🚀 Bitcoin AI Portfolio Optimization System

An advanced AI-powered Bitcoin portfolio management framework integrating LSTM neural networks, Black-Litterman model, Enhanced CVaR risk management, and PPO reinforcement learning for optimal cryptocurrency trading strategies.

📋 Overview

This research presents a comprehensive Bitcoin portfolio optimization system that combines cutting-edge artificial intelligence techniques with modern portfolio theory. The framework addresses Bitcoin's inherent volatility and non-normal return distributions through an integrated approach of predictive modeling, Bayesian portfolio optimization, advanced risk management, and adaptive trading strategies.

🎯 Key Innovation

Unlike traditional backtesting systems, this framework integrates multiple AI methodologies to create a robust, adaptive trading system that can handle the extreme volatility and unique characteristics of cryptocurrency markets.

✨ Key Features

• 🧠 LSTM Price Prediction: Deep learning model capturing temporal dependencies in Bitcoin price movements
• 📊 Black-Litterman Integration: Bayesian portfolio optimization incorporating AI-driven market views
• ⚠️ Enhanced CVaR Risk Management: Multi-component Conditional Value at Risk for tail risk protection
• 🤖 PPO Reinforcement Learning: Adaptive trading agent optimizing risk-adjusted returns
• 📈 Technical Analysis Integration: RSI and Keltner Channel indicators for market momentum analysis
• 📊 Comprehensive Performance Analytics: Detailed backtesting results and risk metrics
• 🔄 Dynamic Portfolio Rebalancing: Real-time position adjustment based on market conditions

🏆 Performance Highlights

Backtesting Results (2019-2025)

• Total Return: 6,280.12%
• Benchmark Outperformance: 3,832.45% alpha over buy-and-hold
• Sharpe Ratio: 1.50
• Maximum Drawdown: 57.00%
• LSTM Prediction Accuracy: RMSE 0.0262, MAE 0.0191
• Trading Efficiency: 148 trades (70% win rate)

Note: Past performance does not guarantee future results

🛠️ Installation

Prerequisites

• Python 3.8 or higher
• CUDA-compatible GPU (recommended for LSTM training)
• 8GB+ RAM recommended

Quick Setup

1. Clone the repository:

git clone https://github.com/Okiled/Backtest-Bitcoin-using-ai.git
cd Backtest-Bitcoin-using-ai

2. Install dependencies

3. (Optional) GPU setup for faster training:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

🚦 Quick Start

Basic Execution

Run the complete optimization pipeline:

python main.py

Custom Parameters

python main.py --start-date 2020-01-01 --end-date 2024-01-01 --initial-capital 100000

Example Implementation

from project_component import BitcoinPortfolioOptimizer

# Initialize the optimization system
optimizer = BitcoinPortfolioOptimizer(
    initial_capital=100000,
    start_date='2020-01-01',
    end_date='2024-01-01',
    lstm_sequence_length=60,
    confidence_level=0.65
)

# Run the complete pipeline
results = optimizer.run_optimization_pipeline()

# Generate comprehensive reports
optimizer.generate_performance_report()
optimizer.visualize_results()

🏗️ System Architecture

1. Data Processing Pipeline

• Historical Bitcoin price data (OHLCV) from Yahoo Finance
• Technical indicator calculation (RSI, Keltner Channels)
• Min-Max normalization and missing value imputation
• Feature engineering for predictive modeling

2. LSTM Prediction Engine

Input Layer (18 features) → LSTM Layers (50 units, 2 layers) → Dropout (0.3) → Dense Output

• Sequence Length: 60 time steps
• Features: Price data + technical indicators
• Optimization: Adam optimizer (lr=0.00005)
• Early Stopping: Implemented at epoch 85

3. Black-Litterman Portfolio Optimization

• Market Equilibrium: δ × Σ × W_market
• Investor Views: LSTM predictions with 65% confidence
• Posterior Returns: Bayesian combination of equilibrium and views
• Risk Aversion Parameter: 2.5

4. Enhanced CVaR Risk Management

Multi-component CVaR calculation:

• Base Model (32%)
• RSI-adjusted (24%)
• Keltner-adjusted (24%)
• Black-Litterman adjusted (20%)

5. PPO Trading Agent

• State Space: 18-dimensional market features
• Action Space: Continuous [0,1] → Buy/Sell/Hold
• Reward Function: 0.8×Returns - 0.1×CVaR_penalty + 0.1×Trade_bonus
• Training: 150,000 timesteps, batch size 128

📊 Model Performance Analysis

LSTM Prediction Accuracy

• RMSE: 0.0262 (2.62% prediction error)
• MAE: 0.0191 (1.91% average error)
• Trend Capture: Successfully identified major trend reversals during 2020-2022 bull run

