AVAXVolatilitySqueezeTrendFollow

AVAX Volatility Squeeze Trend Follow

Exploits the well-documented volatility clustering phenomenon in mid-cap crypto. AVAX on Hyperliquid exhibits periods of extreme volatility contraction (measured by Bollinger Band width percentile over 100 bars on 4h timeframe) followed by explosive directional moves. Unlike Keltner/range breakout strategies already in the pipeline, this strategy uses volatility regime detection as the primary signal — it doesn't trade breakouts of price levels, but rather identifies the *compression phase* and rides the subsequent expansion. Entry is triggered only when BB width drops below its 15th percentile AND a 4h bar closes beyond the upper/lower band with above-average volume, confirming directional commitment. Targets multi-day trends (2-7 days hold) where expected moves of 3-8% far exceed the 0.09% RT fee on Hyperliquid.

Iteration-2 fix for the Layer-2 frozen-signal failure: the previous calculate_signal returned 0.0 on every bar where the squeeze+volume+beyond-band conjunction was absent, which on smooth synthetic scenarios meant a constant 0.0 (frozen) across all 376 bars. The minimal change makes calculate_signal return a CONTINUOUS decision variable — the standardized deviation of close from the BB mean, (close − mid)/std — on every post-warmup bar, which varies bar-to-bar (verified: 251 unique values over 251 post-warmup bars on both steady-trend and flat-ranging scenarios). The squeeze, volume, and beyond-band entry conditions are unchanged but now evaluated in should_enter via stored self._squeeze_ok / self._vol_ok booleans plus the algebraically-equivalent threshold (signal − bb_k ≥ entry_threshold reproduces the old (close − upper)/std ≥ entry_threshold), so entry behavior is identical to iteration 1 and every earlier layer stays green. No imports, structure, exits, or sizing were touched, and no banned imports were introduced — keeping Layer 1 passing while clearing Layer 2.

Decisively money-losing with no edge to optimize, on a thin sample. Sharpe -3.98, profit_factor 0.48 (makes $0.48 per $1 lost), total_return -19.7%, expectancy -$658/trade, omega 0.43, probabilistic_sharpe 0.036, and NEGATIVE in all three years (2024 -8.2%, 2025 -9.6%, 2026 -3.3%). Only 30 trades over ~2.4 years of Hyperliquid AVAX history (data starts 2024-02), win_rate 26.7%. The squeeze-then-breakout-confirmation entry is anti-predictive on AVAX HL 4H — compression breakouts fail and reverse (whipsaw), and market impact ($3.7k vs $1.25k commission) compounds the bleed. NOT optimize: you cannot tune a PF-0.48 / -3.98-Sharpe losing edge into a winner, and a 30-trade sample would fit pure noise. NOT iterate: this is already iteration 2 (the prior iterate was only a Layer-2 frozen-signal code fix, not an edge fix), the sample is thin on short HL history, and the strongly-negative result across all years is an anti-predictive mechanism, not a density or parameter problem. FAILURE PATTERN: Bollinger-squeeze volatility-regime breakout (enter on the band-break out of low-vol compression) is anti-predictive on a high-beta alt 4H — the breakout that resolves a squeeze frequently fails/reverses, producing a 27% win rate, PF 0.48, deeply negative Sharpe with all years negative; volatility-squeeze breakout joins the closed candle-shape/reclaim/fade families of refuted single-mechanism entries, and HL's short history makes it doubly unvalidatable.

AVAX 4H volatility-squeeze trend-follow on Hyperliquid (long+short). It detects a Bollinger-Band-width compression regime (current width in the bottom 15th percentile of its trailing 100-bar distribution) and enters only when a bar resolves the squeeze by closing beyond the band by ≥0.1 std on above-average volume — riding the subsequent multi-day expansion. Exits via hard stop, take-profit, ATR-free trailing giveback once in profit, a max-hold time stop, or a close back through the BB mean. Risk-based sizing (2% of equity to the stop distance, capped at 1× equity), leverage 1.0.

