Probabilistic Limit Order Book Forecasting with Conditional Diffusion Models

By Junyu Chen, Franklin Liu, Moshu Xu, Lijian Yang

Rating

1801
Battle Count: 51

Relevance

9/10
Highly relevant for high-frequency trading and execution algorithms. The paper addresses probabilistic forecasting of LOB dynamics, which is critical for optimal execution strategies and market making. The inclusion of a downstream execution task demonstrates practical utility.

Implementation Complexity

7/10
Requires implementing a conditional DDPM with a recurrent encoder (LSTM) and a noise-prediction MLP. Training involves handling dependent time-series data and ensuring numerical stability with signed-log transformations. Theoretical components are complex but not required for basic implementation.

Reproducibility

5/5
The paper provides a public GitHub repository link (https://github.com/junyuchen-econ/LOBGrad). The datasets used (BTC/USDT from Binance, GOOGL/META from Databento) are standard financial data sources, though Databento requires a subscription. The methodology and hyperparameters are fully described.

Abstract

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.

Links

arXiv:2610.25495 · Code · Paper rankings

The interactive Everscope explorer (charts, battles, favorites) loads below.