[Paper Review] ABIDES: Towards High-Fidelity Market Simulation for AI Research

David R. Byrd, Maria Hybinette+1 more|arXiv (Cornell University)|Apr 26, 2019

Distributed and Parallel Computing Systems7 references35 citations

TL;DR

ABIDES is an agent-based discrete-event market simulator designed for AI research, enabling high-fidelity, nanosecond-precision trading simulations with configurable latency and historical-date replay.

ABSTRACT

We introduce ABIDES, an Agent-Based Interactive Discrete Event Simulation environment. ABIDES is designed from the ground up to support AI agent research in market applications. While simulations are certainly available within trading firms for their own internal use, there are no broadly available high-fidelity market simulation environments. We hope that the availability of such a platform will facilitate AI research in this important area. ABIDES currently enables the simulation of tens of thousands of trading agents interacting with an exchange agent to facilitate transactions. It supports configurable pairwise network latencies between each individual agent as well as the exchange. Our simulator's message-based design is modeled after NASDAQ's published equity trading protocols ITCH and OUCH. We introduce the design of the simulator and illustrate its use and configuration with sample code, validating the environment with example trading scenarios. The utility of ABIDES is illustrated through experiments to develop a market impact model. We close with discussion of future experimental problems it can be used to explore, such as the development of ML-based trading algorithms.

Motivation & Objective

• Provide a configurable, open-source platform for AI research in market applications.
• Enable high-fidelity simulation of continuous double auction markets with realistic latency and computation delays.
• Allow replay and manipulation of historical market dates to enable controlled experiments.
• Support learning and explainability by centralized, time-synced agent logging.
• Demonstrate utility through case studies on background agent realism and market impact of large orders.

Proposed method

• Agent-based discrete-event kernel with nanosecond time resolution and per-agent current time tracking.
• Message-based inter-agent communication through a verified, NASDAQ-like ITCH/OUCH protocol stack.
• Exchange and order-book modeling with full logging, order matching, and slippage emerging from delays.
• Configurable pairwise network latency and computation delays to simulate co-location and latency effects.
• Open-source Python implementation using NumPy and Pandas with a configurable experimental bootstrapper and agent hierarchy.
• Case studies including background agent realism and a market-impact experiment with a single large order.

Experimental results

Research questions

• RQ1 How can we create a high-fidelity, open-source market simulator suitable for AI research?
• RQ2 What latency and computation delay configurations are necessary to produce realistic market microstructure and agent behavior?
• RQ3 Can historical-date replay and controlled perturbations enable causal experimentation on market dynamics?
• RQ4 How does a large market order impact prices in a realistic multi-agent environment and how does profit scale with trade size?
• RQ5 What features enable explainability and analysis of learning agents in market simulations?

Key findings

• ABIDES can reproduce intra-day price behavior of historical days when populated with background agents interacting with an exchange.
• A single large market order (impact agent) creates measurable market impact with impact growing with the trader's ‘greed’ parameter and trade size.
• Profit from an impact trade increases with larger bids, but profit per share can decrease as trade size grows.
• Market impact studies show clear effects on price and subsequent trading patterns, enabling event-aligned analyses across days.
• The platform provides detailed event logging facilitating explainability of learned agent policies.
• The momentum-style and background agents illustrate the feasibility of integrating ML/AI agents into realistic market simulations.