Cryptocurrency Trading Strategies using Mean Reversion and Hurst Exponents

This repository presents a mean-reversion trading strategy developed for BTC/USDT using statistical signal processing and fractal analysis. The strategy was designed with an emphasis on systematic execution, quantitative rigor, and modular architecture. This can also be used on ETH/USDT considering the mean-correlation of BTC and ETH in real-time market.

Overview

The strategy features:

• Kalman Filter for dynamic smoothing and estimation of price signals
• Hurst Exponent to assess trend persistence vs. mean reversion
• Technical indicators (OHLCV) via custom classes
• Integration with Untrade API to source crypto market data

Repository Structure

kalman-hurst-btc-strategy/
│
├── data/                 # Sample data
├── notebook/             # Research and exploration
├── src/                  # Modular strategy code
│   ├── indicators.py     # Indicator calculations
│   ├── strategy.py       # Strategy logic and backtesting
│   └── utils.py          # Helper utilities
├── requirements.txt
└── README.md

Regime Flow

• Zerolag- Cusum based regime prediction, earlier methods used k-means and HMMs.
• F1-Score and TDA used for performance metrics.

Strategy Flow

BTC_Approach

• Pull BTC/USDT, ETH/USDT OHLCV data
• Initialse technical indicators listed in indicators.py file.
• Filter and smoothen using Kalman Filter
• Estimate Hurst Exponent
• Apply mean reversion logic in strategy.py
• Log Performance metrics and PnL
• Actions: Enter Long/Short, Exit Long/Short, Hold
• Reward Function: Profit/loss adjusted for commission and risk

ETH_Correlation_Approach

• Correlation analysis between BTC/USDT and ETH log returns.
• CUSUM for regime shift detection
• ATR-based volatility thresholds
• Hurst Exponent for trend filtering
• Kalman & Gaussian Filters for smoothing

Features

• 📉 Mean-reversion strategy leveraging Hurst Exponent and Kalman Filter smoothing
• 🧠 Kalman Filter-based price estimation for robust signal generation
• 📈 Hurst Exponent analysis to identify trending vs. mean-reverting regimes
• 🔁 Dynamic entry/exit signals based on residual spread from the Kalman filter
• ⚖️ Custom position sizing logic based on volatility-adjusted thresholds
• 🧪 Fully parameterized backtesting with metrics for performance and trade efficiency
• 📊 Clear performance tracking via cumulative returns and trade annotations

Tech Stack

• Python – Core development language
• NumPy, Pandas – Data manipulation and series handling
• Matplotlib – Performance and signal visualization
• Statsmodels – Hurst Exponent and statistical tools
• FilterPy – Kalman filter implementation
• Jupyter Notebooks – Strategy development and prototyping

Getting Started

1. Clone the repository:

   git clone https://github.com/Gradient-7788/BTC-ETH_Stat
   cd BTC-ETH_Stat

2. Install dependencies:

   pip install -r requirements.txt

3. Run simulations:

   • Use Jupyter notebooks in the notebooks/ folder for ETH correlation and BTC RL experiments.

👥 Contributors

Team 67 – Zelta Automations