Portfolio Optimization with NVIDIA cuOpt

GPU-accelerated CVaR portfolio optimization using NVIDIA cuOpt on H200 GPUs.

Quick Start

python flows/portfolio/flow.py --no-pylint --environment=fast-bakery run --with kubernetes

What It Does

Solves CVaR (Conditional Value-at-Risk) portfolio optimization on GPU:

• Fetches historical price data (or generates synthetic data)
• Generates Monte Carlo scenarios for risk estimation
• Solves LP optimization using cuOpt's GPU-accelerated solver
• Sweeps across risk levels to build an efficient frontier
• Produces a Metaflow card with visualizations

Project Structure

optimization-project/
├── flows/portfolio/flow.py   # Main optimization flow
├── src/__init__.py           # Package policy for Metaflow
└── README.md

Parameters

• --n_assets: Number of assets (default: 100)
• --n_scenarios: Monte Carlo scenarios (default: 1000)
• --confidence: CVaR confidence level (default: 0.95)