FinRL-Meta: Market Environments and Benchmarks for Data-Driven Financial Reinforcement Learning

FinRL-Meta is the data and environment layer beneath financial reinforcement learning: it turns real market data into gym-style trading environments and reproduces published DRL trading strategies so they can be compared on common ground. The abstract names the three obstacles the design targets — the low signal-to-noise ratio of financial data, survivorship bias, and backtesting overfitting.

The code is MIT-licensed and still maintained (last commit April 2026). It ships eight environment families under meta/: crypto trading, execution optimizing, futures trading, FX trading, market impact, portfolio allocation, portfolio optimization, and stock trading. The paper’s “hundreds of market environments” counts data-driven instantiations, not distinct environment classes. The PyPI package finrl-meta has not been released since 0.3.6 in February 2023, so the repository — not pip — is the current path in.

An extended journal version appeared as “Dynamic datasets and market environments for financial reinforcement learning” in Machine Learning 113(5), 2024 (doi:10.1007/s10994-023-06511-w). The repo positions FinRL-Meta as the environment and benchmark layer of the FinRL ecosystem, and carries its maintainers’ own disclaimer that nothing in it is financial advice or a recommendation to trade real money.