FABS: an extensible and high-performance digital twin framework of AI-driven financial systems
File(s)
Published version
Author(s)
Leung, Angus
Guo, Ce
Luk, Wayne
Type
Conference Paper
Abstract
he study of complex AI systems in finance requires scalable digital twins, virtual laboratories for model verification, stress-testing, and counterfactual analysis. Agent-based modeling (ABM) is a primary tool for creating these digital twins, offering a bottom-up approach to capturing market-wide emergent phenomena. However, a major computational bottleneck hinders this work: existing simulation tools cannot efficiently model the large populations of interacting agents required for a high-fidelity digital replica of a financial system. The state-of-the-art simulator, MAXE, for instance, cannot accelerate a single, large-scale simulation. To address this, we introduce the Financial Agent-Based Simulator (FABS), an open-source C++ platform specifically designed as a high-performance engine for constructing and operating these large-scale digital twins. FABS combines three key features: a fine-grained parallel architecture to accelerate a single simulation, a dynamic graph-based optimization to reduce communication overhead, and an extensible callback system for simplified AI model integration. When benchmarked against MAXE in a large-scale fire sale scenario, FABS achieves a runtime speed-up of up to 12.52x. We validate its fidelity as a digital twin by showing that its synthetic data reproduces key stylized facts of financial markets, such as volatility clustering and fat-tailed returns. FABS provides a foundational tool for the financial AI community to build and analyze the complex agent-based digital twins needed to understand modern markets, at a scale that was previously computationally prohibitive.
Date Issued
2025-11-14
Date Acceptance
2025-10-19
Citation
ICAIF '25: Proceedings of the 6th ACM International Conference on AI in Finance, 2025
ISBN
9798400722202
Publisher
ACM
Journal / Book Title
ICAIF '25: Proceedings of the 6th ACM International Conference on AI in Finance
Copyright Statement
© 2025 Copyright held by the owner/author(s). This work is licensed under a Creative Commons Attribution International 4.0 License (https://creativecommons.org/licenses/by/4.0/)
License URL
Identifier
10.1145/3768292.3770369
Source
ACM International Conference on AI in Finance
Subjects
CCS Concepts • Computing methodologies → Instance-based learning
• Applied computing → Multi-criterion optimization and decisionmaking
• Information systems → Expert systems
Data analytics Agent-Based Modeling, Simulation, Systemic Risk, High-Performance Computing, Synthetic Data Generation, Graph Partitioning
Publication Status
Published
Start Date
2025-11-15
Finish Date
2025-11-18
Coverage Spatial
Singapore