---
title: Probabilistic Limit Order Book Forecasting with Conditional Diffusion Models
url: https://www.ml-quant.com/papers/arxiv/2610.25495/
site: ML-Quant (https://www.ml-quant.com)
updated: 2026-10-23
license: Summaries CC BY 4.0; links go to the original sources
index: https://www.ml-quant.com/llms.txt
identifier: arXiv:2610.25495
source_url: https://arxiv.org/abs/2610.25495
featured: 2026-10-23
citations: unknown
topic: Trading, Microstructure & Execution
---


# Probabilistic Limit Order Book Forecasting with Conditional Diffusion Models

LOBGrad combines autoregressive encoders with diffusion models for limit-order-book forecasting, reproducing microstructure facts and providing calibrated predictive intervals.

- Source: https://arxiv.org/abs/2610.25495
- Identifier: arXiv:2610.25495
- Released: 2026-10-20
- First featured: Quant Letter No. 136 (2026-10-23): https://www.ml-quant.com/issues/2026-10-23/
- Citations (Semantic Scholar): not tracked
- Published in: not yet
- Topic: Trading, Microstructure & Execution
- Authors: Junyu Chen, Franklin Liu, Moshu Xu, Lijian Yang

## Abstract (arXiv, CC0)

Modeling limit order book (LOB) dynamics is central to market forecasting and execution-strategy evaluation. While diffusion models achieve strong empirical performance, existing LOB snapshot models typically lack a learned recurrent encoder and formal statistical guarantees. We introduce LOBGrad, a conditional DDPM framework for probabilistic LOB forecasting that combines an autoregressive RNN encoder with a diffusion-based generator. In particular, we establish an excess-risk oracle inequality for the one-step DDPM noise-prediction objective under exponential mixing and regularity assumptions. We further show that, under contractive dynamics, the dependence on the conditioning-window length can be reduced to a dependence on the latent-state dimension. Experiments on BTC/USDT, GOOGL, and META show that LOBGrad improves on its encoder-free diffusion ablation across all reported queue metrics and obtains competitive results against other baselines. It also reproduces the key stylized facts of LOB microstructure, and yields reasonably calibrated queue-based predictive intervals. A stylized downstream task illustrates the potential value of the generated forecasts.

## Related

- [Painting the market: generative diffusion models for financial limit order book simulation and forecasting](https://www.ml-quant.com/papers/arxiv/2509.05107/): A new method for simulating financial market data transforms limit order book data into an image format and uses diffusion models to predict future states, showing top performance on LOB-Bench.
- [LOB-Bench: Benchmarking Generative AI for Finance - an Application to Limit Order Book Data](https://www.ml-quant.com/papers/arxiv/2502.09172/): LOB-Bench, a Python benchmark tool, has been launched to assess the quality of generative message-by-order data for limit order books, with the GenAI approach showing superior performance.
- [Algorithmic and high-frequency trading problems for Semi-Markov and Hawkes jump-diffusion models](https://www.ml-quant.com/papers/arxiv/2409.12776/): The paper presents jump-diffusion models to understand limit order book data in algorithmic and high-frequency trading, offering optimal solutions for trading issues within a stochastic optimal control framework.
- [Conditional Generators for Limit Order Book Environments: Explainability, Challenges, and Robustness](https://www.ml-quant.com/papers/arxiv/2306.12806/): Conditional generative models used for order book simulation, enhanced with adversarial attacks.
- [Adversarial Post-Training of Foundation Models for High-Frequency Trading](https://www.ml-quant.com/papers/arxiv/2610.25604/): Wasserstein GAN post-training on limit order book foundation models cuts error accumulation and improves realism by 17%, outperforming continued pre-training with far lower compute cost.
- [Trading Problems with Semi-Markov and Hawkes Models](https://www.ml-quant.com/papers/ssrn/4956752/): The article explores the creation of advanced trading algorithms that replicate Limit Order Book data, with a focus on semi-Markov and Hawkes jump-diffusion models for high-frequency trading.