We moved beyond traditional time-based bars and implemented event-driven bars to extract hidden market structures:
- Volatility Bars → Sampled bars when market volatility exceeded a threshold.
- Dollar Bars → Sampled bars when cumulative traded value reached a set amount.
- Entropy-Based Bars → Sampled bars based on information content (Shannon entropy).
- Neural Spike Bars → Triggered bars when momentum resembled neuron firings.
- Fractal Bars → Sampled bars based on self-similarity patterns (Hurst exponent).
We engineered new financial features from signal processing, information theory, and nonlinear dynamics:
- Shannon Entropy → Measures randomness in price movements.
- Hurst Exponent → Detects whether markets are trending or mean-reverting.
- Wavelet Energy → Captures multi-scale price movement patterns.
- Momentum Spike Indicator → Identifies strong market moves.
- Cumulative Dollar Volume → Tracks liquidity shifts.