Paper: arXiv 2312.14903
Authors: Aaron Wheeler, Jeffrey D. Varner
Abstract
In this study, we developed a computational framework for simulating large-scale agent-based financial markets. Our platform supports trading multiple simultaneous assets and leverages distributed computing to scale the number and complexity of simulated agents. Heterogeneous agents make decisions in parallel, and their orders are processed through a realistic, continuous double auction matching engine. We present a baseline model implementation and show that it captures several known statistical properties of real financial markets (i.e., stylized facts). Further, we demonstrate these results without fitting models to historical financial data. Thus, this framework could be used for direct applications such as human-in-the-loop machine learning or to explore theoretically exciting questions about market microstructure’s role in forming the statistical regularities of real markets. To the best of our knowledge, this study is the first to implement multiple assets, parallel agent decision-making, a continuous double auction mechanism, and intelligent agent types in a scalable real-time environment.
Complexity vs Empirical Score
- Math Complexity: 7.0/10
- Empirical Rigor: 6.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper demonstrates high mathematical complexity through its use of sophisticated stochastic processes, statistical analysis of stylized facts, and complex system dynamics. It shows substantial empirical rigor by implementing a realistic continuous double auction matching engine, running large-scale simulations with distributed computing, and validating results against known market properties without historical data fitting, making it backtest-ready and implementation-heavy.
Research Flowchart
flowchart TD A["Research Goal<br>Develop Scalable Agent-Based<br>Financial Market Simulator"] --> B["Key Methodology"] B --> C["Computational Processes"] C --> D["Key Findings"] B --> B1["Heterogeneous Agent Design<br>Multiple Asset Classes"] B --> B2["Distributed Computing<br>Parallel Decision-Making"] B --> B3["Continuous Double Auction<br>Matching Engine"] C --> C1["Simulate Large-Scale<br>Multi-Asset Markets"] C --> C2["Process Orders in Real-Time<br>without Historical Data"] D --> D1["Captures Stylized Facts<br>of Real Markets"] D --> D2["Scalable Framework for<br>ML & Market Microstructure Research"]
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 proposes a computational framework for simulating large-scale, multi-asset agent-based financial markets with parallel heterogeneous agents and a continuous double auction matching engine.
Source sentence
In this study, we developed a computational framework for simulating large-scale agent-based financial markets.
Assumes 6
- Model parameters are derived from first principles rather than fitted to historical data. model
Source sentence
Rather than fit our agent or environmental models to empirical data, model parameters were derived from first principles.
- The total initial amount of cash and shares is conserved; no new cash or shares are introduced during simulation. execution
Source sentence
Finally, there was no influx of new cash or shares in the simulation, and the total initial amount of cash and shares was conserved. 2.2 Agents We used four agent groups to generate trading activity in our simulated market environment: liquidity takers, liquidity providers, market-making agents, and intelligent agents.
- Agents cannot spend money they do not have or trade shares they do not own. execution
Source sentence
All other agent types cannot spend money they don’t have and cannot trade shares they don’t own.
- An explicit order book structure and price‑time priority matching engine are used to model market microstructure. market
Source sentence
For example, we used an explicit order book structure and matching engine mechanism.
Show 2 more
- The simulation does not include traditional agent behaviors such as fundamental or technical strategies. model
Source sentence
Further, our ABM does all of this without the inclusion of traditional agent behaviors (e.g., fundamental, technical, etc.) or the need for historical market data.
- The baseline ABM is evaluated without calibrating to historical market data. data
Source sentence
Further, we demonstrate these results without fitting models to historical financial data.
Data
- Synthetic financial time series generated by the baseline ABM synthetic unknown
Source sentence
3.2 Validation of synthetic financial time series data The baseline financial ABM captured several empirical properties of financial time series (Fig. 3). To gauge the efficacy of our baseline ABM and validate our approach, we tested our synthetic time series for the presence of several common univariate stylized facts.
Costs
Not applicable
Finds 5
- The baseline ABM captures several known stylized facts of real financial markets, such as heavy‑tailed return distributions and lack of autocorrelation of price returns. statistical
Source sentence
We present a baseline model implementation and show that it captures several known statistical properties of real financial markets (i.e., stylized facts). (a) Heavy-tailed returns distribution is demonstrated by the high and narrow peak at zero return and the slow decay at the margins of the distribution.
- The simulation replicates several, but not all, typical stylized facts without using historical data for calibration. statistical
Source sentence
The simulation was able to replicate several, but not all, of a set of typical stylized facts without using historical data for calibration.
- Introducing an intelligent market‑making agent stabilizes bid‑ask prices and improves market liquidity compared to using only market‑making agents. performance
Source sentence
(a) Two market-making agents induced relatively volatile bid (blue) and ask (orange) prices, characterized by large upward and downward price movements. (b) One market making agent and one intelligent agent induced stable bid (blue) and ask (orange) prices.
- The intelligent agent consistently outperforms other market‑making agents and generates profit in each simulation run despite high noise and a simple linear model. performance
Source sentence
The intelligent agent consistently outperformed other market marking agents and demonstrated a profit in each simulation run. Interestingly, the intelligent agent accomplished this despite the high degree of noise in the environment and the relative simplicity of its linear model.
- The baseline ABM scales to several thousand agents and supports multiple simultaneous assets while using modest computational resources. theoretical
Source sentence
Our results demonstrated how the baseline ABM could support multiple simultaneous assets by only using basic agent types 10 Scalable Agents-Based Financial Market Simulation and modest computational resources. 4 Summary and Conclusion This study introduced a baseline financial ABM that scaled to several thousand agents, supports multiple simultaneous assets, captured intuitive supply and demand dynamics, and
Checked by others
Supported by
- Looking into informal currency markets as Limit Order Books: impact of market makers
Compare the findings
This paper: Introducing an intelligent market‑making agent stabilizes bid‑ask prices and improves market liquidity compared to using only market‑making agents.
