MarketGPT for Generating Limit Order Book Events and Market Simulations

Summary

This work presents a generative pre-trained transformer for modeling financial time series as sequences of order messages. The model serves as an order-generation engine inside a discrete-event simulator, with the aim of reproducing limit order book dynamics and supporting interactive market simulations. It is designed to produce long sequences incrementally, including after the original order-flow prompt has moved outside its context window.

The reported evaluations find that generated order flow reproduces key features of observed order-flow data and captures several statistical patterns associated with real markets and broader macro-scale data. These results concern simulation fidelity, not a demonstrated trading strategy or evidence of profitable execution. The document provides no detailed metrics, datasets, or comparison baselines, so the extent of fidelity and the settings where it holds are unclear from this description. The approach is presented as progress toward high-fidelity market simulation, with practical usefulness depending on validation against real order books and on how well the simulator represents market conditions beyond those evaluated.

Key ideas

• The transformer generates sequences of financial order messages for a discrete-event simulator.
• The system is intended to reproduce limit order book dynamics.
• It can continue generating order flow after the initial prompt leaves its context window.
• The reported evaluations find that it captures key order-flow features and several statistical market patterns.
• Simulation fidelity does not by itself show that the model supports profitable trading.

Tags

# MarketGPT: Developing a Pre-trained transformer (GPT) for Modeling Financial Time Series


# MarketGPT: Developing a Pre-trained transformer (GPT) for Modeling Financial Time Series









This work presents a generative pre-trained transformer (GPT) designed for modeling financial time series. The GPT functions as an order generation engine within a discrete event simulator, enabling realistic replication of limit order book dynamics. Our model leverages recent advancements in large language models to produce long sequences of order messages in a steaming manner. Our results demonstrate that the model successfully reproduces key features of order flow data, even when the initial order flow prompt is no longer present within the model's context window. Moreover, evaluations reveal that the model captures several statistical properties, or 'stylized facts', characteristic of real financial markets and broader macro-scale data distributions. Collectively, this work marks a significant step toward creating high-fidelity, interactive market simulations.

Shown in full with attribution under the source's licence. Licence: abstract CC0

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.