Strategies

LINK BINANCE Futures 4H Range-Bound Mean-Reversion Long-Only with Volatility-Compression Regime Filter (3-Parameter, Chop-Optimized, Explicit Anti-Breakout Mechanism Class)

Outcome: Abandoned

LinkRangeBoundMeanReversionLowVolLong

Outcome Summary

LinkRangeBoundMeanReversionLowVolLong was deliberately designed as the anti-breakout counter-position to the session's abandoned Donchian family, buying the low end of an established range only under a volatility-compression regime gate, betting the 2024-2026 chop that killed breakouts would favor range mean-reversion. But the backtest falsified the premise: it lost 32.2% with a profit factor of 0.77 and negative returns in every year since 2021 — including the very chop window it targeted — because the 4% range-break stop cut losers ($2,767 avg) below the midpoint-target wins ($1,750 avg), netting negative despite a 54.9% win rate. The low-vol filter created no edge, and with locked risk constants and a falsified thesis there was nothing to optimize. The analyst abandoned it at the backtest-review gate, calling it the mean-reversion mirror of the abandoned breakout family — failing in the same regime from the opposite side.

Strategy report

Hypothesis

A long-only single-instrument MEAN-REVERSION strategy on LINKUSDT.BINANCE perpetual futures using 4-HOUR bars and OHLCV-only data, explicitly designed for the 2024-2026 CHOP REGIME where the analyst-confirmed 'Donchian-with-trend mechanism class is system-wide regime-decayed' (4 abandons this session: BTC, AVAX, XRP, SOL). This proposal is the STRUCTURAL OPPOSITE of those failures: instead of buying breakouts (which fail in chop), it buys at the low end of an established range with a volatility-compression regime filter that ONLY trades in confirmed low-vol range-bound markets. Mechanism: (1) Detect range-bound regime via volatility compression: 20-bar ATR <= 200-bar mean ATR (current short-term vol below long-term average — confirms market is in chop, not trending). (2) Identify range: rolling 50-bar high and low (excluding current bar). (3) Enter long when close is within entry_position_pct (15%) of range_low AND in low-vol regime. (4) Exit at range midpoint OR hard time stop OR range-break stop. STRATEGIC RATIONALE: (1) EXPLICIT MECHANISM-CLASS COUNTER-POSITION: the latest SOL HL Donchian failure note explicitly identifies 4 Donchian/breakout failures and recommends 'BLOCK further Donchian-breakout hypotheses'. This proposal is structurally OPPOSITE — it profits FROM chop by trading mean-reversion within ranges, with the regime filter ensuring entries only occur in confirmed chop. The 2024-2026 regime hostility to breakouts is direct evidence of FAVORABLE conditions for range MR. (2) DIRECT IMPLEMENTATION OF ANALYST RECOMMENDATION: 'Future BTC Donchian hypotheses should add a regime filter — (a) volatility regime gate (only trade in high-vol periods)'. Mean-reversion strategies need the INVERSE regime gate (low-vol periods), which this hypothesis implements via the 20-bar/200-bar ATR ratio. (3) ASSET CHOICE — LINK: avoids ALL deprecated/moratorium assets (BNB post-2024 moratorium; AVAX multiple failures including Skewness, Trend-Pullback, Donchian Bull Flag; XRP Premium MR + Donchian Bull Flag failures; SOL Donchian failure; BTC OI Surge + Donchian + Volume Surge failures; ETH Connors Down-Day failure). LINK has the POSITIVE SIBLING SIGNAL: LinkEthRatioZscoreMRLongSpread reached risk_review at Sharpe 0.68 demonstrating LINK exhibits mean-reversion behavior at sub-daily timeframes. The single-asset 4H mean-reversion proposal here is methodologically distinct (range-bound MR vs cointegration MR) but draws on the same underlying LINK behavior. (4) ITERATION-SAFE: only 3 hypothesis-declared tunable parameters (range_lookback, entry_position_pct, position_pct). NO stop_loss parameter, NO take_profit parameter — explicit fix for LINK/ETH iter-2 optimizer-reversal failure mode. Exits are deterministic: range midpoint OR 48-bar time stop OR 4% range-break (all hard-coded). The optimizer cannot push extreme parameter values because there are no risk-tunable parameters. (5) FEE MATH EXPLICITLY VIABLE: typical LINK 4H 50-bar range = 8-15%. Entry at 15% of range from low → midpoint (50%) = 35% of range traversed = ~3-5% gross move per winner. At 30% position sizing: gross per winning trade = 0.9-1.5%. Net of 0.10% RT Binance USD-M fees = 0.8-1.4% per winner. At 55-65% win rate (mean reversion typically wins more than loses by win rate, loses more by avg loss size due to range-break stops): EV = 0.60 × 1.1% − 0.40 × 1.0% = 0.66% − 0.40% = +0.26% per trade NET expected value, comfortably above 0.15% futures viability floor. (6) AVOIDS ALL SYSTEM-WIDE FAILURE PATTERNS: NOT breakout (Donchian, Keltner, range breakout all deprecated); NOT funding/premium/OI/skewness signals; NOT volume-surge daily; NOT pair MR at 1H (LINK/ETH 1H deprecated); NOT cross-venue (single venue); NOT BINANCE_CM (data infrastructure gap); NOT Hyperliquid (analyst says venue diversification thesis falsified for trend mechanism — and this is MR, not trend); NOT Connors down-day (deprecated for daily); NOT Bollinger MR (cliff-prone on BTC 4H — but this is RANGE MR, not BB MR, mechanically different signal). (7) HOLDOUT-FRIENDLY DESIGN: the SOL Donchian failure note emphasizes 'HOLDOUT GATE IS THE STRONGEST PREDICTOR — across the 4 Donchian failures, walk-forward LOOKED acceptable but holdout caught all 4'. Range MR in chop regimes is HOLDOUT-FAVORABLE because the 2024-2026 holdout window is precisely the chop regime where this mechanism has edge — the same regime decay that hurt breakouts HELPS range MR.

