SolHyperliquidVolatilitySpikeExhaustionReversionLS

SOL Hyperliquid Perp 15m Volatility-Spike Exhaustion Reversion (Single-Instrument, Long-Short, Thin-Book Overshoot)

A long-SHORT, SINGLE-INSTRUMENT, single-venue, SHORT-HORIZON mean-reversion strategy on SOLUSD.HYPERLIQUID perpetual futures using 15-MINUTE OHLCV bars only (NO funding/OI/liquidation/IV supplementary data, NO second instrument). The thesis: Hyperliquid's on-chain order book is structurally thinner than Binance's, so aggressive market orders produce larger temporary price impact and overshoot fair value; these overshoots partially revert within a few bars. The strategy fades climactic intraday spikes that are confirmed as exhaustion by a volume climax, and — unlike the fade strategies that failed this session — caps each loss with a tight ATR stop so the move-continues case cannot turn avg_loss > avg_win. Risk profile: risk 1.5% of equity per trade (strong signal 2%, weak 1%); volatility-based stop at 1.2x entry-ATR beyond entry; take-profit on reversion to the 20-bar VWAP/SMA anchor or +1.0x ATR, whichever first; hard time-stop at 8 bars (2h); max single-position exposure 20% of equity; one position at a time. Designed to fill three under-represented buckets simultaneously (HYPERLIQUID venue 6.1%->target 20%, long_short 14.3%, short_1m_15m horizon 2.4%->target 10%) while staying on the only structurally-testable path (single-instrument OHLCV on HL majors, which have dense ~3yr 15m history).

Implements the thin-book overshoot-reversion thesis on Hyperliquid (where the on-chain book is structurally thinner, so impact-driven overshoots are larger and revert). I confirmed SOLUSD.HYPERLIQUID 15-minute bar data exists and that multiple HL SOL strategies (5m, 4H) have already backtested, so the single-instrument HL 15m path is testable. The fade is made robust against the failure mode that killed prior fades (avg_loss > avg_win) by a TIGHT ATR stop (1.2x) that caps the move-continues case below the average reversion win, plus a dual take-profit (anchor reversion OR +1 ATR) and an 8-bar time stop. Selectivity is high: entries require a conjunction of range-climax AND volume-climax AND anchor-overshoot, so it only fades genuine exhaustion spikes, not every wiggle — keeping per-trade edge above HL's cheap ~0.09% RT fee. Risk is capital-relative and strength-scaled (interpolated 1%->2% by spike size in ATRs), stop-anchored so the dollar risk is constant regardless of volatility, and capped at 20% equity notional. calculate_signal returns a CONTINUOUS anchor-deviation score (-(close-anchor)/ATR) that varies every bar (defeats the Layer-2 frozen-signal check) and encodes the fade direction/magnitude, while the discrete climax entry gate lives in should_enter; entry context (price/ATR/direction) is read from the actual fill for robust exit management. OHLCV-only, single instrument, single venue — no funding/OI/liquidation/IV supplementary dependency, sidestepping the recent-only-coverage wall. Fills the HYPERLIQUID, long-short, and short-horizon buckets simultaneously. leverage=1.0 and sizing never multiplies by leverage, so it is not flagged leverage_set_but_unused; HL min-notional floor ($10) enforced.

Fundamentally broken, not a tuning problem, failing on four independent counts: (1) DATA WALL — only 66 days of SOLUSD.HYPERLIQUID 15m data exist (2026-03-16 to 2026-05-24), not the '3yr dense history' the hypothesis assumed; this alone cannot populate a 3-window walk-forward + 15-day holdout (the same HL-15m span limitation that abandoned prior SOL-HL-15m attempts at 38 days). (2) NEGATIVE EDGE — profit_factor 0.52, Sharpe -9.13, expectancy -$36/trade, negative every month across 145 trades. (3) WIN/LOSS INVERSION — the 'tight ATR stop caps losses' fix did not work: avg_loss $160 is 2.2x avg_win $73, because faded climactic spikes continue into the 1.2x ATR stop while reversion take-profits cap small; this is the same broken-fade payoff that killed this session's other 15m fade strategies. (4) FEE DOMINANCE — commission is 45.6% of gross PnL plus $3,672 of impact, so 15m trading on Hyperliquid's thin book costs roughly half the gross; the thin-book overshoot the thesis targets is smaller than the round-trip cost of harvesting it. No parameter sweep can create an edge that is net-negative before costs, carries a 2.2x inverted payoff, and loses ~46% of gross to commissions, and there is not enough data to validate one regardless. FAILURE PATTERN: short-horizon (15m) volatility-spike FADE on a thin-book / high-relative-cost venue (Hyperliquid alt) is structurally unviable — the overshoot edge is consumed by ~46% commission drag, the move-continues case inverts the payoff (avg_loss 2.2x avg_win) despite a tight stop, and HL 15m history (~2 months) is too short to validate; this is the third HL/15m short-horizon fade to fail this way and the family should not be re-proposed.

