DiffVolume: Diffusion Models for Volume Generation in Limit Order Books

2025-08-12

AI Summary

This paper addresses the challenge of jointly modeling strong temporal dependencies and dynamic liquidity couplings in high-dimensional limit-order-book (LOB) volume snapshot generation. We propose DiffVolume, a conditional diffusion model that leverages historical volume sequences and explicit liquidity features—such as bid-ask spread and order-book depth—as conditioning inputs to jointly capture temporal evolution and cross-level liquidity dependencies, enabling controllable counterfactual generation. Compared to existing approaches, DiffVolume achieves significant improvements in statistical distribution fidelity, counterfactual plausibility, and downstream liquidity prediction performance. Empirical evaluation across multiple market datasets shows an average reduction of 12.7%–23.4% in prediction error. By unifying temporal dynamics and multi-level liquidity interactions within a generative framework, DiffVolume establishes a novel paradigm for market microstructure modeling and synthetic LOB data generation.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Modeling limit order books (LOBs) dynamics is a fundamental problem in market microstructure research. In particular, generating high-dimensional volume snapshots with strong temporal and liquidity-dependent patterns remains a challenging task, despite recent work exploring the application of Generative Adversarial Networks to LOBs. In this work, we propose a conditional extbf{Diff}usion model for the generation of future LOB extbf{Volume} snapshots ( extbf{DiffVolume}). We evaluate our model across three axes: (1) extit{Realism}, where we show that DiffVolume, conditioned on past volume history and time of day, better reproduces statistical properties such as marginal distribution, spatial correlation, and autocorrelation decay; (2) extit{Counterfactual generation}, allowing for controllable generation under hypothetical liquidity scenarios by additionally conditioning on a target future liquidity profile; and (3) extit{Downstream prediction}, where we show that the synthetic counterfactual data from our model improves the performance of future liquidity forecasting models. Together, these results suggest that DiffVolume provides a powerful and flexible framework for realistic and controllable LOB volume generation.

Problem

• Modeling high-dimensional LOB volume snapshots with temporal patterns
• Generating realistic LOB data with statistical properties
• Enabling controllable counterfactual generation for liquidity scenarios

Innovation

• Conditional diffusion model for LOB volume generation
• Controllable generation under liquidity scenarios
• Improves liquidity forecasting with synthetic data