Conditional diffusion models improve limit order book forecasts
Probabilistic Limit Order Book Forecasting with Conditional Diffusion Models
Machine Learning
Summary
Predicting how financial markets will behave helps traders make better decisions. The authors developed a new method called LOBGrad that uses a combination of techniques—one that learns from sequences and another that adds noise and then removes it—to model market order data more accurately. Their method produces more reliable predictions and captures important patterns seen in real markets. This approach could help investors understand market dynamics and manage risks better.
What this means in practice
- •For quantitative trading teams: Generate probabilistic forecasts of order book movements to improve trading execution and strategy evaluation.
- •For financial risk managers: Obtain calibrated predictive intervals of market order flow for enhanced risk assessment in trading portfolios.
Authors
Junyu Chen, Franklin Liu, Moshu Xu, Lijian Yang
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.