Black-Litterman Optimization

• Initial Market Return: 9.12% annually
• LSTM-Adjusted Return: 7.00% annually
• Risk Reduction: Volatility decreased from 19.1% to 14.3%
• Posterior Sharpe Ratio: 0.35

CVaR Risk Management

• 90% CVaR: -6.74% (tail risk quantification)
• Dynamic Risk Adjustment: Adapts to market volatility conditions
• Early Warning System: Technical indicators provide pre-emptive risk signals

🔧 Configuration & Customization

Model Hyperparameters

LSTM_CONFIG = {
    'sequence_length': 60,
    'hidden_size': 50,
    'num_layers': 2,
    'dropout_rate': 0.3,
    'learning_rate': 0.00005,
    'batch_size': 32
}

BLACK_LITTERMAN_CONFIG = {
    'confidence_level': 0.65,
    'risk_aversion': 2.5,
    'tau': 0.025
}

PPO_CONFIG = {
    'total_timesteps': 150000,
    'batch_size': 128,
    'learning_rate': 5e-4,
    'clip_range': 0.2
}

Risk Management Settings

RISK_CONFIG = {
    'cvar_confidence': 0.90,
    'max_position_size': 0.95,
    'stop_loss_threshold': -0.15,
    'rebalancing_frequency': 'daily'
}

📈 Key Research Contributions

1. Novel Integration Framework

First comprehensive system combining LSTM, Black-Litterman, Enhanced CVaR, and PPO for cryptocurrency portfolio optimization.

2. Enhanced CVaR Methodology

Multi-component Conditional Value at Risk incorporating technical analysis and model-based adjustments for superior tail risk management.

3. Adaptive Trading Strategy

PPO-based reinforcement learning agent that dynamically adjusts trading decisions based on market conditions and risk metrics.

4. Robust Performance Validation

Comprehensive backtesting across multiple market cycles with detailed performance attribution analysis.

📊 Technical Indicators Implementation

Relative Strength Index (RSI)

RSI = 100 - (100 / (1 + RS))
RS = Average_Gain_14 / Average_Loss_14

Keltner Channel

Middle Line = EMA₂₆(close)
Upper Band = Middle Line + (0.5 × ATR₁₀)
Lower Band = Middle Line - (0.5 × ATR₁₀)

🎯 Use Cases

Institutional Applications

• Hedge Funds: Systematic cryptocurrency trading strategies
• Asset Managers: Portfolio diversification with digital assets
• Pension Funds: Risk-controlled crypto allocation

Individual Investors

• Retail Trading: Automated Bitcoin investment strategies
• Portfolio Management: Professional-grade optimization tools
• Risk Management: Advanced downside protection methods

Academic Research

• Quantitative Finance: Novel AI integration methodologies
• Cryptocurrency Studies: Volatility modeling and prediction
• Reinforcement Learning: Financial applications research

⚠️ Important Disclaimers

Risk Warnings

• High Volatility: Cryptocurrency markets are extremely volatile
• Model Risk: AI predictions are not guaranteed to be accurate
• Market Risk: Past performance does not predict future results
• Regulatory Risk: Cryptocurrency regulations may change
• Technical Risk: System failures can result in losses

Limitations

• Single Asset Focus: Currently optimized for Bitcoin only
• Historical Data Dependency: Performance based on past market conditions
• Computational Requirements: Requires significant computing resources
• Market Assumption: Assumes liquid markets and accurate data feeds

🔮 Future Research Directions

Short-term Enhancements

• Multi-Asset Extension: Include Ethereum, other major cryptocurrencies
• Real-time Implementation: Live trading system development
• Sentiment Integration: Social media and news sentiment analysis
• Alternative Data: On-chain metrics and macroeconomic indicators

Long-term Development

• Transformer Models: Advanced attention-based architectures
• Deep Reinforcement Learning: More sophisticated RL approaches
• Cross-Market Analysis: Integration with traditional financial markets
• Regulatory Compliance: Framework for institutional requirements

Contribution Areas

• Model improvements and new algorithms
• Performance optimization and efficiency
• Additional risk metrics and analysis
• Real-time trading implementation
• Documentation and examples

📊 Performance Dashboard

Metric	Value	Benchmark	Outperformance
Total Return	6,280.12%	2,447.67%	+3,832.45%
Sharpe Ratio	1.50	0.85	+0.65
Max Drawdown	57.00%	84.20%	+27.20%
Win Rate	70%	100%*	N/A
Volatility	14.3%	19.1%	-4.8%

*Buy-and-hold has 100% "win rate" but with higher volatility and drawdown