DeepSeekMath Meets Order Book: Group-Aware Policy Optimization for High-Frequency Directional Trading
About
This paper studies reinforcement learning for high-frequency trading on limit order books by pairing an Order-Flow-based state model with policy-gradient methods. Instead of value-based RL techniques like tabular Q-learning, our approach deploys policy-based methods like vanilla PPO and DeepSeekMath-inspired variants like GRPO and GSPO, that use group-normalized updates and downside-aware shaping. On backtests with financial assets AMZN, AAPL, and GOOG under a simplified backtesting setup based on spread-scaled rewards, these new policies improve net average PnL, profitability, and drawdown over the Q-Learning baseline. Our results show that (1) Order-Flow signals are an adequate state for policy RL and (2) group-aware PPO surrogates are preferable over value-based baselines.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Trading Backtest | AMZN held-out one-hour (test) | Average Return1.82e+3 | 4 | |
| Trading Backtest | AAPL held-out one-hour (test) | Average Return1.93e+3 | 4 | |
| Trading Backtest | GOOG held-out one-hour window (test) | Average Return2.08e+3 | 4 |