The Reinforce.fi team (formerly Overnight.fi) is launching the Market-Action Arena hackathon, offering participants the chance to test their Reinforcement Learning models on real anonymized market data to maximize PnL.
What Happened
The Market-Action Arena project is a competition where developers must build models to select one of 10 discrete market actions (from A1 to A10). The goal for participants is to maximize total PnL based on provided anonymized features. The event is tentatively scheduled to start around June 23, 2026, and the hackathon duration will be between 1.5 to 2 months.
Context
Unlike classic machine learning tasks, which primarily focus on price prediction (regression or classification), this hackathon focuses on sequential decision-making tasks. The use of market sequences with a length of 1,000 steps allows for testing the long-term strategies of agents, and the PnL metric as a target function directly links the quality of the RL policy to economic results.
Why It Matters for the Industry
This event marks a paradigm shift in trading system development: moving from purely informational models to active capital management agents. This facilitates the validation of agentic architectures in conditions as close to live environments as possible and stimulates the emergence of new approaches to sequential decision-making in the fintech sector, which in the long term could lead to deeper integration of RL into high-frequency and adaptive trading platforms.
Why It Matters for Users
For researchers and developers, this is a rare opportunity to test their RL agents on high-quality real-world data outside of standard Kaggle tasks. Participants get a chance to present their solutions to a team of professional algorithmic strategy developers and test their work in a niche, high-yield asset management field.
Sources
Author
Look at AI, Editorial Staff