[Paper Review] Neural Stochastic Agent-Based Limit Order Book Simulation: A Hybrid Methodology

Zijian Shi, John Cartlidge|arXiv (Cornell University)|Feb 28, 2023

Complex Systems and Time Series AnalysisEconomics, Econometrics and Finance3 citations

TL;DR

This paper proposes a hybrid neural stochastic agent-based model (NS-ABM) that integrates a data-driven neural stochastic background trader—trained on real limit order book (LOB) data via a neural point process—into an agent-based simulation framework. The model successfully replicates key stylised facts of real markets and demonstrates realistic order flow impact and herding behavior when interacting with trend and value trading agents.

ABSTRACT

Modern financial exchanges use an electronic limit order book (LOB) to store bid and ask orders for a specific financial asset. As the most fine-grained information depicting the demand and supply of an asset, LOB data is essential in understanding market dynamics. Therefore, realistic LOB simulations offer a valuable methodology for explaining empirical properties of markets. Mainstream simulation models include agent-based models (ABMs) and stochastic models (SMs). However, ABMs tend not to be grounded on real historical data, while SMs tend not to enable dynamic agent-interaction. To overcome these limitations, we propose a novel hybrid LOB simulation paradigm characterised by: (1) representing the aggregation of market events' logic by a neural stochastic background trader that is pre-trained on historical LOB data through a neural point process model; and (2) embedding the background trader in a multi-agent simulation with other trading agents. We instantiate this hybrid NS-ABM model using the ABIDES platform. We first run the background trader in isolation and show that the simulated LOB can recreate a comprehensive list of stylised facts that demonstrate realistic market behaviour. We then introduce a population of `trend' and `value' trading agents, which interact with the background trader. We show that the stylised facts remain and we demonstrate order flow impact and financial herding behaviours that are in accordance with empirical observations of real markets.

Motivation & Objective

• To address the limitations of traditional agent-based models (ABMs) that lack grounding in real market data and stochastic models (SMs) that lack interactivity.
• To develop a simulation framework that combines the realism of data-driven stochastic modeling with the dynamic interaction capabilities of agent-based systems.
• To enable realistic 'dynamic back-testing' of trading strategies by simulating market responses to order submissions.
• To investigate how market structure and behavior emerge from the interaction between a data-driven background trader and heterogeneous trading agents.
• To demonstrate that the model preserves key stylised facts of real financial markets while enabling analysis of order flow impact and herding behavior.

Proposed method

• Train a neural stochastic background trader (BT) using a state-dependent parallel neural Hawkes process on historical level-2 limit order book (LOB) data to model the aggregation of market-wide order events.
• Integrate the pre-trained BT into the ABIDES open-source agent-based simulation framework to simulate a realistic market environment with dynamic agent interactions.
• Model the background trader to stochastically sample realistic order event streams by learning price and volume dynamics from real LOB data.
• Simulate interactions between the BT and a population of 'trend' and 'value' trading agents to assess emergent market behaviors.
• Decompose total price impact into 'plain' impact (from order submission alone) and 'order flow' impact (from market response to order flow) using controlled simulation conditions.
• Use a mean-reverting stochastic process to generate exogenous fundamental values for backtesting, enabling dynamic evaluation of trading strategies.

Experimental results

Research questions

• RQ1 Can a neural stochastic background trader trained on real LOB data reproduce the key stylised facts of real financial markets in isolation?
• RQ2 How does the inclusion of heterogeneous trading agents (trend and value traders) affect the preservation of stylised facts in the simulated market?
• RQ3 To what extent does the market respond dynamically to order flow, and how does this response vary with order volume?
• RQ4 Can the model separate and quantify 'plain' price impact from 'order flow' impact in a way that reflects real market dynamics?
• RQ5 Can the NS-ABM framework serve as a realistic dynamic back-testing environment for trading strategies?

Key findings

• The neural stochastic background trader successfully reproduces ten key stylised facts of real financial markets when operating in isolation.
• The addition of 'trend' and 'value' trading agents preserves the stylised facts, confirming the robustness of the background trader’s realism.
• Plain price impact increases monotonically with order volume (λ), and remains permanent after trading stops, reflecting the impact of order submission alone.
• Order flow impact increases superlinearly with λ, indicating that higher order volumes trigger stronger market responses, pushing prices higher than they would otherwise go.
• The model successfully separates total price impact into two components: 'plain' impact (from order submission) and 'order flow' impact (from market reaction), enabling nuanced analysis of market impact.
• The NS-ABM framework enables dynamic back-testing by allowing market behavior to adapt to strategy actions, offering a more realistic alternative to traditional backtesting.