Portfolio Optimization & Stock Price Prediction

A comprehensive Python-based repository for portfolio optimization and stock price prediction using Modern Portfolio Theory (MPT), machine learning, and deep learning techniques.

📋 Overview

This project combines two major areas of quantitative finance:

1. Portfolio Optimization - Implementing various optimization strategies to construct efficient portfolios
2. Stock Price Prediction - Using machine learning models to forecast stock price movements

The repository provides Jupyter notebooks for both educational purposes and practical implementation, allowing users to:

• Download and analyze historical stock data
• Build optimized portfolios using different risk-return objectives
• Train and evaluate ML models for price prediction
• Visualize portfolio performance and financial metrics

🎯 Key Features

Portfolio Optimization

• Mean-Variance Optimization - Classic Markowitz portfolio theory
• Maximum Sharpe Ratio - Optimize for best risk-adjusted returns
• Minimum Variance - Construct lowest-risk portfolios
• Target Volatility - Achieve specific risk levels
• Efficient Frontier - Visualize risk-return trade-offs
• Backtesting - Test portfolio strategies on historical data

Machine Learning for Price Prediction

• Multiple Algorithms:
  • Deep Neural Networks (Keras/TensorFlow)
  • Random Forest
  • LightGBM (with and without cross-validation)
  • Logistic Regression
• Technical Indicators - Feature engineering using price and volume data
• Model Evaluation - Performance metrics and visualization
• Walk-forward Analysis - Time-series aware validation

📁 Repository Structure

Main Notebooks

Portfolio Optimization

• Portfolio_optimisation.ipynb - Core portfolio optimization implementation using MPT
• Portfolio_Optimisation_SIPP_v2.ipynb - Advanced optimization with multiple strategies (max Sharpe, min variance, target volatility, manual weights)
• Portfolio_optimisation_SIPP.ipynb - Earlier version of SIPP optimization
• Plot_Portfolio.ipynb - Portfolio visualization and performance plotting tools
• riskfolio_example.ipynb - Example implementations using the Riskfolio library
• [Tutorial 6_unique.ipynb](Tutorial 6_unique.ipynb) - Tutorial on portfolio optimization concepts

Machine Learning Price Prediction

• Price_Prediction_ML_v2.ipynb - Main ML price prediction pipeline
• Keras_DNN.ipynb - Deep Neural Network implementation using Keras
• Random_Forest.ipynb - Random Forest classifier/regressor for price prediction
• light_gbm.ipynb - LightGBM gradient boosting model
• light_gbm-cross_val.ipynb - LightGBM with cross-validation
• Logistic_Regression.ipynb - Logistic Regression for directional prediction

🛠️ Dependencies

Core Libraries

# Data manipulation and analysis
numpy
pandas
scipy
statsmodels

# Financial data
yfinance
pandas_datareader
yahoofinancials

# Visualization
matplotlib
mplfinance
seaborn
plotly

# Portfolio optimization
pypfopt
riskfolio-lib
quantstats

# Machine Learning
scikit-learn
tensorflow
keras
lightgbm

# Utilities
ray
timebudget
pickle

🚀 Getting Started

Installation

1. Clone the repository:

git clone https://github.com/ferhat00/PortfolioOptimisation.git
cd PortfolioOptimisation

2. Create a virtual environment (recommended):

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

3. Install required packages:

pip install numpy pandas scipy statsmodels
pip install yfinance pandas-datareader yahoofinancials
pip install matplotlib mplfinance seaborn plotly
pip install PyPortfolioOpt riskfolio-lib quantstats
pip install scikit-learn tensorflow lightgbm
pip install ray timebudget

Quick Start

Portfolio Optimization Example

# Define your stock universe
stocks = ['SPY', 'GLD', 'QQQ', 'TLT', 'EEM']
start_date = '2010-01-01'
end_date = '2024-01-01'

# Run Portfolio_optimisation.ipynb or Portfolio_Optimisation_SIPP_v2.ipynb
# to construct an optimized portfolio

Price Prediction Example

# Choose a stock
stock = ['SPY']
date_start = '1993-01-01'

# Run any of the ML notebooks to train and evaluate models
# - Price_Prediction_ML_v2.ipynb for comprehensive pipeline
# - Keras_DNN.ipynb for deep learning approach
# - light_gbm.ipynb for gradient boosting

📊 Workflow

1. Data Collection

• Notebooks use Yahoo Finance API to download historical price data
• Support for daily, weekly, and monthly sampling
• Automatic data preprocessing and cleaning

2. Portfolio Optimization

Data Download → Calculate Returns → Estimate Covariance → 
Optimize Weights → Backtest → Evaluate Performance