That paper: The market maker improves the quality of the market, reducing bid‑ask spreads and increasing execution rates.
1 paper in the archive
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (70)
- Multifractal analysis of financial markets: a review (2019) — link
- Physics and financial economics (1776 \textendash 2014): puzzles, Ising and agent-based models (2014) — link
- Empirical properties of asset returns: stylized facts and statistical issues (2001) — link
- Econophysics review: I. Empirical facts (2011) — link
- Limit order books (2013) — link
- Get real (2020) — link
- Testing for Common Trends (1988) — link
- Statistical properties of stock order books: empirical results and models (2002) — link
and 62 more
- Radical Complexity (2021) — link
- Challenges, tasks, and opportunities in modeling agent-based complex systems (2021) — link
- Agent-based economic models and econometrics (2012) — link
- Time series properties of an artificial stock market (1999) — link
- The Predictive Power of Zero Intelligence in Financial Markets (2003) — link
- A multi-assets artificial stock market with zero-intelligence traders (2011) — link
- Agent-based simulation of a financial market (2001) — link
- Improving Generalization in Reinforcement Learning \textendash Based Trading by Using a Generative Adversarial Market Model (2021) — link
- Towards a fully rl-based market simulator (2021) — link
- ABIDES: Towards High-Fidelity Market Simulation for AI Research (2019) — link
- On the Design of Agent-Based Artificial Stock Markets (2013) — link
- Stock Price Formation: Precepts from a Multi-Agent Reinforcement Learning Model (2022) — link
- Simulation and estimation of an agent-based market-model with a matching engine (2021) — link
- Agent-Based Model Calibration using Machine Learning Surrogates (2017) — link
- Combining search strategies to improve performance in the calibration of economic ABMs (2023) — link
- Is the Juice Worth the Squeeze? Machine Learning (ML) In and For Agent-Based Modelling (ABM) (2020) — link
- Agent-Based Modeling for Complex Financial Systems (2018) — link
- How markets slowly digest changes in supply and demand (2008) — link
- Understanding flash crash contagion and systemic risk: A micro \textendash macro agent-based approach (2019) — link
- Reinforcement Learning for Market Making in a Multi-agent Dealer Market (2019) — link
- The Flash Crash: High-Frequency Trading in an Electronic Market (2017) — link
- Deep Q-learning market makers in a multi-agent simulated stock market (2021) — link
- An agent based model of the E-Mini SP500 applied to flash crash analysis (2012) — link
- Static and dynamic factors in an information-based multi-asset artificial stock market (2018) — link
- Multi-asset financial bubbles in an agent-based model with noise traders’ herding described by an n-vector Ising model (2021) — link
- DeepABM: Scalable, efficient and differentiable agent-based simulations via graph neural networks (2021) — link
- Th ' e orie de la sp ' e culation (1900) — link
- " U ber die von der molekularkinetischen Theorie der W" a rme geforderte Bewegung von in ruhenden Fl" u ssigkeiten suspendierten Teilchen (1905) — link
- Portfolio Selection (1952) — link
- The Pricing of Options and Corporate Liabilities (1973) — link
- An Intertemporal Capital Asset Pricing Model (1973) — link
- CAPITAL ASSET PRICES : A THEORY OF MARKET EQUILIBRIUM UNDER CONDITIONS OF RISK \ast (1964) — link
- The Variation of Certain Speculative Prices (1963) — link
- ANALYSIS OF LONG-RUN DEPENDENCE IN ECONOMICS-R/S TECHNIQUE (1971)
- Agent based modeling in economics and finance: past, present, and future (2022)
- Foundations of complexity economics (2021) — link
- Theory of games and economic behavior (1947)
- The Art of Computer Programming, The: Seminumerical Algorithms (1997)
- Recursive least squares with forgetting for online estimation of vehicle mass and road grade: theory and experiments (2005) — link
- Some Implementation Aspects of Sliding Window Least Squares Algorithms (2000) — link
- Random matrix approach to cross correlations in financial data (2002) — link
- Distribution of eigenvalues for a class of real symmetric matrices. (1971)
- Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems (2020) — link
- Econophysics: Empirical facts and agent-based models (2009) — link
- Scaling of the distribution of fluctuations of financial market indices (1999) — link
- Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of United Kingdom Inflation (1982) — link
- Long range dependence in financial markets — link
- Gain/loss asymmetry in time series of individual stock prices and its relationship to the leverage effect (2009) — link
- Option pricing: A simplified approach (1979) — link
- GARCH 101: The Use of ARCH / GARCH Models in Applied Econometrics (2001) — link
- ARIMA Models (1990) — link
- Vector Autoregressions (2001) — link
- Noise Dressing of Financial Correlation Matrices (1999) — link
- RANDOM MATRIX THEORY AND FINANCIAL CORRELATIONS (2000) — link
- Financial Applications of Random Matrix Theory: a short review (2009) — link
- Proximal Policy Optimization Algorithms (2017) — link
- Playing Atari with Deep Reinforcement Learning (2013) — link
- High-Frequency Data and Models (2005) — link
- The expected value of the adjusted rescaled Hurst range of independent normal summands (1976) — link
- Julia: A fresh approach to numerical computing (2017) — link
- JuMP: A Modeling Language for Mathematical Optimization (2017) — link
- Market Microstructure Theory (1997)
0 of 70 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.