Liquidity Grab

Very good idea that talks about this idea:

• https://www.youtube.com/watch?v=9REzGB3R6HU

4) Validation (walk-forward, plus “hard-to-fool” tests)

Walk-forward splits

Example:

• train 90 days

• validate 14 days

• test 14 days

• roll forward 14 days

Do this for multiple folds and aggregate.

Two extra “don’t fool yourself” checks

1. Signal shift test (permutation-style)

• take your model’s score series sts_tst​

• within each day, circular-shift sts_tst​ by random k

• rerun backtest
  If real >> shifted distribution, your timing signal matters.

2. Block bootstrap of strategy returns

• take strategy per-bar returns

• resample by day-blocks

• get Sharpe / return distribution
  This tests robustness vs luck.

---

5) Backtesting (simple, realistic for spot-only)

Trading policy (minimal)

• Every 15 min, compute score.

• Only allow 0 or 1 position (start simple).

• Entry: if score in top X% (e.g., top 5–10% of rolling 30d scores) → buy.

• Exit: trailing stop + time stop

  • time stop: max hold 16 bars (4h) or 32 bars (8h)

  • trailing stop: based on recent vol (e.g., 1–1.5× ATR/vol)

• Risk: fixed fraction of equity per trade, or fixed position size initially.

Add fees/slippage later, but keep them in mind when choosing X%.