Iteration 2 fixes the Layer-2 (synthetic) failure 'LinkRangeBoundMeanReversionLowVolLong._true_ranges() takes 1 positional argument but 2 were given'. Root cause: _true_ranges was declared @staticmethod, but the Layer-2 verification proxy rebinds strategy helpers as instance methods and passes self, so the 1-arg staticmethod received 2 args. The minimal fix converts _true_ranges from a staticmethod to an instance method (drop @staticmethod, add self); the call site already invokes it as self._true_ranges(bars), so no other change is needed. Every other line -- the volatility-compression regime gate, range construction, continuous signal, entry threshold, deterministic exits, and capital-relative sizing -- is byte-identical to the previous iteration, preserving all Layer-1 checks (clean imports, subclass, methods) that already passed. This mirrors a known recurring gotcha where @staticmethod helpers invoked as self._helper(...) break only under the synthetic proxy.

LINK 4H range-bound MR with a low-vol regime filter has no edge and its central thesis is falsified by the backtest: total_return -32.2%, Sharpe -0.72 (ci_low -1.91), profit_factor 0.77, expectancy -$286/trade, negative in 6 of 7 years and in EVERY year since 2021. The hypothesis claimed the 2024-2026 chop regime is favorable for range MR, but the strategy loses throughout that exact window (2024 -3.3%, 2025 -10.6%, 2026 -9.8%) — the low-vol regime gate did not create an edge. The mechanism is structurally a long-only knife-catch: it buys the low of a range, and the 4% range-break stop (avg_loss $2,767) cuts losers below the midpoint-target wins (avg_win $1,750), so a 54.9% win rate still nets negative. The 3 tunable parameters cannot fix a falsified premise with PF 0.77 across all recent years, and the asymmetric exits are locked constants — optimizing a negative base would only curve-fit the single positive year (2020). FAILURE PATTERN: long-only range-bound mean-reversion on a high-beta alt (LINK), even with a volatility-compression regime filter, loses money because in crypto a 'range low' in a low-vol window more often precedes a continued breakdown than a reversion — the range-break stop realizes losses larger than the midpoint-reversion wins, producing negative expectancy across regimes. The chop-favors-MR thesis does not hold on a single-asset long-only book; this is the MR-mirror of the abandoned Donchian breakout family, failing in the same regime from the opposite side.

