Skip to content
All strategies

Strategies

LINK USD-M Daily Donchian Breakout, Long-Short with 200-Day Trend-Regime Filter (Single-Instrument BINANCE Perp, Validated Daily-Breakout Family Applied to a Decorrelated, Trend-Persistent DeFi-Infra Major, ATR-Stopped, Low-Parameter)

Outcome: Abandoned

LinkDailyDonchianTrendRegimeLS

Outcome Summary

LinkDailyDonchianTrendRegimeLS applied the factory's validated daily Donchian/trend-regime breakout family to LINKUSDT perpetuals, aiming to add a decorrelated DeFi-infrastructure trend stream while avoiding the multi-instrument aggregation artifacts of prior siblings. The design was mechanically clean — 46 balanced long/short trades with 24-day average holds and low commission drag — but it produced no edge: profit factor 0.9987, Sharpe 0.043, and negative expectancy across a full 6.5-year sample. Its only positive result was a single parabolic day in January 2024; strip that outlier and five of seven years lose money. At the backtest-review gate the analyst abandoned it after 2 iterations, judging the premise dead for LINK and the outlier-dominated 46-trade sample too thin to send to a curve-fitting optimization run.

Strategy report

Hypothesis

A LONG-SHORT, SINGLE-INSTRUMENT, DAILY Donchian-channel breakout (Turtle-style) trend strategy on LINKUSDT.BINANCE (USD-M perpetual). It deliberately uses the factory's VALIDATED daily trend/breakout family — the same edge already promoted via the ADA Donchian / trend-filtered-breakout siblings and the BTC/ETH/SOL TS-momentum books — but applies it to a major the long-short trend book does NOT yet cover and that is structurally DECORRELATED from the existing trend sleeves: LINK is a DeFi/oracle-infrastructure……Show moreShow less

A LONG-SHORT, SINGLE-INSTRUMENT, DAILY Donchian-channel breakout (Turtle-style) trend strategy on LINKUSDT.BINANCE (USD-M perpetual). It deliberately uses the factory's VALIDATED daily trend/breakout family — the same edge already promoted via the ADA Donchian / trend-filtered-breakout siblings and the BTC/ETH/SOL TS-momentum books — but applies it to a major the long-short trend book does NOT yet cover and that is structurally DECORRELATED from the existing trend sleeves: LINK is a DeFi/oracle-infrastructure asset whose return drivers (oracle adoption, DeFi/RWA narrative) differ from the exchange-token beta of BNB and the pure L1 beta of BTC/ETH/SOL, so it adds a diversifying trend stream rather than correlated family risk. It is engineered around every confirmed engine and edge landmine of this session: SINGLE-INSTRUMENT / SINGLE-VENUE / SINGLE-ACCOUNT, so it cannot trip the multi-instrument equity/return aggregation artifact that corrupted every pairs / rotation / fixed-weight-sleeve / cash-carry run; DAILY bars (the validated cadence — NOT the intraday-bar breakout/vol-expansion/impulse variants that are chop/whipsaw-dominated and negative on majors); no options, no funding-carry, no cross-venue. It is built to PASS where LTC trend FAILED: the LTC lesson was to screen for actual trend persistence and to not let a naked signal get chopped on a range-bound asset — LINK has historically produced sustained directional moves (2020 DeFi advance, 2021, the 2024-2025 rallies), and the design adds a 200-DAY TREND-REGIME FILTER (take longs only in a confirmed uptrend regime, shorts only in a confirmed downtrend regime) so the book stays out of the mid-channel chop that destroys breakout systems in ranges. Low parameter count (~4 knobs) to resist overfitting; thresholds set so breakouts fire an ample number of times over deep history (avoiding the no-trade Layer-3 failure).

Iteration-2 fix for the Layer-2 synthetic failure (`_compute_atr() takes 4 positional arguments but 5 were given`). Root cause: the ATR helper was a `@staticmethod`, and the Layer-2 `_StrategyProxy` wrapper invoked it in a way that did not preserve the static descriptor, so `self` was bound as an extra positional argument. Fix: convert `_compute_atr` to a normal instance method (`def _compute_atr(self, highs, lows, closes, period)`) so the `self.`-qualified call site matches the signature under any invocation path. This is the smallest possible change — the helper body, all signal/entry/exit/sizing logic, imports, and config are unchanged, preserving the Layer-1 pass. The strategy still implements the validated daily Donchian-breakout family gated by a 200-day trend-regime filter, applied to a decorrelated DeFi-infra major (LINK), long-short on a single Binance USD-M perp.

No edge and not a code defect — this is a dead premise for LINK at the review gate, so optimization would waste 2 hours fitting noise. profit_factor 0.9987 (breakeven), Sharpe 0.043, negative expectancy (-1.12/trade), and IR -0.54 over a full 6.5-year sample show the regime-gated daily Donchian book has essentially zero edge on LINK. Decisively, the entire +8.9% total return is one fat-tail day (2024-01-03 +19.9%; kurtosis 619, skew 21.5) — strip it and the book is deeply negative, with 5 of 7 years losing (2020……Show moreShow less

No edge and not a code defect — this is a dead premise for LINK at the review gate, so optimization would waste 2 hours fitting noise. profit_factor 0.9987 (breakeven), Sharpe 0.043, negative expectancy (-1.12/trade), and IR -0.54 over a full 6.5-year sample show the regime-gated daily Donchian book has essentially zero edge on LINK. Decisively, the entire +8.9% total return is one fat-tail day (2024-01-03 +19.9%; kurtosis 619, skew 21.5) — strip it and the book is deeply negative, with 5 of 7 years losing (2020 -3.7, 2021 -3.5, 2022 -4.2, 2023 -4.8, 2025 -6.3). The sample is also too thin and outlier-dominated for meaningful optimization (46 trades over 6.5yr; walk-forward windows of ~15 trades will simply select params that capture the single 2024 spike = overfit). The trades correctly implement the hypothesis (balanced 23L/23S, 24-day trend holds), so there is no developer fix that unlocks an edge — LINK's daily trend is not exploitable at this cadence, contradicting the 'validated family transfers to LINK' premise. FAILURE PATTERN: single-asset daily Donchian/trend-regime breakout on a mid-cap major produces a breakeven book (PF ~1.0) whose only positive result is one parabolic day; a validated sibling family does not confer an edge on an asset whose daily trend is too weak, and a 46-trade outlier-carried base should not be sent to a 225-trial sweep that will curve-fit the outlier.

