End-to-End Portfolio Optimization

This project implements an end-to-end deep learning approach for portfolio optimization using PyTorch. The model learns to optimize portfolio weights directly from return series data, providing a data-driven alternative to traditional portfolio optimization methods.

Features

• Deep learning-based portfolio optimization
• End-to-end training with custom loss functions
• Support for multiple evaluation metrics
• Flexible architecture for different asset classes
• Real-time portfolio rebalancing
• Risk-adjusted return optimization

Project Structure

Core Modules

1. Score Block

   • Input: Return series [B, N, T] (Batch, Number of assets, Time steps)
   • Output: Asset scores [B, N]
   • Architecture: LSTM-based neural network
   • Purpose: Learns to score assets based on their historical returns

2. Portfolio Block

   • Input: Asset scores [B, N]
   • Output: Portfolio weights [B, N]
   • Function: Converts scores to valid portfolio weights using softmax
   • Constraint: Weights sum to 1 (fully invested portfolio)

3. Loss Functions

   • Mean-Variance Portfolio Loss
     • Optimizes risk-adjusted returns
     • Balances expected return and portfolio variance
     • Customizable risk aversion parameter
   • Maximum Sharpe Ratio Loss
     • Maximizes risk-adjusted returns
     • Optimizes the Sharpe ratio directly

Evaluation Metrics

1. Expected Return (E(R))

   • Annualized portfolio return
   • Calculated from weighted asset returns

2. Standard Deviation (Std(R))

   • Portfolio risk measure
   • Annualized volatility

3. Sharpe Ratio

   • Risk-adjusted return metric
   • Excess return per unit of risk

4. Sortino Ratio

   • Downside risk-adjusted return
   • Considers only negative returns

5. Maximum Drawdown (MDD)

   • Largest peak-to-trough decline
   • Worst historical loss

6. Positive Return Percentage

   • Percentage of positive returns
   • Win rate metric

7. Frobenius Norm

   • Matrix norm for covariance
   • Risk structure measure

8. Turnover

   • Portfolio rebalancing cost
   • Measures weight changes

Requirements

• Python 3.8+
• PyTorch 1.9.0+
• NumPy 1.21.0+
• Pandas 1.3.0+
• Matplotlib 3.4.0+
• scikit-learn 0.24.0+

Installation

1. Clone the repository:

git clone https://github.com/yourusername/end2end-portfolio-optimization.git
cd end2end-portfolio-optimization

2. Create a virtual environment (recommended):

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

Usage

Basic Usage

1. Run the main script:

python main.py

2. For training:

python train.py

Advanced Usage

1. Customize model parameters:

from portfolio_block import PortfolioBlock
from score_block import ScoreBlock

# Initialize blocks with custom parameters
score_block = ScoreBlock(input_dim=10, hidden_dim=64)
portfolio_block = PortfolioBlock(num_assets=10)

2. Use custom loss functions:

from loss import MeanVarianceLoss, SharpeRatioLoss

# Choose loss function
loss_fn = MeanVarianceLoss(risk_aversion=0.5)
# or
loss_fn = SharpeRatioLoss(risk_free_rate=0.02)

Data

The project uses S&P 500 price data for demonstration. The data includes:

• Daily closing prices
• High prices
• Low prices

Data preprocessing includes:

• Log returns calculation
• Normalization
• Time series windowing

Model Architecture

The model combines:

1. LSTM-based feature extraction
2. Portfolio weight optimization
3. End-to-end training with custom loss functions

Training

The training process:

1. Uses historical return windows
2. Optimizes portfolio weights
3. Evaluates performance metrics
4. Saves training history

License

MIT License# P05_End2End-Portfolio-Optimization