🚀 GARCH–LSTM vs SVJ vs GARCH-SSM

Advanced Portfolio Simulation & CVaR Optimization Framework

---

📌 Overview

This repository implements a robust, research-grade comparison between two state-of-the-art financial scenario generation pipelines for portfolio risk optimization:

• 📉 Stochastic Volatility with Jumps (SVJ)
• 📉 Rgime Update with SVJ
• 📈 GARCH–LSTM Hybrid Model with EVT Tail Smoothing
• **📈 GARCH–SSM Hybrid Model with EVT Tail Smoothing Needs Futher Improvement and Ablation Studies on the Block Size **

The goal is to generate realistic multi-asset return scenarios, optimize portfolios under CVaR constraints, and evaluate out-of-sample performance.

⚠️ Designed for quant research, risk modeling, and agentic AI financial systems

---

🧠 Key Concepts Covered

• Stochastic Volatility (Heston-style)
• Jump Diffusion Models
• GARCH(1,1) Volatility Modeling
• Extreme Value Theory (EVT / POT)
• LSTM Mean Dynamics
• Copula-based Dependence
• Monte Carlo Simulation
• CVaR Portfolio Optimization
• Backtesting & Risk Attribution

---

📂 Repository Structure

.
├── GARCH_LSTM_SVJ.py   # Main experiment pipeline
├── svj_engine.py       # SVJ calibration & simulation engine
├── metric.py           # contains different objective(CVaR,Sharpe,C-Sharpe)
└── README.md