Crypto Forecasting and Backtesting Engine

A robust backtesting system for cryptocurrency trading strategies based on machine learning models with time series cross-validation.

Overview

This project provides a complete pipeline for:

1. Data Processing: Standardize and prepare cryptocurrency data for model training and simulation
2. Model Training: Train machine learning models with hyperparameter optimization and time series cross-validation
3. Backtesting: Simulate trading strategies with realistic assumptions and track performance

Features

• Data Standardization: Ensures consistent datetime formats between training and simulation data
• Feature Engineering: Automatically generates technical indicators and other useful features
• Time Series Cross-Validation: Prevents data leakage when evaluating model performance
• Hyperparameter Optimization: Finds optimal model parameters with grid search
• Realistic Backtesting: Includes transaction costs, position sizing, and risk management
• Performance Metrics: Calculates comprehensive trading metrics including Sharpe ratio, win rate, drawdown, etc.
• Visualization: Generates plots of equity curves, drawdowns, and trade entry/exit points

Installation

# Clone the repository
git clone <repository-url>
cd crypto_forecasting

# Install dependencies
pip install -r requirements.txt

Quick Start

To run a complete example with ADA cryptocurrency:

python run_ada_backtest.py

This will:

1. Process the ADA data
2. Train Random Forest, XGBoost, and Ridge regression models
3. Backtest the models on the simulation data
4. Save results in the results directory

Usage

Data Processing

python -m src.data_processor

This will process all available cryptocurrencies and prepare the data for model training and backtesting.

Training and Backtesting

python -m src.main --coins ada eth btc solana --model-types rf xgb ridge --n-rows 1000

Command-line arguments:

• --coins: List of coins to process and backtest
• --model-types: List of model types to train (rf, xgb, ridge, lasso, bayesian, gbm)
• --n-rows: Number of rows for reduced dataset
• --skip-processing: Skip data processing step
• --skip-training: Skip model training step
• --skip-backtesting: Skip backtesting step
• --initial-balance: Initial balance for backtesting (default: 10000.0)
• --enable-shorting: Enable short positions in backtesting
• --signal-threshold: Signal threshold for entering positions (default: 0.0001)

Data Structure

• /data/train/: Contains training data files
• /data/train/simulation/: Contains simulation data files
• /data/processed/: Contains processed data ready for model training
• /models/: Saved trained models and performance metrics
• /results/: Backtest results, plots, and performance metrics

Project Structure

• src/data_processor.py: Data preprocessing and standardization
• src/datasets.py: Dataset loading and feature engineering
• src/models.py: Model training with time series cross-validation
• src/backtest.py: Backtesting engine for trading strategies
• src/main.py: Main script to run the full pipeline
• run_ada_backtest.py: Example script for ADA cryptocurrency

Customization

You can customize the backtesting parameters in src/config.py:

• TRANSACTION_COSTS: Transaction costs for each cryptocurrency
• INITIAL_BALANCE: Initial account balance for backtesting
• SIGNAL_THRESHOLD: Minimum signal threshold to enter a position
• ENABLE_SHORTING: Whether to allow short positions
• RISK_PER_TRADE: Risk per trade as a fraction of account balance
• MAX_POSITION_SIZE: Maximum position size as a fraction of account balance

Example Results

After running the backtester, you'll find results in the results/ directory:

• <coin>/<model_type>_backtest_results.csv: Detailed backtest results
• <coin>/<model_type>_trades.csv: Individual trade details
• <coin>/<model_type>_performance.json: Performance metrics
• <coin>/<model_type>_backtest_plot.png: Visualization of the backtest

Transaction Costs

Transaction costs are modeled specifically for each coin:

• BTC: 5-10 bps
• ETH: 6-12 bps
• SOL: 12-20 bps
• ADA: 14-25 bps

Running Backtests

from src.runner import run_backtest

# Run backtest with default models (Random Forest, XGBoost, Logistic Regression)
results = run_backtest()

# Or specify custom model types
results = run_backtest(model_types=['rf', 'xgb'])

Output

The backtesting engine generates two key output files in the results/ directory:

1. backtest_results.csv: Performance summary across strategies and coins
2. detailed_backtest_results.csv: Comprehensive trade-level results

Performance Metrics

• Total Return
• Annualized Return
• Annualized Volatility
• Sharpe Ratio
• Maximum Drawdown

Contributing

1. Fork the repository
2. Create your feature branch
3. Commit your changes
4. Push to the branch
5. Create a new Pull Request

License

[Specify your license here]

python_crypto_trading_engine