PaisaFy Trading Bot 🤖📈

An intelligent algorithmic trading bot that combines sentiment analysis, technical indicators, and automated trading execution using the Alpaca API. The bot analyzes news sentiment using FinBERT and incorporates Simple Moving Average (SMA) crossover strategies for enhanced trading decisions.

🌟 Features

Core Trading Features

• Deep Reinforcement Learning: Advanced ML ensemble using A2C, DDPG, and PPO algorithms for trading decisions
• Sentiment Analysis: Uses FinBERT (Financial BERT) to analyze news sentiment for trading decisions
• Technical Analysis: Implements SMA (Simple Moving Average) crossover strategies (50-day vs 200-day)
• Automated Training: Comprehensive ML model training pipeline with FinRL framework
• Automated Trading: Executes buy/sell orders automatically based on combined ML, sentiment and technical signals
• Risk Management: Implements bracket orders with take-profit and stop-loss levels
• Position Sizing: Intelligent cash allocation based on configurable risk parameters

Machine Learning & AI

• Ensemble Methods: Combines multiple DRL algorithms (A2C, DDPG, PPO) for robust decision making
• Feature Engineering: Automated technical indicator calculation and turbulence indexing
• Model Training: Complete pipeline from data preprocessing to model deployment
• TensorBoard Integration: Real-time training monitoring and visualization
• Model Persistence: Trained models saved in .zip format for reuse and deployment

Monitoring & Reliability

• Email Notifications: Automated email alerts for bot status, errors, and crashes
• Auto-Recovery: Automatic restart functionality with configurable retry attempts
• State Persistence: Saves trading state to prevent conflicts during restarts
• Comprehensive Logging: Detailed logs, statistics, and HTML tearsheets for performance analysis
• Backtesting: Historical strategy testing with Yahoo Finance data

Data Sources

• News Data: Real-time financial news via Alpaca API
• Price Data: Historical and real-time stock data via Yahoo Finance and Alpaca
• Market Data: Supports various stocks and ETFs (default: TSLA, SPY)

🏗️ Architecture

├── tradingbot.py          # Main trading strategy implementation (sentiment-based)
├── tradingbot2.py         # Alternative trading strategy implementation
├── tradingbot_train.py    # ML model training pipeline
├── DRLEnsemble.py         # Deep Reinforcement Learning ensemble framework
├── runner.py              # Bot supervisor with auto-restart functionality
├── finbert_utils.py       # FinBERT sentiment analysis utilities
├── news.py                # News data retrieval functions
├── SMA Calc.py           # Simple Moving Average calculation utilities
├── temp.py               # Enhanced strategy with SMA integration
├── mailtest.py           # Email notification testing
├── requirements.txt      # Python dependencies
├── bot_state.pkl         # Trading state persistence file
├── output.txt            # Training execution logs
├── logs/                 # Trading logs, statistics, and tearsheets
├── results/              # Trading performance results and charts
├── trained_models/       # Pre-trained DRL models (A2C, DDPG, PPO)
└── tensorboard_log/      # TensorBoard training logs for monitoring

🛠️ Installation

Prerequisites

• Python 3.8+
• Alpaca Trading Account (paper trading supported)
• Gmail account for notifications (optional)
• CUDA-compatible GPU (optional, for faster sentiment analysis)

Dependencies Installation

pip install -r requirements.txt

Key Dependencies

• lumibot: Trading framework and backtesting engine
• alpaca-trade-api: Alpaca brokerage integration
• transformers: Hugging Face transformers for FinBERT
• torch: PyTorch for machine learning
• stable-baselines3: Deep reinforcement learning algorithms
• finrl: Financial reinforcement learning framework
• tensorboard: ML model training visualization
• yfinance: Yahoo Finance data
• pandas: Data manipulation
• matplotlib/plotly: Data visualization

⚙️ Configuration

1. Alpaca API Setup

Edit the API credentials in tradingbot.py, news.py, and temp.py:

API_KEY = "your_alpaca_api_key"
API_SECRET = "your_alpaca_secret_key"
BASE_URL = "https://paper-api.alpaca.markets/v2"  # Paper trading

2. Email Notifications Setup

Configure email settings in runner.py and mailtest.py:

msg['From'] = 'your_email@gmail.com'
msg['To'] = 'recipient@gmail.com'
# Add your Gmail app password in login()
mailserver.login('your_email@gmail.com', 'your_app_password')

3. Trading Parameters

Customize trading parameters in the strategy initialization:

strategy = PaisaFy(name='mlstrat', broker=broker, 
                   parameters={"symbol":"TSLA",      # Stock symbol
                              "cash_at_risk":.5})   # Risk percentage