Implementation

Turtle-style daily Donchian channel breakout on LINKUSDT USD-M perpetual (BINANCE), single-instrument/single-venue. Enters long on a 20-day Donchian upper-band breakout only when price is above its 200-day SMA (confirmed uptrend regime), and short on a lower-band breakdown only when below the 200-day SMA (confirmed downtrend regime). Risk is managed by a 2x ATR(20) stop-loss and an inner 10-day Donchian exit channel. Position size is ATR-risk-based (1.5% equity risk per trade), capped at 1x equity notional.

Verification Results

Verification failed (Layer 2 — synthetic scenarios): Parameters used: ['risk_pct', 'atr_period', 'min_notional', 'atr_stop_mult', 'donchian_exit', 'regime_period', 'donchian_entry', 'entry_threshold'] Check that __init__ sets all attributes from self.parameters.get(). - steady_uptrend: TypeError: LinkDailyDonchianTrendRegimeLS._compute_atr() takes 4 positional arguments but 5 were given (bar timestamp: 1735701660000) - steady_downtrend: TypeError: LinkDailyDonchianTrendRegimeLS._compute_atr() takes 4 positional……Show moreShow less

Verification failed (Layer 2 — synthetic scenarios): Parameters used: ['risk_pct', 'atr_period', 'min_notional', 'atr_stop_mult', 'donchian_exit', 'regime_period', 'donchian_entry', 'entry_threshold'] Check that __init__ sets all attributes from self.parameters.get(). - steady_uptrend: TypeError: LinkDailyDonchianTrendRegimeLS._compute_atr() takes 4 positional arguments but 5 were given (bar timestamp: 1735701660000) - steady_downtrend: TypeError: LinkDailyDonchianTrendRegimeLS._compute_atr() takes 4 positional arguments but 5 were given (bar timestamp: 1735701660000) - flat_ranging: TypeError: LinkDailyDonchianTrendRegimeLS._compute_atr() takes 4 positional arguments but 5 were given (bar timestamp: 1735701660000) - volatility_spike: TypeError: LinkDailyDonchianTrendRegimeLS._compute_atr() takes 4 positional arguments but 5 were given (bar timestamp: 1735701660000) - zero_volume: TypeError: LinkDailyDonchianTrendRegimeLS._compute_atr() takes 4 positional arguments but 5 were given (bar timestamp: 1735701660000) - price_gap: TypeError: LinkDailyDonchianTrendRegimeLS._compute_atr() takes 4 positional arguments but 5 were given (bar timestamp: 1735701660000)

Backtest Review

Sharpe
0.04
Total return
8.88%
Max drawdown
18.13%
Trades
46
Win rate
32.6%
Profit factor
1.00

Mechanically sound and matches the hypothesis: 46 entries signaled/submitted with zero drops, balanced 23 long / 23 short, 24-day average holds — a genuine regime-gated daily trend-follower, not a broken or degenerate signal.

Single-instrument/single-venue design correctly avoids the multi-instrument aggregation artifacts; low commission drag (1.31% of gross).

No edge: profit_factor 0.9987, Sharpe 0.043, negative expectancy (-1.12/trade), IR -0.54, alpha -0.35% over a full 6.5-year sample.

Entire positive total return (+8.9%) is a single fat-tail day (2024-01-03 +19.9%; kurtosis 619, skew 21.5). 5 of 7 years are negative; the two positive years are outlier-driven.

Thin, outlier-dominated sample (46 trades / 6.5yr, ~7/yr) — walk-forward windows of ~15 trades will fit the one 2024 spike. Tiny capacity_usd ($41.6k) and impact_cost_pct 155% are symptoms of near-zero gross edge.

Contradicts the hypothesis premise: LINK daily trend persistence is not exploitable at these parameters — the 'validated family' does not transfer to this asset/cadence.

Outcome Summary

A validated strategy family does not automatically confer an edge on a new asset — LINK's daily trend was too weak to exploit — and a headline positive return dominated by a single fat-tail day is really a losing book that must be caught before spending hours on optimization.

The analyst issued an 'abandon' verdict at the pre-optimization backtest-review gate: the book showed essentially zero edge (breakeven profit factor, sub-threshold Sharpe and trade count), and the thin, outlier-carried sample would only overfit the one 2024 spike in a walk-forward sweep.

A single-instrument, long-short daily Donchian (Turtle-style) breakout on LINKUSDT.BINANCE perps, gated by a 200-day trend-regime filter and ATR stops, betting the factory's validated daily-trend family would transfer to a decorrelated DeFi-infrastructure major.

Over a 6.5-year sample (2020-01-17 to 2026-06-26) it took 46 trades (23 long / 23 short) with a 32.6% win rate, Sharpe 0.043, profit factor 0.9987, and negative expectancy (-1.12 avg trade return / IR -0.54); its entire +8.9% total return came from a single fat-tail day (2024-01-03, +19.9%; kurtosis 619, skew 21.5), with 5 of 7 years negative.

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