The Impact of Market Informedness on Market Makers' Profitability

Battle Count: 72

Highly relevant to quantitative trading. The paper directly addresses market-making profitability, adverse selection, price discovery, and liquidity provision - all core concerns for quantitative trading firms. The agent-based simulation with RL provides a framework for designing and testing market-making strategies. The finding that informed trading can improve market-maker profitability through price discovery (contrary to some analytical results) has direct implications for strategy design. The Hawkes process modeling of order flow and the LOB simulation are directly applicable to real trading systems.

Very high implementation complexity. Requires: (1) continuous-time Hawkes process simulation with Ogata's thinning method, (2) full limit order book with FIFO queues and price-time priority, (3) multi-agent RL with MAPPO/CTDE paradigm, (4) complex neural network architecture combining MLP, cross-attention, GRU, and Mixture-of-Experts heads, (5) centralized critic with privileged global state, (6) deterministic cancellation mechanisms, (7) state-dependent intensity modulation, (8) warm-start initialization for Hawkes processes. The codebase involves multiple interacting components across continuous and discrete time.

Code is available on GitHub. The paper provides detailed parameter tables (Appendix F) including Hawkes process parameters, agent configurations, PPO hyperparameters, neural network dimensions, and market setup. The simulation environment follows PettingZoo ParallelEnv interface and RLlib. Full mathematical specifications for observations, actions, rewards, and architecture are provided in appendices. Stability proofs are included. However, the paper does not specify exact random seeds for all experiments or provide pre-trained model weights.