A Limit Order Book Simulator that runs on the GPU and can be used for Reinforcement Learning.

What is it about?

Modern financial markets are based on a mechanism called the limit order book (LOB). This work creates a simulator of the LOB which can be run on the GPU allowing many LOBs to be simulated in parallel. This is especially useful for a specific type of Artificial Intelligence called Reinforcement Learning (RL) where an agent needs to interact with an environment millions of times in order to learn how to craft an optimal policy. We show how our GPU simulator allows for faster environment interactions and apply it to a well-known problem in high frequency trading: optimally executing an order.

Featured Image

Photo by m. on Unsplash

Photo by m. on Unsplash

Why is it important?

This is the first GPU-native simulator of the LOB. It uses Google's JAX to allow for easy parallelisation. The applications are far-reaching in high-frequency trading-related problems and could be applied to different tasks. Further, the use of intelligent agents in agent-based models may allow for realistic simulations and synthetic data-generation.