SolSqueezeExpansionLS

SOL USD-M Volatility-Squeeze Expansion Capture, Long-Short (Single-Instrument SOLUSDT.BINANCE Perp, 4h Bars, Trade the Directional Break ONLY After a Confirmed Bollinger-Bandwidth Contraction, Fixed-R Target + Hard Catastrophic Stop, 1x Leverage, Pure OHLCV, 3-Parameter)

A LONG-SHORT, single-instrument volatility-REGIME strategy on SOLUSDT.BINANCE (standard USD-M perp, 4h bars) that arms only after a genuine volatility CONTRACTION and then captures the directional expansion that statistically follows. This is a clean re-test of the squeeze→expansion mechanism (vol autocorrelation) that a prior sibling never actually got to measure — it was mis-coded onto the wrong venue and abandoned on a venue-swap, NOT on edge. Deliberately routed to STANDARD BINANCE USD-M (not COIN-M) to remove that swap failure mode. It fills the under-represented LONG-SHORT direction (14.4%) with pure OHLCV (best-surviving data class), and is designed around the two lessons that killed recent siblings: (1) the intraday momentum-ignition blowup (-100%, account liquidated) is prevented by FIXED 1x leverage, a hard catastrophic stop, and a single-position guard (no pyramiding); (2) trend-chasing whipsaw is cut by the squeeze precondition, which fires rarely and only on high-quality setups, plus a quick fixed-R profit target rather than open-ended trend-riding. Not a carry (all carry except BTC is now confirmed dead), not a rotation (three already pending), not a fade (symmetric reversion pending), not options (trade-count wall), not liquidations (data wall). Only 3 parameters to stay clear of the overfit graveyard.

Implements the squeeze->expansion (volatility-autocorrelation) hypothesis directly and addresses the exact failure modes named in the brief. Venue is STANDARD BINANCE USD-M (linear, long-short) — not COIN-M — removing the venue-swap failure that abandoned the sibling; linear sizing (units = equity*risk/stop, capped at equity/price) is correct here. Blowup prevention (L15): leverage fixed at 1x, notional hard-capped at equity, per-trade risk bounded to risk_frac via the ATR stop, single-position (base class only enters when flat). Whipsaw control: the squeeze precondition fires selectively and a quick fixed-R target replaces open-ended trend-riding. Trade-frequency safety (L21/L25): the arm gate is percentile-based rather than an absolute constant (L5), so on a standalone 4h simulation it produced ~4% entry frequency (24 entries / 581 bars) with a continuously varying signal (520 unique values) — well clear of the zero-trade verification_loop, and warmup is only ~52 bars. Fills the under-represented long-short direction with pure OHLCV. Only three parameters are meaningfully tuned (squeeze_percentile, stop_atr_mult, target_r); the rest are standard structural defaults, keeping the optimization surface small to avoid overfitting.

Clean but too-thin to justify optimization. Unlike the recent siblings this does not blow up (max_drawdown 24.5%, no liquidation, 1x leverage held) and it clears fees (avg_trade_return_pct 0.178% > 0.15% floor, +26% total), but the edge is statistically indistinguishable from noise: Sharpe 0.175 with a 95% CI [-0.48, 0.87] through zero and profit_factor 1.054, below the 1.2 no-edge line. It is a pure-OHLCV directional-breakout (squeeze->expansion) strategy, a class with ~0 survival when the first backtest shows PF < 1.2; optimization has never rescued it. From a PF-1.05 / Sharpe-through-zero starting point the walk-forward optimizer will pick the best of ~225 trials and almost certainly fail the downstream deflated-Sharpe / PBO / holdout gates (the standard failure mode for thin OHLCV edges), burning 2 hours to confirm no significant edge. The mechanism is theoretically reasonable and this is the strongest of the recent batch, so confidence is moderate — but the marginal, noise-level edge does not warrant the optimization budget. Abandon at review.

Long-short volatility-regime strategy on SOLUSDT.BINANCE USD-M perp, 4h bars, pure OHLCV. It arms only after a confirmed Bollinger-bandwidth contraction (current bandwidth in the bottom `squeeze_percentile` of its own last `squeeze_lookback` bars, so the trigger is scale-invariant across symbols/regimes), then trades the directional expansion break: when the close pushes beyond the upper band (signal > entry_thresh) it goes long, beyond the lower band it goes short. calculate_signal returns a continuous normalized band-position (half-bandwidth units, ~ -2..+2) every bar; the squeeze state machine arms on contraction and persists arm_window bars after release. Exits are a hard catastrophic stop at stop_atr_mult*ATR, a fixed-R profit target at target_r * stop distance, and a max_hold_bars time stop. Position sizing risks risk_frac of equity against the ATR stop and is hard-capped at 1x-equity notional (leverage=1) with a single-position guard (no pyramiding).