Long-short single-instrument short-horizon mean-reversion on SOLUSD.HYPERLIQUID perpetual, 15-MINUTE OHLCV-only bars. It fades climactic volatility spikes confirmed as exhaustion by three conjoint conditions: a large bar range (>= range_spike_mult x ATR), a volume climax (>= vol_mult x the 20-bar average volume), AND price extended from the 20-bar SMA anchor (>= dev_mult x ATR). An up-climax extended above the anchor is SHORTED; a down-climax extended below is BOUGHT. Each loss is capped by a TIGHT ATR stop (1.2x entry-ATR); take-profit triggers on reversion to the anchor OR +1.0x ATR favorable (whichever first); a hard time-stop closes after 8 bars (~2h). Risk per trade scales with spike strength from 1% (weak) to 2% (strong) of equity, stop-anchored, capped at 20% equity notional, one position at a time. leverage=1.

Clean execution: 145 trades (70 long / 75 short), 0 dropped, sane sizing (avg_position 20%), no liquidation.

Disciplined design intent (tight ATR stop, time stop, OHLCV-only single-instrument) targeting under-represented HL / long-short / short-horizon buckets.

Only 66 days of data (HL 15m span wall) — the hypothesis's '3yr 15m history' premise is false; cannot support walk-forward + holdout.

Decisively negative edge: profit_factor 0.52, Sharpe -9.13, expectancy -$36/trade, negative every month.

Win/loss inversion the design claimed to prevent: avg_loss $160 is 2.2x avg_win $73 — faded spikes continue into the stop while reversion winners cap small.

Catastrophic cost drag: commission is 45.6% of gross plus $3,672 impact — 15m trading on HL's thin book costs more than the overshoot edge is worth.

SolHyperliquidVolatilitySpikeExhaustionReversionLS faded climactic 15m volatility spikes on SOL Hyperliquid, betting that the venue's thin on-chain book overshoots and reverts, and explicitly adding a tight ATR stop to avoid the avg_loss > avg_win trap that killed prior fades. It failed on every front: a -9.13 Sharpe and 0.52 profit factor over just 145 trades, with the stop fix not working (faded spikes ran into it, inverting the payoff 2.2x), commissions eating 46% of gross on the thin book, and only 66 days of data — far too little to validate. The analyst ruled it structurally unviable, the third HL/15m short-horizon fade to fail identically, and abandoned it with a note to stop proposing the family. It ended after one iteration as abandoned, never advancing to optimization or risk review.

Short-horizon (15m) volatility-spike fades on a thin-book, high-relative-cost venue like Hyperliquid are structurally unviable: the overshoot edge is smaller than the ~46% round-trip commission drag, faded spikes continue into even a tight ATR stop (inverting the payoff to avg_loss > avg_win), and HL 15m history (~2 months) is too short to validate — the third HL/15m short-horizon fade to fail this way, a family that should not be re-proposed.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) as fundamentally broken on four independent counts — only 66 days of HL 15m data (too short for walk-forward + holdout), a decisively negative edge, the tight-stop fix failing to prevent a 2.2x avg_loss/avg_win inversion (faded spikes continue into the stop), and ~46% commission drag — so optimization and all later stages were never reached.

A long/short single-instrument short-horizon mean-reversion on SOLUSD.HYPERLIQUID perps (15-minute bars, OHLCV-only) that faded climactic volatility-spike exhaustion — selling up-climaxes / buying down-climaxes confirmed by a large ATR-unit bar range, a volume climax, and extension from a 20-bar anchor — exiting on reversion to the anchor or +1 ATR, a tight 1.2x-ATR stop, or an 8-bar time stop, on the thesis that Hyperliquid's thin on-chain book overshoots and partially reverts.

Over only 66 days and 145 trades it was decisively unprofitable: total return -1.6%, Sharpe -9.13 (CI [-12.59, -6.50]), profit factor 0.52, expectancy -$36/trade, negative every month, with the win/loss inversion the design claimed to prevent (avg loss $160 = 2.2x avg win $73 despite a 53% win rate) and commissions consuming 45.6% of gross plus $3,672 impact.