Strategies

ATOM Hyperliquid Percentage-Rebound Reversal with Rolling-Midpoint Exit

Outcome: Abandoned

AtomHyperliquidPercentageReboundReversal

Outcome Summary

Across two iterations, the strategy implemented percentage-rebound reversal entries and rolling-midpoint exits on ATOM Hyperliquid 4-hour bars. Its backtest lost 35.14% over 260 closed trades, with negative returns in every reported year and volatility regime. Backtest review abandoned it because the estimated gross edge was below fees and drawdown exceeded the pre-registered limit, so optimization and later reviews were not reached.

Strategy report

Hypothesis

Test a disclosed adaptation of the source's percentage recovery from rolling extremes and midpoint exits on ATOMUSD.HYPERLIQUID, using completed 4-HOUR bars. This is not statistical regression. The prose permits simultaneous long and short conditions; resolve that ambiguity using threshold crossings and mutually exclusive midpoint conditions. Preserve percentage thresholds without adding RSI, moving-average trend filters or cross-sectional ranking. Source performance claims are unverified.

Hyperliquid is 8.5% of experiments against a 20% target, and long-short strategies are underrepresented. ATOM avoids the repeatedly tested BTC series and named failed LTC/BCH spread mechanisms. corpus_concentration is empty, so low ATOM usage cannot be verified. The exact ATOM 4-HOUR catalog directory exists, but its usable coverage still needs inspection. Four-hour bars accommodate Hyperliquid's limited candle history. A single traded instrument preserves the submitted mechanism; adding hedges or options would change it materially.

negative_expectancy (fee_edge/no_edge): The fixed-default 4h ATOMUSD.HYPERLIQUID percentage-rebound reversal with midpoint exit loses money on a reliable sample. avg_trade_return_pct is -0.528% against the +0.15% Hyperliquid floor. PF is 0.697 and Sharpe is -1.79 (95% CI [-3.20, -0.39]) over 260 trades and 790 days. Every calendar year and every volatility regime lost money, and max drawdown of 37.3% is more than twice the pre-registered 15%. The signal IC is -0.02 (t=-0.2), which is noise. With modeled impact and commission added back, the gross edge is only about +0.05% per trade, below the 0.09% round-trip fee, so neither optimizing reversal_pct (2-5) nor lookback_bars (18-48) can lift it above costs. This confirms QA's edge concern: the 3% rebound has already happened, and the rolling midpoint leaves too little distance to capture. Separately, flag for review the inconsistent impact accounting (impact_cost_pct 14199%, capacity $4.96) and holding-bar fields. They do not affect this verdict. This result rejects only this fixed ATOM 4h adaptation, not percentage-recovery reversal on other targets.

Implementation

Trades ATOM Hyperliquid percentage-rebound threshold crossings on completed 4-hour bars: longs below the rolling midpoint and shorts above it. Exits at the rolling midpoint, a 42-bar holding limit, or venue-managed 3% stop/6% target; applies a three-bar cooldown.

Addresses the Layer 1.5 mismatch by explicitly fixing leverage to 1.0 in both config and parameters, retaining the research optimization plan and removing leverage amplification from the 25% equity exposure cap. Previous signal, entry, exit and bracket logic are preserved. Verification remains for the pipeline. The bound outbox file could not be written because this session has read-only filesystem permissions; the complete task-bound JSON is returned directly.

Verification Results

The sandbox reports 260 closed trades, -0.546% average net return and PF 0.688. The 3% rebound threshold measures movement already completed, not remaining profit potential; midpoint exits can offer very little remaining distance. Check full-history midpoint-exit economics before optimization. These results do not establish a coding defect.

Backtest Review

Sharpe: -1.79
Total return: -35.14%
Max drawdown: 37.31%
Trades: 260
Win rate: 50.8%
Profit factor: 0.70

Reliable sample: 260 closed trades over 790 days of clean 4h ATOM Hyperliquid data with no missing bars. metrics_reliable=true and accounting reconciles.

Mechanism is implemented as specified: entries fire on percentage-recovery crossings from the rolling extremes, exits at the midpoint, and long and short trades are balanced (139 long, 121 short). This is not a code defect.

Per-trade return is -0.528%, against the HYPERLIQUID floor of +0.15%.

PF is 0.697. Sharpe is -1.79 with a 95% CI of [-3.20, -0.39], which excludes zero. PSR is 0.022.

Every year loses: 2024 -14.6%, 2025 -13.8%, 2026 -11.8%. Every volatility regime also loses (calm -16.7%, normal -16.3%, stressed -5.3%). Max drawdown is 37.3%, against the pre-registered 15%.

The signal-IC diagnostic rejects the forecast as noise: IC -0.021, t=-0.2, gross edge -0.33%.

Even before costs the edge is roughly zero. Adding back $35.4k of modeled impact and $5.7k of commission to the -$35.1k net leaves about +$5.9k gross on $13.0M traded. That is about 0.05% of notional per trade, below the 0.09% round-trip fee alone.

Accounting oddities to review separately: impact_cost_pct is 14199%, capacity_usd is $4.96, and impact is about 0.27% of notional on positions of about $25-30k. avg_holding_bars is 0.0 while avg_holding_period is 21h. avg_position_pct is 30.7%, above the 25% cap. None of these changes the verdict, because the signal makes no money even before costs.

Iteration History

Verification failed (Layer 1.5 — hypothesis/config consistency) [class=hypothesis_mismatch]: - optimization_plan fixes leverage at 1.0

Outcome Summary

This fixed ATOM 4-hour adaptation left too little remaining price movement after the rebound to cover trading costs.

The backtest-review verdict was abandon because expectancy was negative, estimated gross edge was below round-trip fees, and drawdown exceeded the pre-registered 15% limit. Optimization, post-optimization analysis, and risk review were not reached.

Trade percentage-recovery threshold crossings from rolling extremes on ATOMUSD.HYPERLIQUID 4-hour bars, using mutually exclusive long and short entries and rolling-midpoint exits.

Over 790 days and 260 closed trades, the backtest returned -35.14% with a Sharpe of -1.79 and maximum drawdown of 37.31%. Average per-trade return was -0.528%, and profit factor was 0.697.

Analysis

基于回归幅度的反转交易策略

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