Proceed to full backtest; analyst judges on the multi-year sample. Favorable RR means a modest win-rate improvement flips it positive.

Sandbox edge marginally negative but AMBIGUOUS, not decisively negative: total_return -1.55%, PF 0.97, avg_trade_return_pct -0.012% (below floor), yet Sharpe -0.08 with CI [-1.72, +1.55] straddles zero and RR is FAVORABLE (avg_win 2220 = 1.46× avg_loss 1524, skew +0.33). Only the 40% win rate against a 1.8R target pulls it below break-even (time-stop exits truncate some targets). Within noise for 60 trades/1yr — a legitimate full-backtest candidate, unlike the decisively-negative siblings this session.

Confirm a larger sample in the full multi-year backtest.

Moderate trade count: 60 over 364 days, below the ~100 comfort floor; the percentile squeeze gate fires selectively (~4% of bars), so the near-break-even result can't yet separate edge from noise.

Clean risk profile: no liquidation, 1x leverage held, max_drawdown 24.5% — the sizing/stop guards worked as designed

Positive and fee-clearing: total_return +26%, avg_trade_return_pct 0.178% (above the 0.15% USD-M floor), expectancy +$61.5/trade — not a fee-edge death

Genuinely market-neutral (beta 0.003, corr 0.02) and regime-distributed (positive in 4 of 7 years, negative years mild); well-sampled at 351 trades with healthy $2.3M capacity

Edge indistinguishable from noise: Sharpe 0.175 with CI [-0.48, 0.87] straddling zero, profit_factor 1.054 (< the 1.2 no-edge line), information_ratio -0.78

Pure-OHLCV directional-breakout class (squeeze->expansion), which has ~0 survival when the first backtest shows PF < 1.2 — optimization has not rescued this class

Starting from PF 1.05 / Sharpe-through-zero, walk-forward will select best-of-~225 trials and most likely fail the deflated-Sharpe / PBO / holdout robustness gates

Per-trade edge (0.178%) only marginally clears fees — the kind of thin edge optimization overfits rather than strengthens

SolSqueezeExpansionLS re-tested the squeeze→expansion mechanism a mis-coded sibling never got to measure, routing deliberately to standard BINANCE USD-M with 1x leverage, a hard stop, and a single-position guard to avoid the liquidation and whipsaw deaths of prior attempts. On the risk side it succeeded — no liquidation, only a 24.5% max drawdown, genuine market-neutrality, and a small fee-clearing +26% total return across 351 trades. But the edge itself was statistically indistinguishable from noise (Sharpe 0.175 with a CI through zero, profit factor 1.054, below the 1.2 no-edge line). Judging it the strongest of a weak recent batch but still too thin — a pure-OHLCV breakout class that optimization has never rescued from a sub-1.2 starting point — the analyst abandoned it at the review gate rather than spend hours confirming no significant edge.

A clean, well-behaved, market-neutral risk profile is not enough — a per-trade edge that only marginally clears fees with a profit factor near 1.0 and a Sharpe CI straddling zero is noise-level, and optimization tends to overfit such thin OHLCV breakout edges rather than strengthen them.

The analyst issued an 'abandon' verdict at the pre-optimization backtest-review gate: although the strategy avoided the blowups of its siblings, its profit factor of 1.054 sat below the 1.2 no-edge line with a Sharpe CI through zero, and as a pure-OHLCV squeeze→expansion breakout — a class with near-zero survival from a sub-1.2 PF start — it was judged too thin to justify spending the optimization budget only to fail downstream robustness gates.

A long-short, single-instrument volatility-regime strategy on SOLUSDT.BINANCE USD-M perp (4h bars) that armed only after a confirmed Bollinger-bandwidth contraction (squeeze) and then captured the directional expansion break that statistically follows, using a fixed reward:risk target plus a hard catastrophic ATR stop, 1x leverage, and a single-position guard.

Over ~2020-09 to 2026-07 it took 351 well-sampled trades (166 long / 185 short) with a clean risk profile — no liquidation, 1x leverage held, max drawdown 24.5% — and cleared fees with total_return +26%, avg_trade_return_pct 0.178% (above the 0.15% USD-M floor), and expectancy +$61.5/trade. It was genuinely market-neutral (beta 0.003, correlation 0.02) but its edge was indistinguishable from noise: Sharpe 0.175 (CI [-0.48, 0.87] straddling zero), profit_factor 1.054, and information_ratio -0.78.