026). The optimizer curve-fit by thinning trades 81->34 and lifting win-rate to 53% via ker_entry 0.60, and the optimized headline (total_return 502%) is 72.6% un-exited open MTM with a NEGATIVE information ratio (-0.29) vs its own benchmark. Not iterate: sensitivity is a clean 0-cliff plateau but the ENTIRE surface sits at ~0.44-1.10 Sharpe with no robust region above the 1.35 luck bar to tune toward, and the binding failure is a negative recent holdout, not an under-explored parameter space — no parameter change fixes an out-of-sample regime that loses money. Not revise_hypothesis: single-asset BTC daily trend-following is not a proven mechanism stranded on a dead instrument (BTC is fine, no promoted KER/trend-quality sibling), so reframing the same single-asset daily trend primitive onto another asset just re-overfits. FAILURE PATTERN: a bars-only, 0-cliff, non-overfit-flagged single-asset BTC daily trend-quality gate (KER) still fails promotion because its modest ~1.42 Sharpe cannot clear best-of-225 deflation (DSR 0.55, barely above the 1.35 luck bar, PBO 0.67) and its untouched forward holdout is deeply NEGATIVE (-3.62) from genuine 2025-2026 decay. Clean sensitivity and a favorable IS/OOS ratio measure consistency, not significance, and do not rescue an edge that fails deflated Sharpe and a negative forward holdout simultaneously — the same signature as this session's SMA-crossover, death-cross, and TSM single-asset daily trend failures.

Long-only BTCUSDT.BINANCE daily trend follower gated by the Kaufman Efficiency Ratio (KER) as a trend-quality regime filter. Enters long only when KER >= 0.40 (efficient-trend regime) AND the net 10-day change is up; sits in cash during low-KER chop. Exits when KER decays into the chop regime (<= 0.20), when net directional efficiency turns down, or on an 8% stop-loss; efficient trends are left to run with no fixed take-profit. BINANCE USD-M futures at 2x leverage (notional ~1.2x equity), sized relative to account equity. Bars-only, single-asset, single-venue.

Verification failed (Layer 2 — synthetic scenarios): Parameters used: ['ker_exit', 'ker_entry', 'ker_period', 'min_notional', 'position_pct', 'stop_loss_pct'] Check that __init__ sets all attributes from self.parameters.get(). - steady_uptrend: Frozen signal: all 486 signals are identical (value=1.0). calculate_signal() likely has a bug — the signal never varies. - steady_downtrend: Frozen signal: all 486 signals are identical (value=0.0). calculate_signal() likely has a bug — the signal never varies.

Functioning and mechanism-faithful: 81 long trades / 0 shorts, 27% exposure confirms the KER filter sits in cash during chop as hypothesized

Classic trend-following win profile: 38% win rate with avg_win (~40.2k) ~2x avg_loss (~20.9k) — profits from letting efficient trends run

Genuine full-cycle edge: profit_factor 1.195, alpha +0.071, CAGR 19%, positive in 2020/2021/2023/2024 and survived the 2022 bear

Bars-only single-asset single-venue — avoids the supp-data / multi-instrument / cross-venue failure modes; 3 real parameters give the optimizer a meaningful, low-overfit surface

Ample trade count (81) for statistically meaningful parameter fitting

Thin edge: profit_factor 1.195 and Sharpe 0.69 with CI-low -0.437 (straddles 0) — will face the deflated-Sharpe gate hard after the 225-trial sweep

Recent decay: 2025 -14.6% and 2026 -17.6% (rolling Sharpe negative into 2026) — the OOS/holdout windows land on the weak recent regime

Risky profile: 52% max drawdown, tail_ratio 0.033 (fat left tail), negative information_ratio -0.352 (underperforms buy-hold risk-adjusted)

Clean sensitivity surface: 0 cliffs across all 6 parameters, monotonic and stable

Walk-forward not flagged overfitted (is_overfitted=false; avg IS 2.06 vs avg OOS 1.32, ratio well under 3.0)

Mechanism is faithfully implemented, bars-only, single-asset/single-venue, long-only, low parameter count — infrastructure-compliant with no data-dependency risk

benchmark_meaningful=true with positive alpha (0.143) and modest beta (0.22) on the optimized run

Fails deflated Sharpe: DSR=0.5549 (<0.95), optimized Sharpe 1.4238 barely above the 225-trial expected-max luck bar of 1.3488, is_significant=false, PBO=0.6733 (>0.5), sharpe_ci_low=0.0258 hugging zero

Untouched time-ordered HOLDOUT FAILED and is NEGATIVE: holdout_sharpe=-3.62, ratio=-2.737 vs the 0.70 bar — the recent regime loses money

