An open-source framework for applying deep reinforcement learning to quantitative finance, featuring a train-test-trade pipeline for stock and crypto trading.
FinRL is an open-source framework that applies deep reinforcement learning to quantitative finance. It provides a structured pipeline for developing, training, and backtesting automated trading agents on financial data, solving the problem of creating adaptive, data-driven trading strategies.
Students, researchers, and developers interested in learning about or prototyping financial reinforcement learning strategies for stocks, cryptocurrencies, and portfolio management.
Developers choose FinRL for its status as the first open-source framework in this domain, its comprehensive educational resources, and its integrated workflow that simplifies the process of going from data to a trained trading agent.
FinRL®: Financial Reinforcement Learning. 🔥
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Provides step-by-step examples like the FinRL Stock Trading 2026 tutorial, offering a clear workflow from data download to backtesting, which is ideal for learners and researchers.
Integrates over a dozen sources including Yahoo Finance, Alpaca, and Binance, with automated preprocessing for technical indicators, enabling diverse market experimentation.
Offers a train-test-trade workflow with gym-style environments and agents from libraries like Stable Baselines 3, simplifying the end-to-end process for financial RL prototyping.
As the first open-source framework in this domain, it includes multiple DRL algorithms (e.g., A2C, PPO) for benchmarking, supported by peer-reviewed publications.
The README describes FinRL as a 'coupled monolith' with a three-layer structure, making it less flexible and harder to extend compared to modular alternatives like FinRL-X.
Admits that FinRL-X is the next-generation version for production, indicating this original framework has basic live trading support and lacks advanced risk management features.
Uses hand-rolled evaluation loops instead of professional backtesting libraries, which may limit robustness and performance compared to industry-standard tools.