Machine learning
QuantVision
A studio for experimenting with reinforcement-learning trading strategies on historical market data, evaluated against traditional benchmarks. It is a research sandbox, not investment advice.
What it does
- Proximal Policy Optimization agents (Stable-Baselines3) trained in a custom Gymnasium trading environment.
- Strategies evaluated on Sharpe ratio, Sortino ratio, max drawdown, and equity-curve comparison against buy-and-hold.
- Portfolio allocation explored with Modern Portfolio Theory; risk views include Value at Risk, benchmark beta, and correlation heatmaps.
- Historical stress testing across the 2008, 2020, and 2022 market regimes, with market data pulled via yFinance.
Built with
- Stable-Baselines3
- Gymnasium
- FastAPI
- Next.js
- yFinance
- Pandas