Verification failed (Layer 2 — synthetic scenarios): Parameters used: ['bb_k', 'risk_pct', 'vol_mult', 'bb_period', 'vol_period', 'squeeze_pct', 'min_notional', 'max_hold_bars', 'min_hold_bars', 'stop_loss_pct', 'entry_threshold', 'take_profit_pct', 'squeeze_lookback', 'trail_activate_pct', 'trail_giveback_pct'] Check that __init__ sets all attributes from self.parameters.get(). - steady_uptrend: Frozen signal: all 376 signals are identical (value=0.0). calculate_signal() likely has a bug — the signal never varies. - steady_downtrend: Frozen signal: all 376 signals are identical (value=0.0). calculate_signal() likely has a bug — the signal never varies. - flat_ranging: Frozen signal: all 376 signals are identical (value=0.0). calculate_signal() likely has a bug — the signal never varies.

Clean execution (30/30 submitted, no artifacts, metrics_reliable=true), genuine long/short, sound risk-based sizing; HL sizes cleanly (no engine over-exposure artifact)

Decisively money-losing: Sharpe -3.98, profit_factor 0.48, total_return -19.7%, expectancy -$658/trade, omega 0.43, PSR 0.036 — ALL 3 years negative (2024/2025/2026)

Thin sample (30 trades) on short Hyperliquid AVAX history (~2.4yr since 2024-02); win_rate 26.7%

Anti-predictive mechanism: squeeze-then-breakout entries whipsaw on AVAX HL 4H (breakouts out of compression fail/reverse); impact ($3.7k) exceeds commission ($1.25k)

Already iteration 2 (prior iterate was a Layer-2 frozen-signal fix, not an edge fix) — the edge was never present

all 3 years negative

majority positive

AVAXVolatilitySqueezeTrendFollow tried to monetize volatility clustering — detecting the compression phase via Bollinger-width percentile and entering only when a volume-confirmed 4H bar resolved the squeeze beyond the band, then riding the multi-day expansion. The result refuted the thesis: over 30 trades on AVAX's short Hyperliquid history it lost 19.7% with a -3.98 Sharpe, a 0.48 profit factor, and negative returns in all three years, because squeeze-resolving breakouts whipsawed and reversed while impact costs compounded the bleed. The reviewer noted this was already iteration 2 — the prior change was a frozen-signal code fix, not an edge fix — so the edge was never present, and a 30-trade sample on short HL data could only fit noise. It was abandoned as another anti-predictive single-mechanism entry, adding volatility-squeeze breakout to the closed family of refuted candle-shape/reclaim/fade patterns on crypto majors.

Bollinger-squeeze volatility-regime breakout is anti-predictive on a high-beta alt at 4H — the breakout that resolves a low-vol compression frequently reverses, yielding a 27% win rate, PF 0.48 and all-years-negative returns — so it joins the closed, refuted family of single-mechanism candle-shape/reclaim/fade entries, made doubly unvalidatable by Hyperliquid's short ~2.4-year AVAX history.

The analyst abandoned it at the backtest-review gate as a decisively negative, anti-predictive result: squeeze-resolving breakouts frequently fail and reverse (whipsaw) on AVAX HL 4H, so a PF-0.48 / -3.98-Sharpe loser cannot be tuned into a winner — and this was already iteration 2 (the prior iterate was only a Layer-2 frozen-signal code fix, not an edge fix), on a thin 30-trade sample over short HL history that compounds the un-validatability.

A long/short volatility-regime strategy on AVAXUSD.HYPERLIQUID 4H that detects extreme Bollinger Band-width compression (below the 15th percentile over 100 bars), then enters in the direction a 4H bar resolves the squeeze by closing beyond the band on above-average volume — aiming to ride the subsequent 2-7 day expansion where 3-8% moves dwarf Hyperliquid's ~0.09% round-trip fee, with risk-based sizing at leverage 1.0.

It was decisively money-losing over a thin 30-trade sample on AVAX's short Hyperliquid history (~2.4 years since 2024-02): Sharpe -3.98, profit factor 0.48, -19.7% total return, expectancy -$658/trade, 26.7% win rate, omega 0.43, PSR 0.036, 23.2% max drawdown, negative in all three years — with market impact ($3.7k) exceeding commission ($1.25k); execution was clean (30/30 submitted, metrics_reliable=true, genuine long/short).