ETH Volatility-Contraction Squeeze Breakout, Long-Short (Binance USD-M, 4H Bars, Low-Vol Coil to Expansion Trigger, ATR-Trailing Winners, Low-Parameter)

Hypothesis

A structure variant on ETH (the factory proven directional asset) with a trigger mechanistically distinct from the impulse-continuation and opening-range families: it conditions on the volatility-regime cycle (realized vol is mean-reverting/autocorrelated, so a sustained low-vol coil releases as a directional expansion leg) rather than on a single impulse bar or a time-anchored range. Long-short to capture both upside and downside expansions and to reduce the portfolio long-only skew. Kept to 4 tunables (coil_lookback, squeeze_ratio, stop_atr_mult, risk_frac) with atr_period and the baseline multiplier frozen, to resist overfitting. Per-move targets are multi-percent vol-expansion legs, an order of magnitude above the ~0.10% USD-M round-trip fee, so it is not fee-marginal. Engine-safe: discrete enter-once/exit-once, single position, hard-capped sizing, no continuous rebalancing. Routed to BINANCE USD-M futures because the strategy is genuinely long+short; leverage 1.0 with notional capped at equity, so no unused-leverage gate triggers.

Implementation

Long-short directional breakout on ETHUSDT.BINANCE USD-M perpetual (4-HOUR bars) gated by a VOLATILITY CONTRACTION. A coil is detected when short-window realized vol (std of returns over coil_lookback bars) is below squeeze_ratio x the longer baseline vol (over coil_lookback*3 bars), both measured EXCLUDING the current bar so the breakout bar does not mask the contraction. When a squeeze is active and price closes beyond the coil Donchian range (max-high/min-low of the prior coil_lookback bars), the strategy enters in the breakout direction. calculate_signal returns a continuous signed breakout score = (close - coil_mid)/ATR(14) that varies every bar; should_enter applies the squeeze + channel-break gates. Exit is a pure ATR trailing stop (stop_atr_mult x ATR off the favorable extreme, seeded at entry so it also serves as the initial stop). Single position, flat between signals. Sizing risks risk_frac of equity over the stop distance, notional capped at equity.

Backtest Review

Sharpe

0.08

Total return

14.53%

Max drawdown

27.80%

Trades

246

Win rate

30.1%

Profit factor

1.01

Clean, well-formed implementation on the proven asset (ETH): full ~6yr history, 245 trades, balanced long-short (126/119), reasonable capacity ($552k), headline mostly realized (end_unrealized 14.5%), no code bug.

Correct trend-breakout asymmetry (avg_win $2360 = 2.4x avg_loss $986) and a legitimate, distinct vol-regime-cycle premise.

Entire 6-year profit is one outlier day: 2020-08-07 = +27.2% vs total_return +17.2% — that single day exceeds the whole backtest; remove it and the strategy is net negative.

6 of 7 years negative or flat (only 2020, the outlier year, is positive). Baseline Sharpe 0.089, profit_factor 1.015 — no risk-adjusted edge.

Extreme tail dependence (return_skew 13.7, kurtosis 277): the edge is fat-tail luck, not a stable body of trades — rolling Sharpe deeply negative across 2022-2025.

impact_cost_pct 42.6% eats nearly half the already-thin gross edge.

This profile reliably fails post-optimization deflated-Sharpe: best-of-225-trial selection on a 0.09 base Sharpe will not clear the expected-max-under-noise bar, and outlier-carried edges collapse in walk-forward OOS/holdout.

6 of 7 years negative/flat; +17% total is one Aug-2020 day

edge in the body, not one outlier

skew 13.7 / kurtosis 277; single day > total return

distributed edge

Outcome Summary

Volatility-contraction squeeze breakout is not the structure variant that carries ETH's trend-persistence edge (impulse-continuation is): it produces a lottery profile whose multi-year profit lives in one or two extreme-tail days with 6/7 years negative, and a ~0.09 base Sharpe cannot clear the best-of-N deflated-Sharpe bar — an outlier-carried breakout offers nothing for optimization to surface.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) because the result is fat-tail luck, not a tradable signal — strip the single +27% day and it is net negative — a near-zero base Sharpe with no distributed edge that would reliably fail post-optimization deflated-Sharpe and walk-forward, so optimization and all later stages were never reached.

A single-instrument, pure-OHLCV long/short volatility-contraction squeeze breakout on ETHUSDT.BINANCE USD-M 4H perps — detecting an extended low-realized-vol 'coil' (short-window vol below a fraction of its longer baseline), then entering in the direction price breaks the coil's Donchian range and riding the vol-expansion leg with an ATR trailing stop — a structure variant on the proven directional asset (ETH) keyed to the volatility-regime cycle, kept to ~4 parameters.

Over ~6 years and 245 trades (126 long / 119 short) it had effectively no risk-adjusted edge: Sharpe 0.089 (CI [-1.21, 0.80]), profit factor 1.015, expectancy $10.7/trade, win rate 30%, total return +17.2% with avg_win ~2.4x avg_loss, 27.5% max drawdown — but the entire six-year profit came from one day (2020-08-07, +27.2%, exceeding the whole backtest), with 6 of 7 years negative or flat, return skew 13.7 / kurtosis 277, and impact cost 42.6% of gross.