Track 09Closed
Cascade Tick Bars
tick/imbalance bars catch the start of a liquidation cascade for a directional entry
Key metric
3 days: WR 100% (false) → 90 days: WR 47%, net −0.10%/trade, Sharpe −3.4
- Why an edge should exist / hypothesis
- Information-driven bars (de Prado) should detect a flow surge earlier than time bars.
- Construction
- Tick bars (500) + imbalance bars → cascade trigger → directional entry.
- Test data
- 5 coins; 3 days (pilot) and 90 days (144 trades).
- Results
- On 3 days WR=100%, Sharpe 1.27 — a small-sample false signal. On 90 days WR=47%, net −0.10%/trade, Sharpe −3.4. MFE≈MAE → direction after the trigger is random.
- Validation
- Overfitting on a short sample; on the full sample — noise.
- Root cause
- No directional information after the cascade trigger; a small sample fooled us.
- Verdict
- Dead. A showcase case: why a 3-day 100% WR means nothing. The infra (streaming aggTrades parser) was reused.
- Source
archive/CASCADE_TICK_BARS_RESEARCH.md
Want the same check on your strategy?
Validation →