Research on Multi-Agent Reinforcement Learning for Prediction Market Market Making Released
Get up to speed
Researcher Skander Lejmi has released a research paper on cooperative multi-agent reinforcement learning for market making in Polymarket prediction market order books. Based on the JaxMARL-HFT framework, the market making model was trained on 417 real markets, with the market making agent split into a quoter and a risk manager, trained using recurrent independent proximal policy optimization combined with RNN policies. Tests show that the improved single-agent market making model quotes more frequently and has smaller out-of-sample losses, but none of the models achieved positive average portfolio value in out-of-sample evaluation, and each configuration was trained with only a single seed.
AI-generated from reports · 3 hr agoUpdated
Coverage timeline
Follow the coverage to understand different sides of the event.
- arXiv q-finResearch on cooperative multi-agent reinforcement learning for market making in prediction market order books
The study addresses market-making challenges in Polymarket's thin order books, using the JaxMARL-HFT framework to train cooperative multi-agent market-making models on 417 real markets. The market-making agents are split into quoters and risk managers, trained using recurrent independent proximal policy optimization (IPPO) combined with RNN policies. Testing showed that the improved single-agent market-making model quoted more frequently and had smaller out-of-sample losses, but none of the models achieved positive average portfolio value in out-of-sample evaluation, with only single-seed training completed.
Trend of this event
Not enough consecutive observations yet; trend is not drawn.