OrderFlowOnFutures

Research project for order flow analysis on futures markets using VectorBT Pro.

Overview

This project explores order flow dynamics in futures markets, leveraging high-frequency data, advanced indicators, and portfolio simulation techniques. Built on Python with GPU acceleration support (MPS on Apple Silicon).

Features

• Order Flow Analysis: Imbalance detection, footprint charts, volume profile analysis
• VectorBT Pro Integration: Signal generation, backtesting, portfolio simulation, parameter optimization
• Multi-Timeframe Support: Tick, minute, and higher timeframe analysis
• GPU Acceleration: PyTorch MPS backend enabled for Apple Silicon (torch 2.12.0)
• Numba-Optimized Indicators: Fast computation of 500+ built-in and TA-Lib indicators
• Risk Management: Stop orders, cross-validation, deflated Sharpe ratio analysis

Tech Stack

• Python 3.11+ via Conda/Mamba
• VectorBT Pro — backtesting and portfolio simulation
• PyTorch 2.12.0 — ML with MPS GPU support
• TA-Lib — technical analysis indicators
• Numba — JIT compilation for performance
• direnv — automatic environment activation
• uv — fast Python package management

Installation

Prerequisites

• macOS (Apple Silicon recommended for MPS GPU)
• miniforge3 (or Anaconda/Miniconda)
• direnv
• uv

Setup

# 1. Clone the repository
git clone https://github.com/davidleconte/OrderFlowOnFutures.git
cd OrderFlowOnFutures

# 2. Create conda environment (uses local .conda/ directory)
conda create --prefix .conda python=3.11

# 3. Install dependencies
uv pip install vectorbtpro torch torchvision

# 4. Install TA-Lib (macOS)
brew install ta-lib
pip install TA-Lib

# 5. Configure direnv for auto-activation
cp .envrc .envrc.backup  # if you want to keep a copy
# Edit .envrc paths if your conda base differs from /Users/david.leconte/miniforge3
direnv allow

Environment Variables

The .envrc file automatically:

• Activates the local conda environment
• Sets KMP_DUPLICATE_LIB_OK=TRUE to resolve torch/libomp conflicts on macOS ARM

Usage

import vectorbtpro as vbt

# Load futures data
data = vbt.YFData.download('ES=F', start='2024-01-01')

# Generate signals
short_ma = vbt.MA.run(data.get('Close'), 10)
long_ma = vbt.MA.run(data.get('Close'), 50)
entries = short_ma.ma_crossed_above(long_ma)
exits = short_ma.ma_crossed_below(long_ma)

# Backtest
portfolio = vbt.Portfolio.from_signals(
    data.get('Close'),
    entries,
    exits,
    init_cash=100_000,
    fees=0.0001
)

print(portfolio.stats())

Development

Project Structure

OrderFlowOnFutures/
├── .envrc                 # direnv auto-activation
├── .gitignore             # Git ignore rules
├── .conda/                # Local conda environment (gitignored)
├── LICENSE                # MIT License
├── README.md              # This file
├── CONTRIBUTING.md        # Development guidelines
└── .github/
    └── workflows/
        └── python-ci.yml  # CI/CD pipeline

Running Tests

# Activate environment (direnv should auto-activate)
direnv allow

# Run the test suite
python -m pytest tests/ -v

Linting

ruff check .
ruff format --check .

CI/CD

GitHub Actions workflows are configured to run on every push and pull request:

• Linting: ruff for code style and quality checks
• Testing: pytest for unit and integration tests
• Build Validation: Ensures the conda environment spec is valid

Contributing

Contributions are welcome! Please read CONTRIBUTING.md for guidelines on:

• Setting up the development environment
• Code style and linting standards
• Testing requirements
• Pull request process

License

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

Acknowledgments

• VectorBT Pro for the backtesting framework
• PyPortfolioOpt for portfolio optimization
• Riskfolio-Lib for risk management