Strategies

SOL Volume-Confirmed Momentum, Long-Short on HYPERLIQUID (SOLUSD.HYPERLIQUID, 4H — same edge ported to the under-represented DEX venue)

Outcome: Abandoned

SolHyperliquidVolumeConfirmedMomentumLS

Outcome Summary

A momentum edge that barely clears fees on a CEX doesn't survive a venue port; net per-trade return must be positive before optimization is worthwhile.

Strategy report

Hypothesis

Negative expectancy on a reliable 208-trade / 955-day sample: avg_trade_return_pct -0.027% (below ZERO, far under the Hyperliquid 0.15% fee floor), profit_factor 1.07 (fee-drag band), Sharpe 0.229 with CI straddling zero. This is NOT optimizable — a per-trade edge already below trading costs cannot be tuned above them, and the pipeline downgrades an 'optimize' here anyway. The positive headline (+16.5%, +$47/trade USD) is a size-weighting artifact concentrated in a single day (2026-08-27 +29.78%); impact consumes 31.76% of gross PnL. Critically, the strategy LOSES in the stressed vol tercile (-15.1%, Sharpe -0.63) — the exact high-participation regime the volume filter is meant to exploit — so the participation gate adds no edge, confirming QA's concern. This is the ETH/Binance sibling ported to Hyperliquid, and single-TF/dual-TF momentum ports to the DEX venue have failed repeatedly (hyperliquid 0/N). Not worth 2 hours of optimization. Failure pattern: fee_edge / edge-below-cost.

Implementation

Long/short 4H momentum on the SOLUSD Hyperliquid perpetual, gated on volume participation. Every bar it computes a continuous signal = (price change over mom_lookback bars, in ATR units) x min(1, current volume / median volume of the prior 60 bars / vol_surge_mult). It enters long when that signal crosses up through +1.0 ATR with the surge genuinely met (participation >= 1), short on the mirrored cross down, and is flat otherwise. Exits are structural: momentum reversing 0.25 ATR against the position (hysteresis so an intra-leg pause does not truncate the winner and bill a second round trip), or a trail_atr_mult ATR trailing stop from the best close since entry. Sizing risks a fixed 2% of equity to the trailing-stop distance, capped at 1x equity notional, so position size shrinks as SOL volatility expands; leverage 1.0 and no margin is required.

The hypothesis is an explicit venue port of the volume-confirmed momentum mechanism to Hyperliquid, so the implementation keeps the mechanism identical to the ETH/Binance sibling and changes only what the venue forces: instrument SOLUSD.HYPERLIQUID, 4H bars (mid/long-term horizon), SOL's 0.01 size step, and Hyperliquid's $10 minimum notional. HYPERLIQUID (MARGIN) is the required venue for two independent reasons - the strategy takes SHORT momentum as well as long, which a CASH spot account cannot do, and the hypothesis is specifically about the DEX venue; the port is also fee-favourable, since HL's ~0.09% round-trip taker cost is below Binance USD-M's ~0.10% while a confirmed 4H SOL leg traverses 2-5%. Participation is measured against a MEDIAN baseline because volume is right-skewed and a mean baseline inflated by one prior spike hides the next genuine surge. The signal is returned continuously in its natural ATR units every bar (never gated behind the entry condition), with the surge requirement expressed as a weight inside it and re-checked as a hard gate at entry so a large move on thin volume is never traded. The cross requirement prevents re-entering the same high-volume leg every bar. Only 3 parameters are tunable (mom_lookback, vol_surge_mult, trail_atr_mult); the entry threshold, exit hysteresis, baseline window, ATR period and risk_pct are locked so the optimizer cannot buy Sharpe with notional instead of edge. No supplementary feeds, no extra legs, no process-relative counters: every quantity is a pure function of the bar buffer, so restarts, backtests and paper replay agree. Per-bar work is bounded (~90 element reads plus one 60-element sort) and independent of accumulated history, keeping the sandbox well inside its wall clock. SOLUSD.HYPERLIQUID 4-HOUR bars are confirmed present in the catalog from 2024-01-25 to today (~3,600 bars), so the sandbox window has ample data.

Verification Results

Verify FactoryStrategy uses the subclass min_bars_required() for the warm-up gate; if it ignores overrides, the intended warmup window would not be honored.

min_bars_required(self) overrides a base framework method. Appears deliberate (returns max of vol_window/atr_period/mom_lookback warmup needs) and static analysis flagged it only as informational. No functional problem, but confirm the base template consults this override rather than its own value for warmup gating.

Unoptimized sandbox is marginal: avg_trade_return_pct -0.027% (below zero), PF 1.12, Sharpe 0.23 over 208 trades, and the strategy LOSES in the stressed vol tercile (return -15.1%, Sharpe -0.63) — the exact high-volatility regime a volume-confirmed momentum filter is meant to exploit. Headline +16.5% total return is carried by the calm/normal terciles. Short leg is weaker (short_win_rate 0.31 vs long 0.45). This is a smoke test on one param draw so it does not block, but the analyst should confirm the participation gate actually adds edge over plain 4H momentum on full history and check whether the negative stressed-regime result is structural before optimizing.

Backtest Review

Sharpe: 0.23
Total return: 16.48%
Max drawdown: 32.32%
Trades: 208
Win rate: 36.5%
Profit factor: 1.07

Clean, well-formed single-instrument OHLCV strategy; trades correctly implement the volume-confirmed momentum hypothesis (balanced 97 long / 111 short, cross-gated entries, 208 trades over 955 days — a reliable sample).

Metrics_reliable=true, so this is a fair, decisive test of the ported edge.

avg_trade_return_pct = -0.027% — BELOW ZERO and far under the Hyperliquid 0.15% fee floor. Per-trade edge (the fee-comparable unit) is negative; optimization cannot lift an edge that is already below costs.

profit_factor 1.07 sits in the classic fee-drag band — gross edge is barely present and is consumed by costs. Impact is 31.76% of gross PnL (impact_cost_pct) and commission 4.33% — trading costs dominate.

The positive headline (+16.5% total, +$47/trade USD) is a size-weighting artifact concentrated in a single day (2026-08-27 = +29.78% daily return); the equal-weighted per-trade return is negative, so the book is essentially cost-dead ex that print.

Loses in the STRESSED vol tercile (-15.1%, Sharpe -0.63) — the exact high-participation regime a volume-confirmed momentum filter is meant to exploit; the participation gate adds no edge where it should. QA's edge concern is borne out on full history.

Weak short leg (short_win_rate 0.31), Sharpe 0.229 with CI [-0.96, 1.35] straddling zero, negative 2025 year (-12.7%), 555-day max-DD duration, information_ratio -0.38 vs benchmark.

Outcome Summary

Abandoned at the pre-optimization backtest-review gate — negative expectancy isn't optimizable, the positive headline was a single-day (2026-08-27) size-weighting artifact, and it lost in the stressed vol tercile the volume filter was meant to exploit.

Take long/short 4H SOL momentum on the HYPERLIQUID perpetual only when a concurrent volume surge confirms real participation, porting the ETH/Binance sibling's participation-filter edge to the DEX venue.

Over 955 days and 208 trades, headline +16.5% total return but Sharpe 0.229 (CI straddling zero), profit factor 1.07, win rate 36.5%, and a negative per-trade edge (avg_trade_return_pct -0.027%) with impact eating 31.76% of gross PnL.

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