FracJax 🚀

High-Performance Fractal & Econophysics Tools for Financial Time Series using JAX.

FracJax is a production-grade Python library designed for quantitative finance and econophysics. It leverages Google JAX to perform heavy fractal and multifractal analysis on financial time series with GPU acceleration, achieving speeds up to 300x faster than standard CPU-based libraries (like nolds or hurst).

Key Feature: FracJax combines modern statistical robustness (Theil-Sen Estimator) with fractal theory (DFA, Wavelet Leaders) to extract noise-free, lag-free market regime signals.

⚡ Why FracJax?

📦 Features

FracJax calculates 10+ advanced market features across multiple dimensions:

1. Fractal & Memory (The "Market Microscope")

• DFA (Detrended Fluctuation Analysis): Robust long-term memory estimation (Trend vs. Mean Reversion).
• Wavelet Leaders: Multifractal singularity spectrum analysis (Detects structural shocks).
• Higuchi Fractal Dimension: Measures market roughness and complexity.

2. Microstructure & Liquidity

• Rolling CVD Proxy: Estimates aggressive buy/sell pressure from OHLCV.
• Amihud Illiquidity: Measures price impact per unit of volume.

3. Volatility & Risk

• GARCH(1,1) Forecast: Forward-looking volatility prediction.
• Realized Semivariance (RSV): Downside-specific volatility measure.
• Hill Estimator: Tail risk and fat-tail index estimation.

4. Information Theory & Inter-market

• Permutation Entropy: Measures time-series chaos and unpredictability.
• Cointegration Z-Score: Robust spread analysis between asset pairs.
• Lead-Lag Mutual Information: Measures non-linear information flow between assets (using Gaussian KDE).

🚀 Quick Start

FracJax provides a high-level API create_market_microscope that JIT-compiles kernels for maximum speed.

import numpy as np
from fracjax import create_market_microscope

# Generate dummy price data (Random Walk)
prices = np.cumsum(np.random.randn(10000)) + 1000

# 1. Initialize Kernels (Compiles once)
# Method options: 'dfa', 'wavelet', 'higuchi', 'garch', 'cvd', etc.
calc_dfa = create_market_microscope(
    method='dfa', 
    window_size=100, 
    batch_size=4096
)

calc_wavelet = create_market_microscope(
    method='wavelet', 
    window_size=100,
    max_level=4
)

# 2. Run on Data (Ultra Fast)
hurst_dfa = calc_dfa(prices)
hurst_wavelet = calc_wavelet(prices)

print(f"DFA Signal (Last): {hurst_dfa[-1]:.4f}")

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.