Show HN: RL agent trained only in synthetic markets got better on real 2020crash
Category: ai-ml
Tags: reinforcement-learning, synthetic-data, finance, transfer-learning, reproducibility
Score: 6.5/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 4)
This project demonstrates that a reinforcement learning agent trained exclusively on synthetic market data can transfer to real historical market crashes (2020 and 2022), improving hedging performance. The work is pre-registered, fully reproducible, and includes a robustness check with five seeds and a larger model, making it a valuable case study for synthetic-to-real transfer in RL environments.
Target audience: RL researchers, quantitative finance practitioners, ML engineers interested in sim-to-real transfer
Repository: https://github.com/giack-hack/synthetic-markets-transfer · Python · MIT · 5 stars
View on Hacker News