Parallel Market Environments for FinRL Contests

Financial reinforcement learning has attracted lots of attention recently. From 2023 to 2025, we have organized three FinRL Contests featuring different financial tasks. Large language models have a strong capability to process financial documents. By integrating LLM-generated signals into the state, trading agents can take smarter actions based on both structured market data and unstructured financial documents. In this paper, we summarize the parallel market environments for tasks used in FinRL Contests 2023-2025. To address the sampling bottleneck during training, we introduce GPU-optimized parallel market environments to address the sampling bottleneck. In particular, two new tasks incorporate LLM-generated signals and all tasks support massively parallel simulation. Contestants have used these market environments to train robust and powerful trading agents for both stock and cryptocurrency trading tasks.