Key optimization objectives:

• Max Sharpe: Maximize risk-adjusted returns
• Min Variance: Minimize portfolio volatility
• Target Volatility: Achieve specific risk level
• Manual Weights: Custom allocations

3. Price Prediction

Data Download → Feature Engineering → Train/Test Split → 
Model Training → Evaluation → Prediction

Models available:

• Neural Networks: Multi-layer perceptrons with dropout
• Tree-based: Random Forest, LightGBM
• Linear: Logistic Regression

📈 Key Metrics & Outputs

Portfolio Optimization

• Sharpe Ratio - Risk-adjusted return metric
• Sortino Ratio - Downside risk-adjusted return
• Maximum Drawdown - Largest peak-to-trough decline
• Volatility - Standard deviation of returns
• Cumulative Returns - Total portfolio performance
• Efficient Frontier - Risk-return visualization
• Portfolio Weights - Optimal asset allocations

Price Prediction

• Accuracy - Classification accuracy for directional prediction
• Precision/Recall - Model prediction quality
• RMSE/MAE - Regression error metrics
• ROC Curve - Model discrimination ability
• Feature Importance - Key predictive variables
• Predicted vs Actual - Visualization of model performance

🎓 Use Cases

1. Academic Research - Study portfolio theory and ML in finance
2. Investment Strategy Development - Design and backtest portfolio strategies
3. Risk Management - Analyze portfolio risk characteristics
4. Algorithmic Trading - Build predictive models for trading signals
5. Financial Education - Learn about quantitative finance techniques

⚙️ Configuration

Most notebooks include configuration sections at the top:

# Portfolio Optimization Settings
stock = ['SPY', 'GLD', 'QQQ', 'TLT']
date_start = '2010-01-01'
date_end = '2024-01-01'
max_sharpe = True
min_variance = True
target_vol = False

# ML Settings
sampling = 'daily'  # or 'weekly'
train_start = '2010-01-01'
train_end = '2021-12-01'
test_start = '2021-12-02'
test_end = date.today()

📝 Notable Features

Portfolio_Optimisation_SIPP_v2.ipynb

• Comprehensive optimization framework
• Multiple optimization objectives
• L2 regularization support
• Semi-variance and CVaR options
• Quantstats integration for detailed analytics
• Historical bear/bull market dates reference

Price_Prediction_ML_v2.ipynb

• Complete ML pipeline
• Feature engineering with technical indicators
• Multiple model comparison
• Walk-forward validation
• Extensive visualizations

🔬 Technical Details

Data Sources

• Yahoo Finance - Primary data source via yfinance and pandas_datareader
• Support for stocks, ETFs, indices, commodities, futures
• Global market coverage (US, UK, Europe, Asia)

Optimization Methods

• Convex Optimization - Using cvxpy via pypfopt
• Covariance Estimation - Ledoit-Wolf shrinkage, sample covariance
• Expected Returns - Historical mean, CAPM, custom estimates

ML Techniques

• Cross-validation - Time-series split to prevent look-ahead bias
• Hyperparameter Tuning - Grid search, random search
• Feature Scaling - StandardScaler, MinMaxScaler
• Regularization - L1/L2 for preventing overfitting

⚠️ Important Notes

1. Past Performance: Historical returns do not guarantee future results
2. Data Quality: Results depend on data accuracy and completeness
3. Market Conditions: Models trained on specific periods may not generalize
4. Transaction Costs: Backtests do not include trading fees and slippage
5. Rebalancing: Portfolio optimization assumes periodic rebalancing
6. API Limitations: Yahoo Finance API has rate limits and occasional data gaps

🤝 Contributing

Contributions are welcome! Areas for improvement:

• Additional optimization algorithms
• More ML models
• Enhanced risk metrics
• Real-time data integration
• Transaction cost modeling
• Multi-period optimization

📄 License

This project is available for educational and research purposes.

📧 Contact

For questions or collaboration:

• GitHub: @ferhat00

🙏 Acknowledgments

This project uses several excellent open-source libraries:

• PyPortfolioOpt - Portfolio optimization
• Riskfolio-Lib - Risk-based portfolio optimization
• QuantStats - Portfolio analytics
• LightGBM - Gradient boosting framework
• Keras/TensorFlow - Deep learning

📚 References

Portfolio Theory

• Markowitz, H. (1952). "Portfolio Selection"
• Sharpe, W. (1964). "Capital Asset Prices: A Theory of Market Equilibrium"

Machine Learning in Finance

• Advances in Financial Machine Learning (Marcos López de Prado)
• Machine Learning for Asset Managers (Marcos López de Prado)

---

Last Updated: February 2026

Status: Active Development

Python Version: 3