← Back
Evidence source 4914Spot Checked

Deep Reinforcement Learning in Non-Markov Market-Making

Risks2025-02-25Paper
Executive summary

This paper presents a deep reinforcement learning framework for optimal market-making using non-Markov jump-diffusion models with simulated and real data.

What it examines

This paper examines an optimal market-making problem in high-frequency trading using deep reinforcement learning. It employs non-Markovian price models—semi-Markov and Hawkes jump-diffusion processes—to simulate limit order book dynamics, offering a modern alternative to traditional stochastic optimal control approaches in volatile trading environments.

What it concludes

This study shows that deep reinforcement learning with the Soft Actor--Critic algorithm can enhance market-making by addressing non-Markovian dynamics and adverse fills. The robust strategies may be applied in optimal liquidation, statistical arbitrage, and other algorithmic trading areas, with future work aimed at improved order book modeling.

Extracted from this source

Evidence objects

Evidence 340378% extraction confidence
Researchers unveil a novel deep reinforcement learning framework that leverages a Soft Actor-Critic algorithm with semi-Markov and Hawkes jump-diffusion models to simulate limit order book dynamics in non-Markovian financial markets.

key_findings bullet 1 · key_findings · validation V0

Evidence 340478% extraction confidence
Incorporating adverse fill modeling and non-adverse fill probabilities boosts trading performance by enabling deep RL agents to navigate complex state-action spaces, minimize inventory risks, and outperform traditional stochastic control methods.

key_findings bullet 2 · key_findings · validation V0

Evidence 340578% extraction confidence
Extensive simulations using real ES futures data verify the frameworks efficacy despite constant spread and midprice simplifications; findings highlight deep RL's promise yielding adaptive, realistic trading strategies to mitigate risk.

key_findings bullet 3 · key_findings · validation V0

Evidence 340678% extraction confidence
This paper introduces an innovative approach by applying deep reinforcement learning to market-making, circumventing the typical Markov assumption. This novel method aligns with real-world complexities in electronic markets and limit order book research. Its originality and fresh perspective offer compelling insights into market microstructure dynamics, inspiring future research advancements significantly.

key_findings bullet 4 · key_findings · validation V0

Raw abstract and provenance

- … a deep reinforcement learning (RL) framework for an optimal market-making (MM) trading … In this way, the model can discover interesting trading strategies that are not …

Source row: 563 · abstract type: snippet