Adaptive Regime Portfolio Optimization

This repository contains the implementation of the thesis project:
"Regime-Dependent Portfolio Optimization: An Integrated Framework of Statistics, Risk-Parity and Deep Learning"

📘 Abstract

In an era of rapidly shifting market conditions, static portfolio optimization models fall short in capturing structural market transitions. This project introduces a regime-switching portfolio optimization framework using Hidden Markov Models (HMMs) to detect market regimes and adaptively rebalance portfolios using a combination of traditional financial models and deep learning approaches.

🧠 Key Features

• Regime Detection using HMMs
• Dynamic Portfolio Switching based on market regimes
• Integration of Multiple Portfolio Strategies:
  • Markowitz Mean-Variance (MVP)
  • Hierarchical Risk Parity (HRP)
  • Autoencoder-based Deep Learning Optimization
  • Black-Litterman blended with Conditional Value-at-Risk (CVaR)

📊 Methodology Overview

1. Data Acquisition: NSE sectoral data (2018–2022) via Yahoo Finance APIs using pandas-datareader.
2. Static Portfolio Strategies: Implemented and benchmarked MVP, HRP, and Autoencoder-based methods.
3. Adaptive Strategy:
   • Regime identification via HMM
   • Regime-specific portfolio allocation using CVaR/MVP
   • Walk-forward testing approach for rebalancing
4. Performance Metrics:
   • Annual Return
   • Annual Volatility
   • Sharpe Ratio

📈 Sectors Covered

Portfolios are constructed for the following 10 NSE Thematic Sectors:

• NIFTY Commodities
• NIFTY Energy
• NIFTY Manufacturing
• NIFTY Services
• NIFTY MNC
• NIFTY Transportation & Logistics
• NIFTY Infrastructure
• NIFTY Housing
• NIFTY Consumption
• NIFTY 100 ESG

🛠️ Technologies Used

• Python 3.10+
• Libraries: numpy, pandas, scikit-learn, hmmlearn, keras, matplotlib, seaborn, pypfopt
• Data: Yahoo Finance API, NSE sectoral compositions

📬 Contact

For questions or collaborations, reach out to: 📧 [email protected]