OrderFlowDiT: Transport-Regularized Diffusion Transformer for Probabilistic Limit-Order Supply Forecasting
We introduce OrderFlowDiT, a conditional Diffusion Transformer for unnormalized bid and ask volume-at-price fields, trained with an unbalanced optimal-transport (UOT) auxiliary loss.
ProofPaper ↗
Key points
- We study the conditional distribution of limit-order volume submitted over the next five minutes, given the current order book and recent order flow.
- The task requires capturing price placement, total volume, and predictive uncertainty jointly.
- Coordinate-wise diffusion losses do not explicitly account for price displacement, while normalizing fields to unit mass removes information about total volume.
- In the primary comparison on 1950 validation conditions across 30 U.S. equities, adding UOT reduces full-field transport cost by 45.8% and the transport energy score by 26.1% relative to training with mean squared error alone.
Sources (1)
- [1]OrderFlowDiT: Transport-Regularized Diffusion Transformer for Probabilistic Limit-Order Supply ForecastingarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 10:56 PM
We introduce OrderFlowDiT, a conditional Diffusion Transformer for unnormalized bid and ask volume-at-price fields, trained with an unbalanced optimal-transport (UOT) auxiliary loss.
We study the conditional distribution of limit-order volume submitted over the next five minutes, given the current order book and recent order flow.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 15, 2026FlashControl: One-Step Controllable Generator via Distillation-Friendly Single-Stream Teachers
- Oct 15, 2026Designing Transformers for Continuous Diffusion Language Models
- Oct 15, 2026OmniXRef: A Unified Framework for Multimodal Multi-Reference Video Editing
- Oct 15, 2026Velox: Spatial Memory Makes World Models Efficient
- Oct 14, 2026Vertical Sliced Wasserstein Distances for Meta-Measures
- Oct 14, 2026NVIDIA/TensorRT-LLM v1.4.0rc0