Reproduce “DeepSeekMath Meets Order Book: Group-Aware Policy Optimization for High-Frequency Directional Trading” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sayak Charabarty, Souradip Pal
The system compresses limit-order-book changes into order-flow imbalance features, predicts returns at six event horizons, and trains trading policies with GRPO and GSPO using comparisons among episodes. On a one-hour 2013 backtest for AMZN, AAPL, and GOOG, both methods report higher average return and profitability, and less severe drawdown, than PPO and Q-learning; this result is limited to that test.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.