PSR 0.9956 vs DSR 0.5549 is the textbook best-of-225 selection-inflation trap

Genuine recent decay: annual 2025 -15.2%, 2026 -8.6%; most-recent WF-OOS window negative (-0.260); rolling Sharpe negative into 2026

Curve-fit signature: optimization thinned trades 81->34 and lifted win-rate 38%->53% by pushing ker_entry to 0.60 — fitting the in-sample trend regimes

Headline optimized total_return 502% is heavily mark-to-market (end_unrealized_pct 72.6%), and information_ratio is negative (-0.29) — not active alpha vs its own benchmark

Base run max_drawdown 52% with 871-day drawdown duration

Do NOT optimize yet — the +324.7% headline is inflated by over-exposure and a large un-exited position. Fix and re-assess first: 1) OVER-EXPOSURE / PYRAMIDING: avg_position_pct is 183.8% but the design intends ~120% (equity*0.6*leverage 2.0). The KER entry (KER>=0.4 AND net-up) stays true for many consecutive days during an efficient trend, and should_enter returns BUY each time with no check for an existing position, so positions stack. FIX: in should_enter, return None when a position is already open for this instrument (enter only from flat) so exposure caps at ~1.2x. Verify with the trades in the detail_file. 2) IMPLAUSIBLE DAYS confirm the bug: 2023-01-30 +84.8%, 2021-01-12 +52.7%, 2020-11-15 +51.1% daily returns are impossible for a 1.2x long (would require BTC +28% to +46% on those days). After the flat-only fix, confirm no single-day return exceeds what ~1.2x leverage permits. 3) LARGE UNREALIZED: end_unrealized_pct 183.3 means ~56% of the headline is an open position marked to market at the last bar (realized ~141%). Confirm the exit logic (KER<=0.2, net-down, stop) actually fires; judge the strategy on its REALIZED track record, not the MTM headline. 4) AFTER FIXING, RE-ASSESS: with exposure capped, check whether the edge holds. It is modest (profit_factor 1.14, win_rate 40%, though payoff 1.70 is healthy) and bull-concentrated (2022 -62%, 2025 -20%, 2026 -17% are losing). If the fixed-exposure rerun still shows a positive, less tail-dependent edge with reasonable drawdown, proceed to optimization (where it will still need to clear the deflated-Sharpe/walk-forward/holdout gates). If it collapses toward the losing-year profile, abandon.

BtcDailyKaufmanEfficiencyRatioTrendRegimeLong deliberately targeted every documented factory failure mode — bars-only, single-asset, single-venue, long-only, three parameters — using the Kaufman Efficiency Ratio to trade only in statistically efficient uptrends, and its base backtest was promising enough (81 trades, 19% CAGR, profit factor 1.195, positive alpha through the 2022 bear) to earn an 'optimize' verdict despite a 52% drawdown. Optimization produced a flattering headline (Sharpe 1.42, +503% return) but did so by curve-fitting: thinning trades 81→34 and pushing ker_entry to 0.60 to fit in-sample regimes, with 72.6% of the return sitting as unrealized open MTM and a negative information ratio. The analyst abandoned it after 3 iterations because the config failed deflated Sharpe (0.55, barely above the 1.35 luck bar, PBO 0.67) and its untouched forward holdout was deeply negative (-3.62) from real 2025–2026 decay — with no robust parameter region to tune toward. It joined the session's SMA-crossover, death-cross, and TSM strategies as another single-asset daily trend follower defeated by the same best-of-N deflation and negative-holdout signature.

A clean sensitivity surface (0 cliffs) and a non-overfitted walk-forward measure consistency, not significance — a modest ~1.4 Sharpe with no parameter region above the luck bar and a negative forward holdout cannot be tuned to promotion, so single-asset daily trend-quality gates should be pre-screened against recent-regime holdout performance before optimization.

The analyst returned 'abandon' at the post-optimization ANALYZING stage because the strategy failed deflated Sharpe (DSR 0.55 vs 0.95 bar; optimized Sharpe 1.42 barely above the 225-trial luck bar of 1.35; is_significant=false, PBO 0.67) and, decisively, its untouched time-ordered holdout was deeply negative (holdout Sharpe -3.62, ratio -2.737) amid genuine 2025–2026 decay.

A bars-only, single-asset long-only BTC daily trend follower that uses the Kaufman Efficiency Ratio (KER) as a trend-quality regime filter — entering long only in efficient uptrends (KER ≥ 0.40, net direction up) and sitting in cash during low-KER chop that historically destroys trend strategies.

The base backtest showed a modest full-cycle edge — 81 long trades, 19% CAGR, Sharpe 0.69 (CI-low -0.437), profit factor 1.195, 38% win rate with avg win ~2x avg loss, but a heavy 52% max drawdown; after optimization it reported Sharpe 1.42, profit factor 2.29 and +503% total return on 34 trades, though 72.6% of that was unrealized open MTM.