Paper: arXiv 2308.13289

Authors: Sascha Frey, Kang Li, Peer Nagy, Silvia Sapora, Chris Lu, Stefan Zohren, Jakob Foerster, Anisoara Calinescu

Abstract

Financial exchanges across the world use limit order books (LOBs) to process orders and match trades. For research purposes it is important to have large scale efficient simulators of LOB dynamics. LOB simulators have previously been implemented in the context of agent-based models (ABMs), reinforcement learning (RL) environments, and generative models, processing order flows from historical data sets and hand-crafted agents alike. For many applications, there is a requirement for processing multiple books, either for the calibration of ABMs or for the training of RL agents. We showcase the first GPU-enabled LOB simulator designed to process thousands of books in parallel, with a notably reduced per-message processing time. The implementation of our simulator - JAX-LOB - is based on design choices that aim to best exploit the powers of JAX without compromising on the realism of LOB-related mechanisms. We integrate JAX-LOB with other JAX packages, to provide an example of how one may address an optimal execution problem with reinforcement learning, and to share some preliminary results from end-to-end RL training on GPUs.

Complexity vs Empirical Score

  • Math Complexity: 3.0/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper presents a GPU-accelerated LOB simulator (JAX-LOB) with a focus on implementation, parallelism, and RL training speedups, featuring code-like algorithms and performance metrics but minimal novel mathematical derivations.

Research Flowchart

  flowchart TD
  A["Research Goal: <br>Develop GPU-Accelerated LOB Simulator"] --> B["Key Methodology: <br>Design JAX-LOB Architecture"]
  
  B --> C{"Data/Input: <br>Market Data & Agents"}
  C --> D["Computational Process: <br>Parallel GPU Processing"]
  
  D --> E["Computational Process: <br>Reinforcement Learning Integration"]
  E --> F["Outcome: <br>1000x Speed Improvement"]
  
  F --> G["Outcome: <br>Large Scale RL Training Enabled"]
  G --> H["Outcome: <br>Optimal Execution Results"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper introduces JAX-LOB, the first GPU‑enabled limit order book simulator designed to process thousands of books in parallel for reinforcement learning applications.

Source sentence
We showcase the first GPU-enabled LOB simulator designed to process thousands of books in parallel, with a notably reduced per-message processing time.

Assumes

No explicit assumptions found in the text.

Data

  • LOBSTER order-book public
    Source sentence
    JAX-LOB, while originally designed to process data from LOBSTER [18], may process messages from any source, includ-ing strategic agents.

Recognised sources: LOBSTER

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • Using JAX-LOB within a gymnax environment yields at least a 5× speedup compared to a comparable CPU implementation. performance
    Source sentence
    Using this as part of a gymnax environ-ment for reinforcement learning we obtain at least a 5x speedup compared to a comparable CPU implementation.
  • When training a RL agent with PureJaxRL, a 7× speed‑up is observed relative to the comparable CPU implementation. performance
    Source sentence
    When using the environment to train a RL agent with PureJaxRL we even see a 7x speed-up with respect to our comparable CPU implementation.
  • The PPO agent trained on JAX-LOB outperforms the TWAP benchmark in training, though out‑of‑sample validation is not provided. performance
    Source sentence
    We set 𝜆= 0 to make training easier, as the drift part of the reward function is much harder to learn, and compare the training curve to the TWAP benchmark strategy, which is to execute the desired volume linearly over the episode duration.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 10 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 16 papers in the archive (13 by independent authors): Mixture-of-Experts for Cryptocurrency Order Execution: … (uses its method or data); Neural Estimation of Irreversibility in Real and Simulated … (uses its method or data); Prospects of Imitating Trading Agents in the Stock Market (uses its method or data)

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.