Ali Habibnia's team releases the StochLOB limit order book trend prediction framework paper
Get up to speed
Ali Habibnia and other paper author teams released the StochLOB heavy-tailed score matching limit order book trend prediction framework research paper. This framework is a training-only score matching framework that replaces diffusion forward perturbation from Gaussian Brownian noise with jump Lévy noise of compound Poisson variance mixture and truncated α-stable noise of positive stable Gaussian scale mixture, while still using a deterministic predictor during the inference phase. On the time-split Coinbase limit order book held-out dataset, relevant variants of this framework remain competitive in undamaged Bitcoin and altcoin test cycles, and improve multiple realized tail window stress diagnostic metrics. The team also conducted an exploratory Feishu A-share migration study to test whether the framework's probabilistic outputs remain usable under transaction cost and trading rule constraints. Currently, the framework's code and configuration files are publicly available.
AI-generated from reports · 2 hr agoUpdated
Coverage timeline
Follow the coverage to understand different sides of the event.
- arXiv q-finStochLOB: Robust Limit Order Book Trend Prediction via Heavy-Tailed Score Matching
An arXiv paper proposes the STOCHLOB training-specific score matching framework, which replaces Gaussian Brownian noise in diffusion forward perturbation with jump Lévy noise from compound Poisson variance mixtures and truncated α-stable noise from positive stable Gaussian scale mixtures, while still using a deterministic predictor during inference.
Trend of this event
Not enough consecutive observations yet; trend is not drawn.