Implementation

Long-only LINKUSDT.BINANCE 4H range-bound mean-reversion with a volatility-compression regime gate. Each bar it checks a chop filter (ATR(20) <= mean true range over 200 bars), builds the rolling 50-bar high/low range excluding the current bar, and enters long when the close is within 15% of the range low AND the low-vol regime holds. It exits deterministically at the range midpoint, on a 4% range-break below the low, or after a 48-bar time cap. Three tunable parameters (range_lookback, entry_position_pct, position_pct); all risk/exit constants are hard-coded. Long-only, leverage 1.0, OHLCV-only.

Verification Results

Verification failed (Layer 2 — synthetic scenarios): Parameters used: ['min_notional', 'position_pct', 'range_lookback', 'entry_position_pct'] Check that __init__ sets all attributes from self.parameters.get(). - steady_uptrend: TypeError: LinkRangeBoundMeanReversionLowVolLong._true_ranges() takes 1 positional argument but 2 were given (bar timestamp: 1735701720000) - steady_downtrend: TypeError: LinkRangeBoundMeanReversionLowVolLong._true_ranges() takes 1 positional argument but 2 were given (bar timestamp: 1735701720000) - flat_ranging: TypeError: LinkRangeBoundMeanReversionLowVolLong._true_ranges() takes 1 positional argument but 2 were given (bar timestamp: 1735701720000) - volatility_spike: TypeError: LinkRangeBoundMeanReversionLowVolLong._true_ranges() takes 1 positional argument but 2 were given (bar timestamp: 1735701720000) - zero_volume: TypeError: LinkRangeBoundMeanReversionLowVolLong._true_ranges() takes 1 positional argument but 2 were given (bar timestamp: 1735701720000) - price_gap: TypeError: LinkRangeBoundMeanReversionLowVolLong._true_ranges() takes 1 positional argument but 2 were given (bar timestamp: 1735701720000)

Backtest Review

Sharpe: -0.72
Total return: -32.18%
Max drawdown: 42.59%
Trades: 122
Win rate: 54.9%
Profit factor: 0.77

Clean OHLCV-only implementation; signal fires a reasonable number of times (122 trades); no sizing/min-notional drops

Thoughtful anti-breakout design with a low-vol regime gate — a genuinely distinct mechanism class

Net-losing with no edge: total_return -32.2%, Sharpe -0.72 (ci_low -1.91), PF 0.77, expectancy -$286/trade, PSR 0.14

Thesis falsified: negative in EVERY year since 2021 (2024 -3.3%, 2025 -10.6%, 2026 -9.8%) — it loses in the exact 2024-2026 chop regime it claimed to profit from

Asymmetric exits bleed it: 54.9% win rate but avg_loss $2,767 > avg_win $1,750 (range-break stops cut losers below wins)

Severe risk: max_drawdown 42.6% (ci_high 70%), drawdown duration 2,219 days; long-only range-buying = catching falling knives

Outcome Summary

The chop-favors-mean-reversion thesis does not hold on a single-asset long-only book: in crypto a 'range low' in a low-vol window more often precedes a continued breakdown than a reversion, so the range-break stop realizes losses larger than the midpoint-reversion wins — making long-only range-buying a falling-knife catch with negative expectancy across regimes.

The analyst abandoned it at the backtest-review gate before optimization because its central thesis was falsified by its own data — it lost throughout the exact 2024-2026 chop regime it claimed to profit from — and it had negative expectancy with no edge for optimization to surface, the asymmetric exits being locked constants rather than tunable parameters.

A long-only, OHLCV-only range-bound mean-reversion strategy on LINKUSDT 4H futures — the structural opposite of a breakout — buying within 15% of a 50-bar range low only when a volatility-compression filter (20-bar ATR ≤ 200-bar mean true range) confirmed a low-vol chop regime, exiting at the range midpoint, a 4% range-break stop, or a 48-bar time cap, with 3 parameters and no tunable risk knobs.

Over 122 long trades the strategy was net-losing with no edge: -32.2% total return (CAGR -6.6%), Sharpe -0.72 (CI low -1.91), profit factor 0.77, per-trade expectancy -$286, a 42.6% max drawdown, and negative returns in 6 of 7 years — including every year since 2021 (2024 -3.3%, 2025 -10.6%, 2026 -9.8%). Its 54.9% win rate was undone by asymmetric exits, with avg loss ($2,767) larger than avg win ($1,750).

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