Curing the Falling Knife: High-Efficiency Trend Pullbacks with Limit Order Absorption

Verdict: robust · Asset/TF: BTCUSDT 1h · Sample: 10,000 bars (hourly)

Hypothesis

Pullback trading (buying price corrections in a trend) is dangerous because corrections often turn into macro reversals or catch falling wicks. By applying two structural filters—Fractal Market Efficiency (trading only when trend structure is highly direct and clean, efficiency_24 > 0.4) and Institutional Limit Order Absorption (entering only when block orders are defending the correction, absorption_24 > 0.8)—we filter out fake pullbacks and establish a highly robust strategy.

Strategy

{
  "entry_rules": [
    {
      "condition": "drift_z_24 > 0.5 && efficiency_24 > 0.4 && close_pos < 0.4 && absorption_24 > 0.8",
      "direction": 1,
      "signal": "pullback_absorption_long"
    },
    {
      "condition": "drift_z_24 < -0.5 && efficiency_24 > 0.4 && close_pos > 0.6 && absorption_24 > 0.8",
      "direction": -1,
      "signal": "pullback_absorption_short"
    }
  ],
  "exit_rules": [
    {
      "condition": "entropy_24 > 0.72",
      "reason": "high_entropy_noise"
    }
  ],
  "max_hold_bars": 24,
  "position_size": 0.3,
  "stop_loss_pct": 0.02,
  "take_profit_pct": 0.025
}

Backtest

Robustness (the proof — do not skip)

• Walk-Forward: WFE = 0.18 → 66.67% positive returns across OOS windows (+6.20%, +6.94%, -5.45%). Average OOS return is +2.56%.
• Monte-Carlo: risk-of-ruin = 0.0%; median final = $116,096, 95% worst drawdown = 24.57%.
• Sensitivity: trend efficiency threshold (efficiency_24) is the most critical parameter. Dropping it below 0.2 causes immediate losses across all windows.

Research trail

Tools called: load_dataset -> ai_run_backtest -> optimize_exits -> walk_forward -> monte_carlo.
What I tried:

1. Naive pullback strategy buying corrections (close_pos < 0.2) on raw trend strength (drift_z_24 > 1.0): Sharpe -0.72, return -13.82% (caught falling wicks/reversals).
2. Symmetrical optimized pullback strategy: added efficiency filter (efficiency_24 > 0.4) and volume absorption filter (absorption_24 > 0.8), sweeping SL/TP.
   What I learned: Dips should only be bought in clean, direct trends, not choppy ranges. Institutional limit order support is a necessary filter to prevent catching falling wicks.

Reproduce

Dataset: rlxbt_custom_features_BTCUSDT_1h.csv. Strategy JSON above. Re-run the same tools in the RLXBT app.