Painting the market: generative diffusion models for financial limit order book simulation and forecasting

2025-09-05

AI Summary

To address the dual challenges of performance degradation of deep generative models under high noise and error accumulation in autoregressive long-horizon forecasting for limit order book (LOB) modeling, this paper pioneers a novel paradigm: representing LOB time series as spatiotemporal images and proposing a diffusion-based generative framework leveraging image inpainting. By encoding explicit spatiotemporal inductive biases, the method enforces global structural consistency and enables parallel, non-autoregressive generation of long sequences. Evaluated on the LOB-Bench benchmark, it achieves state-of-the-art performance—particularly excelling in structural coherence metrics—and maintains robust superiority even when trained solely on low-resolution Level-2/3 data. The core innovation lies in the principled integration of LOB image representation with diffusion-based generative modeling, bridging computer vision and financial time-series synthesis.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Diffusion Models for Vision
• 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: Graph embeddings and representation learning for Web-related graphs
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Simulating limit order books (LOBs) has important applications across forecasting and backtesting for financial market data. However, deep generative models struggle in this context due to the high noise and complexity of the data. Previous work uses autoregressive models, although these experience error accumulation over longer-time sequences. We introduce a novel approach, converting LOB data into a structured image format, and applying diffusion models with inpainting to generate future LOB states. This method leverages spatio-temporal inductive biases in the order book and enables parallel generation of long sequences overcoming issues with error accumulation. We also publicly contribute to LOB-Bench, the industry benchmark for LOB generative models, to allow fair comparison between models using Level-2 and Level-3 order book data (with or without message level data respectively). We show that our model achieves state-of-the-art performance on LOB-Bench, despite using lower fidelity data as input. We also show that our method prioritises coherent global structures over local, high-fidelity details, providing significant improvements over existing methods on certain metrics. Overall, our method lays a strong foundation for future research into generative diffusion approaches to LOB modelling.

Problem

• Simulating limit order books for financial forecasting and backtesting
• Overcoming error accumulation in autoregressive models for long sequences
• Generating future LOB states using diffusion models with image inpainting

Innovation

• Convert LOB data into structured image format
• Apply diffusion models with inpainting technique
• Enable parallel generation of long sequences