🚀 Usage

Training ML Models

# Train DRL ensemble models (A2C, DDPG, PPO)
python tradingbot_train.py

# Monitor training with TensorBoard
tensorboard --logdir=tensorboard_log

Running the Bot

# Direct execution (sentiment-based strategy)
python tradingbot.py

# Alternative strategy
python tradingbot2.py

# With auto-restart functionality and training
python runner.py

Backtesting

The bot includes backtesting capabilities. Modify the date range in tradingbot.py:

start_date = datetime(2023,1,1)
end_date = datetime(2024,4,11)

Testing Individual Components

# Test sentiment analysis
python finbert_utils.py

# Test SMA calculations
python "SMA Calc.py"

# Test news retrieval
python news.py

# Test email notifications
python mailtest.py

📊 Trading Strategies

Deep Reinforcement Learning Strategy (DRLEnsemble)

• Algorithms: A2C (Advantage Actor-Critic), DDPG (Deep Deterministic Policy Gradient), PPO (Proximal Policy Optimization)
• Training Data: Historical stock data with technical indicators (SMA, RSI, MACD, etc.)
• Features: 11-dimensional state space including price, volume, and technical indicators
• Ensemble Approach: Combines predictions from multiple DRL models for robust decision making
• Turbulence Detection: Incorporates market turbulence index for risk assessment

Sentiment-Based Strategy (tradingbot.py)

Buy Signals

• Sentiment: Positive news sentiment with >97% confidence
• Technical: 50-day SMA above 200-day SMA (bullish trend)
• Execution: Market buy with 20% take-profit and 5% stop-loss

Sell Signals

• Sentiment: Negative news sentiment with >97% confidence
• Technical: 50-day SMA below 200-day SMA (bearish trend)
• Execution: Market sell with 20% take-profit and 5% stop-loss

Risk Management

• Configurable cash-at-risk percentage (default: 50%)
• Bracket orders with automatic take-profit and stop-loss
• Position conflict prevention through state tracking
• Turbulence-based position sizing for DRL models

📈 Performance Monitoring

Log Files

The logs/ directory contains:

• CSV files: Detailed trading logs and statistics
• HTML files: Visual tearsheets and trade summaries
• JSON files: Strategy settings and parameters

Log File Naming Convention

[BotName]_YYYY-MM-DD_HH-MM-SS_[type].[extension]

Key Metrics Tracked

• Total returns and Sharpe ratio
• Win/loss ratios and maximum drawdown
• Trade frequency and position sizes
• Strategy parameters and market conditions

🔧 File Descriptions

File	Purpose
tradingbot.py	Main trading strategy (TSLA focused, sentiment-based)
tradingbot2.py	Alternative trading strategy implementation
tradingbot_train.py	ML model training pipeline using FinRL and DRL ensemble
DRLEnsemble.py	Deep Reinforcement Learning ensemble framework (A2C, DDPG, PPO)
temp.py	Enhanced strategy with SMA integration (SPY focused)
runner.py	Supervisor script with crash recovery and ML training
finbert_utils.py	FinBERT sentiment analysis implementation
news.py	News data retrieval and processing
SMA Calc.py	Technical analysis utilities
mailtest.py	Email notification testing
bot_state.pkl	Persistent state storage
output.txt	ML training execution logs and performance metrics

Directory Structure

Directory	Purpose
logs/	Trading execution logs, statistics, and HTML tearsheets
results/	Trading performance results and visualization charts
trained_models/	Pre-trained DRL models in .zip format (48 models)
tensorboard_log/	TensorBoard logs for training monitoring (A2C, DDPG, PPO)

⚠️ Important Notes

Security

• Never commit API keys to version control
• Use environment variables for sensitive credentials
• Enable 2FA on your Alpaca account

Trading Risks

• This bot trades with real money when live trading is enabled
• Always test thoroughly with paper trading first
• Monitor positions regularly, especially during high volatility
• Understand that past performance doesn't guarantee future results

Technical Considerations

• FinBERT requires significant computational resources
• GPU acceleration recommended for faster sentiment analysis and DRL training
• DRL model training can take several hours depending on hardware
• Ensure stable internet connection for real-time trading
• Monitor API rate limits to avoid service interruptions
• TensorBoard logs and trained models are safe for version control (no sensitive data)

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Test thoroughly with paper trading
4. Submit a pull request with detailed descriptions

📄 License

This project is for educational purposes. Use at your own risk when trading with real money.

🔗 Resources

• Alpaca API Documentation
• Lumibot Documentation
• FinBERT Model
• Yahoo Finance API

---

Disclaimer: This software is for educational and research purposes only. Trading involves significant financial risk, and you should never trade with money you cannot afford to lose. The authors are not responsible for any financial losses incurred through the use of this software.